Ontwerp van LCOS microbeeldscherm backplanes voor projectietoepassingen Design of LCOS Microdisplay Backplanes for Projection Applications Jean

Van den Steen

Promotor: prof. dr. ir. H. De Smet Proefschrift ingediend tot het behalen van de graad van Doctor in de Ingenieurswetenschappen: Elektrotechniek Vakgroep Elektronica en Informatiesystemen Voorzitter: prof. dr. ir. J. Van Campenhout Faculteit Ingenieurswetenschappen Academiejaar 2005 - 2006

ISBN 90-8578-062-4 NUR 959 Wettelijk depot: D/2006/10.500/20

motions overwhelm me when thinking about the many efforts, contributions, laughter, hint and help from so many people. Amongst others, the fear not to be grateful enough for the opportunities that were created. A pretty large number of people effectively contributed to the realization of this book. In particular, I thank Koen Van Laere and my brothers Etienne and Philippe Van den Steen for their reading efforts and constructive comments. This thesis is one written part about the results of teamwork. Not everyone did contribute directly to the technical aspects neither am I just a cold research machine. Some contributed with a smile, a listening ear, a tasty meal, challenging cycle tours, regular exercising, direct pep-talk or with a rough hike through snowy landscapes. To all of you, I hereby whish to express my deepest feelings of gratitude. The research was supported by the University of Gent and IMEC vzw. University projects, extensive european projects and direct bilateral contracts made it possible to conduct research on microdisplays. I am very grateful to live in a society that makes research possible. The cooperation with our project partners has been most enjoyable: Alcatel Microelectronics (now AMIS, Oudenaarde, Belgium), Barco (Kortrijk, Belgium), Merck KgaA (Darmstadt, Germany), Thomson Multimedia (Rennes, France), Thales Avionics LCD (Grenoble, France), Taiwan Micro Display Corporation (tmdc, Hsin-Chu, Taiwan), University College of London (London, United Kingdom), University of Stuttgart (Stuttgart, Germany). Maybe the most enjoyable was the exchange of research results with the world-wide community of display researchers. Four excellent computer engineers patiently provided me with efficient help at times when the computers got tired of me: Ronny Blomme, Marnik Brunfaut, Wim Meeus and Peter Sebrechts thank you so much! Amongst the research groups colleagues, special thanks go to Geert Van Doorselaer, Miguel Vermandel, Johan De Baets, Nadine Carchon, Dieter Cuypers, Igor Popov, Miroslav Vrana, Jan Vanfleteren, Joeri De Vos, Jan Doutreloigne, Peter Sebrechts, Katrien Vanneste, Andr Van Calster and Herbert De Smet. I want to thank Geert in particular, because of his patient elaboration of the system electronics and Dieter for his persistent quest for improved assembly methods. They also provided me with insights in what would otherwise be the dark side of microdisplays. Geert and Dieter, your efforts have been most instrumental! Especially, I want to thank Professor Andr Van Calster and Professor Herbert De Smet for their guidance and clear-cut analysis, also in the more difficult times. I also thank them for the successful creation of a consistent string of projects, as well as for the many efforts in the sometimes exhausting creation and follow-up of projects. Gentlemen, I cannot say anything else but chapeau!

Acknowledgements

History of my research

n November 1994 the thin-film-components-group (TFCG) started a small feasibility study on crystalline silicon based display systems. In those pioneering times of liquid-crystal-on-silicon (lcos), the design and manufacturing were indeed questionable. This small-scale experiment included basic possibilities for measurements and provided data about attainable pixel pitches and about the issue of light sensitivity of silicon. Time window of my research : November 1994 December 2003
End 09/1995 1998 03/2000 12/2003 2005 08/2004 List of projects Internal feasability study Internal REFLEC European MOSAREL, EP25340 Bilateral TMDC European LCOS4LCOS, IST_2001_34591 Incubation GEMIDIS Writing my PhD

Start 11/1994 01/1995 09/1997 04/2000 04/2002 01/2004 2003

09/2005

Table of projects over the years

95 96 97 98 99 00 01 02 03 04 05

94 Project Feasibility REFLEC

x-Si IWT Sigma

MOSAREL TMDC LCOS4LCOS Incubation Writing my PhD

Timeline of primary activities Soon after (beginning of 1995), a full microdisplay project (REFLEC) was born. The standard low-cost foundry process chosen implied the light shielding and top mirror layers had to be processed within laboratorys clean-room. Assembly into a real display cell and the design and implementation of peripheral electronics also required a lot of precision and effort from a team of colleagues. Two years of chip design, assembly with peripherals and problem-solving allowed to finally reach the first milestone. It consists of a working microdisplay (august 1997) controlled through a personal computer.

In the mean time, the author worked on a project on SOI (silicon-on-insulator) based displays and on the Copernicus Sigma project. The purpose of the Sigma project is the development of intelligent gas-sensors. Next, the European Esprit-IV project MOSAREL (EP-25340) started: 270 times as many pixels as with the REFLEC project. Continued work showed off in march 2000: the consortium was able to see a projected 2560x2048 image. From an organizational and technical perspective, I learned most about silicon design during this project. Basically it allowed me to finetune the design procedures up to a high level of perfection. These successes led a Taiwanese business man to come along and to order a full microdisplay menu. Starting with XGA, development of SXGA and full HDTV formats were the future. This is the cooperation with the Taiwan Micro Display Corporation (tmdc), headed by Mr. L.Y. Tseng. In 2002, TFCG joined the LCOS4LCOS project led by Thomson Multimedia, France and financed by the EC. The last chapter introduces this work. For the LCOS4LCOS project, I did not perform the silicon implementations. My task here was to develop and simulate smart pixel architectures to allow for an implementation of a single-panel system. As such, IMECs participation to the project concerned the high-level design of the microdisplay chip. The last years also, a plan has been worked out to throw the GEMIDIS spin-off company into the real world (august 2004) of high-tech companies. My active research work effectively ended at the beginning of 2004. The first months of 2004, I actively participated in the creation of the company; the rest of the year I monitored Gemidiss first chip design and the LCOS4LCOS project (on a distance) and finally started writing my thesis. Since december 2005 I am with the DSL Experts Team of Alcatel Bell in Antwerp, Belgium.

Executive summary
he research work presented in this book concerns the design of lcos microdisplay backplanes. The aim is to clarify the design, because lcos backplanes are not standard chips. Microdisplays are displays with very small dimensions; this small that optical magnification is required to view the image. The acronym lcos stands for liquid-crystal-on-silicon and is the name of one of the competing technologies. Lcos microdisplays consist of a layer of liquid crystal, encapsulated between a transparent glass plate and a crystalline silicon chip. The term backplane refers to this chip. The prime applications of these displays are large-diagonal projection systems. They are however not limited to that. The author believes the research has been very successful considering the publications, the world-wide patent awarded and of course, the team-work that yielded several successful implementations. Mid-august 2004, a spin-off company was born GEMIDIS, which stands for Gent microdisplays. Its core business is the commercialization of lcos panels. In the beginning, lcos technology seemed excessively attractive because of its apparent simplicity. The overall system in which lcos microdisplays are applied, is rather complex however and requires the efforts of a team of specialists. It is almost an impossible task to compare lcos with all other innovating display technologies. Maybe because the whole team believes the developed solutions just are the best. Seriously, the author thought it was thrilling enough to write down a concise story about his own contributions to lcos. My responsibility covered the entire silicon design of the backplane Ics. With entire silicon design I mean the simulations, the layouting, the functional testing of the chips, the datasheets and specifications that was my job. Incidentally, I got help from Herbert for measurement setups and had many discussions with him regarding the smart pixel ideas. The desing of the backplanes is the area where I contributed most to. In the technology plane, I developed an original design procedure for stitching. The implementations during the MOSAREL project have shown how well this approach works. In the circuit plane I implemented the classic AM architecture for different display formats. The successes with these design ultimately led to the creation of the GEMIDIS startup company. I also developed smart pixel circuits for multi-million pixel arrays these ideas form an important part of the patent we were granted. Some of the ideas have been implemented for single-panel HDTV backplanes and have also been tested during the LCOS4LCOS project. The results hereof will be presented at SID06. Roughly, the book splits into four parts. Chapter one is the first part and places the work presented in this book, in a broader context. The first paragraph of this chapter recalls three essential facts about human vision; three facts that are fundamental to understand the operation of most display systems. The second paragraph briefly mentions economical aspects and gives an indicative value of the

"real world" market. After all, what's the practical interest in research on displays? The third paragraph is entitled application specific displays. It is indeed justified to some degree to state that each display type fits a particular application. A classification attempt is made to help distinguishing between the several display technologies, applications, etc, etc. Subsequently, the text focuses on examples of liquid-crystal-on-silicon (lcos) projection architectures. Finally, a glance at the anatomy of an lcos micro-display backplane wraps up this chapter and by then provides more insight in the meaning of the thesiss title Part two chapters two and three is about lcos technology, it deals with rules and procedures to ensure first-time-right designs. Lcos technology is restricted here to what is needed from the point of view of the backplane-chip designer. Chapter two combines the requirements of LC assembly technology with standard IC technology into a set of relatively straightforward design rules. This chapter is a collection of crucial background information concerning the constraints that must be dealt with for the design of a micro-display backplane. A first paragraph called a designers perspective, summarizes the lcos product chain. The second paragraph called lcos cell assembly technology highlights, discusses the constraints from cell assembly technology. The third paragraph IC processing technology highlights discusses constraints directly related to the Si processing itself. Chapter three proposes a solution for the design of very large ICs with dimensions so large that standard design techniques fall short. It is entirely devoted to stitching. The discussion starts with paragraph 3.1 three considerations about stitching, a comparison between three different ways of partitioning the mask sets. Ease of processing and cost of prototyping make that multiple-module reticule sets are the preferred option. The two other though are technically feasible. The second paragraph implications for ic design describes the implications of the multiplemodule reticule set approach on the way to construct the backplane IC design. Particularly with the mask layout verification there are some difficulties. There are not only consequences at the layout stage of the design cycle, but also at the schematic stage. Ultimately, the difficulties encountered are addressed by the virtual stitching procedure. It consists of a way to organize the mask layout and of a few simple routines. The reader gets insights in a tool that helps to ensure first time right designs. The third part about circuits is covered by chapters four and five. Chapter four introduces the classic active matrix and reviews the implementations. The heart of lcos devices is known as the pixel matrix; it designates the area where the electro-optic signal conversion occurs. Paragraph 4.1 describes the classic analog active matrix (AM) architecture. The description starts with the single transistor pixel circuit. A SPICE model is proposed to help with the simulation of the LC capacitor. The addressing of the AM with integrated drivers is looked at on the grounds of speed and redundancy considerations. Paragraph 4.2 summarizes the

implementations, along with the results obtained. It is not the aim to give detailed reports of the optical performance, rather is this chapter limited to the chips functionality and peculiarities. As such, it concludes the discussion on micro-display designs that use the classic AM architecture. It also forms the starting ground for chapter five, where improvements to the pixel circuit are presented. These improvements effectively add a second frame-memory to the AM. These circuit ideas are covered by our patent, essential for both cheaper displays and so-called single panel systems. Paragraph 5.1 covers the CE switching principle, the compatibility with continuous light sources and estimates the maximum transistor count per pixel. Paragraph 5.2 describes a couple of original circuit ideas that effectively implement the functionality of an additional frame memory. The resulting pixels are labeled smart pixels maybe because of the functionality offered notwithstanding the limited area available. The fourth and last part of the book is composed of conclusions and appendices actually they contain information that is not only useful for the previous chapters. They contain concise datasheets of the designs that were made as well as a beginner's hands-on manual of verification software. This way, I hope this book can also be useful for anyone desiring to learn about IC design, in particular about the design of lcos microdisplays.

Nederlandse samenvatting
Het werk dat in dit boek neergeschreven staat, behelst het onderzoek van het ontwerp van zogenaamde lcos micro-beeldschermen. De doelstelling van het boek is het toelichten van het ontwerp van lcos microbeeldscherm backplanes. Microbeeldschermen zijn beeldschermen van dermate kleine afmetingen dat optische vergroting noodzakelijk is om het beeld zichtbaar te maken. De afkorting lcos staat voor vloeibaar kristal op silicium en geldt tevens als naam voor een van de meest competitieve beeldscherm-technologien. Lcos microbeeldschermen bestaan uit een dun laagje vloeibaar kristal dat ingekapseld is tussen een doorzichtig glasplaatje en een kristallijne silicium chip. De term backplane verwijst precies naar deze chip. De voornaamste toepassingen van deze soort beeldschermen zijn projectiesystemen met grote beelddiagonaal. De auteur gelooft dat het onderzoek meer dan geslaagd mag genoemd worden, gezien de vele resultaten, publicaties, het wereldwijde patent en natuurlijk ook gezien het feit dat verscheidene succesvolle implementaties tot stand gekomen zijn. Op 15 augustus 2004 zag zelfs een nieuw spin-off bedrijf het licht: GEMIDIS (staat voor Gent microdisplays). De hoofdaktiviteit van deze spin-off is de commercializatie van lcos micro-beeldschermen. Bij het begin leek lcos technologie bijzonder aantrekkelijk, omwille van diens schijnbare eenvoud. Het systeem waarin dergelijke micro-beeldschermen toegepast worden, is vrij ingewikkeld en vereist inspanningen van een ploeg specialisten. Het is niet eenvoudig om lcos met alle andere vernieuwende technologien te vergelijken. Misschien komt dit doordat ons team gelooft dat haar eigen oplossingen simpelweg de beste zijn... In elk geval was het voor de auteur spannend genoeg een beknopt verhaal neer te schrijven over de eigen bijdragen tot het hele lcos-gebeuren. Grofweg kan het boek in vier delen opgesplitst worden. Het eerste hoofdstuk plaatst het werk in een bredere context. De eerste paragraaf beschrijft kort drie essentile feiten over het menselijk gezichtsvermogen; deze liggen aan de basis van de opbouw van de meeste beeldsystemen. De tweede paragraaf haalt kort een paar economische gegevens aan, uit de rele wereld. Wat zou immers het nut kunnen zijn van intensieve onderzoeksprojecten omtrent nieuwe beeldsystemen. De derde paragraaf heeft als titel beeldsystemen op maat van de toepassing. Het is inderdaad min of meer gerechtvaardigd te stellen dat elk type beeldscherm past bij een welbepaald type toepassing. Er wordt een poging ondernomen om een klassificatie tot stand te brengen. Deze heeft tot doel de verschillende technologien, toepassingen, enz. te helpen onderscheiden. Hierop volgen enkele voorbeelden van lcos projectie architecturen. Ten slotte richt een doorsnede het licht op de interne structuur van lcos microbeeldscherm backplanes. Hopelijk is hiermee dan ook de titel van het proefschrift duidelijk. Het tweede gedeelte omvat de hoofdstukken twee en drie; deze gaan over lcos technologie. M.a.w. ze beschrijven ontwerpsregels en -procedures die een efficint

chip-ontwerp helpen bekomen. Met lcos technologie beperkt de auteur zich uiteraard tot hetgeen nodig is voor de microbeeldscherm chip-ontwerpers. Hoofdstuk twee brengt de vereisten samen die voortvloeien uit de vloeibaar kristal (LC) assemblage technologie en de integrated circuit (IC) ofte chip-technologie. Dit resulteert in een vrij eenvoudige lijst ontwerpsregels. Heel beknopt herleidt het tweede hoofdstuk zich dus tot een verzameling informatie die fundamenteel is voor het ontwerp van microbeeldscherm backplanes. Een eerste paragraaf (2.1) vat het productieschema samen, gezien vanuit het perspectief van de chip-ontwerper. In een zelfde perspectief beschrijft de tweede paragraaf een aantal relevante aspecten van de LC assemblage technologie. Tenslotte beschrijft de derde paragraaf de relevante aspecten van de IC technologie. Een beperking treedt op bij het maken van naar verhouding zeer grote chips; deze kan omzeild worden met een techniek die in het engels stitching genoemd wordt (to stitch betekent letterlijk: naaien). Hoofdstuk drie is geheel gewijd aan dit stitchen dat dus enkel nodig is voor het ontwerpen en maken van zeer grote chips. Paragraaf 3.1 maakt een vergelijking tussen drie verschillende benaderingen m.b.t. het stitchen van microbeeldscherm backplanes. Een van de benaderingen multiple module reticule sets blijkt de meest interessante, om redenen van kost en productie-eenvoud. De tweede paragraaf geeft een gedetailleerde beschrijving van de gevolgen van deze benadering voor de ontwerpsprocedure. I.h.b. zijn er een aantal delicate punten met betrekking tot de verificatie van het maskerontwerp. Tenslotte beschrijft de virtuele stitching procedure, een systematische aanpak die de auteur opgezet heeft om op trefzekere wijze de maskersets te ontwerpen. Het derde gedeelte omvat de hoofdstukken vier en vijf. Op een zeer analoge manier, gaat hoofdstuk vier over de klassieke circuitoplossing en gaat hoofdstuk vijf over uitbreidingen hierop. Hoofdstuk vier bespreekt de klassieke aktieve matrix en geeft een overzicht van de geimplementeerde chips. De verzameling beeldpunten of pixel matrix vormt het kloppend hart van een microbeeldscherm; de pixel matrix is het laatste stukje schakeling dat de electro-optische signaal conversie stuurt. Paragraaf 4.1 handelt uitsluitend over de klassieke, analoge aktieve matrix architectuur. Deze architectuur houdt in de praktijk in dat drie microbeeldschermen nodig zijn om alle drie de kleurenbundels te sturen. De analyse van de pixel schakeling wordt verder uitgediept dankzij een origineel computer model voor de LC capaciteit. De gentegreerde sturing van de aktieve matrix wordt bekeken op grond van snelheidsbeschouwingen en van redundantie. Paragraaf 4.2 geeft dan een overzicht van de realizaties, tezamen met de bekomen resultaten. Het is hier niet de bedoeling een gedetailleerde beschrijving te geven van de optische performanties; het gaat hier eerder om een beschrijving van de bekomen functionaliteit en eventuele bijzonderheden. Als dusdanig legt hoofdstuk vier ook de basis voor hoofdstuk vijf waar zgn. smart pixels als verbeteringen voorgesteld worden. De essentie van deze verbeteringen is dat een tweede beeldgeheugen toegevoegd wordt aan de aktieve matrix. De schakelingen die uitgevonden werden, zijn beschermd door een patent op naam van de auteur. Ondertussen heeft de spin-off GEMIDIS de eigenaarsrechten hierop overgenomen. De uitvinding is cruciaal voor het bekomen van goedkopere chips en voor het implementeren van optische

architecturen die slechts n enkel microbeeldscherm nodig hebben. De eerste paragraaf beschrijft het principe achter de schakelende tegenelektrode en diens compatibiliteit met continu-lichtbronnen en geeft tenslotte een schatting van het maximum aantal transistoren per pixel. Deze schatting laat toe in te zien dat het aantal mogelijke pixel-schakelingen eerder beperkt is. Paragraaf 5.2 beschrijft een aantal originele pixel-schakelingen die de idee van een tweede beeldgeheugen implementeren. De resulterende pixels worden smart pixels genoemd, allicht omwille van de verhouding functionaliteit tot beschikbare oppervlakte. Het vierde en laaste gedeelte bevat hetgeen overblijft: het besluit en appendices. Deze laatste bevatten meer informatie dan nodig voor de andere hoofdstukken: ze bevatten beknopte datasheets van de gerealizeerde ontwerpen alsook een handleiding voor het gebruik van de verificatie-software. Hiermee hoopt de auteur dat dit proefschrift als nuttig naslagwerk kan dienen voor eenieder die iets wil leren over chip-ontwerp, i.h.b. over het ontwerp van microbeeldscherm chips.

Table of contents
Cover page Dedication Acknowledgements History of my research Executive summary Nederlandse samenvatting Table of contents List of abbreviations Publication list Chapter 1 : Welcome to Display World 1.1 How we see colored, moving images 1.2 Market significance 1.3 Application specific displays
1.3.1 Classification attempts 1.3.2 Lcos systems

1 2 3 4 6 9 12 15 21 25 25 29 31

31 32 38

1.3.3 Lcos micro-displays: definition and physical structure

Chapter 2 : Constraints from lcos technology 2.1 An IC designer's perspective

2.1.1 A system flow and global constraints 2.1.2 Assembly flow of lcos panels 2.1.3 An (lcos) IC design cycle 2.2.1 The lcos cell structure

41 41

41 42 42 45 53 60

2.2 Lcos cell assembly technology highlights 2.3 IC processing technology highlights
2.2.2 The lcos cell assembly

45 59

2.3.1 Synopsis of 4 lcos-IC processing flows

2.3.2 Photolithography I: some basics

62

Chapter 3 : Photolithography (II) - stitching

3.1 Three considerations about stitching 3.1.1 Multiple-reticule reticule sets: cost versus complexity 3.1.2 Multiple-module reticule sets: MMI tables, multiple display formats 3.1.3 (Stitching)2 or the 1X-5X mix&match 3.2.1 Anatomy of a module 3.2.2 Double exposures

71 71

72 77 81 85 87 90 90 92 94 95

3.2 Implications for IC design

85

3.2.3 Split the schematics as well?

3.2.4 Overcoming problems with mask generation 3.3 Virtual stitching procedure

3.2.5 Overcoming problems with mask verification 3.3.1 Module layout organization for virtual stitching 3.3.2 Computing stitched patterns

94

Chapter 4 : The pixel matrix(I) the basics 4.1 The pixel matrix (I): the classic AM
4.1.1 One, single picture element (pixel) 4.1.2 Millions of pixels 4.1.3 AM driver circuits

99 100

100 109 111 113 115 122 128

4.2 Micro-display implementations

4.2.2 5Mpix and stitching

4.1.4 VAN pixel design: constraint listing 4.2.1 REFLEC chips: 160x120 pixels 4.2.3 Tmdc prototypes: XGA, WXGA

115

Chapter 5 : The pixel matrix(II) smart pixels 5.1 Counter electrode switching

5.1.1 The need for a second in-pixel frame memory

133 134

134

5.2 AMs with an additional frame memory

5.2.1 Two-transistor pixels 5.2.2 The bucket brigade architecture 5.2.3 The double dram architecture 5.2.4 PWM for fast gray scales

5.1.2 Area estimation for smart pixels

139

144

144 147 149 150

Conclusions Appendices 6.1 Datasheets of three microdisplay chips

157 159 159

159 165 175 200 235 237 255

6.1.1 Feasibility: a 32x18 matrix with integrated column scanners 6.1.2 HBiMOS 1/16th VGA (160x120 display) 6.1.3 Mosarel GXGA, SXGA and XGAp 6.1.4 TMDC XGA

6.2 Hands-on dracula

6.3 The skill interpreter

6.2.1 Introduction: general remarks 6.2.2 The input deck: 'description', 'input layer' and 'operation blocks 6.2.3 Evolution of a dracula job: a UNIX command sequence 6.3.1 A brief introduction 6.3.2 Function and/or structure and/or source code of some useful routines

216

258

258 260

References Further reading

265 270

List of abbreviations
D 1x, 5x AC ADC Al AM AMI AMIS AMLC AMLCD AR a-Si ASIC AX B BJT C CDL CE CH CIE CMOS CMP CMY Cr CRT D DAC DARC DC DF micro-display paragraph micro- or micron 1x, 5x magnification alternating current analog-to-digital converter Aluminum active matrix American Microsystems Incorporated AMI Semiconductors active matrix liquid crystal active matrix liquid crystal display aperture ratio or assembly rule (depending on context) amorphous silicon application-specific integrated circuit appendix bulk terminal of transistor bipolar junction transistor preferred symbol for capacitance component description language counter-electrode chapter Commission Internationale de l'Eclairage complementary MOS chemical-mechanical polishing cyan-magenta-yellow chromium cathode ray tube drain terminal of transistor digital-to-analog converter Dielectric anti reflective coating; trademark of Brewer Science direct current dark field Directorate General 13:Information Society and Media

DGXIII

DLP DMD DRAM DRC EC ELD EO ERC ESSDERC FED FET FLC FPD FPTV FTW G GDS GEMIDIS GLV GOA GXGA HBiMOS HDTV HMD HR HTPS Hz I IBM IC IDRC IDW IEEE IMAPS IMEC IPS IR ISL

digital light processor digital mirror device dynamic random access memory design rule check European Commision electro-luminiscent display electro-optical electrical rule check European Solid-State Device Research Conference field emission display field effect transistor ferro-electric LC flat panel display front projection TV Faculteit Toegepaste Wetenschappen gate terminal of transistor Generalized Data Stream; here, Calma GDSII stream format database Ghent Microdisplays; the TFCG microdisplay spin-off grating light valve Gezamelijke OnderzoeksAktie G XGA; some call it QSXGA; 2560x2048 pixel array High voltage bipolar MOS technology from AMIS high definition TV head-mounted device holding ratio high temperature polysilicon Hertz; unit of frequency prefered symbol for current International Business Machines Corporation integrated circuit; 'chip' International Display Research Conference International Display Workshop Institute of Electrical and Electronics Engineers International Microelectronics And Packaging Society Interuniversitair Microelektronica Centrum in-plane switching infra red independent stitch layouts

IST

IT ITO JVC k L4L LC LCD LCOS LCOS4LCOS LED LEOS LF LPE LTPS LUT LVS maskgen MCM MEBES MEMS MMI MOS MOSAREL Mpix MPW MSA MV NC NDA nMOS NTE OLED pA PBS PC PDA

Information Society Technologies (European Framework Programmes) information technology indium-tin-oxide Japan Victor Company thousands of euros short for LCOS4LCOS liquid crystal liquid crystal display liquid crystal on silicon name of the project on single panel microdisplays light emitting diode Laser and Electro-Optics Society light field layout parameter extraction low temperature polysilicon look-up table layout versus schematic mask generation multi-chip module Manufacturing Electron Beam Engraving System micro electro-mechanical system mask-module-information metal-oxide semiconductor Monocrystalline Silicon Active Matrix Reflective Light Valve million pixels multi-project wafers module stitching area medium voltage not connected Non-disclosure agreement n-type MOSFET near-to-the-eye organic light emitting diode Pico ampere polarizing beam splitter personal computer personal digital assistant

PDLC PDP pF pMOS PNLC PWM QSXGA QVGA QXGA R RCA REFLEC RGB RMS RPTV RUG S SDEMOS SEM Si SID SID-ME Skill SMD SMIC

SOA SOI specs SPICE SPIE STN SXGA SXGA+ Sw TFCG

polymer dispersed LC plasma discharge panel Pico farad p-type MOSFET polymer network LC pulse-width modulation quad SXGA; same as GXGA; 2560x2048 pixel array quarter VGA; 320x240 pixel array quad-XGA; 2048x1536 pixel array preferred symbol for resistance Radio Corporation of America name of the first project on reflective displays red-green-blue root mean square rear projection TV Universiteit Gent source terminal of transistor symmetric drain-extended MOS scanning electron microscope Silicon Society of Information Displays SID Mid-Europe chapter interpreter language proper to the Cadence electronic design software suite surface-mounted-device Semiconductor Manufacturing International Corporation stitching overlap area silicon-on-insulator Specifications Simulation Program with Integrated Circuit Emphasis The International Society for Optical Engineering super TN Super-XGA; 1280x1024 pixel array SXGA 'plus'; 1400x1050 pixel array Switch Thin Film Components Group; the research group the author was with at the Ghent University's Faculty of Engineering, affilated with IMEC vzw., now TFCG Microsystems

TFT TiN TI TMA tmdc TN

TSMC

TV UMC UNIX UV UXGA V VAN VFD Vpp VGA Vt WSXGA WUXGA WXGA WYSIWYG XGA x-Si t V

thin film transistor Titane nitride Texas Instruments inc. tilted mirror arrays Taiwan MicroDisplay Corporation twisted nematic Taiwan Semiconductor Manufacturing Corporation Limited Television United Microelectronics Corporation operating system for workstations ultra violet ultra XGA; 1600x1200 pixel array volt, preferred symbol for voltage vertically aligned nematic; (type of LC arrangement) vacuum fluorescent display peak-to-peak voltage Video graphics array; 640x480 pixel array threshold voltage wide SXGA; 1280x1050 pixel array wide UXGA; 1920x1200 pixel array wide XGA; 1280x768 pixel array what you see is what you get eXtended Graphics Array; 1024x768 pixel array crystalline silicon delta-t or a time difference delta-V or a voltage difference

List of publications
1. J. De Baets, A. Van Calster, J. Van den Steen, G. Van Doorselaer, D. Wojciechowski, G. Schols, and J. Witters, Design of an x-Si active matrix for high resolution reflective displays, in Proc. of the 15th Int. Display Research Conf., (Hamamatsu, Japan), pp. 477-479, SID, October 1995. J. Van den Steen, G. Van Doorselaer, J. De Baets, and A. Van Calster, Custom design and system integration, in Proc. of the 20th International Spring Seminar on Electronic Technology, Education and Research in Microelectronics (M. Lukaszewicz and M. Kramkowska, eds.), (Szklarska Poreba, Poland), pp. 263-266, June 1997. J. Van den Steen, N. Carchon, G. Van Doorselaer, C. De Backere, J. De Baets, H. De Smet, J. De Vos, J. Lernout, J. Vanfleteren, and A. Van Calster, Technology and circuit aspects of reflective PNLC microdisplays, in Proc. of the 1997 International Display Research Conference, (Toronto, Canada), pp. 195-198, SID, September 1997. M. Rczey, R. Dobay, G. Harsanyi, Z. Illyefalvi-Vitz, J. Van den Steen, A. Vervaet, W. Reinert, J. Urbancik, A. Guljajev, C. Visy, and I. Barsony, ASIC chip, hybrid multisensor, and package co-design for smart gas monitoring module, in Proc. of the Internl Workshop on Chip Package CoDesign, (Zurich, Switzerland), pp. 132-139, IEEE, March 1998. G. Van Doorselaer, N. Carchon, J. Van den Steen, D. Cuypers, J. Vanfleteren, H. De Smet, and A. Van Calster, A reflective polymer dispersed information display made on CMOS, in Proc. of the ICPS, (Antwerpen, Belgium), pp. 222-225, September 1998. G. Van Doorselaer, N. Carchon, J. Van den Steen, J. Vanfleteren, H. De Smet, D. Cuypers, and A. Van Calster, Characterization of a paperwhite reflective PDLC microdisplay for portable IT applications, in Proc. of the 1998 International Display Research Conference, (Seoul, Korea), pp. 5558, SID, September 1998. G. Van Doorselaer, B. Dobbelaere, M. Vrana, X. Xie, N. Carchon, J. Van den Steen, J. Vanfleteren, H. De Smet, D. Cuypers, and A. Van Calster, A paper-white chip-based display MCM package for portable IT products, in Proc. of the IMAPS'98, (San Diego, USA), pp. 543-547, IMAPS, November 1998. G. Van Doorselaer, N. Carchon, J. Van den Steen, J. Vanfleteren, H. De Smet, D. Cuypers, and A. Van Calster, A silicon based reflective polymer dispersed LC display for portable low power applications, in Proc.

2.

3.

4.

5.

6.

7.

8.

of the 11th international symposium on Electronic Imaging, (San Jose, USA), pp. 95-102, SPIE, January 1999. 9. G. Harsanyi, M. Rczey, R. Dobay, I. Lepsnyi, Z. Illyefalvi-Vitz, J. Van den Steen, A. Vervaet, W. Reinert, J. Urbancik, A. Guljajev, C. Visy, G. Inzelt, and I. Barsony, Combining inorganic and organic sensors elements: a new approach for multi- component sensing, Sensor Review, vol. 19, no. 2, pp. 128-134, 1999.

10. M. Rczey, I. Lepsnyi, A. Reichardt, G. Harsanyi, R. Dobay, A. Schon, Z. Illyefalvi-Vitz, J. Van den Steen, A. Vervaet, W. Reinert, J. Urbancik, A. Guljajev, C. Visy, G. Inzelt, and I. Barsony, Combining inorganic and organic gas sensor elements: a new approach for multicomponent sensing, in Proc. of the 12th European Microelectronics and Packaging Conference, (Harrogate, England), pp. 189-195, IMAPS-Europe, June 1999. 11. H. De Smet, J. Van den Steen, N. Carchon, D. Cuypers, C. De Backere, M. Vermandel, A. Van Calster, A. De Caussemaeker, A. Witvrouw, H. Ziad, K. Baert, P. Colson, G. Schols, and M. Tack, The design and fabrication of a 2560x2048 pixel microdisplay chip, in Proc. of the 19th International Display Research Conference (EuroDisplay99), (Berlin, Germany), pp. 493496, SID, September 1999. 12. C. De Backere, M. Vermandel, J. Van den Steen, H. De Smet, and A. Van Calster, A silicon backplane technology for microdisplays, in Proceedings of the 29th European Solid-State Device Research Conference ESSDERC`99, pp. 712-715, 1999. 13. G. Harsanyi, I. Lepsnyi, A. Reichardt, R. Rczey, M an Dobay, A. Schon, Z. Illyefalvi-Vitz, J. Van den Steen, A. Vervaet, W. Reinert, J. Urbancik, L. L, A. Petrikova, A. Guljajev, C. Visy, G. Inzelt, and I. Barsony, New perspectives of selective gas sensing: Combining electroconducting polymers with thick and thin films, in Proc. of the IMAPS'99, (Chicago, USA), pp. 207-212, IMAPS, October 1999. 14. J. Doutreloigne, H. De Smet, J. Van den Steen, and G. Van Doorselaer, Low-power high-voltage CMOS level-shifters for liquid crystal display drivers, in Proc. of the 11th International Conference on Microelectronics, (State of Kuwait), pp. 213-216, November 1999. 15. J. Van den Steen, H. De Smet, G. Van Doorselaer, A. Van Calster, and Colson, Cost effective reticule design for very high resolution Si backplane prototypes, in Proc. of IDW, (Sendai, Japan), pp. 235-238, ITE, December 1999. 16. A. Guljaev, I. Warlashov, O. Muchina, I. Miroshnikova, O. Sarach, A. Titov, J. Van den Steen, A. Vervaet, M. Reczey, R. Dobay, G. Horsanyi, Z. Illjefalvi-Vitez, J. Urbanchik, and W. Reinert, Sigma sensors intellectual

gas monitoring applications, in Proc. of the XXX international seminar: Noise and degradation processes in semiconductor devices, (Moscou), pp. 402-407, November-December 1999. 17. P. Colson, F. De Pestel, M. Tack, G. Schols, H. De Smet, J. Van den Steen, and A. Van Calster, The design and fabrication of a GXGA microdisplay chip, in Proceedings of SPIE on Micromachining, Micromanufacturing and Microelectronic Manufacturing, Vol.4181, (Santa Clara, California), pp. 315-323, September 2000. 18. G. Van Doorselaer, J. Van den Steen, H. De Smet, D. Cuypers, and A. Van Calster, Flicker reduction in AMLC displays by individual pixel voltage correction, in Proceedings of the 20th International Display Research Conference (IDRC), (Palm Beach, Florida), pp. 447-450, September 2000. 19. H. De Smet, J. Van den Steen, and J. Doutreloigne, Custom display driver design (invited speech), in Proceedings of the SID-ME Spring meeting, (Delft, The Netherlands), pp. -, April 2001. 20. H. De Smet, J. Van den Steen, and A. Van Calster, Microdisplays with high pixel counts (invited), in Proceedings of the International Symposium, Seminar and Exhibition (SID), (San Jose, California), pp. 968-971, June 2001. 21. J. Van den Steen, G. Van Doorselaer, D. Cuypers, H. De Smet, A. Van Calster, F. Chu, and L. Tseng, A 0.9 XGA LCoS backplane for projection applications, in Proceedings of the Microdisplay 2001 conference, (Westminster, Colorado), pp. 87-90, August 2001. 22. H. De Smet, J. Van den Steen, D. Cuypers, N. Carchon, and A. Van Calster, Monocrystalline silicon active matrix reflective light valve (invited speech), in Proceedings of Microdisplay Conference, (Edinburg), pp. -, September 2001. 23. J. Van den Steen, G. Van Doorselaer, H. De Smet, and A. Van Calster, On the design of LCoS backplanes for large information content displays, in Proceedings of the 2nd RUG-FTW PhD Symposium, (Gent, Belgium), p. paper 79, December 2001. 24. H. De Smet, J. Van den Steen, and P. Colson, Use of stitching in microdisplay fabrication, in Proc. of SPIE, Vol. 4657: Projection Displays VIII, (San Jose, USA), pp. 23-30, SPIE, January 2002. 25. H. De Smet, D. Cuypers, A. Van Calster, J. Van den Steen, and G. Van Doorselaer, Design, fabrication and evaluation of a highperformance XGA VAN-LCOS microdisplay, Displays, vol. 23, no. 3, pp. 89-98, 2002.

26. G. Van Doorselaer, D. Cuypers, H. De Smet, J. Van den Steen, A. Van Calster, K.-S. Ten, and L.-Y. Tseng, A XGA VAN-LCoS projector, in Proceedings of Eurodisplay 2002, (Nice, France), pp. 205-208, October 2002. 27. D. Cuypers, G. Van Doorselaer, J. Van den Steen, H. De Smet, and A. Van Calster, Assembly of an XGA 0.9" LCoS display using inorganic alignment layers for VAN LC, in Proceedings of Eurodisplay 2002, (Nice, France), pp. 551-554, October 2002. 28. D. Cuypers, G. Van Doorselaer, J. Van den Steen, H. De Smet, and A. Van Calster, A projection system using vertically aligned nematic liquid crystal on silicon panels, in Proceedings of IDW 2002, (Hiroshima, Japan), pp. 45-48, December 2002. 29. D. Cuypers, H. De Smet, G. Van Doorselaer, J. Van den Steen, and A. Van Calster, Measurement methodology for vertically aligned nematic reflective displays, in Proceedings of SPIE, Vol. 5002, (Santa Clara), pp. 6272, January 2003. 30. H. De Smet, J. Van den Steen, G. Van Doorselaer, D. Cuypers, N. Carchon, and A. Van Calster, On the development of VAN LCOS microdisplays (invited), in Proceedings of the IEEE LEOS Annual Meeting Conference, (Tucson, Arizona), pp. 814-815, October 2003. 31. H. De Smet, J. Van den Steen, and D. Cuypers, Spice model for a dynamic liquid crystal pixel capacitance, in Proceedings of the 10th international Display Workshops (IDW), (Fukuoka,Japan), pp. 53-56, December 2003. 32. D. Cuypers, J. Van den Steen, G. Van Doorselaer, H. De Smet, and A. Van Calster, WXGA LCOS projection panel with vertically aligned nematic LC, in Proceedings of the 10th international Display Workshops (IDW), (Fukuoka,Japan), pp. 1541-1544, December 2003. 33. H. De Smet, J. Van den Steen, and D. Cuypers, Electrical model of a liquid crystal pixel with dynamic, voltage history-dependent capacitance value, Liquid Crystals, vol. 31, no. 5, pp. 705-711, 2004. 34. T. Borel, H. De Smet, J. Van den Steen, et al. , Report on the results of theLCOS4LCOS project, SID06, to be published

Chapter 1 : Welcome to Display World

CH1 - 25

Chapter 1 : Welcome to Display World

he target of this book is to describe some of the issues concerning the design of liquid-crystal-on-silicon (lcos) micro-display backplanes. This first chapter introduces the research topic in more detail. An lcos micro-display backplane comprises a special chip. Its behavior is much that of a programmable slide. Essentially, it is an electro-optical device; lcos micro-displays are often referred to as light valves: electrical signals steer the modulation of a beam of light in a few million points. It is useful to place the lcos display case in a broader context and to try to assess the following questions: is this particular system the display system that will conquer our houses, is it meaningful to spend research resources to such a subject and which are the basic principles behind display operation? The first paragraph of this chapter recalls three essential facts about human vision; three facts that are fundamental to understand the operation of most display systems. The second paragraph briefly mentions economical aspects and gives an indicative value of the "real world" market. After all, what's the practical interest in research on displays? The third paragraph is entitled application specific displays. It is indeed justified to some degree to state that each display type fits a particular application. A classification attempt is made to help distinguishing between the several display technologies, applications, etc, etc. Subsequently, the text focuses on examples of liquid-crystal-on-silicon (lcos) projection architectures. Finally, a glance at the anatomy of an lcos micro-display backplane wraps up this chapter and by then provides more insight in the meaning of the thesiss title Welcome to display world!

1.1 How we see colored, moving images

The first paragraph recalls three facts about human vision, essential to understand the operation of any display system. Todays state of evolution is marked by exchanges of increasingly massive quantities of information. In this respect, display systems play a prominent role in everyday communications. But how did mankind come to invent display systems? Lets look at some historical facts: some 200 years BC, the famous Aristote noted how the suns image remained visible for a while, after turning his eyes away from it! Around 1650 the magic chamber shows up; it is not clear whether it should be attributed to Athanasius Kircher or Christiaan Huygens. Later, amongst other discoveries in optics, Joseph Plateau (1801-1883) studied phenomena of image CH1 - 25

CH1 - 26

Chapter 1 : Welcome to Display World

retention by the retina and invented a precursor of animated films in 1831: the phenakistiscope. Eventually in 1895, the brothers Lumire came up with the first movie camera. On november 18th 1929 Vladimir Kosma Zworykin demonstrates a TV receiver containing his version of the Braun tube (1897). After the war, David Sarnoff (chairman of RCA) said about color that it added sight to sound. Color TV was born somewhere in the fifties. It took mankind to 1965 before color specialization of the retinal cones was demonstrated. In the early 70s Peter Brody brilliantly demonstrated the concept of active-matrix liquid-crystal displays [1],[2],[3]. In 2000, displays continue to invade and possess the entire world. Someone said seeing is believing and everyone appears to need to believe in something Compared to the other physiological senses, the eyesight is maybe the most amazing and probably the most instrumental of our senses. As such, an accordingly important place is to be given to the eye. Honestly, it is hard looking into display systems without a little understanding of our own vision system. The human eye is a marvelous living-tissue optical system. In [4] plenty of information can be found just about the perception of contrast; however, for this introduction, a few facts will suffice. Consider figure 1-1 depicting a cross section of the eye.

The retina is definitely one of the masterpieces of this extraordinary optical system. It is composed of ~130 million light sensors. Most of these sensors are specialized in sensing low brightness levels there are about 120 million rods for so-called scotopic vision. The remaining ~10 million cones concentrate in the fovea centralis a small spot of the retina where images project onto. Figure 1-2 depicts the distribution of the cones and rods. The cones allow us to discriminate between colors at higher brightness levels (photopic vision). Of them 64% are specialized in red light, 32% in green and a mere 2% in blue light! The dynamic sensing range of the retina is sensational. Note that a description of the human eye system is not complete without mentioning the visual cortex the portion of our brains that is said to treat all the visual information. Our brains anticipate and interpret images, sometimes leading to illusions. CH1 - 26

Figure 1-1 : the human eye

Chapter 1 : Welcome to Display World

CH1 - 27

Figure 1-2 : cone and rod distribution over the retina Three elements are essential to understand how electronic displays work. It is the eye's capability of mixing colors (1) over time (2) and in space (3) that should retain our attention here. The text below only provides a rough overview and the reader is referred to literature for extensive discussions see e.g. [5]. 1) Most color impressions perceived by the human eye can be reproduced by summing red, green and blue primary colors. In addition, many color combinations can yield one and the same color impression; these combinations are called metamers. An excellent tool to represent this is the 'chromaticity diagram'. This is the 1931 chart, enough for the purpose of the illustration.

green yellow cyan magenta red

(nm)

(nm)

blue

Figure 1-3 : the CIE chromaticity diagram CH1 - 27

CH1 - 28

Chapter 1 : Welcome to Display World

The chromaticity diagram shown in figure 1-3 is a systematic collection of color impressions a standard human eye is able to detect1. It is an instrument that allows to quantify and to compare color display systems. Many color displays use three primary colors (red, green and blue). These points define a triangle of color impressions that can be emulated with weighted combinations of the three primary colors. The larger the triangle, the more color impressions can be reproduced. For different display systems the coordinates of the primary colors can differ and thus are a performance parameter. As a result, one color picture element (color pixel) often contains three sub-pixels (red-green-blue RGB, or sometimes cyan-magenta-yellow CMY). Because of the spatial frequency of the phosphor primaries of a CRT display, one can say the color reproduction is obtained through spatial-multiplexing. Another technique relies on time-multiplexing to reproduce color. With time-multiplexing, our brains interpret quick successions of images as full color images. The latency of the eye is at the basis of color perception and of the perception of video. 2) Sufficiently fast successions of still images are seen as 'moving images' see e.g. the phenakistiscope mentioned above for a rather ancient example. Later in history one of the first electronic displays shows up, the 'Braun tube'. It is the pre-runner of our everyday, omnipresent television (TV) sets. As electronics enable ever higher switching speeds, the booming of electronic displays is unavoidable. However, even nowadays, flat panel displays reproduce moving images by displaying ... (very) quick successions of still images. The number of images displayed per second is called the frame rate. Note that if the frame rate of a display system is too low, the human eye will detect it. This is just one case of display flicker. The eye can be fooled with sufficiently high frame rates, and cannot detect flicker. This is easily verified by comparing the images of 100Hz TV sets with their older 50/60Hz variants. 3) Below a threshold viewing angle, intensities and colors will be averaged out. This happens when fine details become too close to each other, or when the magnification level (zoom level) is small enough. It means that details in a small area will be averaged into a single 'spot'. A group of spatially close red, green and blue spots yields a 'white' impression, provided the viewing distance to the spots is large enough. This characteristic is at the very heart of many color display systems. Below a threshold resolution, the eye will detect no extra information. Thus, the viewing distance determines the number of pixels a display must count to make image discretization very hard to detect. Current color display systems for moving images rely upon each of these features. In particular, this is demonstrated by the single cell color sequential optical architecture (see 1.3.2) and by the color thin-film-transistor (TFT) active matrix displays in laptops. It is nice to know all this; however, as we do have working displays, why does display research still make sense?
Note that at most the colors within an cyan-magenta-yellow (CMY) triangle can be correct, because the printer used for the reproduction of this figure is not capable of more. A display or printer produces color impressions limited to the inside of the triangle formed by the so-called color primaries.
1

CH1 - 28

Chapter 1 : Welcome to Display World

CH1 - 29

1.2 Market significance

The second paragraph briefly mentions economical aspects and gives an indicative value of the "real world" market. After all, why research on displays? Todays state of evolution is marked by exchanges of increasingly massive quantities of information. In this respect, display systems play a prominent role in everyday communications. Maybe this is a sufficient incentive to have a closer look at what display systems are sold around the planet. For sure, market estimations are neither a goal nor an essential tool for the research oriented work presented here. Rather does such exercise provide a means to evaluate an aspect of the relevance of research in this field. Looking at market data is another interesting way to learn about "display world". Markets only exist because there are customers. Fortunately for the display community, there are plenty of customers desiring more and better display systems. There is a growing demand for large, flat displays. McLaughlins 2005 market estimate for microdisplay-based rear-projection TVs (RPTVs) is 13 million units by 2009 (5 million units being the conservative estimate). The sales prices for liquid-crystal-on-silicon based RPTVs is around 2600$ according to Microdisplay Report of august 2005.
Projected FPD market (all technologies, Displaysearch) 70 60 50 Billion $ 40 30 20 10 0 1998 1999 2000 2001 2002 2003 2004 2005 2006

Year

Figure 1-4 : estimated total flat panel display market Note that the market figures are maybe optimistic estimates to attract investors they give an indication of what several manufacturers believe to be the market potential. Depending on who ordered the market study, numbers and proportions will vary this isn't exact science anymore. Anyway, the billion euro figures on displays can help appreciating the drive behind micro-display research [6],[7],[8]. Another evidence of the role played by displays in real life, is the large number of displays that operate at all time of the day and that show up in a multitude of equipment. Maybe it all started with vacuum fluorescence displays (VFDs) in hand-calculators; for sure many people remember the emergence of liquid crystal displays (LCDs) in CH1 - 29

CH1 - 30

Chapter 1 : Welcome to Display World

wristwatches. Next to LCDs, various other display technologies exist. Electroluminescent displays show up in car radio displays, plasma displays show up in airports and luxury living rooms, people want personal digital assistants (PDAs) and mobile phones with color displays, home theatre systems, etc. The more new technologies emerge, the more applications show up. Whether lcos displays will be the global winner remains to be shown... but a trend has definitely been set. Someone even said about liquid crystal materials: "Liquid crystal has a habit, if you look historically, of winning race after race. It does it in a systematic way, and conquers each market completely." [9]. Interestingly, the history of LCDs is reviewed in [10].

CH1 - 30

Chapter 1 : Welcome to Display World

CH1 - 31

1.3 Application specific displays

1.3.1 Classification attempts

Do you speak CRT, LCD, PDP, OLED, (O)EL, FED, VFD, H/LTPS, X-SI, lcos, DMD, DLP, TMA, MEMS, or...? And how do you like VAN, STN, FLC, cholesteric texture LC, TN, IPS,...? Name of the display technology Amorphous silicon Cathode ray tube (Braun tube) Digital direct drive image light amplifier Digital light processor Digital mirror device (In/organic) electro-luminescent displays Field emission displays Flat panel display Grating light valve High/low temperature poly-silicon Liquid crystal display Liquid crystal on silicon Micro electro mechanical systems (Organic) light emitting diodes Organic light emitting polymers OLED on silicon Plasma addressed liquid crystal Plasma discharge panels Tilted mirror array Vacuum fluorescent displays Vacuum fluorescence on silicon Crystalline silicon Abbreviation a-Si CRT D-ILA DLP DMD (I/O)ELD FED FPD GLV H/LTPS LCD lcos MEMS (O)LED OLEP OLEDOS PALC PDP TMA VFD VFOS x-SI

Table 1-1 : some abbreviations relating to display technologies The following lines attempt to classify the different types of display technologies. It is an attempt, because there is such a variety of differentiating parameters that it is hard to get all of them within a single, simple, comprehensive catalog. In other words, there is no place for a binary classification tree.

CH1 - 31

CH1 - 32

Chapter 1 : Welcome to Display World

Classification 1: direct view vs. projection Direct view: a) CRT (TV sets, PC monitors, video walls, measuring equipment) b) FPD (PDP, LCD, ELD, VFD, FED) Projection: a) NTE or HMD b) Front/rear projectors (MEMS, LCD (HTPS, lcos), CRT) Note: CRTs have been used both in direct view and projection systems. Classification 2: emissive vs. valve Emissive: a) Phosphor excitation: e.g. CRT, FED, VFD, PDP b) Electron excitation: OLED/P, ELD Valve: a) MEMS (reflective: DMD, GLV, TMA) b) LCD (transmissive: LTPS, HTPS, x-Si; reflective: x-Si - lcos) Note: Crystalline silicon (x-Si) technology has the advantage of being mature and is used either for reflective (lcos) or for transmissive panels (Kopin) Note: D-ILA: digital direct drive image light amplifier, originates from a combination of an emissive and a valve display! The thesis focuses on the design of a part of lcos projection systems: the silicon backplane. But what do lcos systems look like; what are they composed of and how can it work?

1.3.2 Lcos systems

Near-to-the-eye and projector applications Lcos micro-displays are small, lightweight and very thin displays (~2mm). They enclose a liquid crystal layer as such, they are quite small LCDs. Typical display diagonals are between 0.5 and 1; a diagonal of 0.7 is an unofficial standard. Some people do call them mini-LCDs. They really are very flat and lightweight; the overall system needs magnification optics. As cost increases dramatically with the size of the optical components, it is clear why everyone wants small display diagonals. There is also a lower limit set by diffraction. Diffraction causes the outgoing beam to split into several lobes. The energy lost in these sidelobes reduces the light throughput. This loss in throughput increases with decreasing display 2 dimensions and becomes unacceptable at some point. It can only be countered with again larger and thus more expensive optical components or, by imposing a lower limit on the display dimension (see footnote 2 and also [11]).

Actually, it is the small dimensions of the pixels that cause diffraction; at equal pixel counts, smaller diagonals imply smaller pixel dimensions.

CH1 - 32

Chapter 1 : Welcome to Display World

CH1 - 33

Very compact optics are well tailored for use in NTE applications like HMDs. HMDs are used by pilots, firefighters, gamers, maintenance personel, hi-tech freaks,... NTE systems must be very lightweight and indeed do not need high-power, bulky illumination systems. On the other hand, lcos based front/rear projectors allow for high performance optics and much larger brightness levels. From an optical point of view, lcos micro-displays either a) form real images on a screen (front/rear projectors) or b) form virtual images in so-called near-to-the-eye applications The primary focus of the work presented in the next chapters is on micro-displays for front/rear projection applications. Indeed, at the start of the research, no specific plans were made to tailor the backplane for NTE applications. The next (brief) presentation of the monochrome projection architecture serves as introduction to the more practical color architectures. A single panel monochrome projector The overall system components can be grouped into electronics, optics and an lcos micro-display panel bridging the two. Both the electronics and optics can be decomposed into chains of components. The data input of the projector can be any kind of image source like a video player, TV receiver or computer video card. Because of the wild variety of image formats and the specific requirements of the lcos panel, conversion of the input image data is necessary. Usually, the conversion comprises several operations; it can include (re)sampling, de-interlacing, frame rate conversion and keystone correction. Specific chipsets must be developed for this and an image memory is necessary. In the examples of this book, the digital data are converted into analog signals by means of DACs and look-up tables. This conversion can be done by the lcos panel, or by the external system electronics. The look-up table is necessary to adjust for the nonlinear response of the eye to lower gray shades. One can question whether all electronics should be on a single chip. The advantage of a single chip system is cost, especially with larger volumes. However, as chip technology is tuned to the electronic function to implement, cost effective solutions imply different chips. E.g. Microdisplay Corp. sells lcos chipsets consisting of a few chips. Early micro-display processes are known to suffer from relatively low yields, so additional functionality would imply a further decrease of the yield (except if the extra circuits include redundancy). Any projection system uses some kind of lamp as light source. Lets try to explain roughly what happens to the light from the lamp. Of course, this does not constitute a manual for the optical design of a projection system. The micro-display (D) modulates a rectangular-shaped beam of light and thus determines the image displayed. By comparison with a slide projector, the slide is the light modulating element. The speed at which a sequence of images can be

CH1 - 33

CH1 - 34

Chapter 1 : Welcome to Display World

displayed is comparable or higher to what cinema projectors or TV sets achieve. In other words, lcos micro-displays are fast enough to display moving images (video). A basic monochrome projection setup is shown in figure 1-5. Note that every component has limited efficiency and every component therefore decreases the amount of light that reaches the screen. A pair of components is fundamental to the operation of this system: the polarizing beam splitter (PBS) and the micro-display (D). Let's follow a ray of light from the lamp down to the projection screen. A lamp converts electrical power into light or optical power with an efficiency of about 20%. With so-called arc lamps, the light is generated inside a continuously changing arc shaped volume between two electrodes. The first requirement is the elimination of undesired IR (infra red) and UV (ultra violet) wavelengths out of the light beam. These wavelengths are generated by the lamp as undesirable by-products. IR waves must be stopped because they generate useless heat in the projection system and UV waves also, because they have been shown to shorten the lifetime of lcos panels [12],[13]. A significant portion of the light is thrown out here.
IR, UV filters pre-polarizer, condensing lens PBS D electronics LUT, DAC frame memory
image data converter

Lamp

Integrator post-polarizer, projection lens

Figure 1-5 : a basic monochrome lcos setup The second 'operation', is the averaging of the (remainder of the) lamps' arc over the rectangular image area of the LCD panel. In other words: it is about the conversion of the arc-shaped beam into a uniform rectangular beam. This is done by an optical integrator; usually its optical efficiency is close to 100%, so light power losses are limited at this stage. A further tailoring of the light beam is polarization in this example. This is done with polarizing filters. As shown in figure 1-5, the polarization is done by two subsequent components: a high efficiency pre-polarizer and a subsequent polarizing beam splitter (PBS). The pre-polarizer is necessary because of the insufficient polarization efficiency of PBSs. Note that light with the other polarization direction is lost. That is a 50% loss huge enough for people to develop smart polarization converters (sp converters: senkrecht/parallel converters) that limit the loss. The benefit of sp converters is offset by tendue matching problems the loss is limited to about 35%. CH1 - 34

Chapter 1 : Welcome to Display World

CH1 - 35

The polarizing beam splitter is a key component. For one polarization it behaves as a transparent block of glass, but for the other it behaves as a 45 degree mirror. The path of the latter polarization is bent over 90 degrees. If the (reflective) lcos panel does not modify the polarization, the light returns to the lamp; otherwise the rays bounce towards the projection lens. A 90% efficiency is a typical figure for PBSs.
PBS out: s out: p

in: p&s polarization

Figure 1-6: PBS operation The second key component is the light valve. In the case of lcos this, is often a polarization converter. The lcos panel controls the voltages of ~1 million pixel electrodes; each of these electrodes controls the polarization conversion rate in corresponding point of the image. The light is reflected backwards into the PBS; the polarization of the reflected light being modified by the D. Practical implementations of lcos panels rarely obtain throughput efficiencies above 85%. Depending on the polarization state, the PBS will send the reflected light either to the projection lens or back to the lamp. A post polarizer overcomes the limited selectivity of the PBS. Here also the light throughput is diminished further. At last the projection lens enlarges the images with some efficiency. As a result the light throughput (only taking the visible spectrum into account) is computed by the product of the efficiencies of the components: 0.99 (integrator) x 0.65 (prepolarization) x 0.9 (condensing lens) x 0.9 (PBS) x 0.85 (lcos) x 0.9 (post-polarizer) x 0.95 (projection lens) = 0.37. Note that the efficiency values are optimistic and nothing is said about tendue mismatches; in real life, color projectors do not reach 20% throughput! With color systems, the optical architecture gets slightly more complicated. Several architectures have been developed and have been successfully patented. The next paragraph outlines the structure of the representatives of two important families of optical architectures: the triple-panel and single-panel architectures. Colors with lcos: the triple-panel and single-panel architectures One way to build a color projector is shown in figure 1-7 [14]. The beam of white light is first split into two beams by dichroic mirrors. The blue beam travels to one of the three Ds. The yellow beam is split into a green and a red one. Each beam is simultaneously modulated by a D. An X-cube combines the output of the three CH1 - 35

CH1 - 36

Chapter 1 : Welcome to Display World

PBSs back into a single beam. Note that the three color sub-frames are renewed synchronously. D mirror R+G W Lamp mirror dichroic mirrors Figure 1-7: three-panel reflective architecture The next architecture is different, as it generates color sub-frames one after the other. Hence its name: color-sequential architecture. It has the advantage of a lower component count; hence one can expect lower cost. There are not only advantages to this approach. E.g. rgb color wheels have a maximum throughput of 33%; optical performance of the system is hampered by lower throughput and perceivable color break-up phenomena. To obtain good video quality the frame rate must be multiplied at least by a factor of three. In [15], frame rates well in excess of 400Hz are used. These disadvantages can be compensated for, at the cost of larger display diagonals however. This raises the cost of optical components and of the D panel itself. B PBS PBS pic proj. lens D R PBS

CH1 - 36

Chapter 1 : Welcome to Display World

CH1 - 37

Color wheel

proj. lens

D W Lamp, polarization filters, PBS

Figure 1-8: single panel, frame sequential color architecture

Figure 1-9 is a picture of the inside of the labs three-panel WXGA color projector. It uses a very compact Unaxis cube. There are many different architectures, this is just one more example [16],[17].

red D

blue D green D

white in rotated out

Figure 1-9: Unaxiss ColorCorner architecture

CH1 - 37

CH1 - 38

Chapter 1 : Welcome to Display World

1.3.3 Lcos micro-displays: definition and physical structure

A glance at the anatomy of an lcos micro-display backplane wraps up this chapter and provides more insight in the meaning of the thesiss title. The definition of the term lcos micro-display partially reveals the physical structure. Definitions of the terms 'micro-display' and lcos Micro-displays are displays that have such small dimensions that magnifying optics are needed to use them. The display device diagonal is typically in the range of a few cm (~ 1 inch). In essence, micro-displays are electro-optical devices: the modulation of beam of lightrays is controlled by electronic signals. The exact nature of the electro-optical effect acts as a differentiator between the several competing micro-display technologies. Each of these technologies has specific advantages, yet none has come out as the sole winning combination. Lcos or liquid crystal on silicon is one of the competing electro-optical effects. The many supporters believe it is the cheapest technology. As the name suggests, lcos has two components: a thin layer of liquid crystal and a silicon chip. Just another chip? Silicon technology is well established (cost, yield, manufacturing techniques ...) and can be used with minor modifications. This fact is important, because there is no need for massive investments in a dedicated processing plant. Chips used in lcos micro-display assembly are just another diversification of ICs (integrated circuits). The micro-display chip or silicon backplane produces a pattern of electrical force fields inside the LC layer. This pattern corresponds with the image to be displayed. Often this chip will be referred to as the (micro-display) backplane. The electro-optical interface: liquid crystals The discovery of liquid crystals (LCs) is attributed to an Austrian scientist (Reinitzer, 1898). The aggregation phase of liquid crystals is literally in between liquid and solid phase. This peculiar aggregation phase is referred to as mesophase. On a macroscopic scale, this phase resembles to the liquid phase; while on a microscopic scale, the molecules show some kind of long range ordering. The type and amount of molecular ordering characterizes LC materials into several classes (nematic phase, smectic phase, cholesteric or chiral nematic phase, ). LCs contain long molecules that try to align to each other. Some LC effects are highly dependent upon temperature, making them applicable in thermometers or for CH1 - 38

Chapter 1 : Welcome to Display World

CH1 - 39

thermography. Others are birefringent, i.e. they exhibit different optical refraction indices. This means that a light wave traveling through a LC layer will travel at a speed that depends on the orientation of the wave. The two indices are referred to as the ordinary and extraordinary refraction indices: no and ne respectively.
entering the LC leaving the LC (m)

component along no axis component along ne axis resulting vector, rotated over 90o on exit

Figure 1-10: birefringence in a layer of LC Figure 1-10 depicts a linearly polarized input wave that enters a LC layer. Two components make up this wave, each facing different refraction indices. The full arrow represents the wave vector, the dashed and dash-double dot arrows the two components. On entry, both components are in phase. The dashed component travels faster as it needs only two periods to reach the exit; the second component needs exactly two and a half periods. This makes them to be in opposite phase on exit. The outgoing wave is again linearly polarized (in this example), however rotated over 90 compared to the incoming wave. The layer thickness and the difference in refraction index play a substantial role in the determination of the resulting phase rotation. The phase difference can be modified by an electric field, because the LC dipoles reorient themselves and this on turn causes a change in n. As described earlier, the combination of a chip that provides a pattern of electrical fields with an LC layer and a polarization-selective optical component yields the core structure of an lcos micro-display. The lcos sandwich Figure 1-11 shows a cross section of an lcos micro-display. The glass plate and the backplane encapsulate the LC layer. The glass plate also functions as common counter-electrode (CE) to create the electrical fields. Spacer balls assure a minimum thickness and help to maximize the uniformity of the thickness. A hermetic seal ring protects the LC against chemical alteration by moisture or oxygen, because this would destroy the LC. The orientation of the LC director is initiated along both the CH1 - 39

CH1 - 40

Chapter 1 : Welcome to Display World

glass plate and backplane by means of an align layer. The way to do this, however, is probably enough for another thesis. The dimension ratios in figure 1-10 were not taken into account for the sake of clarity; the following figures give a better understanding of the differences in order of magnitude: Align layer : nanometers LC layer and spacer thickness : 3 micrometers Backplane contact area : 100m x100m Backplane and CE thickness : 0.8mm Chip size : 2cm x 1.5cm Because it involves only a few steps, LC assembly seems relatively simple, yet many work very hard to get a robust, yield-effective LC technology. Many parameters control the properties of the thin layers and this makes LCs the subject of a branch of science on its own. The backplane steers the (average) orientation of the LC molecules and thus controls the amount to which the waves are affected. How the light waves are affected, depends on the type of LC layer they travel trough. The electro-optical effect is usually one out of the following list: polarization rotation effect, scattering effect or diffraction effect.
Backplane connections Seal Spacers LC Counter electrode glass Cross-section Silicon chip Cross-section Align layer CE connection

Figure 1-11: a first glance at an lcos micro-display The control electronics (figure 1-5) need to make electrical contacts with both the counter-electrode (CE) and the backplane. Room for this is provided with the overhanging areas of both the CE and backplane. Finally note that for mechanical robustness, it is absolutely necessary to mount this assembly onto a much more solid carrier and to provide an easy and reliable way to establish reliable connections. Figures 2-4 to -13 provide pictures of implementations chapter two as a whole is about lcos technology. So, this is where my introduction to Display World ends and where the real presentation starts.

CH1 - 40

Chapter 2: Constraints from lcos technology

CH2 - 41

Chapter 2: Constraints from lcos technology

What constraints must be dealt with for the design of a micro-display backplane? This chapter describes the technological boundary conditions to live with; it is a collection of crucial background information. The topics cover panel assembly and IC processing technology. A first section a designers perspective summarizes the lcos product chain. The second section lcos cell assembly technology highlights discusses the constraints from cell assembly technology. The third section discusses constraints directly related to the Si processing itself. Stitching takes a particular place with the processing of very large dies.

2.1 An IC designer's perspective

2.1.1 A system flow and global constraints
A simple organigram partitions an lcos system into three separate sub-systems (figure 2-1). The optical and electronic subsystems strongly relate to one another. The mechanical sub-system serves the purpose of providing mechanical robustness, along with a marketing body. It is not of primary concern here. Lcos panels actually form a bridge between the first two subsystems. The optimization of the optical sub-system is not treated in this thesis; rather a portion of the electronic sub-system is focused on. It is however impossible to consider the design of an lcos backplane chip without considering the relationships to the other components (in casu: the interface electronics and the optical engine architecture). This also reflects in the impact some global constraints have on both sub-systems. general performance specifications (pixel count, frame rate, power) optical engine architecture (color architecture, component size) LC effect (pixel architecture, IC voltage requirements)
LCOS system: LCOS RPTV, LCOS FP

Electronic sub-system: micro-display module

Optical sub-system: optical engine

Mechanical sub-system: the apparatus

Figure 2-1 : a simple organigram of a LCOS system CH2 - 41

CH2 -42

Chapter 2: Constraints from lcos technology

Some system level specs, such as the type of optical architecture, have a direct impact on the specifications of the electrical sub-system. The way the system has to communicate with the outside world (signal sources) will also affect the overall circuit architecture. The interfacing electronics requires a specific signaling 'protocol' for communication with the panel. As a last example, the mechanical construction of the optical system can request the bond pads to be on the short/long side of the chip. Chapter three addresses the electrical specifications and presents the resulting backplane circuits. The current chapter discusses restrictions inherent to lcos panel manufacturing.

2.1.2 Assembly flow of lcos panels

The manufacturing of an lcos system requires successful implementation of a long chain of fabrication steps. In Figure 2-2 below, system assembly represents the final stage where the optical engine, interface electronics and lcos panel combine into e.g. an lcos projector. Simplified, the panel manufacturing steps are lcos IC design, lcos IC processing and lcos panel assembly.
IC Design Circuit design Mask design Verification IC processing Mask making CMOS processing Lcos back end Panel assembly Alignment layer LC filling, sealing Dicing, packaging System assembly

Mechanical assembly Optical engine Interface electronics

Figure 2-2 : lcos manufacturing chain(s) Each of these steps has limitations and requirements that a designer must take into consideration; in addition to the boundary conditions set by the other system components (like e.g. optical engine architecture, interface electronics). The most complex panel manufacturing step probably is IC processing. This step is hard to modify on-demand, both from a cost point of view as from a technical point of view. This explains why not all Si foundries have dedicated lcos process modules in their roadmaps. One of the attractive aspects of lcos technology is that it fills the production capacity of existing IC foundries it does not need investments in dedicated processing capabilities. This is completely different from the situation in TFT AMLCD manufacturing, where huge investments are required for every new generation of production facilities. Although seemingly not the most complex, the CH2 - 42

Chapter 2: Constraints from lcos technology

CH2 - 43

panel assembly processes become a major hurdle with volume production ironically [18]-[22].

2.1.3 An (lcos) IC design cycle

The design flow is full custom design for the implementations discussed in chapter 3. For sure, full custom design is not the only way to build an lcos mask layout. This paragraph reviews the design flow in six points. The description starts with arrow 6 in Figure 2-3 below as it is common to all IC design cycles. The last step is indeed the transfer of mask data for the mask making; in this dissertation, the transfer data format was always the widely known MEBES1 fracture format. In this format, the data directly steer the electron beam engine for the fabrication of the masks. Note that often some of the mask data like process related alignment markers, testkey patterns, foundry mask identification, etc. are out of control of the IC designer; they are 'transparent' to the designer. As a result, a designer's job usually ends with the creation of an intermediate data file that does not contain these 'transparent' data. The transfer of this intermediate file is referred to as 'the' tape out the data file format used at this stage is most often GDSII. In other words, the 'mask and chip layout verification' box includes the translation into the fracture format, although this translation is transparent to the designer.
Schematic circuit description
1) spice netlist

Physical circuit layout 3) GDSII and/or

Dracula data

2) full custom

Simulation of circuit behavior

4) CDL netlist

Mask and chip layout 6) fractured data verification

5) extracted netlist

Mask making

Figure 2-3: an (lcos) IC design cycle Because of the relatively high cost of the subsequent mask making and IC manufacturing, it is essential for the mask layout to be error free. This is the most critical issue with IC design: mask layouts are so complex that bugs can easily hide in them. The problem here compares best with the problem of finding all the needles in a large haystack without knowing the exact number of needles, unfortunately. Experience shows that, even with the aid of special purpose software, verification does not necessarily guarantee 100% error free layouts. By the way, each manipulation of mask data can induce errors. A simple example is the translation from the GDSII data format into a fracture format!

for ETEC electron beam writers CH2 - 43

CH2 -44

Chapter 2: Constraints from lcos technology

The verification stage comprises three different checks: topological or geometric rule checks2 (arrow 3), computation of the electrical behavior from the chip layout (extraction: arrows 3, 4, 5) and cross checking the final fractured data (arrow 6). The verification software packages name is Dracula see paragraph 5.2 for a hands-on introduction. This well established (good old) software does more than verification alone. It is indeed capable of computing additional layout data (e.g. for planarization purposes) or even of automatically generating complete mask layouts. This feature can simplify the layout job; it often makes the layout easier to understand, to handle, etc. Getting back to the fracturing of mask data (arrow 6), fracturing indeed implies that a number of rounding errors are generated. Most often, these rounding errors are negligible however, an additional visual inspection to increase the odds further represents a justifiable cost. The extraction of layout circuit components is the basis for a thorough check of the chip's functionality (layout simulation: arrow 5, with/without parasitic components). However, the check depth depends on how the designer instructs the software to extract devices. The final extraction step occurs after a series of preliminary checks. Without such a sequenced approach, the verification software most often outputs an error message jungle that is nearly impossible to untangle, and at least at the cost of big heaps of time. The breakdown into smaller steps is necessary for efficient operation of the software and for efficient interpretation of the error reports. Full parasitic extraction is only meaningful once the layout simulations match with the initial schematic's behavior. Before layout extraction and simulation (arrow 5), the layout passes preliminary checks to remove 'gross' errors that would otherwise compromise or even halt the final extraction process. The preliminary checks include design rule checks (DRCs), electrical rule checks (ERCs) and layout versus schematic comparisons (LVS). DRCs make sure only legal combinations of mask patterns are formed from a wafer-processing point of view. They make sure the wafer processing results in physically correct devices and reliable connections. Without DRCs, the behavior of some devices may fail to match the behavior of the device models used by the circuit simulator. ERCs are interconnection consistency checks (shorts, open circuits, device connection...). ERCs require extraction of merely the main circuit components and of electrical conductors; ERCs are pre-runners for LVS. LVS compares the implemented layout circuitry with the originally planned schematic circuitry. The LVS software therefore requires reading in a schematic circuit description to compare with the layout circuit. The schematic description format used for this was always CDL (component description language, arrow 4); yet other formats are accepted as well (see software manuals). The layout file format used was GDSII (arrow 3), as it is also the file format for tape out. A LVS check with positive outcome is a solid basis to start doing layout simulations.

includes 'design rule checks' (DRCs) among others

CH2 - 44

Chapter 2: Constraints from lcos technology

CH2 - 45

With full custom design, the layout is constructed through manual placement of devices, interconnections (arrow 2). It is needless3 to say that even for moderately complex circuits, the layout job quickly becomes tedious. Three different (Cadence) tools speed up the layout job, especially for circuits involving large amounts of repetition. A first tool is the 'structure compiler, which helps generating aligned arrays of cells through parameterization of the arrays. The second tool comes with parameterized cells the so-called pCells. Pcells represent programmable layout blocks. The third tool is the Skill language interpreter that allows to access, to modify and to create data directly in the design database. Section 5.3 gives scores of examples. Yet another tool is the verification software; it can automatically generate some (parts of) mask layouts from the data for other masks and/or from data on extra help layers. The basis for the creation of a mask layout is the schematic circuit description. Spice netlists are systematic textual description of a circuit (arrow 1). Graphical editors like Cadence-Composer or simple text editors are used to create spice netlists. A spice circuit simulator checks the intended behavior of the schematic. At this simulation stage, it is necessary to evaluate the behavior taking into account the tolerances that exist on component values, on component models, on temperature... Moreover, eventually before anything can start, a sound set of technical specifications is the ultimate input to start up the whole design cycle. One last point is missing: the design cycle is used both for sub-blocks of the chip circuit as for the entire chip circuit. In other words, this cycle is repeated again and again, starting with small sub-blocks and ending with the complete chip layout.

2.2 Lcos cell assembly technology highlights

This section describes design restrictions imposed by the cell assembly processes. The first paragraph presents an lcos cell cross section along with the issues of spacer technology and chip planarity. The second paragraph gives a quick look at the ways reflective LC cells can be packaged and how electrical contact can be established. The last paragraph collects a set of mask design rules intended to prevent design errors that would jeopardize the yield of the panel assembly process.

2.2.1 The lcos cell structure

2.2.1.1 Cell cross section Cell cross sections tell a lot about the structure of lcos panels. The cross sections include cross sections of the silicon backplane; the text however must not disclose
3

not the ones from the haystack CH2 - 45

CH2 -46

Chapter 2: Constraints from lcos technology

details hereof4, so only typical drawings are presented. The dimensions in the drawings are for the illustration solely, but still are realistic. With lcos panels, the spacers are necessary for controlling the uniformity and the thickness of the LC layer. The LC layers local birefringence variations allow for light modulation. Electrical fields between two electrodes control the birefringence variations. Any additional, uncontrolled birefringence variations are undesirable for evident reasons. The CE glass serves the purpose of common counter electrode, while the Si chip controls the voltages on the individual pixel electrodes. The first item in the cross section (Figure 2-4) is the (common) counter electrode. This electrode is made of a standard glass (e.g. Corning 1737F) covered by a transparent conductor (Indium Tin Oxide or ITO). This may very well be the simplest component. It is essential for the thermal expansion coefficient of the CE glass to match with the backplanes (the chip) thermal expansion coefficient as closely as possible. This conceals part of the mechanical robustness concerns. The ITO conductor layer is at the LC side of the cell.
1 : counter electrode 2 : LC layer, seal 3 : silicon backplane 4 : panel carrier board 4 1 2 3

1 2 3

Figure 2-4 : cell cross section

This information is protected by non-disclosure agreements (NDAs)

CH2 - 46

Chapter 2: Constraints from lcos technology

CH2 - 47

(N,M)

Figure 2-5 : Si backplane cross section The second item consists of the LC layer most often with spacers embedded in it (2). A sealing ring delimits the area covered by LC material and protects the LC layer inside from moisture, oxygen... LC materials are good solvents and lose their electro-optical properties once polluted. The next item (3) is the Si backplane. A look-through view from the top and a cross section template are given in figures 2-6 and 2-5 respectively.

Figure 2-6 : layout (or top-) view of a Si backplane (3x3 pixels approximately)

CH2 - 47

CH2 -48

Chapter 2: Constraints from lcos technology

2cm

Figure 2-7 : 160x120 80 PDLC panel Finally, there is the panel carrier board (item 4) with the connections between a robust connector and both the Si chip and the CE. Impedance matching components and a portion of the peripheral electronics can be integrated on the carrier board as well. Figure 2-7 till Figure 2-10 represent (working) lcos panels.

5cm Figure 2-8 : 2560x2048 15 MOSAREL GXGA panel (note the two PCB connectors and flex interconnect)

CH2 - 48

Chapter 2: Constraints from lcos technology

CH2 - 49

2cm

Figure 2-9 : 1024x768 17 engineering panel

Figure 2-10 : commercial 12 WXGA panel prototype

CH2 - 49

CH2 -50 2.2.1.2 Spacer technology

Chapter 2: Constraints from lcos technology

The paragraph briefly discusses spacer technology, because it happens to imply the design of a mask. Some panel manufacturers try so-called spacer-less assembly to avoid visual artifacts resulting from the spacers. Spacer-less assembly means no spacers are present in the display area; however, uniformity is not easy to control with such approaches [23].

Figure 2-11 : spacers on 15m VAN pixels

Figure 2-12 : an undesirable cluster of spacers in a MOSAREL cell With spacers inside the display area, two types show up. One solution consists of placing spacers on predefined positions before the addition of the alignment layer. It is best to have random spacer positions, as the human vision is extremely good at pattern recognition. The spacer positions do not change from chip to chip. A disadvantage typical with placed spacers is their interaction with the formation of the alignment layer. The alignment layer exhibits direction dependent (anisotropic) properties; it is its sole reason of existence. Two common examples of anisotropy inducing techniques are oblique evaporation of inorganic alignment layers and velvet-rubbing of spun organic alignment layers. These two processes can lead to shadow respectively comet forming when placed spacers are present on one of the substrates. Figure 2-13 below is taken from a 2001 SID Seminar by M. Pfeiffer [24]. CH2 - 50

Chapter 2: Constraints from lcos technology

CH2 - 51

Finally, with the placed spacer solution, the designer must produce a dedicated mask layout.

Figure 2-13 : placed spacer comets The second solution consists of spraying tiny spacer balls on top of the alignment layer. Their positions are random and differ from chip to chip. To its advantage, the second solution does not exhibit shadows or comets. Spacer thickness depends on the LC effect; the precision of the LC layers thickness also depends on the planarity of the Si chip. Note that the visibility of spacers can depend on the state of the LC director. It is important for good contrast ratios to have no spacer visibility in the dark state. On the other hand, when spacers are visible in the bright state, it will affect the brightness of the display.

2.2.1.3 Chip planarity

Some LC modes require extremely well controlled thickness of the LC layer. Others are more tolerant to thickness variations. Fact is, without precautions, a chip's surface topography presents 'hills' and 'dales' of the order of micrometer. This is not surprising, considering the thickness of the patterned layers (m range). As the topography variations are large enough compared with the cell gap, it can be expected that the Si planarity affects the quality of the assembled cell. A first technique to smoothen the chip surface consists of the addition of a planarizing poly-imide layer just below the top mirror layer. Furthermore, this layer can be made opaque so that it results in an excellent light shield [23]. Fortunately, advanced sub micron processes already suffer this much from excessive surface topography that foundries anyway had to introduce a polishing technique known as chemical mechanical polishing (CMP), [59].The processing step consists of the deposition of an insulator layer that reduces the topography variations; next, this layer is polished to yield a very flat surface. The CMP step here comes after the patterning of a conductor layer (poly and/or metal layers) and covering it with an insulator. It is incapable to yield a mathematically flat surface; it works as an integrator that averages 'fast' topography variations over a small distance range. It cannot cancel topography variations that are too steep on short range or that slowly CH2 - 51

CH2 -52

Chapter 2: Constraints from lcos technology

vary over too long ranges. Figure 2-14 below is the result of a topography measurement after patterning of the last poly and prior to the CMP step.

target planarity figure! AM region

(both axes in arbitrary units)

Figure 2-14 : typical display chip topography prior to the metal (aluminum) depositions

-strips

Dummy polygons

Figure 2-15 : dummy polygons and de-coupling of signals with -trips On the mask layout side, homogenizing the fill-factor is meaningful for the conductor layers (i.e. the polys and metals) and gate area layers. It relieves some of the difficulties associated with obtaining a flat surface. Automatic procedures (see chapter 5.2) allow a designer to add dummy polygons to the layout. Addition of the dummy polygons is done in the last stages before mask fracturing (tape-out). It does not interfere with the design as far as the signals integrity is not affected. For the micro-display design case, the use of micro-strip structures safeguards the integrity for the most sensitive signals. This is most welcome e.g. for the analog video signal lines, the active matrix and long wires. A sample from a real mask-set illustrates this in Figure 2-15.

CH2 - 52

Chapter 2: Constraints from lcos technology

CH2 - 53

2.2.2 The lcos cell assembly

2.2.2.1 LC filling and cell sealing LC materials are extremely sensitive to contaminations; therefore, they need a hermetical seal. This protects the LC material from deterioration due to moisture, oxygen The concept of cell sealing is simple, its implementation however is more delicate than many would expect. The quest for good sealing processes results in cookbook secrets know-how that LC assembly houses are keen to hide. In [18], a description is given of problems with the performance of the alignment layer as a result from the sealing process. The physical dimensions and tolerances of the seal width affect the size of the d chip. A typical dimension is 1.5mm; smaller dimensions are desirable as this greatly affects the silicon cost (area/chip). In the MOSAREL project [23], a spacer-less assembly process has been developed at the cost of a double sealing ring. This double ring is necessary to help achieving good cell uniformity.

Pads for flex interconnect

LC fill openings

Double sealing rings

Figure 2-16 : MOSAREL double sealing ring design The following process steps roughly describe the LC filling and sealing (ordered): - alignment layers on both CE glass and Si backplane chip - processing of sealing ring, spacers; assembly of an empty cell - LC filling in vacuum - sealing of LC fill opening CH2 - 53

CH2 -54

Chapter 2: Constraints from lcos technology

2.2.2.2 Panel interconnection and packaging

The second item that affects the size of the backplane chip is the interconnection. With the MOSAREL project, excess Si area was available, because no additional chips would fit onto the wafer. This allowed for flex interconnection. A flexible flat cable directly attaches to the Si backplane. Such large Si area surplus rarely occurs; more usual is the situation where wire bonding is used for interconnection to an intermediate carrier substrate [25]. Wire bonding is more area-efficient. The packaging of micro-display panels is special, because high illumination levels must reach the AM (but not the rest of the chip(s)) and because of mounting requirements inside an optical engine. Packaging includes the mounting of a black (illumination-absorbing) cover with a rectangular window just above the AM (see Figure 2-10). To account for positioning tolerances of this cover, the window opening must be slightly larger than the AM and a border electrode all around the AM is foreseen. This AMborder electrode allows e.g. to produce a neat black border around the (projected) pixel array. Similarly to other chip designs, there are design rules for the bonding pads. With more conventional chip designs, it is customary to place bondpads all around the periphery of the chip. In the lcos case this would not be the optimal choice for two reasons. First, the bondpad area requirements add up to the sealing area requirements and second, wafer-scale assembly is much easier to do when the CE glass dimensions match those of the Si backplane. Finally note that the number of bondpads is expected to be much lower than the sum of the number of columns and rows. The first micro-display panel developed within TFCG had 80 bondpads for a resolution of 160x120; the 280 busbar connections are reduced to 80 (~29%). With the MOSAREL project 128 connector pins are available for a matrix of 2560x2048 pixels (~3%); with the tmdc XGA panel, still 43 connections are needed (~2.4%)

Figure 2-17 : 120x160 wire bonded to plated substrate CH2 - 54

Chapter 2: Constraints from lcos technology

CH2 - 55

Figure 2-18 : mask for 6 128-pin flexes, see also Figure 2-8 ceramic

glob top edge

bond wires

chip edge

CE glass edge

Sealing ring AM border AM

Figure 2-19 : XGA wire bonding to flex substrate with glob top protection The last item mentioned about cell assembly is the CE contact. Either the CE contact is formed between the micro-display chip and the cover-glass or it is formed between the carrier and the cover-glass. The former method requires a relatively large contact pad on the backplane chip and a conductive glue glasschip. The glue forms a conductive bridge between the CE and the pad. With this method, design rules must be defined for the size of the contacting pad, position on the chip, etc. CH2 - 55

CH2 -56

Chapter 2: Constraints from lcos technology

The other method requires a contact pad on the carrier; the contact must bridge the thickness of the backplane chip and LC layer. Figure 2-20 and Figure 2-21 below show examples of both types of CE contacts.

chip CE glass edges

carrier

active matrix

conductive glue

chip + CE glass

LCOS chip edge

Figure 2-20 : 160x120 example of CE contact towards the chip

a good CEcarrier contact and a short: CEcarrierchip!

carrier

chip

conductive glue

LCOS chip edges

Figure 2-21 : examples of a good (left) and bad (right) CE contact directly between the carrier and the CE

CH2 - 56

Chapter 2: Constraints from lcos technology 2.2.2.3 A basic set of mask design rules for cell assembly

CH2 - 57

The assembly of the lcos panel is at the origin of a set of assembly design rules. These rules apply to lcos IC designs in general. The values in Table 2-1 are indicative. The chip centre is the reference point for coordinate lists.
a b c b d a e b

b a

Figure 2-22 : micro-display floorplanning pCell [Table 2-2] below lists the parameter expressions associated with the different stretch lines and corresponding sample values. Some of the parameters are used for more than one stretch line; i.e. a parameter called pixelPitch is used in the expressions that determine the AM size, etc.
AR01. Location of bond pads (along which side) AR02. Width of sawing lane AR03. Width of sealing lane AR04. Width of connection pads AR05. Length of connections pads AR06. Pitch of connection pads AR07. Center to center spacing (bond pad to sawing lane) AR08. CE glass edge spacing to bond pad center (min) AR09. Center coordinates of CE contact AR10. Size of CE contact zone AR11. Coordinates of assembly markers (chip/wafer level) : wide/narrow : 100m : 1500m : 100m : 100m : 250m : 250m : 1000m : (,) : 1x3mm : (,) [, (,)]

Table 2-1 : a basic set of assembly rules

CH2 - 57

CH2 -58

Chapter 2: Constraints from lcos technology

The definition of a programmable cell (a pcell) helps to quickly obtain a chip floorplan that is immune to assembly process tolerances.
Stretch a b c d e f Parameter expression width of sawing lane/2.0 width of sealing width of sawing lane/2.0 width of AM or pixelPitch x nr. of cols (horiz.l) IF(bpLngSde == false) width of bonding ELSE 0 IF( bpLngSde == true) width of bonding ELSE 0 height of AM or pixelPitch x nr. of rows (vertical) Typ.val 50 1450 17mm 1mm 1mm 15mm Typ.val 150 1500 1000 true 1032 776 10

Parameter wSawLne wSealRng wBpArea bpLngSde pixCntX pixCntY pixPitch

Parameter meaning width of sawing lane width of sealing ring width (/height) of bonding pad area [binary] true for bonding pads on the long side pixel count in X direction pixel count in Y direction pixel pitch dimension (square pixels assumed)

Table 2-2 : List of parameters of floorplanning p-cell

Figure 2-23 : sample lcos die floorplan CH2 - 58

Chapter 2: Constraints from lcos technology

CH2 - 59

2.3 IC processing technology highlights

The goal of this section is to give an overview of different IC process flows. These are the flows used for the manufacturing of the lcos panels the author designed. The processing flow determines the mask set and its inherent limitations. One of the limitations is the maximum field size of stepper reticules. It ultimately obliges to the use of a technique called stitching which will be introduced in the second paragraph. The subject of stitching is treated in detail in the next chapter (chapter 3). The first paragraph consists of a listing of four flows the aim is to provide the reader with an overview and a number of difficulties associated with each flow. The flows can be summarized as follows: the first REFLECI flow is a lab-scale demonstration flow [26],[27]. The MOSAREL flows are upgrades applicable at an industrial level. The fourth flow is a virtual flow; it groups the flows available with some of the largest foundries in the world. Planarization of the silicon chips is an important step as it helps with the optimization of the yield of the assembled cells. OHara et al. [59] report on chemical-mechanical polishing (CMP) as a method to improve the quality of the mirrors and to improve the efficiency of their spatial light modulators. This article refers also to prior art on how to avoid hillock formation. Hillocks can badly affect the quality of the mirror electrodes. CMP fortunately has become a standard processing step with sub-micron technologies. The reason it is needed relates to the processing of stacks of metallization layers. This fits with what is needed. However, we started with the use of standard cmos chips, i.e. with no microdisplay specific layers whatsoever. Initial work in the lab has been done by Nadine Carchon, as she did the planarization and back-end processing of the REFLEC chips. The second paragraph Photolithography I: some basics is again a list that provides the reader with an introduction to stitching. Stitching is discussed in detail in section 2.4. The introduction starts with reviewing some essentials of photolithography. The terms light-field and dark-field masks are defined thereafter. Next, 1X and 5X lithography are described; the very last sub-paragraph explains why and in what circumstances stitching is required.

CH2 - 59

CH2 -60

Chapter 2: Constraints from lcos technology

2.3.1 Synopsis of 4 lcos-IC processing flows

The basis is formed by the standard 2.4m CMOS 2p-2m 18V technology from AMIS-Oudenaarde5 (Belgium). Planarization of the chip is done through spinning of a transparent polyimide. The essential blocking of light is done with Brewer Science DARC. As the upper side of the DARC layer is rather rough, smoothening with an extra (transparent) layer is needed to guarantee smooth pixel surfaces. The mirrors are obtained by Al sputtering and patterning, with an adapted sintering step to avoid hillock formation [28]. 2.3.1.1 REFLECI HBiMOS 160x120

Note 1: the lab-scale lithography process has a minimum feature size of about 3~5m. This seriously affects the maximum pixel aperture ratio and thus the total light throughput. Note 2: the lcos post-processing (planarization, light blocking and mirror layer) needs to handle rectangular dies, instead of complete wafers. This is because the 160x120 silicon is processed through Europractice multi-project wafer (MPW ) runs: wafers from MPW runs contain chip designs from others and only the labs part of the wafer is delivered to us. Note 3: lab scale developments demonstrate the performance of the planarization and light blocking layers [29]; nevertheless, they are not standard industrial process steps. The very advantage of the approach with the DARC material is that a cheap, standard CMOS process can be used. However, fears for contamination by dye particles from the DARC material itself blocks industrialization by an IC foundry. An alternative could be the investment in a dedicated clean-room facility (expensive). Note 4: the wafers are only 4 inches in diameter, which significantly limits the number of chips per wafer. It is also a 2m process which limits the level of integration (the density and/or amount of circuitry).

2.3.1.2 MOSAREL DARC

Standard 0.7m 1p-2m 5V C, DMOS + proprietary SDEMOS pixel switch + proprietary lcos back-end. Planarization (OLIN) + light blocking (DARC) + planarization (polyimide). Mirrors (un-sintered Al)

Note 1: there arefears for fab contamination by particles in DARC material what is the portability of this approach? Note 2: the flex interconnection selected, requires patterns only for m1 over a large area Note 3: mix-and-match of wafer and stepper lithography needed [see section 2.4] Note 4: the life expectancy of the SDEMOS gate is unknown

previously Alcatel-Mietec, then Alcatel Microelectronics, now AMIS.

CH2 - 60

Chapter 2: Constraints from lcos technology

CH2 - 61

Note 5: the die size is very large; only 4 chips fit on the 6 wafers; with 8 wafers the number of chips rises to only seven.

2.3.1.3 MOSAREL TiN

Standard i2t 0.7m CMOS 5V/15V 1p-4m + proprietary SDEMOS pixel switch Planarization (CMP) Light blocking (conductive TiN) Mirrors (un-sintered Al)

Note 1: there is only single CMP step after four conductor layers (after metal2); Note 2: the flex interconnection selected, requires patterns only for m1 over a large area Note 3: lithography of saw lanes is don by blading [refer to 2.4.1.3] Note 4: the life expectancy of the SDEMOS gate is unknown Note 5: the die size is very large; only 4 chips fit on the 6 wafers; with 8 wafers the number of chips rises to only seven.

2.3.1.4 Micro-display foundries in the world E.g. UMC 0.35m LCOS 3.3V/18V 2p-5m
Note 1: there are specific planarization requirements for cell assembly Note 2: still uses mixed gate oxide thicknesses, is this the most cost effective technology what about design portability (to other processes)? Note 3: eventually, stitching is not available. As far as known, only 6 processes allow for stitching, hereby limiting the overall yield and increasing the cost of the panel assembly.

Known lcos foundries are Amkor, Fujitsu, SMIC, TSMC and UMC. None of these foundries officially offers stitching as option; JVC however, supposedly uses stitching to obtain the large dies needed for their D-ILA product range. As far as known to the author, silicon foundries that offer stitching in their roadmap are limited to AMIS Inc. and TOWER Semiconductor Ltd. TOWER has a roadmap specific for camera chips the dual counterpart of micro-displays?!

CH2 - 61

CH2 -62

Chapter 2: Constraints from lcos technology

2.3.2 Photolithography I: some basics

2.3.2.1 Stacks of patterned layers Chips are thin (~10-3m) rectangular slices of silicon, covered with stacks of patterned layers. The patterns arise through local modification of the physical properties of the layer (insulator/conductor removal, ion implants e.g.). These treatments are applied to a complete batch of wafers at once; therefore, the layer areas to remain unchanged are protected from further processing by means of a resist layer. This protection layer is obtained by spinning photo-resist' over the wafers. Patterning of the photo-resist with photolithography defines all the areas on the wafer(s) to undergo processing. Photolithography is the cornerstone of Si technology. A simple etymological breakdown of photolithography tells about the exact meaning of it. There are three parts: (genitive form of ) or 'light', or 'stone' and the verb or 'to write'. It indeed is a technique that yields patterned layers on silicon crystals with the help of light. The photo-sensitive resist layer is patterned by means of light, the layers-to-bepatterned not being photosensitive. The stepper engine copies (by optical projection) the master image on a mask onto the photosensitive layer. Illuminated areas of the resist are (are not6) easy to develop away, making the underlying layer (un)available for processing. The mask is often a quartz plate that serves as support for a chromium pattern. E.g. the mask making process utilizes electron beam writers to implement the physical copy of the software mask layout. A photolithography sequence could be as follows: 1. Deposit the layer to pattern (by epitaxy, oxidation, spinning, sputtering...) 2. Spin the photo-resist layer 3. Print the mask pattern (illumination) 4. Develop the patterned photo-resist layer (removes illuminated areas) 5. Batch processing (etching, oxidation, ion implantation...) 6. Strip or remove the (patterned) photo-resist layer. The resulting stack of patterned layers quickly yields a rather rough surface. This reduces the reliability of a subsequent layer. Therefore, chemical-mechanicalpolishing (CMP) is applied in between layer pairs. CMP flattens the topography before deposition of a subsequent conductor layer (poly-silicon, metal). See also paragraphs 2.2.1.3 and 2.4.1.3.

so-called negative resists

CH2 - 62

Chapter 2: Constraints from lcos technology 2.3.2.2 Layer polarity and other inversions

CH2 - 63

The following discussion originates from the issue of how to interpret mask layouts. A mask layout is a collection of closed polygons; these polygons either represent areas to undergo or not a process treatment. Crudely stated, do the polygonenclosures correspond with opaque Cr regions on the quartz plate or do they correspond with clear, transparent regions on the quartz plate? Resist sign is theoretically an extra source of confusion, it is easy to eliminate though. Figure 2-24 below depicts the behavior of both positive and negative resists and shows schematically the 'simple' photolithography sequence described previously. A resist material is defined as positive when non-illuminated resist areas remain on chip as a (temporary) protection. The inverse is true for negative resists: only illuminated areas remain protected. Submicron technologies only use positive resists, because of their higher patterning performance. One can rather safely assume only positive resists are used. When drawing a mask pattern with design software, one is free to choose whether the area enclosed by a (software) polygon ultimately represents an area to be illuminated or not. The designer makes a choice, so that the mask layout job is easiest. A mask tooling request form communicates this information to the mask shop. E.g. when a polygon represents a metal connection (wire), the polygon enclosure corresponds with a metal area that must be left untouched. When a polygon represents a via hole (contact) between two conductor layers, it corresponds with an area in the insulator layer that must be etched away. To make the distinction, software layers have a property called either '(working) field tone' or 'digitized area tone' these are each other's inverse! For both, the property value is either 'dark' or 'light'. The digitized area tone refers to the areas enclosed by polygons; the (working) field is the area outside of the polygons. The adjectives 'dark field' and 'light field' are abbreviated as DF respectively LF. Layers representing connection wires are usually LF layers (dark digitized area tone), those representing contacts or vias are usually DF layers (light digitized area tone). To avoid confusion it is best to always use one of the two properties; the property values LF and DF will be used by convention for the rest of the text. Why bothering about LF/DF layers and layer inversions one can ask whether this is a problem the mask shop alone should deal with? Isnt it possible to use a fixed convention? The combination of layer inversions, mask bias and the unconventional stitching is prone to errors. Without special attention, mask bias and layer inversion result in wrong reticule layouts! Nevertheless, let us have a brief look at wafer and stepper lithography, before dealing with stitching in greater detail.

CH2 - 63

CH2 -64

Chapter 2: Constraints from lcos technology

Positive resist
0) substrate 1) e.g. sputtering

Negative resist

2) resist spinning

3) lithography

4) developing

5) e. g. etching

6) resist stripping

Corresponding top view

Figure 2-24 : lithography with positive or negative resist 2.3.2.3 Wafer- vs. stepper- or 1X- vs. 5X- lithography As an introduction to the differences between 1X (no magnification) and 5X (5x magnification) lithography, let us look at the anatomy of wafers, chips and masks in some detail. Wafers are thin disks of crystalline silicon (about 1mm thickness). They are the starting material for the manufacturing of chips. The wafers are processed in batches of e.g. 25. Wafers come in different sizes; only the most advanced X-Si processes use the largest 12 ones. The processes that best fit the needs of micro-display backplanes use 8 wafers; in some cases only 6 wafer processes are available. The wafer size impacts on the cost per chip (Table 2-3): for a typical (0.9 diagonal) backplane chip, the count on a 6 wafer halves compared with an 8 wafer. Assuming equal processing yields, the chip cost is also halved when comparing 8 processes to 6 processes.

CH2 - 64

Chapter 2: Constraints from lcos technology

Nr. of 20x15 mm2 chips as function of wafer diameter 16 6=150mm 60.8 40 8=200mm 108.1 74% 80

CH2 - 65

Wafer diameter Chips/Wafer Area ratio

4=100mm 27.0

12 =300mm 243.2 79% 194

% W area used Wafer diameter Chips/Wafer Area ratio

59%

4=100mm 3.1 1

Nr. of 2x2 chips as function of wafer diameter 6=150mm 7.1 4 8=200mm 12.6 7

65%

12 =300mm 28.3 21

% W area used

31%

56%

55%

74%

Table 2-3 : micro-display chip counts for several wafer sizes

12" 194#

12" 21#

8" 6" 4"

80# 40#

8" 6" 4"

7# 4#

16#

0.9" 20x15mm2

1#

GXGA 2x2 in2

Figure 2-25: relative wafer sizes and corresponding chip counts The wafer area utilization is never 100% for two reasons. First, the wafer handling requires a ring of ~5mm wide at the edge of the wafer. It is not guaranteed that any circuitry within this handling zone is functional. For circuitry, it therefore is a forbidden area that is systematically lost. Second, the fill factor of rectangleshaped chips within a circle cannot be 100%. This additional loss tends to decrease with larger wafer sizes and smaller chip sizes. The chip array must be correctly positioned to obtain the maximum number of potentially functional chips. This is why the large 2x2 MOSAREL chip count does not double when comparing 6 to 8 counts. Therefore also, one can expect a higher chip count increase (>2x) for chips much smaller than 20x15mm2. Figure 2-25 depicts the wafer layouts that correspond with the data in Table 2-3. After processing, a saw slices the wafers into individual chips in a process step called dicing. Usually, the chips are indeed arranged in a single, regular array. The sawing process removes some silicon from the wafer. The sawing lane too is an area free of electronic circuitry, because any patterns in there are destroyed later on. The sawing blade is a few tens of microns wide; combined with the saw alignment tolerance, this value yields a minimum saw lane width [see also rule AR02 in Table 2-1 and Figure 2-23]. Foundries fill the saw lane area with proprietary structures for CH2 - 65

CH2 -66

Chapter 2: Constraints from lcos technology

process control. As soon as the wafer is diced, this information is gone. Chip dimensions, saw lane included, determine the stepping distance for stepper lithography. Assembly rule AR11 in Table 2-1 describes the position of cell assembly align markers. Align markers are needed for most processing steps to control the position of layers with respect to one another. The capability to align precisely is crucial for the processing. Processing-align-markers can be included on each chip (stepper or 5X lithography) or just a few on the wafer (1X lithography, wafer scale cell assembly). Figure 2-26 shows a 1X mask designed for the top mirror layer of the MOSAREL test vehicle. The nature of the circuits contained on each chip is defined by the patterns on the several masks of the mask set. A particular situation occurs when several different chip designs are processed within the same wafer batch: this is what happens with MPW or multi-project wafer runs. It means that each mask layout contains the patterns for all the chips.

Align markers for wafer scale cell assembly: mirror layer only A pair of 1X align markers per layer and per wafer

5X align marker windows: combination of 5X and 1X

Figure 2-26 : picture of a 1X mirror layer mask (6 wafers) CH2 - 66

Chapter 2: Constraints from lcos technology

CH2 - 67

Figure 2-27 shows pictures of a metal1 mask layout and a photograph of an actual 5X reticule. The Cr pattern is enclosed by protection films on a safe distance so that dust particles are out of focus. Note that this trick cannot be used with 1X masks (contact printing).

> 300 m
5X align markers

upper border module

5X 1X overlap edge

lower border module

5X 5X overlap edge

special-purpose AM module
boxes = module borders > 300 m

regular AM module

Figure 2-27 : 5X mask drawing details 1X lithography is sometimes described as contact printing or proximity printing: the mask touches or nearly touches the photo-resist layer. With 1X lithography, the patterns on the mask and the resulting patterns in the photo-resist have the same dimensions. The mask pattern is copied to the wafer at once. Contact printing and extremely small feature sizes are not compatible. With 5X or stepper lithography much smaller feature sizes are possible. A lens system reduces the effective field size to an area of 2x2 cm2. Steppers illuminate the CH2 - 67

CH2 -68

Chapter 2: Constraints from lcos technology

wafer substrate step by step, one 'shot' at a time. One shot can contain several chips, if at least one chip dimension is smaller than half the maximum shot dimension. Note that the chip includes sawing lane, align and testkey patterns. 1X or wafer lithography
1

5X or stepper lithography
1 2 3 4

1 : light source 2 : blades

Legend 3 : mask 4 : lens(es)

5 : spin-coated wafers 6 : step directions

Figure 2-28 : contact printing and stepper mechanisms Figure 2-28 compares the mechanisms of both contact printing and stepper lithography. The grids drawn on the wafer show the sawing lane centers. There are no blades with 1X printing; blades allow shielding a portion of a reticule with a precision in the order of 100m. With 5X printing, it is the wafer that is being stepped in x and y directions. Finally note that the stepping process itself is fast. Up to now, mask polarity and lithography mechanisms have been looked at in more detail. They are important details as they serve as a foundation to the coming paragraphs about stitching.

2.3.2.4 The need for stitching With the demand for higher display formats and with optics-imposed minimum pixel dimensions, micro-display backplane chips can grow quite large. In the MOSAREL project a pixel array of 2560x2048 15 pixels is implemented in a 0.7m process (stepper lithography). The resulting active matrix size is 38.40mmx30.72mm way larger than the maximum dimensions of the shot window CH2 - 68

Chapter 2: Constraints from lcos technology

CH2 - 69

(20x20mm2 is the maximum field size for many stepper engines). Inclusion of the 5mm LC sealing width, further explodes the chip size to 48.4x40.72mm2. The 6 process therefore provides only 4 2x2 chips per wafer. But, can chip dimensions exceed those from the (projected) reticule? This question of course only applies to processes involving stepper lithography. Current stepper technology allows for extremely high precision at the positioning of shots. In 2000, steppers achieve positioning accuracies better than 50nm; in other words, the positioning error between two adjacent shots is less than a mere 100nm. A chip pattern is then obtained by two or more neighboring shots. Because of the high precision, it is technically feasible to accommodate for the sub-micron feature sizes in the overlap areas between two neighboring shots. Usually one tries to minimize the chip area to minimize the cost and maximize the yield. For micro-displays however, a combination of cost, image format, optical efficiency and assembly technology requirements dictates the chip dimensions. To minimize the cost when stitching seems unavoidable, two solutions show up: a) Look for a foundry that uses larger reticules; however, by 2002 the maximum reticule dimensions readily available are 2cm by 2cm. Even if available, larger reticules are more expensive as well as the corresponding stepper engine. b) Use the stitching technique; however, this requires longer design times and more complex reticule sets. And CMOS foundries accepting to do stitching are more likely to be an exception than a rule. Stitching is non-standard with many CMOS foundries. Fortunately, lcos microdisplay backplanes are not the only application that requires stitching. CMOScamera chips are also known to require stitching for the higher pixel counts! [30][32]. When stitching is used, it is necessary for the layout to anticipate the effects of stepper positioning inaccuracies. The positioning error is not at all a constant; instead it randomizes the patterns along the stitching lines. As a result, the printed patterns will slightly differ from chip to chip. The next chapter treats the subject of stitching in greater detail.

CH2 - 69

CH2 -70

Chapter 2: Constraints from lcos technology

CH2 - 70

Chapter 3: Photolithography (II) stitching

CH3 - 71

Chapter 3 : Photolithography (II) - stitching

This chapter is entirely devoted to stitching and adresses the question Is it possible to process chips that are larger then the reticule field size?. The work presented here is original material, as no prior art could be found on stitching for crystalline silicon processing. The work was carried out to meet the targets as defined in the MOSAREL Esprit project EP25340. I developed the stitching procedure, because I felt that a secure design flow is absolutely necessary considering cost aspects as well as the fact that here, the mask layout procedure is not at all standard. The discussion starts with a comparison between three different ways of partitioning the mask sets. Ease of processing and cost of prototyping make multiple-module reticule sets the preferred option, although the two other options are technically feasible as well. Retaining only the preferred option the multiple-module reticule set approach has implications on the way to construct the chip design. Particularly with the layout verification there are some difficulties. This impacts both the layout stage of the design cycle, as well as the schematic stage. Ultimately, the difficulties encountered are dealed with by following the virtual stitching procedure, described at the end of this chapter. This procedure consists of a way to organize the mask layout and of a few software routines. The reader will get insights in a tool that helps to ensure first time right designs.

3.1 Three considerations about stitching

Paragraph 3.1 compares three different ways of organizing the reticule layouts for stitching. First, 3.1.1 introduces the three approaches and explains why one of them is really not viable. Paragraph 3.1.2 details the preferred multiple-module approach. Here, the introduction of MMI-tables helps with the description of the complexity of the stepper-job. At last, 3.1.3 describes the problems encountered with the mixand-match approach. Although this last approach works technically, the use of 1X masks reduces the level of circuit integration. The reason to adopt this last approach is that one of the milestones of the MOSAREL project is the testvehicle chip. The aim of this chip is to evaluate the assembly process. The evaluation is done by addressing blocks of rows and columns in parallel, i.e. there are no integrated row or column drivers. This of course means there are practically no feature size requirements outside the active matrix area. Note by the way that, considering the current evolution of lcos microdisplays towards smaller pixel sizes, 0.35 technologies are necessary to get the row and column driver circuits integrated. In other words, the mix-and-match approach is unlikely to be adopted for current lcos backplanes.

CH3-71

CH3 - 72

Chapter 3: Photolithography (II) stitching

3.1.1 Multiple-reticule reticule sets: cost versus complexity

With very large chips, two or an even larger number of reticules can be necessary for a single lithography step. Very large means larger than the field size of the stepper engine (see e.g. figure 3-1). In such a case, the number of reticules per mask layer is multiplied by two or more. Therefore, the overall cost of the full reticule set is multiplied by the number of unique reticules needed per mask layer.

L-H

R-H

The chip measures 3.9 x 1.9 cm2

Maximum reticule field of 2x2 cm2

Two reticules can do the job: - nr.1 for the left-half chip patterns and - nr.2 for the right-half chip patterns

Thinking about prototyping costs, it means the cost literally explodes. A good example is given by the MOSAREL project: the display (pixel) area occupies 38.40x30.72mm2. With a maximum field opening limited to 20x20mm2, 4 reticules are needed. However, the size of integrated scanners and interconnection blocks is not taken into account. Extra area is also needed for sealing of the LC cell. In the end, per layer, 9 reticules are needed to pattern the full 2x2 chip area.
R1 R4 R7 R2 R5 R8 R3 R6 R9

Figure 3-1: two reticules for one masking step

5.08x5.08cm2 GXGA chip dimensions For 5 out of 9 reticules: area utilization < 100%

GXGA chip

Figure 3-2: many reticules per MOSAREL masking step As a single reticule costs about 1500; a process normally utilizing 20 masks and 9 reticules per layer would bear a mask cost of 180x1500=270k. This is a large amount, at least large enough for it to be considered a penalty in project plans. An alternative solution is most welcome. The first MOSAREL design milestone consists of a test-vehicle chip. Basically, this test-vehicle chip contains a GXGA-format active matrix (AM) with no integrated drivers. See 5.1.3 for more specifications and schematics. Special-purpose external CH3-72

Chapter 3: Photolithography (II) stitching

CH3 - 73

electronics address the matrix data lines in groups. This results in quite some wiring between the active matrix (AM) and the interconnection pads of the chip. The wiring consists of groups of relatively coarse tracks. This fact leads to a second approach where the lithography is done with both (coarse) 1X and (fine) 5X reticules: this is the mix&match approach. The point is to split the patterns into fine feature patterns of the display area and larger feature patterns outside of it. This fits particularly well the design requirements of the MOSAREL test-vehicle chip. See e.g. figure 3-3 below:
interconnection pads 5X, fine pitch precision

5X stepper field=2x2cm2 Figure 3-3: the MOSAREL test-vehicle chip layout and the wafer layout. Coarse, non-repetitive features occur outside the fine-pitch display area The field of a 1X reticule covers the entire wafer at once. The 1X mask is designed to pattern the entire wafer except the fine-pitch display areas. Per layer, this reduces the number of reticules to five: four 5X reticules for the AM areas and one 1X for the rest of the wafer. The cost now is at 150k down from 270k; a saving of a mere 120k. This is still large enough for it to be considered a penalty. Furthermore, it is clear that the 1X masks cannot provide sufficient accuracy to pattern the column and row driver circuits. If this were the case, the 5X litho would be superfluous. In other words, the cost issue does not look any brighter as it stands at 270k for the GXGA demonstrator chip. The experiences gained with the 1X-5X mix&match are described in 3.1.3. As explained there, the mix&match approach is not evident as the mask design and the equipment needed for such mix are not standard. A striking feature of micro-displays is the high degree of repetition in the chip patterns. This is what can be used to our advantage. As introduction to the multiplemodule reticule set approach, look at figure 3-4 above. The desired chip pattern in the example is simplified to a maximum; it consists of a wire running from one side of the chip to the other end, without crossing the chip edge. Partitioning the chip into arrays of relatively small blocks yields the approach with mask v4 depicted in figure 3-4. CH3-73

1X pattern precision

CH3 - 74

Chapter 3: Photolithography (II) stitching

The desired result is a 20mm by 40mm chip with a stripe pattern running from the left half to the right half of the chip. Mask versions 1(a), 2 or 3 alone cannot produce the correct result. Either the resulting pattern is continuous at the edges (mask2, 1(a) or 1(b)) or the pattern is interrupted in the middle (mask3, 1(a) or 1(b)). Two options are to complement mask1(a) with mask1(b) or to use mask4.
desired chip pattern
4x2 cm2 chip edge

2x2 cm2 max. field

non-exposed Si

exposed Si

mask v1(a)

bad
mask v1(b)

ok!
mask v2

bad
mask v3

bad
MOD2

MOD1

mask v4
module field < 2x2 cm2

?
ok?

Combined use of mask1(a) and mask1(b) implies that the reticule cost doubles. This is precisely what is meant with multiple-reticule reticule sets. Note that in this particular case, all of the available reticule area is used. Also, note the dimple in the middle of the resulting stripe pattern. A very slight mechanical alignment error of the stepper causes this dimple. The drawing is somehow misleading, as the dimple actually also shows up at the edges of the chip. There, however, it is of little concern, because the layout patterns at the chips edges are in the range of tens of m whereas the stepper mis-alignment is in the range below 0.1m. The CH3-74

Figure 3-4 : multiple-reticule and multiple-module approach

Chapter 3: Photolithography (II) stitching

CH3 - 75

misalignment can cause problems for the sub-micrometer patterns of the chip circuitry this topic is discussed in full in 3.2. The use of mask v4 needs some explanation. With mask v4, the chip pattern is obtained trough successive use of only portions of the reticule. These portions are called modules. Modules contain unique pieces of the final chip pattern. The field of a module is necessarily smaller than the field of a reticule. In the stitching process, modules are successively printed next to one another. This way, repeated prints of the same module results in a larger repetitive pattern. With a sufficiently large number of prints, the size of the resulting chip pattern grows larger than the field size of a reticule. The process is depicted in figure 3-5 and is explained below.
desired chip pattern
Blades reduce the field to the field of MOD1 4x2 cm2 chip edge The chip pattern is obtained through a series of prints

print1, mod1

print2, mod1

print3, mod2

print4,mod2

print6,mod2

ok!

Mask v4 is organized around two so-called modules, MOD1 and MOD2. So-called blades hide large portions of a reticule. Blading allows to project a single module. The blade positioning is rather coarse it is of the order of tens of m. Because of the limited blade positioning accuracy, a safely wide buffer zone between modules is to be foreseen. CH3-75

Figure 3-5 : multiple-module sequence

CH3 - 76

Chapter 3: Photolithography (II) stitching

While blades block the projection of MOD2 patterns, module MOD1 is projected at both ends to make sure the stripe pattern does not cross the edges of the chip. Once this is done, the blades are moved to block projection of the MOD1 pattern and to allow projection of the MOD2 patterns. MOD2 is repeatedly stitched to itself and in between the MOD1 prints until the final stripe pattern is obtained. Changing the position of the blades takes time; therefore, once a module is selected, it is printed for all the chips on the wafer. Because the positioning error is random, the printed patterns will slightly differ from chip to chip. This time, through stitching of several modules, exposure of a chip larger than the (projected) reticule is possible with only one reticule. This way, the reticule cost is minimized, although the chips are larger than the reticule field. Of course, this only makes sense for chips with sufficient repetition in the layout patterns. The layout patterns split into module patterns with a lower degree of repetition. MOD1 is a so-called border module; it is required to obtain correct patterns at the chip edge. Border modules function as a stop for repetitive patterns. Without border modules, wrong patterns result like the ones obtained with mask1(a), 1(b), 2 or 3 alone (see figure 3-4 above). For correct patterns at the edges of the chip, border modules are a must. Actually, for a correct finish of any array, border modules are a must. With the mix&match approach, the 1X mask functions as a single border module that prints all the chips edge patterns at once. Several dimples result from the repeated stepper positioning error. Both the multiple-reticule as the multiple-module approaches are subject to this. This is also true for the mix&match approach: the 1X align precision (~0.5m) is much worse than the 5X align precision. As a result, the 5X modules must bear wider structures along their edges, so that 1X alignment errors are compensated for. Also, overlap of 1X patterns inwards a 5X region must be compensated for. An example of this can be seen from figure 2-27 and also from figure 3-6:
upper AM border module
5X 1X overlap edge

lower AM border module

5X-5X overlap edge

Figure 3-6 : correction for 1X-5X and 5X-5X positioning errors; the two modules serve as link between the 5X active matrix and the 1X periphery The next two paragraphs describes some experiences gained with the mix&match approach and give examples of so-called mask-module-information tables. These tables help organizing and maintaining the (1X/)5X reticule set. They provide key data to steer the stitching process. CH3-76

Chapter 3: Photolithography (II) stitching

CH3 - 77

3.1.2 Multiple-module reticule sets: MMI tables, multiple display formats

Where the multiple-reticule approach is stuck in rather high costs it has the advantage of a relatively simple stepper job: each reticule is stepped as often as there are chips on the wafer. The approach with multiple modules greatly reduces the cost of the reticule set. The price to pay however is the complexity of the stepper-job. A supplementary advantage of working with multiple-modules is that different array dimensions can be realized with the same reticule set. In the micro-display backplane case, this means that several display formats can be printed with the same reticule set: several different chips share the same reticule set [33],[34], [ 6.1.3]! For prototyping, this means a further reduction in reticule cost per micro-display version. There is a stepper-job per display format. In the specific case of the MOSAREL design, there is even no other option but to still provide two reticules for the metal2 masks. This relates to the layout of the interconnection pads. Additionally, some modules effectively contain data on only a few layers. Thus, for some layers, there are fewer modules than for other layers. A complication to deal with is the fact that the array of printed modules does not cover the entire chip. In other words, some areas are not patterned at all. Because this includes the sawing lanes and because problems can be expected when dicing through metal layers, blading shots are used on some layers. These empty modules uniformly pattern large areas with a relatively low accuracy. In other words, on some layers, there are more modules than can be seen from the reticule layout! As a result, the stepper program must take the layer into account so that only the necessary modules are printed. Furthermore, any stepper program needs the following inputs: the module field/window size, the position of the module centre on the reticule, the module position(s) with respect to the chip, the module count per chip and the chip count per wafer. Paragraph 3.3 presents more info on how to easily collect all this crucial information. The setup of a mask-module-information table (MMI-table) makes the stepper-job somewhat easier to manage. The MMI-table gives an overview of the use of modules, on a layer-by-layer basis and for each chip variant. The Excel tables below show the MMI tables for the MOSAREL test vehicle and demonstrator reticule sets. Question marks and dummy names have been used to avoid revealing too much classified information and still fully illustrating the complexity of the MOSAREL stepper-jobs. In 2.4.1.3 the mix&match of 1X and 5X lithography is reported. Such a mix&match further complicates the stepper-job, because two different litho engines must be made to work together. With the MOSAREL test vehicle, the stepper-job complexity due to mix&match is somewhat eased by two facts: the fact there is only one format for the test vehicle (GXGA format) and the fact the technology route is based on a pure 0.7 CMOS process (fewer masks). See also 2.3.1.2 and 2.3.1.3.

CH3-77

CH3 - 78

Chapter 3: Photolithography (II) stitching

Table 3-1: 1X&5X MMI table

CH3-78

Chapter 3: Photolithography (II) stitching

CH3 - 79

Table 3-2 : MOSAREL chipset MMI table

CH3-79

CH3 - 80

Chapter 3: Photolithography (II) stitching

Although the MMI-table gives an excellent overview of what must happen during Si processing, a second set of data is required to steer the stepper. The term print-centre coordinate lists is a concise name for this second set of data. In some sense, someone has to do the bookkeeping of module position, sizes... This is a rather tedious little job; fortunately, a computer program can help the designer to secure this task [see the sample routines in 6.3, see also the stitching procedure in 3.3]. The inputs to this software routine are the module dimensions and a description of how the chip is divided in modules. The output consists of a list of print centre coordinates. This list is used as input both for the stepper program as for the automatic generation of the chip layout. The layout allows for a visual check of the print centre lists. See 3.3 for more information. The bookkeeping relies on a convention about the location of the reference coordinate (0.0, 0.0). E.g. the layouts of modules, chips, reticules and wafers are all centered on coordinate (0.0, 0.0) by convention. Cadence users understand this as all cell bboxes centers are at (0.0, 0.0).
reticules M1 M2 M3 R-I R-II M1 M4 M4 M2 M5 M5 chip M2 M5 M5 M3 M3 M6 M6 M6 M6

M4 M7 M12

M5 M8

M6 M9 M13

M9 M8 M13

M8 M8 M7 M13 M11

With the MOSAREL project an empty module is needed to remove all conductors in the sawing lane regions. Because relatively large tolerances on the sizes of these empty modules can be accepted, no high-precision reticule is needed. In fact, no reticule is needed at all. In other words, there are more modules than can be seen from the reticule layout(s), because these empty modules are obtained through bladings-without-reticule. These are marked as B[?] in tables 3-1 and 3-2. The stepper program must take into account: - the layer to be processed - the positions of the chips on the wafer - the module-array dimensions and position(s) with respect to the chip centre - the module window size and centre position on the reticule (for blade positioning), - the reticules to be used (e.g. in the cases more than one reticule is used for a layer, or when blading is necessary) CH3-80

Figure 3-7 : modular stitching principle

Chapter 3: Photolithography (II) stitching

CH3 - 81

3.1.3 (Stitching)2 or the 1X-5X mix&match

In this paragraph, the problems encountered with mixing 1X-lithography and 5X stitching are reported. The issues are the design rules, a processing hurdle (stringers) and the homogeneity of chip planarity. As a conclusion, this approach is difficult to recommend, because of the non-standard 1X-5X litho mix. Design rule mix: Both the positioning accuracy and the minimum feature size are quite different when comparing 1X to 5X lithography. Some modules indeed form the boundary between the 1X and 5X regions of the chip. The 5X border modules must take into account the more relaxed 1X align accuracy. The stitching overlap area is much larger than with 5X-5X stitching. Fig 3-6 shows a picture of this. Also the alignment precision of e.g. the 1X via to the 1X metal1 patterns is more relaxed of course see also figure 3-8. In a first design, only the difference in feature size was taken into account. As a result the alignment of the 1X via mask to the metal1 layer proved to be quite difficult. An improved version of the 1X mask-set had to be designed to address this aspect. It is rather evident that the mask layouts need different design rule check files. Because this methodology is not standard, the designer must carefully use sets of DRC files. The optical microscope picture below shows a portion of resulting 1X patterns (substrate metal1[vertical lines] via1 metal2[horizontal lines] ).

Note the relative size of the dust particle in the lower left corner or is it a large short? Some contacts appear to be missing e.g. ellipse at the top) and note how close the contacts are to the metal2 edge every two rows. The stringer problem: The major problem encountered with the mix&match approach in the MOSAREL project is the occurrence of stringers. The mix&match is used for the metal1-via1CH3-81

Figure 3-8: microscope picture of 1X patterns

CH3 - 82

Chapter 3: Photolithography (II) stitching

metal2 layers. In a first approach, the etching of the metal layers is done in two lithography sequences. First comes the etching of the 5X areas and second, the etching of the 1X areas. Even with double-checked, correct masks layouts, massive shorts show up with this approach. All connectivity towards the pixel matrix is established after the processing of the metal2 layer. About every experimental trick was used on wafers processed up to metal2: repair of the chip by blowing up shorts, forcing relatively high currents to find hot-spots, making a statistical analysis of the shorts (resistance, characteristic, distribution). The large number of connection pads (128 flex connection pads) did not make this job easier But at first, whatever trick used, no conclusions could be drawn from the numerous, timeconsuming experiments. Optical microscope images finally dragged suspicions to the 1X-5X stitching zone:

weird lines

SEM picture: idem

Figure 3-9 : a first optical indication, confirmed with SEM, top view
1X area
Al wire A thin upright wall running from one metal wire to the other: a stringer!

stitching overlap area

[both 1X and 5X area]

Via to M1

1 m

Even more detailed SEM pictures reveal the presence of stringers shorting all wires that cross the 1X->5X stitching overlap area. Stringers are extraordinary, thin CH3-82

Figure 3-10 : the culprit pictured with SEM, 3D view

Chapter 3: Photolithography (II) stitching

CH3 - 83

wall-like residues that are extremely resistant to the etching process. They are formed after the 5X etch. The fact that the 1X area totally encloses the 5X area somehow guarantees that all connections to the test-vehicle matrix are shorted together on both the metal1 and the metal2 layers. The lack of control over the stringer width, height, and resistance probably best explain the unpredictable nature of the shorts.

SEM picture of similar area: the left (right) arrow points at an m1 (m2) stringer.

Figure 3-11 : stringers along the edges of the stitching overlap area
stitching area ok:

5X area

1X area

perfect patterning

Figure 3-12 : corrected processing for a perfect result Planarity issue: In paragraph 2.2.1.3 the importance of a homogenous fill factor is described; it is linked to the control over the LC thickness a crucial parameter. LC thickness is of major concern in the AM area of the chip. In the beginning of the MOSAREL project, the AM and close surroundings is the only area that is looked after in this respect. The quality of the chemical mechanical polishing (CMP) process step is problematic because of too large topography variations just beyond this area. Severe yield problems are the logical consequence of this. Figure 3-13 shows optical microscope pictures (polarized light condition) of the upper-right AM corner of two different chips. To counter this, the following rule applies: the fill factor of all conductor layers must be homogeneous all over the chip area. In principle, it is insufficient to homogenize the fill factor of the modules alone, because the chip size can be larger than the area covered with module patterns. With the MOSAREL mix of 1X and 5X lithography, this problem shows up: the areas outside the AM are patterned for metal1, via1 and CH3-83

CH3 - 84

Chapter 3: Photolithography (II) stitching

metal2 only. There is no control over the fill factor of the poly layer(s). The same holds too for the metal3 and mirror layers. The last two however have very high fill factors within the modules, so that the absence of patterns does not cause large variations in fill factor. As a result non-patterned areas cannot be tolerated dummy patterns are necessary on all conductor layers (polys and metals).
video drivers video drivers

! problem !

! ok ! row drivers

AM

AM

row drivers

Now it is time to look at the implications for the design of stitched chips. From the above considerations, the multiple-module reticule sets seem to offer the best approach in terms of cost. Additional data are necessary to steer the stitching process, and one must take special care of the design rules used to certify the chip layout.

Figure 3-13: chips with (left) and without (right) planarity problems

CH3-84

Chapter 3: Photolithography (II) stitching

CH3 - 85

3.2 Implications for IC design

In this paragraph implications from stitching on the IC design are reported. Four items affect the design of the micro-display backplane chip, particularly at the layout stage. As a continuation of 3.1, 3.2.1 first discusses the anatomy of a module with some more precision. This inevitably leads to the issues of double exposure and stitching overlap. Patents by Tower Semiconductor [31],[32] covers this subject; apparently, they were applied for by the end of the MOSAREL project as they simultaneously worked out a solution for ultra-large resolution camera chips. The patent protects a set of design rules which provide a safeguard against the effects of alignment precision and double exposures. It is not always necessary to stick to these rules for a successful design. Two ways of creating a design are looked at in a third paragraph. The reason why stitching can affect the schematic structure is explained. Finally, mask generation and mask verification are discussed.

3.2.1 Anatomy of a module

The approach with multiple-module reticule sets implies that arrays of basic building blocks partition the chip for all layers. The building blocks are called modules. The reticule sets discussed here thus consist of a collection of such modules. Figure 3-7 shows how a double reticule set is organized and how a related chip layout might look like. The chip in this figure is assembled with modules from two distinct reticule sets. A chip does not necessarily use all of the modules available from the reticule set(s). But what does a module look like? Figure 3-14 below looks at the internals of a module and defines a few names.
1 2 1 6 2 5 5 5 4

5 6

}
4

1 : module interior (4) 2 : ideal stitching line

Legend 3 : stitching area 5 : runs across modules 4 : inter-module area 6 : completely inside

Figure 3-14 : anatomy of a module's layout

CH3-85

CH3 - 86

Chapter 3: Photolithography (II) stitching

It shows a fictitious drawing and a picture of the final m1 mask layout in a lowerright corner of a module from the MOSAREL project. For clarity, only one LF layer is represented. One can tell the layout is for a LF layer as a polygon occupies the inter-module area (marked 4 in figure 3-14): with stitching no area outside the active module must be printed and with LF masks, only the polygon enclosures remain non-exposed. The inter-module areas are filled polygons for LF masks or are left empty for DF masks. Note that whatever the resist sign, the inter-module areas must block the energy from the lamp (see also 2.3.2)! The dash-dot line marks the ideal bounding box (marked 2) of the module (marked 1). Normally, a bounding box (bbox) encloses all features/patterns; here the term ideal bounding box refers to the situation where infinite align precision is available. In reality, such situation will never occur. Further on in the text, the edge of the ideal bounding box will be referred to as the stitching line/edge. The real bbox must be larger because of the finite align precision. Thus, some module patterns spill out of the ideal bounding box (bbox), because of the finite (non-ideal) alignment precision. An overlap between stitched modules is absolutely necessary. The drawing in 3-15 clearly shows what can happen when there is no such spill-out
ideal alignment module_1 edge

module_1

stitching line

ideal alignment ideal alignment

module_2 edge

module_2 module_1

!ok - ideal

! shorted tracks !

module_2 edge

module_2

!bad - realistic alignment error

The amount of overlap needed is mainly defined by the stepper alignment tolerance. Typical tolerance values are below 100nm and therefore, as a rule of thumb, values above 0.2m are enough. An overlap of 0.5m is taken as example for the rest of the paragraph. The module bbox equals the ideal bbox plus the overlap. Patterns extending further than these 0.5m outside the ideal bbox, all are blanked. The layout picture in Figure 3-14 shows how patterns (marked 5) extend further into the inter-module area (marked 4). Because two complementary fill-stipples are used for the inter-module polygons and intra-module polygons, these additional extensions naturally show up as 100% saturated areas. This trick is useful to easily spot where connections cross the stitching lines.

Figure 3-15 : ideal bounding boxes vs. realistic alignment

CH3-86

Chapter 3: Photolithography (II) stitching

CH3 - 87

On the other side, a perfectly aligned adjacent module will also overlap for 0.5m inwards into the module. As a result, between two adjacent modules there is a strip, 2x0.5m=1m wide for ideal alignment, where double exposures occur. In the case the adjacent module is not perfectly aligned, the width of the double exposure strip can either grow or shrink. The worst case situation is with maximum width of the double exposure strip: when stitching side-effects affect the largest possible area. This strip is called the stitching overlap area or SOA. The portion of the SOA inside the module bbox is called module stitching area (marked 3, abbreviated as MSA). Unaffected patterns (6) do not overlap with the MSA. By definition, the MSA cannot overlap with the inter-module area.

3.2.2 Double exposures

Adjacent prints of modules must overlap to accommodate for the stepper alignment tolerance; this results in the SOA. The overlap causes some resist material to be exposed twice. Small patterns that lie inside or that cross the stitching area can be badly affected by this. E.g. contacts inside the SOA can roughly double in size. Another example is with a pixel matrix where the minimum separation between the reflective electrodes falls inside the SOA. Figure 3-16 and Figure 3-17 below illustrate this; module_1 is being stitched to itself. The edges of exposed resist enclose areas that received at least the threshold energy. The threshold energy is the amount of energy needed for the photo-resist to develop away. A second exposure pushes the edges of exposed resist into areas that first did not reach the threshold: pattern widths change.
shot_n stitching line module_1 photo-resist > threshold energy wafer reticule blades module_2 module_0 photo-resist > threshold energy wafer shot_n+1 stitching line module_1

Figure 3-17 shows a module being stitched to itself: the previously printed pattern is shown in light gray (shot_n), the position of the mask in dark gray (shot_n+1). Figure 3-18 below shows a real mask layout and the corresponding microscope picture of a developed layer of photo-resist. It represents a poly masking level somewhere on a chip. The vertical spacing on the right-hand side is noticeably wider than the left-hand side. One might argue it could be due to the microscope lamp, CH3-87

Figure 3-16 : cross section view of an area exposed two times

CH3 - 88

Chapter 3: Photolithography (II) stitching

because the right-hand side of the picture is brighter. Look at the horizontal spacing at the top of the picture; they look similar both on the right-hand and the left-hand side!

shot _ n

shot _ n+1
pattern pitch shot pitch module bbox/window

stitching overlap causes corresponding changes of the width of some patterns

Figure 3-17 : top view of an area exposed two times

(layout drawing)

1m

1.5m 30m

(optical microscope)

Figure 3-18 : a microscope picture of developed photo-resist shows the clearly noticeable effect of double exposures

CH3-88

absorbs 2x energy ~1.3m

Chapter 3: Photolithography (II) stitching

CH3 - 89

Double exposure is an issue for the finest patterns only. The finest patterns indeed have dimensions that are only a few multiples of the distance affected by double exposures. Some common pattern structures require attention: the contact/via holes, the pixel electrode spacing and minimum width tracks. As a rule of thumb, the widening by double exposure can be estimated to be within the range of the thickness of the photo-resist layer (~0.5m). Because contacts/vias are so tiny (typically 0.4x0.4m2), problems can be expected when they are inside the SOA. The dimensions in the microscope picture in Figure 3-18 are obtained by comparison with the layout. Apparently, the double exposure makes the right-hand spacing widen by an estimated 0.15m in all directions. This number corresponds with typical minimum metal-contact overlap rules! As a result, contacts/vias that are exposed twice, can fall beside the underlying (stopping) layer. This can have dramatic consequences. The only way to avoid double exposure of contacts, is to make sure only one module prints a given contact/via, at once and only once. E.g. with a module that is stitched to itself, eventual contacts/vias inside the SOA are only permitted in one corner of the SOA and in the two adjacent sides of the SOA. They cannot be allowed along the other edges of the module. Minimum spacing between pixel electrodes can be controlled the same way. The major reason for the spacings to be minimal at all times relates to the minimization of the photo-currents in the active matrix. It is also best to protect fine wires that run across the SOA. The way to do this is to oversize wire widths in the MSA to pre-compensate for the narrowing effect of double exposures and misalignments. One could state a general rule for all layers: LF polygons in the MSA need oversizing, while DF polygons need under-sizing. This is referred to as MSA precorrection. There is an important exception to this general rule. Indeed, note that with undersize operations, complete arrays of small patterns can disappear (see also 5.2). Examples of this are arrays of contacts and vias large under-sizing is not indicated for contact/via layers! The inter-module area does not serve the same purpose as a polygon from a modules inner layout. The function of the inter-module area does not permit any sizing for this reason! See 3.2.4 problems with maskgen. To summarize, note that half the SOA width is the sum of three numbers: the distance affected by double exposures, plus the maximum align tolerance plus the distance between the modules ideal and real bboxes. The MSA width equals the sum of the distance affected by double exposures, plus the maximum align tolerance plus double the distance between the modules ideal and real bboxes. The implications for IC design go somewhat further than this. Up to here, only the requirements at mask level have been discussed. The following paragraphs present a few issues on why/how to split the entire design into modules essential for stitching and a few more issues on how to verify the design of the reticule set. Verification is an operation that is really necessary for any chip design.

CH3-89

CH3 - 90

Chapter 3: Photolithography (II) stitching

3.2.3 Split the schematics as well?

The fact that the chip layout must be split in modules is not a sufficient reason to have a corresponding schematic. Ideally, one could think of a software routine that automatically slices through the layout to create the modules, without taking into account any intelligence from the schematic. A concordant split of the schematic hierarchy costs some time; this looks like a reason to keep the schematic structure as it is. In other words, the chip layout alone is partitioned in modules at the end of the design cycle, just before mask fracturing (refer to 2.1.3). Tower Semiconductor even offers such a service to its customers. Foundries like AMIS and X-FABe.g., do not. Three points in particular merit our attention. First, with this method, there are fewer problems with mask generation. Second, the partitioning in modules of the final layout can cost quite some computer resources. It means flattening of the entire design. Usually, the computer available lacks sufficient resources for this kind of job... Third, in the event layout corrections are necessary, there are good chances that the partitioning must be completely re-done. Furthermore, the design hierarchy is lost at layout level; this does not help, neither for the ease of verification tasks nor for saving computer resources. Consequently, there are a few difficulties deserving solutions. The solution proposed consists of planning the module content and size at schematic level already. To some extent, the designer first draws a reticule floorplan by estimating module dimensions. The schematic hierarchy is then arranged in concordance. One advantage is that hierarchy is maintained, even with the reticule layout. Also, several designers can design modules simultaneously. Design verification is possible at module level as well. This can shorten verification cycles significantly. Eventual design upgrades can use existing module layouts; one can realistically envisage libraries of module cells. For these reasons, this approach was used with the MOSAREL project.

3.2.4 Overcoming problems with mask generation

Two problems show up with automatic mask generation routines. A first problem is related to the layout of the inter-module areas; actually, mask generation routines are not valid for them, they are only valid for the intra-module areas. The second problem is with some mask generation routines that generate patterns up to a few m away from existing patterns. As a result, some generated patterns can fall outside a modules bbox. Mask inversion, mask bias and inter-module areas The layout of the inter-module area is a source of concern. To understand this, consider the following classic mask inversion operation: pwell=not(nwell). Indeed the pwell mask very often is the complement of the nwell. Both however are DF layers. E.g. for the nwell reticule layout, the inter-module area must be empty. So, the same holds for the pwell reticule layout. The empty inter-module area of the nwell layer must not be inverted to get the layout for the pwell layer. Therefore, CH3-90

Chapter 3: Photolithography (II) stitching

CH3 - 91

there are cases where the inter-module area layout cannot be obtained with a simple mask inversion command. Here, it seems that the inversion operation is valid only inside the module bboxes. There is a second source of concern regarding the layout of the inter-module areas. The standard mask making process performs mask biasing operations just before the mask fracturing (e.g. conversion to MEBES data format). When the bias values are known, pre-correction of the inter-module fill is possible and mandatory. Otherwise, the dimensions of the MSAs and SOAs change! The alternative is to steer the mask making process in a way that the inter-module fill is done all at the end after the mask bias has been applied. The inter-module fill blanks all data outside the module bbox. This seems the easiest approach; it requires cooperation with the foundry or mask house though. A potentially very insidious situation is when the MSA pre-correction and mask bias size operations add up destructively. Alone none of the two harms, but the combined under-sizing/over-sizing can wipe out fine pattern features. Extreme care is recommended! Eventually a few more structures must be banned from overlap with the MSA. E.g. see the example about contacts/vias inside the MSA in paragraph 3.2.2. Micro-summary: classic mask inversion and mask bias commands are not compatible with the layout of the inter-module areas. In other words, they are not applicable as is on complete reticule layouts. the mask bias value can be equal to or an order of magnitude larger than the stepper alignment tolerance; the MSAs thus must be carefully checked for. A solution is proposed: the inter-module layout is to be finalized just before the mask fracturing step. Generated patterns that fall outside the module bbox From the previous paragraph it is clear that mask generation procedures (maskgen in short) cannot be used to obtain the layout of the inter-module areas. As a result, the mask generation procedures can only apply to the intra-module areas any patterns falling outside the module bbox must be cleared. This means some of the module patterns are wiped out that otherwise would and should interact with patterns from an adjacent module! Clearly, there is a problem: patterns inside the inter-module areas are prohibited, the design/process rule based maskgen and the module layout forces the presence of patterns in the inter-module areas. The solution is to clear the inter-module areas just before mask fracturing; this is compatible with the solution to the problem of mask bias on inter-module layout. For a better understanding of the problem, look e.g. at the generation of the p-field implant mask from nwell mask data. Typically, no p-field implant is allowed in nwell areas. For high/medium voltage wells, it is often required to keep the p-field implant at a certain distance (e.g. 10m). This is essential to safeguard voltagebreakdown thresholds. Imagine e.g. an nwell at 5m inside distance from a modules bbox edge. The p-implant must be prohibited up to 5m outside the bbox. CH3-91

CH3 - 92

Chapter 3: Photolithography (II) stitching

This implies no p-implant allowed over 6m1 inside the modules that will stitched along that edge. Either one has to include a portion of the generated patterns in the adjacent module, or one includes the generating polygons in the adjacent modules layout. This way, the mask generation procedure can be run on the module layout alone, hereby taking the information needed from other modules into account to produce the right patterns! Fortunately, there are but a few layers for which the maskgen procedure generates patterns out of the module bbox. This means that a clever layout avoids most of the trouble. Micro-summary: Add a minimum set of artificial patterns in the inter-module area to include patterns generated by data in other modules. Avoid high voltage wells close to the stitching edge, whenever possible. This reduces the number of artificial patterns to add. Restrictions are unavoidable when more than one module is stitched along a modules edge : sometimes the conflict cannot be solved, and patterns must be included in the inter-module area The proposed solution is compatible with the previous solution for the mask bias and mask inversion problems. The solution to the problems with mask generation and mask bias implies that the mask layout process breaks up into several stages: the reticule layouts with and without mask generation must be stored in different files. This is because of the artificial data in the inter-module areas that are wiped out before the mask fracturing process. The artificial data are needed for correct operation of the mask generation routines. Finally, only the generated data that fall inside module bboxes are valid. These valid generated data are the last addition to the module layout before the reticule verification jobs can start.

3.2.5 Overcoming problems with mask verification

The logical step after mask generation is mask verification. Reticules designed for stitching have the particularity that there is no matching schematic. Therefore a LVS check at reticule level is not possible. Because of the partitioning into modules of both the layout and schematic, LVS is possible at module level. Nevertheless, LVS-error free module layouts alone do not guarantee chip functionality. And what about DRC? The most basic DRC rules relate to minimum widths of and spacings between patterns on the same layer. One can easily guess what the problem is: when two modules are stitched next to each other, there is no guarantee that the resulting pattern satisfies even the most basic DRC rules. A good hands-on approach is to apply the following rule of thumb: the minimum separation of a polygon edge
1

5m+1m because of the stitching overlap; the stitching line is not in the middle between the edges of the nwell and the p-field implant (to be generated). CH3-92

Chapter 3: Photolithography (II) stitching

CH3 - 93

to the stitching line should be half the minimum width. Very smart mask designers probably are capable of taking slightly more complex rules also into account like e.g. minimum nplus to nwell edge separation, etc However, here also there is no guarantee that the stitched chip layout is ok. The necessity for verification of stitched chip patterns led to the development of the virtual stitching procedure: paragraph 3.3 describes this practical tool.

CH3-93

CH3 - 94

Chapter 3: Photolithography (II) stitching

3.3 Virtual stitching procedure

The description splits into three topics. The first topic describes how to organize the mask layout. Second, an automated layout correction procedure for the stitching overlap areas (SOAs) is detailed. At last, a tool is described that predicts the patterns resulting from stitching. The layout is organized to allow for relatively short computing times. This is important for the virtual stitching procedure to be quick and efficient. There is indeed a straightforward way to check the stitched patterns for errors: just look at wafers with developed photo-resist. The sequence of actions is quite simple: first, order the mask set at the mask shop. Next, print the patterns on a wafer covered with photo-resist. Then develop the photo-resist. At last, look at the photo-resist patterns (with a microscope) and more specifically check the SOAs for mistakes. The advantage is that the stepper program is checked for as well. The disadvantages are the masks cost, the long correction cycle and the rather special processing. There is indeed a need for only one wafer per unique layer pattern and there is no real processing of the wafers. The mask making steps and the stitching of photoresist patterns must be repeated in the eventual case of a mistake. This eventually takes more time and money than with the virtual stitching procedure. The virtual stitching procedure is the method I developed to perform standard mask verification (DRC, ERC, LVS, and PRE/LPE) on the expected results of the stitching process.

3.3.1 Module layout organization for virtual stitching

A fixed cell naming convention is recommended to avoid confusion for the following. A suggestion is to have module cell names starting with M_; to have the second field pointing at the cell purpose, e.g. prep_; the last suggestion is to have a meaningful name as third field, e.g. RDblock for row-driver block. The layout of a module is organized around two major cells. One cell is the preparation cell (the M_prep_ cell), the other is the final cell used for tape out and for the virtual stitching procedure (the M_4TO_ cell). Both share one cell with the main data, say, the data in the intra-module area (the M_mainData_cell). No data in this cell fall outside the intra-module area. It contains e.g. hierarchical data corresponding with a portion of the circuit schematic. The M_prep_ cell contains two more sub-cells: M_prep_MSA_ and M_prepMGen_. M_prep_MSA_ contains MSA data prior to pre-correction; this cell is kept separately in the case the pre-correction routine is to be re-run after correction of a bug. The output of the pre-correction routine is stored in the M_finalMSA_ cell. M_prepMGen_ contains the artificial polygons needed to have the mask generation routine work properly. The output splits in data added to M_finalMSA_ and intra-module data stored in M_final_MGen_. MSA precorrection routines run after maskgen. The tape out cell M_4TO_ additionally contains M_finalMSA_ and M_finalMGen_. In short: M_prep_ M_mainData_, M_prepMSA_, M_prepMGen_ M_4TO_ M_mainData_, M_finalMSA_, M_finalMGen_ CH3-94

Chapter 3: Photolithography (II) stitching

CH3 - 95

3.3.2 Computing stitched patterns

With all the above correctly implemented, it now is very easy to compute the patterns resulting from stitching. The aim is to emulate the stitching of a single chip. The computation only takes the MSA into account; the main module data are added at the end. Three keys open all doors to the procedure: 1. Instead of stitching module by module, the software can stitch groups of non-overlapping modules. These are referred to as independent-stitch layouts (ISLs). Typically 4 ISLs suffice to simulate the stitching of all modules. Actually, the module inner contents are left out of the ISLs to save computing time. Only the M_final_MSA_ subcells are taken into account. The layout of these ISLs can be easily obtained with a slightly modified version of the chip layout generator program: worst-case stepper alignments can be emulated. The ISLs must be orthogonal in that no module-instance must present in more than one ISL. The areas in between modules of an ISL must be empty for DF layers and solid-filled for LF layers. There is a link with the discussion on inter-module areas
M1 M2
empty DF, solid LF

M3

M2

M3

empty DF, solid LF

M5

M6

M5

M6

M13

M11

M13

M4

M5
empty DF, solid LF

M6

M5

M6

empty DF, solid LF

M9

M8

M7

M8

M8

2. 3.

The ISL combine into the emulated stitching pattern. Combine means that LF layers are AND-ed and that DF layers are OR-ed. The DRACULA software runs in 4 loops; where the output of one loop serves as input to the next loop. This has to do with the fact that the software reads only one GDSII file at a time. A second input relies on the import feature see 5.2. The output after four loops is then stuffed with the module inner contents to yield the final chip layout. This stuffing can again be done with the chip layout generator program, this time only to

Figure 3-19 : four ISLs for the chip layout of Figure 3-7

CH3-95

CH3 - 96

Chapter 3: Photolithography (II) stitching

instantiate the modules inner contents the M_mainData and M_finalMGen_ cells. The diagrams below summarize the operation of the stitching procedure.
Chip/reticule dimension constraints stitching reticule f.p. generator module partitioning chip fl.plan generator wafer fl.plan generator SOA precorrection module layout ISLs generator random align error chip-level checks module-level checks SOA pattern checks stitched chip layout calc. SOA patterns

Verifications reticule-level checks

Figure 3-20 : organigram of the virtual stitching procedure

Layout generators Reticule floorplan Die floorplan Wafer floorplan reticule centre module centre offset module centre die centre offset die centre wafer centre offset print-centre, -window tables

Figure 3-21 : stitching procedure output data for the stepper-job

CH3-96

Chapter 3: Photolithography (II) stitching

CH3 - 97

The same procedure can be used to emulate stepper positioning errors by adding random offsets in the print centre coordinate pairs. The patterns printed will indeed slightly differ from chip to chip. Note that it is now possible to perform the classic verifications (DRC, ERC, LVS, LPE/PRE) on the resulting patterns. At last, the layout generators are small pieces of code that produce print-window and print-centre coordinate lists. These numbers are combined to steer the stepper engine.

CH3-97

CH3 - 98

Chapter 3: Photolithography (II) stitching

CH3-98

Chapter 4: The pixel matrix I the basics

CH4 - 99

Chapter 4 : The pixel matrix(I) the basics

he focus in this fourth chapter is on a classic lcos circuit and on the resulting implementations. The heart of lcos devices is known as the pixel matrix; it designates the area where the electro-optic signal conversion occurs.

Paragraph 4.1 describes the classic analog active matrix (AM) architecture. The aim is to analyze the design of such AMs in detail. The description starts with a single pixel. A SPICE model is proposed to help with the simulation of the LC capacitor. The addressing of the AM with integrated drivers is looked at on the grounds of speed and redundancy considerations. A listing of important design constraints concludes the discussion on the classic AM architecture. Paragraph 4.2 summarizes the implementations, along with the results obtained. It is not the aim to give detailed reports of the optical performance, rather is this chapter limited to the chips functionality and peculiarities. As such, it concludes the discussion on micro-display designs that use the classic AM architecture. It also forms the starting ground for chapter five, where improvements to the pixel circuit are presented. As discussed in that chapter, these improvements effectively add a second frame-memory to the AM. In this chapter the implemented chips are described, together with a minimum of measurement results. To prove by measurements that all of the chips functionality is ok, is difficult, because of the extremely small size of the electrical charges to be measured in the matrix. As a consequence, the measurements were split in several parts. One part the part I performed tests the functionality of the driver blocks. The successful outcome of these measurements formed a milestone; it indeed had no sense to assemble non-functional microdisplay backplanes. The verification of the pixel matrix itself can hardly be done better than by measuring the optical output of the cell. On turn the optical measurements are complex, considering their dependence on the optical architecture used. The main contributors to the matrix evaluations are Geert Van Doorselaer (CTO Gemidis) for the system electronics and Dieter Cuypers who did the cell assembly and optical testing [60]. My work consisted of the definition of the circuit blocks, the spice simulations, the layouting (including the verification and layout simulations).

CH4-99

CH4 - 100

Chapter 4: The pixel matrix I the basics

4.1 The pixel matrix (I): the classic AM

his section describes the classic analog AM architecture. The aim is to analyze its design along with integrated drivers. The pixel capacitor is the starting point as it plays a central role. The next point discusses important electro-optical properties of the pixel. Driving signal characteristics are looked at for VAN-LC in particular. An initial model is proposed to allow for the inclusion of the electrical behavior of the LC capacitor into the spice simulations. After investigating a single pixel, it is natural to look at arrays of pixels. The bandwidth of an array of pixels and subsequent parallelism in the column driver is an important parameter. Next, some topics on circuit redundancy and -testing are discussed. The addressing of the AM is explained with a presentation of scanner circuits and frame speed requirements. A last paragraph systematically lists design constraints in three groups: optical, technological and electrical constraints. It gives an overview of the most important constraints that directly affect the design of a backplane.

4.1.1 One, single picture element (pixel)

The pixel capacitor is the starting point as it plays a central role. The roles played by the LC and the storage capacitor are discussed to come to a presentation of the classic AM pixel circuit. The discussion continues with the electro-optical properties of the pixel. The relation between optical output and electrical input is crucial for the design of all of the driving electronics, including the pixel circuit. Important driving signal characteristics are looked at for PDLC and VAN-LC in particular. Finally, an initial model is proposed for more accurate simulation of the electrical behavior of the LC capacitor. The pixel simulations of a-Si TFT pixels in [35] gives a good introduction to the issue. More details can be found in [36],[37], papers that present a well motivated and much more accurate model, that fits the lcos case particularly well.

4.1.1.1 The pixel LC capacitor

The term pixel originates from the contraction of the words picture and element; maybe this is the best explanation. Clearly, the pixel is at the heart of the operation of displays in general. Each pixel is capable of adjusting the brightness and color in a corresponding screen spot. Apart from this evidence, what is a pixel composed of in the case of the classic AM architecture? The most basic structure of a LC pixel is a capacitor: two electrodes produce an electrical force field inside a thin, insulating layer of liquid crystal (LC). With lcos, one electrode is provided by the silicon chip, the other is provided by the counterelectrode (CE) common to all pixels. By changing the strength of the force field the orientation of the LC molecules will change accordingly. On turn, light rays that travel through the LC layer see their path modified see chapter one for an elementary introduction. Two aspects deserve our attention here: we are dealing CH4-100

Chapter 4: The pixel matrix I the basics

CH4 - 101

with a peculiar LC material that somehow steers the path of light rays and we are dealing with an electrical component (the capacitor). The liquid crystal as dielectric Materials that display just some orientation order, but no position order are somewhere in between the solid and liquid phases. Therefore they are called liquid crystals. The orientation order is a fundamental characteristic of the liquid crystal phase. The chemicals concerned consist of elongated molecules; on average, each molecule naturally tries to orient itself in concordance with its neighbors (minimum energy state). By contrast, the liquid phase features no orientation or position order whatsoever, whilst the solid phase features both. The LC functions as a peculiar dielectric in the basic pixel structure. The labs implementations use a thin layer of vertically aligned nematic (VAN) LC. The optical response to a change in the electric field can be quite slow (several milliseconds), and also depends upon the strength of the electrical field. The response speed corresponds to the speed at which the average molecule orientation changes. The electro-optical response is often depicted as function of RMS voltage; see for example figure 4-1 below. It shows a typical optical output of a single LC cell as function of the RMS voltage applied. Note the EOresponse relates to the permittivity of the LC: both respond to changes of VRMS.
optical output (arbitrary units)

100 50 0

EOresponse
VRMS 1 3 5 (VLC)

Figure 4-1: the electro-optical response as function of RMS voltage A sample and hold capacitor Driver circuitry is needed to change the voltage across the LC capacitor, i.e. the charge the LC capacitor holds. The capacitor is capable of storing an electrical charge for some time, and thus capable of maintaining an electric field across the LC layer. On turn, this capacitor is capable of maintaining the optical response for a while. This amounts to a memory effect. Chapter five refers to it as the primary memory or as the pixel memory. Often, pixel circuits feature an additional storage capacitor that augments this pixel memory. One reason to do this, is that parasitic leakage currents are capable of significantly modifying the charge stored, hence of modifying the optical output. At CH4-101

CH4 - 102

Chapter 4: The pixel matrix I the basics

equal leakage currents, large storage capacitors slow down this undesirable effect, or even make it impossible to notice. A 10 pixel, with a 3 thick LC and a relative dielectric constant of 4 has a capacitance CLC=1.18 x 10-15 Farad. For a storage capacitor of 1 pF, a leakage current of 1pA results in a V/t of 1V/second. Fortunately, the LC memory is refreshed at least every 20ms (frames are updated at least 50 times per second). The numbers cited here, give the order of magnitude for real lcos pixel designs.

Typical specifications require an HR > 90%. For the example above, at 50Hz and for a Vpix of 5V, HR amounts to 99.6%. Note that a storage of 1pF is optimistic; on the other hand, the magnitude of the leakage current is hard to measure with very high precision, considering the temperature influence as well as the fact that the contribution of photo-currents is of unpredictable nature. Only with a sufficiently good HR, one can disconnect the driver circuit from the pixel capacitor temporarily to use it for driving other pixels. This approach is the basis of a class of pixel arrays known as active matrices, invented by Peter Brody [2]. Indeed this architecture foresees a transistor switch between the LC capacitor and the video data line. Because the switch is implemented with an active component (the transistor), the name active matrix was given. AM pixel circuit Figure 4-2 shows the circuit used in the so-called classic active matrix configuration. The matrix contains crossed busbars: the row select wires and the column data wires. Because of the similarity with a class of memory chips, this configuration is often referred to as the DRAM architecture. It has, however, not everything in common with memory chips: although the image data written to the active matrix need continuous refreshing, they are never retrieved though. The active matrix therefore is more something of a write-only-memory (WOM). It is up to the viewer to remember the (hopefully beautiful) images!

The voltage holding ratio (HR) of an lcos device is a measure for the accuracy with which image data are maintained in between two updates. It is a measure for the quality of the LC memory. By definition, HR == 100% x (1 - |V| / |Vpix| ) [in percent] with |V| == absolute voltage loss across the pixel capacitor, measured from the instant right after an update to just before the next update. Vpix == desired voltage across the pixel capacitor

CH4-102

Chapter 4: The pixel matrix I the basics

column (data) line row (select) line switch (transistor) pixel electrode storage capacitor CS constant voltage CE

CH4 - 103

LC capacitor CLC

The switch allows to modify the pixel electrode voltage, hence the state of the LC director for that pixel alone. All switches of a row are activated simultaneously with the row select line. All along a selected row, the pixel electrode voltages are made equal to the column voltage. The voltages of the columns overwrite the pixel voltages. The storage capacitor CS is needed to safeguard the voltage holding ratio. There are indeed various possible parasitic signal sources that can alter the electrode voltage in between updates, i.e. when the electrode is disconnected from the column. The storage capacitor is a preventive measure against them.
col row switch Cp2 pixel electrode Cp1 CS Il1 Il2 CE CLC Il3

Figure 4-2: the classic 'DRAM' pixel architecture

Cp1 Cp2 Il1 Il2 Il3

: column storage coupling : row storage coupling : Cs leakage current : terminal leakage : LC leakage

Figure 4-3: parasitic components in the DRAM pixel CH4-103

CH4 - 104

Chapter 4: The pixel matrix I the basics

The two groups of sources of parasitics are: coupling capacitors and leakage currents. Parasitic couplings come from spatial proximity of signal tracks in and around the pixel, e.g. from overlapping wires. Leakage currents alter the charge stored at the pixel electrode node, hence its voltage. They result from natural imperfections of dielectrics or from light induced photo-currents [38]. The relevant leakage currents show up at the switch terminal, through the insulator of the storage capacitor and through the LC layer. One cannot completely eliminate leakage sources; correct processing minimizes these. The leakage at the switch terminal is a junction or diode leakage. It is highly sensitive to photons and for this reason no light should reach the junction(s). A correctly designed light shield and panel package take care of this (see chapter two). Many of the parasitic couplings between signals can be turned into couplings with ground. This is done by systematically providing guard tracks between signal tracks on the same layer and between layers. A good pixel layout will minimize the overlap between the row and column electrodes; guarding with a metal layer is possible, however at the cost of either an extra metal layer or at the cost of a less optimal light shield (see e.g. the REFLEC pixel implementation in figures 4-10 and 4-11). For the dimensions of the two components (the switch and the storage), please see the discussion on transistor count per pixel in paragraph 5.1. In practice, the smallest dimensions possible are chosen for the switch, while the layout of the storage optimizes the usage of the remaining pixel area.

4.1.1.2 Electro-optical properties

The laboratory developed polymer dispersed (PDLC) and vertically aligned nematic (VAN) technology for the implementations, respectively a scattering effect and a polarization rotation effect. These two LCs are similar from the point of view of the required voltage range of the driving signals. The VAN mode yields the best optical performance, and therefore much of the text must be placed in the framework of VAN LC. In [12], the intrinsic contrast properties of VAN LCs are studied in detail and compared with those of other LC materials. These papers suggest that VAN LC material represents an excellent, maybe the best, choice for lcos micro-displays. This being said, much of the following relates to VAN LC; similar considerations however, are valid for other LC modes as well. Issues concerning the voltage range, the AC driving and LC response speed are detailed hereunder. Voltage range of driving signals The driving voltage range depends on the LC (effect). E.g. VAN and PDLC need medium voltage ranges (10Vpp is not unusual); others like ferro-electric LCs work with lower voltage levels (0V, 5V). Modern deep sub-micron ICs have limited voltage ranges between 1V and 2.5V (anno 2001); they offer incredible circuit densities (number of transistors per sq. micron). Older technologies rely on larger CH4-104

Chapter 4: The pixel matrix I the basics

CH4 - 105

transistor structures that withstand higher voltages (e.g. 15V for Alcatel Microelectronics 2.4 CMOS, 100V for AMIS I2T 0.7/DMOS). Many foundries do offer double gate processes or processes extended with high/medium voltage devices (e.g. DEMOS, BJT ) so that both low and high voltage range transistors are integrated on one chip. This inevitably comes at the cost of a few extra masks, and thus at the cost of extra processing time. Several circuit techniques exist to convert low-voltage (control) signals to higher voltage (level shifters). There will always be a need for this conversion, as the real world does not shrink while on contrary, transistor dimensions do. It is even more difficult to find foundries that offer dedicated lcos processes. I designed in a 2.4-15V and a 0.7/15V SDEMOS both from AMIS, Oudenaarde, Belgium and 0.35/18V technologies from UMC, Taiwan and SMIC, R.O.China. The reason for 18V and not just 10V lies with the so-called bulk-effect of MOS transistors (see figure 4-4 below). Figure 4-4 shows the evolution of the drain-source current in function of the gate-source voltage, at a fixed drain-source voltage and for the most extreme parameter variations (temperature, transistor models). As explained, a single transistor is used to switch the correct voltage onto a pixels electrode. The amplitude of the gate signal minus the threshold voltage determines the maximum voltage that the transistor can pass on. The threshold voltage depends on the voltage switched this is what the bulk effect is about. Computations and equations of this are found in basic textbooks on MOS circuits; the figure below shows a corner simulation of the bulk effect for a typical 3 technology. It allows to draw conclusions, whatever the IC process variations and within the temperature range of the device.
10-3

IDS (A)

10-6

0C 100C all simulation corners VDB = 18V

10-9

VGS (V)
0.0 3.0 6.0 9.0 12 15 18

Figure 4-4: bulk effect simulation for a 3 NMOS CH4-105

CH4 - 106

Chapter 4: The pixel matrix I the basics

No DC across the LC A DC signal across the LC cell destroys the LC material. Thus, the signal across the LC must equally alternate between negative and positive. Figure 4-5 shows a typical evolution of a pixel voltage. The pixel voltage is updated every new frame. The start of each frame is marked F[index] on the horizontal time axis; the vertical axis represents volts. The first set of traces represents the voltage on the CE and pixel electrode with respect to the lowest chip supply voltage. Note how the CE voltage is constant and at about the middle of the peak-to-peak range. It not necessarily exactly in the middle, as VCE is adjusted to cancel DC offsets. The DC offsets often result from workfunction differences at the interfaces of the LC with the Al mirror electrodes and with the ITO CE electrode. The second trace in figure 4-5 shows the net voltage across the liquid crystal: VLC=Vmirror - VCE Note the rectangular shape of Vmirror; it is very much a square wave with an amplitude that can change every half period. For square waves, the value of the amplitude corresponds to the RMS value. This is how figure 4-6 makes sense: the EO curve is represented symmetrically around the vertical axis (compare with figure 4-1). The LC voltage swaps polarity at the transition from frame0 to frame1. In this example, the amplitude of the voltage remains constant during two subsequent frames. Thus, the optical output remains unchanged. The amplitude changes at the start of frame 2 the arrows indicate the voltage sequence for a succession of five frames. The optical output changes accordingly after some time.
volts VLC(F0) constant VCE F1 F2 F3 F4 Vmirror

10V 5V 0V

F0 volts VLC(F0)

time

5V 0V -5V

LC voltage

VLC time

VLC(F1) F1 F2 F3 F4

F0

Figure 4-5: Vmirror with respect to chip ground and resulting VLC

CH4-106

Chapter 4: The pixel matrix I the basics

a.u.

CH4 - 107

100 50 0 VLC(F1)

EOresponse VLC

100 50 0

a.u.

Optical output response ideal response time

F0

F1

F2

VLC(F2) VLC(F4) VLC(F0) F3

VLC(F3)

F4

Figure 4-6: optical response to changes in VRMS LC response speed Ideally, the optical response should be immediate read: as fast as the voltage change across the LC. In practice, transitions between optical on and off states take several milliseconds. In some cases, the response speed interacts with the frame rate: the LC will act as an integrator i.e. it will average successive frames, when the LC response speed is slower than the frame rate. For a frame rate of 50Hz and response speeds of 20ms, the result is a display producing blurred video images. For still images this can be more or less acceptable, as the desired image data remain the same for a succession of several frames. For crisp video images, it is harder to motivate. There is already a blurring effect from the sample and hold effect each image stays constant for the complete duration of the frame. This visibility of this effect can be reduced by using higher frame rates, or by using pulsed light sources (cfr. the good old CRTs). In practice, LC response times of ~10ms are acceptable for frame rates of 60Hz. Frame sequential color architectures and flicker considerations push for higher frame rates (120Hz and way up). In these cases, faster LC response is necessary. There are two ways to obtain faster LC response. One way is called the voltage overdrive technique. The strength of the electrical field is exaggerated at the start of the refresh, as it has been shown to speed up the reorganization of the LC molecules [39]-[42]. The other way is to have new LC mixtures developed. Some approaches limit the LC states to two extreme states, because the transitions between these two states are the fastest. However, this assumes the digital driving of the CH4-107

CH4 - 108

Chapter 4: The pixel matrix I the basics

LC; this has quite some consequences for the overall electronic subsystem. It is not uncommon to see digital systems on display that suffer from contouring. This effect is seen on still images containing e.g. a background with fading colors, but also around a moving object, because of the difficulty to control the spacing between dark gray shades. In the frame of the discussion of the classic AM, digital driving will not be further expanded on. A spice model for the LC response is proposed in the next paragraph. It is a very useful tool, in particular in those cases where the storage capacitance gets about as large as the LC capacitance itself: the peculiar behavior of the LC capacitor can have a significant impact on the optical output. Spice model of the LC capacitor, of the LC response The reaction speed of LC crystals to changes in electrical stress (RMS voltage e.g.) can be characterized by 10%-90% rise/fall times. Remarkably, these response times depend on the start and final gray level. Transitions between middle gray levels are typically the slowest [37][39]. LCs cannot be characterized with a single rise/fall time pair therefore. The transition times are not necessarily symmetrical. The reaction speed depends on the states the LC transits from and to. Some report that transitions to one of the dark/bright states occur faster than to any other. This fact is exploited by pulse-width-modulation (PWM) schemes (see chapter five). From the point of view of the pixel circuit, transitions between gray levels correspond to changes in the capacitance of the LC capacitor. The LC capacitor is voltage-dependent. Like the optical response, the LC capacitor also is non-linear. To emulate the behavior of the LC capacitor, a first spice model was developed that uses ideal components and math functions to match gray level transitions with response times. A much better model with a formal physics background has been presented in [36],[37].

CH4-108

Chapter 4: The pixel matrix I the basics

CH4 - 109

4.1.2 Millions of pixels

Display formats The display (image) format defines the number of points a display (image respectively) is partitioned in. The table and picture below shows some of the most common standards in use today. Note that for the larger formats, no standards exist(ed) and each manufacturer had the freedom to define one. With defined standards, manufacturers can build sub-system components independent of each other. Of course, manufacturers will try to turn their format into a standard Note that display format can be described as a sort of maximum, intrinsic display content. E.g. addressing of only a sub-window of the display reduces the effective image content, but not the display format.
Format Nr. of rows Nr. of cols Pixel count
2048

VGA 480 640 307.2k

SVGA XGA SXGA QXGA 600 768 1024 1536 800 1024 1280 2048 480.0k 786.4k 1311k 3146k Table 4-1 : display formats

GXGA 2048 2560 5243k

gxga

1536 1200 1024 768 600 480 240 120 0 qvga 1024 1280 1600 vga svga xga-p xga sxga wxga

qxga uxga wsxga wuxga

row count

1920

2048

The term resolution is used to quantify the ability of a device/system to reproduce an image. The resolution of a micro-display can be defined as the number of pixels per unit of length. It is clear that the visual appreciation of resolution depends on the distance the display is looked at. In other words a caveat must be issued when display format is used as a display quality criterion. E.g. with lcos projectors, the micro-displays resolution is an upper limit, because both image processing and the optics can further deteriorate the quality of the perceived image. Again, the image CH4-109

Figure 4-7: image format comparison at constant pixel pitch

column count

2560

160

320

640 768 800

CH4 - 110

Chapter 4: The pixel matrix I the basics

can be captured with a lower resolution device... hereby limiting the resolution of any subsequent reproduction. Now that we know how many pixels we should be able to address, lets look at two peculiarities proper to AMs: polarity inversion schemes and AM bandwidth. Polarity inversion schemes A DC component causes irreversible degradation of the LC material. During one frame, the polarity of the signal VLC can be identical for all pixels (frame inversion scheme), can alternate between rows or columns (row, respectively column inversion scheme), or swaps every next pixel (pixel inversion scheme, following a checkerboard pattern). Each frame refresh is accompanied by a polarity inversion. Frame inversion is often recommended as it minimizes fringe fields, hence throughput losses. At modest frame rates, a viewer will perceive DC components as flicker the amplitude of the driving signal being asymmetrical for positive and negative frames. This leads to a noticeable and sometimes rather irritating variation of the light output. For frame inversion schemes, the flicker base frequency is half the frame frequency. Because the human eye cannot detect frequencies above 30~40Hz, some display manufacturers increase the frame frequency to 100Hz so that the flicker base frequency is well above the detection limit. I.e. the phenomenon cannot be detected by the eye. Cosmetically this is a good trick. Unfortunately, another effect resulting from the DC component mars the cosmetics: image sticking. Image sticking causes a previous image to remain visible during a few frames. Image sticking is much harder to hide. At last, even a small DC offset can considerably shorten the lifetime of the LC material a scary situation for consumer products. AM bandwidth column wire model A second property of AM is the maximum addressing speed. The pixel data get to a pixel via the column busbar. It is a wire that connects all pixel transistors on a single row. This wire thus is rather long, thin, and connects to about 1000 transistor terminals1. The model used to estimate the signal propagation delays along this wire is a chain of RC filters. One could model it with 1000 RC filters, where R is the resistance corresponding to a piece of wire as long as one pixel, and C the capacitance of it plus the capacitance of the pixel transistor terminal. Summing all Rs and Cs together yields a lumped model that speeds up the spice simulations. However, the lumped model lacks precision; i.e. the calculated delays are much more important than when a chain of RC filters is used in the model. A better model uses a chain of 10 RC filters, where R corresponds to the lumped resistance of a wire 100 pixels long, etc. Simulations indicate there is only a marginal difference with a chain of 100 RC
1

This is approximately the number of rows.

CH4-110

Chapter 4: The pixel matrix I the basics

CH4 - 111

filters. In [43] is shown how minimum width wires yield larger RC values than wires that are somewhat wider. The paper thus gives a rule of thumb for the layout of column busbars with minimal propagation delays: a width of 3m yields a lower RC product per pixel, and is small enough to fit inside the pixel layout.
10V 5V 0V F0
pixel (001,0001), F0 update; valid during 1 frame period

[volts]

F1

[time]
pixel (1,1), F1 update

pixel (768,1024), F0 update; valid during 1 frame period

Frame1 data for pixel (1,1) overlap with frame0 data for pixel (768,1024) this is exactly what should occur with progressive scanning

The column propagation delay is large enough to require parallelism for addressing of the complete matrix at frame durations of 10ms. Thus more than one column is addressed simultaneously; otherwise, the column voltage would not have enough time to settle to the desired value. A column addressing time window of 40ns is not unusual. Progressive scanning involves the sequenced addressing of e.g. one million pixels. The addressing of a complete AM would take 40ms. Clearly, 4-fold parallelism is needed to achieve the 100Hz frame rate. This also means that the updating is done sequentially: first the first (set of) pixel(s) on the first row, then the next (set), until all pixels of the first row have been updated. The same mechanism applied sequentially for the other rows as well. As a result, not all pixels are updated at the same instant (see figure 4-8).

Figure 4-8: frame updating mechanism

4.1.3 AM driver circuits

All-in-one chips? Until here, the text gave details of the heart of an lcos backplane: the active matrix. One of the big advantages that lcos technology offers is the possibility of highdensity, integrated drivers. The rules for panel assembly largely determine the size of the lcos backplane chip: the circuit density is so high, that besides the most essential matrix drivers, extra space is available to integrate even more of the system electronics. CH4-111

CH4 - 112

Chapter 4: The pixel matrix I the basics

Many think that the micro-display chip can integrate the entire electronic subsystem. A lot can be integrated onto the backplane chips; however, specific system functions are better implemented in a more adequate IC technology for cost reasons. In particular, the frame memory is best implemented in a dedicated, deep-submicron memory technology, rather than in the backplanes 0.35/3 technology. In other words, there will always be several chips; one should not expect all electronic functions to be stowed onto the micro-display chip. Commercial players very often offer chip-sets to accommodate for most of the system electronics. Furthermore, next to the essential functions, test circuitry can help with the identification of critical assembly steps. This can be quite useful to minimize the production cost, if the test circuitry allows for the elimination of unacceptable backplanes as early as possible in the panel assembly process. Of course, test circuitry also provides a means to monitor the process yield over a longer time. Scanner circuits It is time to expand somewhat on the most essential matrix driver circuits: the row and column drivers. The appendix gives extended examples. In essence, the column and row drivers are scanners; i.e. their combined operation systematically scans the matrix by addressing one pixel after the other. Within the frame time, all of the pixels need to be addressed once. The row scan frequency thus is the frame time divided by the number of rows, assuming there is no dead time, or frame porch time. In exactly the same way, the column scan frequency can be determined: row time divided by the number of columns addressed. Note that only one row can be addressed at a time, while several columns or all columns can be addressed simultaneously. This depends on the degree of parallelism that is built into the column driver circuit. Sometimes this is referred to as line-at-a-time addressing to make the difference with pixel-at-a-time addressing where just one column is addressed per write sequence. There is a lot of similarity between the column and row scanners. The circuits are very similar, for VAN cells both need medium voltage level shifters, the pitch constraint is similar, the loads are similar there are some important differences of course. Because of the parallelism in the column scanner, the low-voltage part of the column driver will often be less dense compared to the row driver. The scan frequency is higher for the column driver, and the load is actually composed of a bank of video switches. The row drivers load is composed of the row busbars and all of the pixel transistor gates on one row. See e.g. appendix 1, AX1.3, point d) Schematic of main circuit. The scanners are built using a shift register and level shifters. The length of the shift register relates to the number of rows or number of columns of course. Note that the clock signals that drive the shift registers are buffered using a symmetric buffer tree to avoid differences in signal timing, e.g. differences when writing to the upper side of the AM compared to when writing to the lower side. The level shifters are needed to provide the pixel electrodes with the necessary voltage levels. Note that some extra functionality is necessary: hold logic that force the driver outputs into a given state; bi-directional scanning, so that the image orientation does CH4-112

Chapter 4: The pixel matrix I the basics

CH4 - 113

not depend on the orientation of the display panel; mode bits for display format selection, etc. Until now, redundancy has not been discussed. Matrix defects are likely to occur considering the physical dimensions of the AM. There are very few defects that are tolerated for commercial displays. As a result, some might dream of pixel redundancy: the idea of pixel replacement does not work for display systems. It works for memory chips, it works for camera chips, because there is no need for an exact match between the physical structure of these arrays and the real world they are supposed to represent. A non-working pixel cannot be replaced by another. Thus, when one has a perfect AM, the drivers should be operational. And at the very start of this research, no information at all was available e.g. to estimate the chances of having a working chip. This is the reason why in the first stages of the research work, as much redundancy as possible was foreseen. As the group gained experience, redundancy lost some of its importance. Another good practice, is to try using somewhat stricter layout rules to lessen the defect probability wherever possible, and certainly for the design of the highly sensitive AM. A typical example is the minimum separation of wires: keeping long wires further apart from each other where possible seems a good idea. The 160x120 panels have double row and double column drivers; clearly, enable circuits are needed to enable the correctly working drivers and to disconnect the non-working ones from the AM. The five million pixels of the MOSAREL GXGA demonstrator definitely seem to be the most sensitive to defects (considering AM size). However, double column drivers are not possible here considering the interconnection method and timing issues (long connections to the other side of the chip for the most sensitive signals: the video signals). Column and row shift registers are composed of chains of blocks of 32 registers; each block is foreseen twice. As a result, there are two sets of register blocks, setA and setB. One can decide externally which one to use. At least that was the original plan; in practice, a missing buffer does not allow selecting any combination of register blocks, it only allows selecting one of the two sets. The tmdc XGA still had two row drivers; practice shows the second row driver has little impact. Therefore, the later versions bear no more redundancy.

4.1.4 VAN pixel design: constraint listing

Optical Diffraction Aperture

Less light loss with larger pixels; this requirement is opposite to cost considerations [11] Less light loss with larger pixels; this requirement is opposite to cost considerations [11]

CH4-113

CH4 - 114 Photoconduction

Chapter 4: The pixel matrix I the basics A light shield requires an extra layer in the IC stack. It is also recommendable to use minimum inter-electrode gaps to maximize the aperture, to limit the light beam to the sole AM area and to block light that reaches the back or edges of the chip! This has primarily consequences for the package design. Can sometimes be bent; this is at the designers risk and often need agreement from the foundry. See chapter 2; too stringent and these will greatly affect panel yield Minimal for frame inversion and with larger pixels; counterbalanced by cost considerations [11] and dependent upon the LC effect used [45] Depends on LC, for our VAN developments, 10Vpp signal ranges are more than sufficient Considering the body effect, the Vrange of the pixel transistor further expands to 15-18V Need for parallelism: depends on the drive scheme (analog/digital), on the color scheme (3-valve, 1valve) and on the AM bandwidth The use of simple guarding avoids cross-talk from one video signal to the next: take care of the layout of the analog switches and of the AM

Technological

IC design rules Assembly design rules

Electrical

Fringe losses

V-range AC-drive Frame speed

(Electrical)

Cross-talk

Table 4-2: constraint listing

Table 4-2 lists the design constraints specific to the design of pixels. In paragraph 4.1 some important issues regarding the design of classic AMs are presented. The next paragraph gives an overview of the three generations of micro-display chips I designed and that have been successfully assembled into display demonstrators by a very skilled team of specialists from the TFCGroup.

CH4-114

Chapter 4: The pixel matrix I the basics

CH4 - 115

4.2 Micro-display implementations

The section describes the implemented chips, along with the results obtained. Three generations of lcos displays are described. Each implements the classic DRAM described previously; the differences lie with the IC technology, display format, and the amount of circuit redundancy. Along with the research advances, more and more knowledge was gained, also regarding the system electronics and regarding the optical architecture. The first implementation is the result of a thorough, first shot. The second implementation features three different display formats, one of them resulting in the largest lcos known to date: the GXGA MOSAREL chip. The tmdc prototypes comprise the XGA and WXGA formats. They served the purpose of developing high volume lcos process chains.

4.2.1 REFLEC chips: 160x120 pixels

Trading standard cell design versus area cost The development of a digital column driver IC paved the way to the micro-display design activities in the research group. This design1 used standard cells in a 2.4m CMOS technology from Alcatel, Oudenaarde, Belgium (now AMIS). The most striking disadvantage of the standard cell approach is the size of the digital-to-analog converter (DAC). A test IC with four column outputs needs about 25 sq. mm. At that time, the automatic place and route tools available did not make use of the symmetry in the design schematics and had only two metal layers for routing manual layout of at least some of the sub-cells appeared to be a necessity. Finally, it was the pitch of the standard cells that determined the pitch of the output pads... which is far larger than any acceptable lcos pixel pitch. Another point is that the standard cells were general purpose cells not exactly optimized for lcos micro-display backplanes. Eventually, the column outputs can be rewired to match the smaller pixel pitch; the advantage of this idea is thwarted by the area cost of rewiring and by the area cost resulting from stacked driver cells. After a first, successful mini feasibility study [AX1.1], a project was born (REFLEC) to design a digital quarter-VGA micro-display chip in Alcatels 2.5um HBiMOS technology. One of the aims was to compare line-at-a-time and pixel-at-atime addressing schemes. The former updates the AM row by row, while the latter updates the AM pixel by pixel. Quite unfortunately, redesign resources only allowed for 1/16th VGA format. The experience with the first design showed an area need of pixelPitch x 1000m per bit. A decent 8-bit digital driver would need at least 8mm x chipWidth of silicon. This is the consequence of the conservative 2.5m feature size and limited routing density. The integration of 8-bit digital drivers with a lineat-a-time addressing scheme just is too costly considering the CMOS technology available.
1

J.-P. Voortman, 1994 CH4-115

CH4 - 116

Chapter 4: The pixel matrix I the basics

The goal narrowed therefore to the design of twin 1/16th VGA backplanes, without integrated digital line-at-a-time drivers, but each with opposite pixel transistor types. This trial stems from the difference in leakage currents between NMOS and PMOS transistors. The chips are baptized AMN and AMP respectively. Part of the project goal is to provide sufficient means for debugging, as well as means to optimize both the circuitry and the technology. This is anyway a necessity for any future display, because of the prototyping cost and because of the cycle time. Technology of the 160 x 120 backplanes The silicon backplane technology has a feature size of 2.5m, offers 2 metallization layers with 15V n-and p-MOSFETs. To comply with the optical requirements, this technology must be extended with a number of back-end processing steps. The TFCGroup developed this back-end entirely; it provides the necessary planarization, light blocking and reflective layers. The pixel pitch is 80m, partially because of the limited capabilities of the available cleanroom equipment. For a first design, it is rewarding to work with loose constraints to ensure successful back-end processing. A schematic vertical cross section of the pixel structure is shown in the figure below.
Reflec: technology cross section
Mirror (N,M)

Mirror (N-1,M)

Mirror (N+1,M)

ColumnM = vid.signal RowN-1 Gate D Pixeltransistor (N-1,M) OFF S Vpixel (N,M) Cstorage (N,M) D

ColumnM RowN Gate Pixeltransistor (n,M) ON S

Passivation Metal2 Via1 Metal1 Contact Poly

Substrate

The number of back-end layers is kept to a minimum for evident reasons of process yield and also because the cost of tuning these extra non-standard processing steps. E.g. one could think of adding a plug fill or extra metal layers for light shielding and/or interconnection. The performance of the back-end is limited from two viewpoints. First, there is the thick planarization and shielding layer causing large via sizes for the connection between the mirror electrodes and the pixel circuitry. Second, lithography limitation results in a lower limit on the achievable inter-pixel distance. This means that the geometric aperture is far from maximal, hereby limiting the optical throughput of the device. CH4-116

Figure 4-9: schematic 'reflec' backplane cross section

Chapter 4: The pixel matrix I the basics

CH4 - 117

In addition to the DARC absorbing polyimide layer, the metal layers are connected to one another to form some kind of a labyrinth. Each micro-display chip has double column and row drivers. The aim is to provide driver redundancy, and therefore each driver can be independently activated. The bondpads are located on two opposite sides of the chip. The assembly process is done on a die basis with a cover glass that is smaller than the chip itself. Otherwise the bondpads have to be located along one side of the chip. The bondpads are arranged such that they exhibit point-symmetry around the AM centre. At the assembly stage, this allows the chip to be rotated over 180 degrees without introducing wrong connections. For details on the row and column driver circuits, please refer to appendix AX1.2.

row pad

pixels

row pad

column pads

light block around AM

Test circuits are foreseen separately to validate both circuit blocks and the back-end technology. As they are separated from the main AMP and AMN chips, testing can be done prior to the back-end. That allows distinguishing eventual problems with the design of the raw CMOS chip from eventual problems stemming from the back-end process. Among the test circuits foreseen, individual transistors are foreseen to be able to match the real transistor characteristics with the simulation models used. Note that the implementation of the test circuits is time consuming, because of the many features to test, but also because each test block needs an adapted layout. And for each layout adaptation, the risk of introducing errors increases. Because the vast majority of test circuits are separate from the main circuits, there is practically no CH4-117

Figure 4-10: a 'reflec' chip with colored pixels

CH4 - 118

Chapter 4: The pixel matrix I the basics

risk that eventual errors in the test circuit propagate into the main AMN and AMP circuits. Testing of AMN and AMP chips themselves comprises two tests. One is the testing of column and row voltages; this can only be done before cell assembly and with so-called pico-probes. The second test allows checking the functionality of the bidirectional shift registers. This check can still be executed after cell assembly see for example the _SOL pad (bondpad nr. 42, appendix AX1.2). The picture 4-10 above reveals the elongated pixels all around the AM, originally foreseen to be able to verify the row and column voltages. Thanks to Filip for making this picture [46]! Mask Set The mask set is composed of the classic mask set for the 2.5u CMOS process, plus masks for the back-end layers. A layered top-view of the pixel layout is shown in figure 4-11. The wafers are very small compared to current standards; namely 4 in diagonal (to be compared with the latest 12 or 300mm wafers). The second picture below shows the rough topology of the AM, before the back-end processing. The pixels are not yet present, as can be verified by comparison with the layout.
mirrors mirror via row N labyrinth via row N+1 pixel switch storage capa contact column M Figure 4-11: 'reflec' pixel layout - layered view

CH4-118

Chapter 4: The pixel matrix I the basics

passivation opening row N labyrinth via row N+1 pixel switch storage capa contact column M

CH4 - 119

100 m

Figure 4-12: 'reflec' pixel quad seen by a SEM

The polyimide masks are needed to contact the non-passivated metal2 areas. Subsequently, the mirror mask covers the polyimide vias wherever needed. The mirror layer is used in the pixel matrix, in a donut shape around the AM to provide a single dark zone as margin for the mechanical assembly, for the on-chip counterelectrode contact pad, on the bondpads and for labeling. The remainder of the chip thus looks dark, because of the light absorbing polyimide. Performances The performance of the REFLEC chips is as good as can be expected from a PDLC micro-display. I.e. the projection setup either optimizes the contrast or the brightness. The holding ratio is excellent, which demonstrates the quality and effectiveness of the light shield. As an overall statement the result is excellent and a superb starting point to develop the next generation of micro-displays. The next pictures show the first working micro-display mounted on a ceramic carrier and with the first version of the electronic interface between the micro-display and a pc. One can spot the cable connector to the pc at the top right.

CH4-119

CH4 - 120

Chapter 4: The pixel matrix I the basics

Slimmed-down versions are shown hereunder. The last carrier version is hardly larger than the micro-display chip. It provides mechanical support for the microdisplay chip and for the electronic components that need to be very close. The rest of the driving electronics is located on an intermediate board. This setup is more useful for building a projection setup.

Figure 4-13: early version of the 'reflec' demonstrator

CH4-120

Chapter 4: The pixel matrix I the basics

CH4 - 121

Figure 4-14:'reflec' on a ceramic carrier with SMD components

FR4 front side

uDisplay comes here

connector FR4 back side

Figure 4-15: 'reflec' demonstrator with components on the back of the carrier

CH4-121

CH4 - 122

Chapter 4: The pixel matrix I the basics

4.2.2 5Mpix and stitching

MOSAREL project synopsis The work described here was executed in the framework of the MOSAREL Esprit project EP-25340, financed by the European Commission. MOSAREL is short for Mono-crystalline Silicon Active Matrix Reflective Light Valve. The project is extremely ambitious in terms of specifications. The large number of pixels has a tremendous impact on the entire project. The optical system needs to be adapted for the large cell diagonal; the video source and format conversion electronics need a specific design; the spacer-less cell assembly is another major hurdle and finally, the Si design is not standard at all because of the stitching issue. Because of the complexity and the amount of hurdles, there are two major design tasks involved. A first task consists of the design of a so-called test vehicle. The intention behind the test vehicle is to elaborate, to fine-tune and to provide a means for validation of all the steps involved. The second task is to design the final demonstrator chip; this chip combines with the results from the test vehicle into the desired GXGA or 5 million pixel light valve [23].

The design of the final demonstrator results in the creation of a reticule set that provides for three different chipsets: GXGA, SXGA and a somewhat peculiar XGAp (portrait mode). The lack of thorough software verification procedures led to a number of minor design errors; most importantly, the lcos back-end Si processing had to tackle difficulties with chip planarity. However, the chip functionality was sufficient to prove the concept. A few cells fitted into a projection system and allowed for a limited series of tests.

Figure 4-16: GXGA projector in action (SXGA data source)

CH4-122

Chapter 4: The pixel matrix I the basics

CH4 - 123

4.2.2.1 Technology and mask set

Several considerations made the consortium to abandon the back-end processing developed during the REFLEC project. Fears for contamination of the processing plant by particles from the DARC layer are the main reason for this. As replacement for the planarization layers, a CMP step is added before the patterning of the light shield and mirror layers. With MOSAREL, the light shield is conductive; i.e. it is composed of a very thin, brownish metal layer (TiN). As the process is based on a 0.7m process, the finest metal3 (and metal4) features can be made much smaller than with the back-end steps from the REFLEC project. These layers thicknesses are low compared to the usual metal layers. Therefore the metal3 and 4 layers are used only for the light shielding and for the mirror electrodes. Large portions of the chip are covered with metal3 and metal4, because of the way the stitching is done and because it does not harm the chips planarity. The test vehicle chip showed what you can do and what you cannot with stitching. Please refer to chapter three and to the appendix for details about the demonstrator mask set. In particular note the triple display format that can be achieved with just one mask set. Note also the layout of the two row driver sub-blocks that can be stitched either at the left or right hand side of the AM. Below is a picture of the layout of the MOSAREL pixel (figure 4-17).

bulk bias

storage capacitor Col(k) gate Contact to mirror Figure 4-17: MOSAREL pixel layout contact to row line Col(k+1)

CH4-123

CH4 - 124

Chapter 4: The pixel matrix I the basics

The first figure (4-17) shows pixels organized in a group of 4 pixels. This building block is to be mirrorred to get an elementary building block of the active matrix layout array: a 2x4 pixel subblock (figure 4-18). The gates of four consecutive pixels are shorted at poly level; as can been seen from the figure 4-19b:

Figure 4-18: eight-pixel block This figure also shows the metal3 layout; the via from the mirror layer to metal3 cannot be stacked on top of th via from metal3 to metal2. That is why an elongated metal3 island is needed. Finally the group of pictures below (Figures 4-19 a, b, c, d, e, f, g) gives an overview of the drawn layers for the pixel layout:

[a) nwell mask, source/drains of trts] [b)with active area and poly masks] (Figures 4-19 continued on next page)

CH4-124

Chapter 4: The pixel matrix I the basics

CH4 - 125

[c) metal and via1 layers]

[d) metal2 and via2 masks]

[e) light shield and vias to mirror layer, and f) at last the mirror layer]

Figures 4-19: [f) SEM picture of MOSAREL pixel array] CH4-125

CH4 - 126

Chapter 4: The pixel matrix I the basics

4.2.2.2 Circuit and evaluation

This paragraph just highlights some of the most significant circuit figures and briefly says something about the evaluation of the chipsets. Because of the large pixel count, and because of the wide scanners in the column and row drivers, it is a good idea to foresee some redundancy in the circuits. Redundancy is supposed to help at having chips that pass the acceptance criteria. As the matrix cannot be foreseen with redundant pixels, the only circuits that can bear redundancy are the column and row drivers. This way, the chances that a chip with a perfect AM also has working drivers is maximized. Signal delay considerations push to allow for only one column driver, while two row drivers seem relatively easy to implement. Because of the size of the level shifting circuits (3 devices), no redundancy can be foreseen. Thus, a fault in the medium-voltage (MV) part of the column driver is fatal; while one in the MV part of the row driver is not, because there are two row drivers. The logic part of the drivers uses much more compact transistors (0.7 devices), and allows much more easily to incorporate redundancy.

Figure 4-20: MOSAREL backplane under test

Figure 4-21: usage of probe-needles to measure inside the chip CH4-126

Chapter 4: The pixel matrix I the basics

CH4 - 127

In practice, both the column and row scanners are implemented as double chains of blocks of shift registers (e.g. chain a and b). A block from chain a can be made to connect to a next block from chain b, and so forth. This way a very high degree of redundancy can be achieved. That was the plan; in practice, a problem with signal timings prohibited us to mix both chains (figure 4-22). In the end, the chips thus had double shift registers.

Figure 4-22: functional testing reveals a problem in the clock buffer tree The measurements we could perform, have been hampered a long time by a planarity issue that resulted in shorts. Functional testing was done by taking wafers out after the processing of metal2. By using the right probes, we had access to virtually all of the chips functions (figures 4-20 till 4-24). Because of the absence of the mirror and light-shielding layers, rather high sensitivity to lighting conditions have been experienced (figure 4-24). When a good chip was finally available, no time and no money were available to continue the project. Figure 4-16 shows an operational GXGA setup. Yield of the complete process could not be evaluated since the markets did not open up to real volume production.

CH4-127

CH4 - 128

Chapter 4: The pixel matrix I the basics

Figure 4-23: testing of shift registers

but this time with the lights on: photocurrents!

Figure 4-24: testing the same shift register,

4.2.3 Tmdc prototypes: XGA, WXGA

4.2.3.1 Project goal and general specifications
The project goal is to develop a prototype for an industrial volume manufacturer. From a technical viewpoint, this concerns questions on how to slim-down excess redundancy and chip area from the previous architectures, while still improving lcos cell performance. It also concerns questions on how to scale-up circuit functionality, so that the lower component count reduces the system cost. One thing CH4-128

Chapter 4: The pixel matrix I the basics

CH4 - 129

though is very clear: because of the link with a volume producer, all must go quick and all must be perfect. Fortunately, the training endured with the MOSAREL project, provides some insights on how to achieve that. The display formats aimed for are XGA (1024x768 pixels) and WXGA (1280x768). The panels aimed for, follow the classic AM architecture described above. Several means are foreseen for testing the chip functionality. Parameter Chip size (rounded) AM area ( ) AM diagonal Display format Display aspect ratio Pixel pitch Inter pixel gap Geometric aperture Frame rate (nominal) Response speed (off-on-off) Contrast ratio Chip supply voltages Video driver On-chip video clamp Bidirectional r&c scanners XGA 21mm x 19mm 14mmx 18mm 0.9 1024 x 768 4:3 17.6 0.6 93% 60Hz <20ms >500:1 3.3V, 18V 4-fold analog Y Y WXGA 19mm x 14mm 9mm x 15mm 0.7 1280 x 768 4:3, 16:9 11.9 0.6 90% 75Hz <20ms > 800:1 3.3V, 18V 4-fold analog Y Y

4.2.3.2 Technology and mask set

The backplane technology relies on a specific lcos route available at the time from UMC. UMC performed a customer screening since, leading our project to drop out of their business scope. Fortunately, other foundries are eager to fill the gap. Anyway, the technology from UMC is fully certified and documented; a fact that somehow lessens the design burden. Essentially, the route offers a double gate technology, with two poly layers and 4 metal layers. See also chapter two. Furthermore, as it concerns an extended 0.35 technology, the technology includes the CMP processing steps needed for the planarization of the chips. The design kit offered some support. Because code needed to be written to implement all rules. The mask set is characterized by both a simplification and a further refinement compared to the MOSAREL mask sets. No more stitching, because this time the foundry does not allow for it. Most large foundries do not allow for process changes at the wish of small customers. Ten thousand wafers per year is a small volume! With some yield and size factors in mind, this means that 100k display sets per year CH4-129

CH4 - 130

Chapter 4: The pixel matrix I the basics

is a small volume. Further refinement comes with the accurate experimental and instrumental evaluation of the chips planarity. To ensure chip planarity, the layer pattern density is equalized for the most contributing layers. I.e. the conduction layers: the polys and the metals. A second refinement is the trimming down of the sealing area around the AM for the sealing process; i.e. tougher assembly rules. This of course puts extra pressure on the quality of the cell assembly process. However, the gain in wafer chip count is significant when trimming down from an excessive 5mm to below 2mm. Whereas with the two previous generations, the 160x120 and MOSAREL chips, the mask making process was taken over by specialized teams (EUROPRACTICE and Mietec teams); this time the mask making process had to be monitored at much closer range. The final mask patterns were indeed visually inspected in a WYSIWYG approach. It allowed correcting both layout mistakes (XGA design) and mistakes made during the mask making process (WXGA)! It appears one can never be sure enough.

4.2.3.3 Circuit and performance

The circuit redundancy that characterizes the MOSAREL designs, is reduced to a minimum for reasons of area: the XGA chip still has a double row driver; however, the WXGA chip does not have this feature any longer. The reason for this is that the XGA yield appears primarily determined by the LC assembly process rather than by the circuit. For circuit descriptions please refer to the appendix. The operation of the scanners can be checked as the output of the last shift register is available as output on a bondpad. Verification of the pixel array is best done by visual evaluation of a completely assembled lcos panel. The column busbars can be checked in pairs; this is done with test switches that short pairs of columns: the column scanner activates a block of four video switches. Two of the four video inputs serve as input, the other two as output. The test switches at the other end of the active matrix allow to pass the input signal to the output columns. A fault in the scanner, level shifters, video switches or busbars is detected when the outputs do not match the inputs. The issue of testing the busbars is useful, because line defects cannot be tolerated and one wants to minimize the financial loss associated with the LC assembly of faulty chips. Debugging and functionality test circuitry is located in the four corners of the chip. The peculiarity is that this normally unused circuitry lies outside the chips guard ring. It has the particularity that, whether the test circuitry is present or not, it does not affect the operation of the backplane. Eventually, these parts of the chip can lie very close or eventually inside the wafer handling zone. This yields a few extra chips per wafer. The XGA pixel pitch is maximized for maximum throughput; the WXGA pixel pitch is even slightly below the minimum pitch of the pixel transistors. A pitch of CH4-130

Chapter 4: The pixel matrix I the basics

CH4 - 131

12 is about the best that can be achieved with 3 MOSFET pixel switches. The XGA and WXGA pixel designs are optimized for maximum light blocking, maximum storage capacitance and maximum guarding of signals. Finally, the chips contain a clamping circuit that allows for 5V external video amplifiers. In the future, the look-up tables, DACs and video amplifiers should/would be integrated on the lcos chip directly.

The results obtained with both chips are very good. The successful XGA design paved the way for the WXGA design. At this moment (2005), the WXGA format is the only left to still matter commercially speaking. The projection setup was developed by the TFCGroup, with the aid of tmdc and as a continuation of the experiences gained in previous projects (see also figure 1-9) [18],[47]-[49].

Figure 4-25: a first monochrome XGA projection setup

Figure 4-26: WXGA projector in action

CH4-131

CH4 - 132

Chapter 4: The pixel matrix I the basics

CH4-132

Chapter 5: The pixel matrix II smart pixels

CH5 133

Chapter 5 : The pixel matrix(II)smart pixels

he focus in this fifth chapter is on the development of new pixel circuits. The heart of lcos devices is the pixel matrix, as it is the location where the electrooptical conversion occurs. The classic active matrix circuit suffers from important drawbacks regarding voltage range and addressing speed. The impact hereof can be significantly reduced with new pixel circuits. Paragraph 5.1 initiates the reader to the CE switching principle, discusses the compatibility with continuous light sources and estimates the maximum transistor count per pixel. Indeed, to use the CE switching approach, more complex pixel circuits are necessary. The reason behind the need for a second in-pixel frame memory is explained. Actually, this is the core of chapter five. A second point is to figure out how complex pixel circuits can grow. The maximum transistor count is estimated in the second part of paragraph 5.1; its purpose is to serve as rule of thumb for the selection of new pixel circuits. Paragraph 5.2 addresses the implementation of AMs with frame memory; it describes a couple of original circuit ideas that effectively implement the functionality of an additional frame memory. The resulting pixels are labeled smart pixels maybe because of the functionality offered notwithstanding the limited area available. The authors thoughts indeed came across the double-DRAM circuit, not totally by accident. Along with other original circuit ideas a patent was successfully applied for [50]. This is partially the basis for the creation of the GEMIDIS startup company. At last, a very brief introduction to pulse-widthmodulation or PWM addressing is presented. At this point, it is noteworthy that in the mid 80s, spatial light modulators using LC-on-silicon were developed. The arrays then were relatively small compared to the million-pixel displays described here. Shortly later, the term smart pixel probably was coined in 1993, when K.M. Johnson, D.J. McKnight and I. Underwood publish Smart spatial light modulators using liquid crystals on Silicon [56]. This article is very interesting as a similar progressive buildup is used for the description of smart pixels, i.e. with memory. More recently, Lee et al. [58] and Blalock et al. [57] wrote articles about the subject as well. The article from Blalock very accurately describes the combination of dual-memory and PWM circuits as described in paragraph 5.2. Lee et al. describe a very compact frame buffering pixel circuit, starting from the work by Shields and McKnight [54],[55].

CH5-133

CH5 134

Chapter 5: The pixel matrix II smart pixels

5.1 Counter electrode switching

Why should we be looking at more complex pixel circuits? Let me sum up three fundamental reasons. First, there is the quest for ever smaller pixels a cost driven argument. Actually, Smaller pixels to get more of them within the boundaries of a reticule probably best describes the situation. A second reason is the rather limited availability of foundry processes that combine high-density sub-micron technology with medium-voltage technology. As long as mainstream processes are used, there is little to fear about silicon supply for micro-display backplanes. The third reason is that among system builders, some opt for single-panel projection optics. As such, maybe not the univocally best option regarding optical performances, but there should be at least an lcos alternative to Texas Instruments DMDs (digital micromirror devices). Most probably, system builders wish to have alternatives to TIs product to diversify their component supply chain. TI effectively is the sole supplier of DMDs in the world. DMDs fit right away into single-panel projection architectures, whereas the classic AM displays do not. Clearly, there is a need for smaller pixels operating with lower supply ranges and allowing for high frame rates. The classic pixel circuit with its single transistor switch, medium supply voltage range and moderate frame refresh speed does not fit the picture any longer... The new requirements can be met with the design of smart pixels that incorporate a second frame memory. Lets see why.

5.1.1 The need for a second in-pixel frame memory

Counter electrode toggling the principle The pixel matrix and the pixel driver circuits described until here operate with a rather high supply voltage, i.e. about 15V (see also chapter four). This is because of the LCs characteristics and because a single MOSFET operates as pixel switch. On its own, the bulk effect of the MOSFET-switch is not the only cause. The bulk-effect effectively augments the required range by about 5V; however, the LCs used in the lab indeed require a signal range of ~10Vpp. Such a range is labeled medium voltage or MV in the text. Thus even with a complementary switch that does not suffer from the bulk effect, an MV range of 10V is necessary. This is still way above the supply range of the current deep-submicron technologies. On-going research [European IST project LCOS4LCOS] tries to combine new LC mixtures and reduced cell gaps to allow for a reduction of signal ranges. In the mean time, we have to deal with the LC materials available and thus, we need a work-around. A few foundries offer high-voltage and medium-voltage process options. These processes consist of a combination of a low-voltage technology (e.g. 0.35/3.3V) CH5-134

Chapter 5: The pixel matrix II smart pixels

CH5 135

augmented with MV devices from an ancient 3 technology. An example is the 0.35 3.3V/18V process from UMC. The processing of additional MV devices requires additional masks, because of the different gate thickness and the different doping levels (implants and diffusions). The increased mask cost and the fact that these processes are not mainstream processes, reduce the cost pressure and also raise fears of supply continuity.
a.u. Optical output

100 50 0

F0

F1

F2 100 50 0 a.u.

F3

F4 EOresponse VLC

time

Figure 5-1: AC driving of VAN LC The discussion starts with the mechanism for driving the LC material. The horizontal axis of the first curve in figure 5.1 represents time; a single subdivision corresponds with the duration of a frame. Hence, during frame zero, a ~90% output is desired idem for frame one. The next two frames an approximate ~5% output is desired, etc. The dashed line in figure 5-1 represents the desired light output; the full line a realistic output. The vertical axis represents the optical output and is expressed in arbitrary units (a.u.): it can be understood as percentage of maximum light output. The second curve shows a typical electro-optical response, with labeled arrows indicating the sequence of LC voltages required for each frame. Here, the horizontal axis represents voltage across the LC. Crucial is the fact that AC driving is required, meaning that the voltage across the LC must change sign. Figure 5.2 shows the polarity swap each new frame; VLC(F0) has the opposite polarity of VLC(F1), etc.

VLC(F1)

VLC(F2) VLC(F4) VLC(F0)

VLC(F3)

CH5-135

CH5 136
10V 5V 0V F0 volts VLC(F0) volts VLC(F0)

Chapter 5: The pixel matrix II smart pixels

constant VCE F1 F2 F3 F4

Vmirror

time

5V 0V -5V

LC voltage

VLC time

VLC(F1) F1 F2 F3 F4

F0

Figure 5-2: Vmirror with respect to chip ground and resulting VLC An AC signal across the LC is obtained by switching the pixel electrode from a voltage level above to one below the constant counter-electrode (CE) voltage. This requires a mirror voltage range of typically 10V and a CE voltage around the middle between the lowest voltage and the ranges maximum (~5V).
5V 0V volts VLC(F0) toggling VCE Vmirror F1 F2 F3 F4 time

F0

5V 0V -5V

volts VLC(F0)

LC voltage

VLC time

VLC(F1) F1 F2 F3 F4

F0

Figure 5-3: identical LC voltage obtained with CE switching Instead of a constant CE voltage in the middle of the voltage range, one can decide to switch the CE voltage to 0V for positive frames and back to 5V for negative frames [53]. As seen from figure 5-3, to obtain the voltage drop across the LC, this trick requires having Vmirror adjusted accordingly. The signal range of Vmirror is halved. As a result, not only the pixel transistor can be smaller, but all the up-front electronics in the video path see this reduction in signal range. CH5-136

Chapter 5: The pixel matrix II smart pixels

CH5 137

The principle vs. continuous light sources CE switching is a major key to reduce the voltage span on the mirror electrodes. However, without extra measures this has negative consequences for the optical throughput of a projection system. The trouble is that the CE is common to all pixels. Consequently, at the very instant the CE is switched, the voltage across the LC layer changes for all pixels at once. This conflicts with the serial refresh used with the classic AM: a complete frame period is used to update the AM a few pixels at a time. Right after the CE voltage switch, all the AM data are wrong; without adjusted measures it would take a complete frame time to get the last pixel updated. In other words, some pixels (the last written to) have less time to drive the LC with the correct signal. Obviously, this is not ok.
5V 0V [volts]

F0
pixel (1stR,1stC), F0 update ; valid during 98% of frame period

F1

[time]
pixel (1,1), F1 update

pixel (lastR,lastC), F0 update; valid during 5% of frame period

pixel(lastR,lastC) data validity pixel(1,1) data validity frame period

toggle CE

toggle CE

Figure 5-4: issue of data validity after CE toggling A partial solution consists of splitting frame periods into a write period right after the CE switch, and a display period. Line-at-a-time addressing schemes help to achieve short write cycles [15],[51],[52]. However, the shortest write cycle equals the minimum time to address a single column. The time needed for line-at-a-time addressing is the product of the number of lines with this minimum column write duration. As an example, a column write time of 50ns results in an overall 50s write cycle. During the write cycle, the light output is blocked, e.g. by switching off the lamp. The power lamps used with projection systems usually cannot be switched, so a shutter is necessary. On the contrary, those opting for light emitting diodes (LEDs) as light source, can switch the lamp. This has been demonstrated by NTE applications. However, diodes lack brightness for projection applications.

CH5-137

CH5 138
[volts]
block lamp

Chapter 5: The pixel matrix II smart pixels

5V 0V

block lamp

F0

pixel (1stR,1stC), F0 update ; valid during 98% of frame period

F1

[time]

pixel (1,1), F1 update

pixel (lastR,lastC), F0 update; valid during 65% of frame period

block lamp

60% lamp duty cycle pixel(lastR,lastC) data validity pixel(1,1) data validity frame period

toggle CE

toggle CE

Figure 5-5: 'concentrated' write cycles and light flow shutter The result is an illumination duty cycle lower than 100%; depending on the duty cycle, the optical output of the system can still be sufficient. Because of the limited power output needed, it can be expected that this reasoning is valid for NTE applications. The reduced illumination duty cycle however is no good outcome for projection systems, where arc size (lamp power) is already traded against tendue. For good tendue matching, small arc sizes are required, usually leading to limited lamp power and thus limiting the optical output. A reduced lamp duty cycle further reduces the systems output. A solution consists of finding a way to address all of the AM very quickly, i.e. reducing the duration of the write cycle to the minimum. E.g. a very high degree of parallelism in the video data drivers allows for this, at the cost of increased interconnection and increased data speeds. So, the shortest write cycle indeed limits the loss of brightness to the minimum. This is exactly where a second frame memory comes into the picture. Remember indeed that the classic DRAM circuit is designed to hold the image data for one frame. As such the AM thus already constitutes a frame memory; the text refers to this memory as the pixel memory, because the LC capacitor effectively contributes to the holding ratio. If we succeed to incorporate a second memory that can instantly flood the primary pixel memory, the problem due to the limited illumination duty cycle is solved. Note that with a second frame memory, the term progressive scanning is not totally at its place: the refresh of the secondary memory is done progressively, but the data are displayed frame-at-once. A special remark on frame sequential color schemes is at its place here. With frame sequential color every next frame changes color at once for all of the AM. In other CH5-138

Chapter 5: The pixel matrix II smart pixels

CH5 139

words, like with CE toggling, when the transition to the next color takes place all image data are corrupt. There thus is a similarity between these two schemes both with CE toggling and frame sequential color a frame-at-once refresh is necessary. For AMs without a second frame memory, it is necessary to insert dead periods or black segments in the color wheel to avoid color blending between successive subframes. This leads to an even further reduction of the light output of the optical system. The three arguments given at the start of paragraph 5.1 point in the same direction: CE toggling allows for lower voltage transistors, which cost less area. And more complex pixel circuits can be thought of e.g. circuits providing the AM with a second frame memory. This is exactly what is needed to implement CE toggling and to implement color sequential schemes without trading light output. Now that it is clear more transistors are needed underneath the pixel mirror, a good question is: just how many fit within the pixel boundaries? The next paragraph gives an answer to this question.

5.1.2 Area estimation for smart pixels

The implementation of CE switching requires the pixel circuit to count more transistors i.e. pixels get smart. This paragraph tells how many transistors fit inside the pixel area hereby defining a practical criterion for the selection of new pixel circuits. Two components occupy the silicon area of a classic DRAM pixel: a switch transistor and a storage capacitor. This is a very small component count considering the fact that the most advanced technologies allow to integrate far more components within the same area. Some people tried estimating the minimum pixel area needed for the circuit to be implemented. A 3 technology roughly yields MOSFET dimensions of around 3 x 6 or 182. Therefore, in a first approach, one could state that a 10 pixel can hold up to 5 transistors. Apparently, slightly more complex pixel circuits fit into a 10 pixel. A 0.35 technology is expected to allow for much denser circuits. The gate length ratio is 3/0.35=8.57; the area ratio would then be the square of this, about 73. This suggests that a monumental 73x as many transistors could fit inside a pixel! Apparently, the pixel circuit complexity can increase quite a bit. Unfortunately, the 73x figure is a fallacy. Circuit area is not determined by gate area alone. Next to the gate area, there are necessarily spacings and connections. These do not scale at the same rate as the gate length. Area is needed for interconnections; lets take the example of substrate connections. The area cost of a substrate connection is larger than the area of the contact itself. Not only does one have to respect the minimum spacing rules, but also a metal track is needed to establish the connection throughout the AM. There cannot be an unlimited amount CH5-139

CH5 140

Chapter 5: The pixel matrix II smart pixels

of metal wires on the metal layers. If the transistors are not of the same type (all NMOS or all PMOS), two wells and corresponding bulk contacts need to be foreseen. Eventually, well contacts can be shared by neighboring pixels; however, the two wells still need different bias voltages. Moreover, each transistor requires three connections (plus the well contact, eventually shared with other transistors). In other words, it could be that the pixel area is determined by the number of metal connections, rather than by the gate count. Additionally, the circuit configuration can make some of the contacts to be common to several transistors, e.g. two transistors connected in series. So, maximum pixel circuit complexity cannot be estimated just with a too simple extrapolation of the number of gate areas per pixel. Therefore, with a 3 technology, the area effectively occupied by a minimum size transistor is larger than 18 sq., because of the spacings and interconnection requirements to be respected. The text below lists a few assumptions made to come to a much more realistic estimation. It still yields an optimistic figure. However, this already gives a far better insight on how complex the pixel circuit can grow. For the calculations, no transistors are shared between adjacent pixels; the figures represent the maximum number of transistors per 10 pixel The wiring on metal1 is not taken into account; this artificially boosts the maximum number of source/drain/gate contacts allowed No substrate contacts are accounted for as their impact is minimal for the layers considered; the area estimation counts only the cost on diffusion and poly layers. If the first metal layer is taken into account, the relative importance of substrate contacts would increase Only one transistor type is taken into consideration; complementary circuits are less area efficient in this case, because an extra well spacing rule is to be observed. The extra spacing causes a noticeable reduction in the maximum transistor count The presence of one or more storage capacitors reduces the pixel area available for transistors. Therefore, two cases are looked at, one with half of the area reserved for capacitors; a second which foresees no capacitors at all The number of transistors is computed as a ratio of areas. Their precise shape is not taken into account nor is the way they are stacked. Therefore, one should multiply the net count by a 'packing' efficiency a number between 0 and 1. The packing efficiency figure depends on the amount of time spent by the layouter and will rarely reach the maximum value Common source/drain terminals often show up in MOS circuits; only three cases are studied. The cases with common gate connections are not considered. A common s/d connection eventually connects to the metal1 layer, but this is not always the case. The calculations assume three theoretical circuit configurations. The first has all transistors isolated from one another, i.e. there is no geometric overlap (titled 0% overlap in figure 56). This yields the lowest transistor count of the three configurations. The next two assume overlap of source and drain: all transistors are connected in CH5-140

Chapter 5: The pixel matrix II smart pixels

CH5 141

series. In one case all common source/drain contacts systematically connect to metal1 (marked conn com s/d) and in the last case they do not connect to metal1 (marked -NC- com s/d). Figure 5-6 below illustrates the above. The case without capacitors and NC-com s/d can be expected to yield the highest counts. 0% overlap conn com s/d -NC- com s/d

contact (ctct) window spacing between diffusions spacing between conductors of 1 layer spacing between gate ply and ctct gate length overlap of ctct by diffusion overlap of ctct by conductor (mt1/ply)

Typical design rules 0.35 tech:

== 0.4 x 0.4 >= 0.6 >= 0.5 >= 0.4 >= 0.35 >= 0.3 >= 0.2

Figure 5-6: impact of circuit configurations on layout area The next two pages show the transistor count as function of the pixel pitch, the circuit configurations and percentage of storage capacitors. As a conclusion: with 50% storage area, roughly 10 0.35 transistors fit into a 10x10 pixel. A packing efficiency of 80% further reduces the count to 8. Transistors cannot be molded into just any shape. Note that 0.35 transistors have a supply range of 3.3V. With CE switching, this results in a maximum 6.6Vpp swing. For many LC modes this is insufficient to saturate the LC directors in one of the two states. E.g. it becomes impossible to obtain both a good dark and a good bright state. Thus, at least one additional 0.5 transistor (5V) is needed and most probably, even more than one. My guess is that 4 0.35 plus two 0.5 transistors can fit into half of a 10x10 pixel, the rest being used by storage capacitors. By now, it is time to look at the developement of smart pixel circuits. CH5-141

CH5 142

Chapter 5: The pixel matrix II smart pixels

Figure 5-7: estimated count as function of pitch and circuit configuration CH5-142

Chapter 5: The pixel matrix II smart pixels

CH5 143

Figure 5-8: transistor count as function of pixel pitch for 0.35 tech

CH5-143

CH5 144

Chapter 5: The pixel matrix II smart pixels

5.2 AMs with an additional frame memory

The starting point for this development is the quest for low-voltage AM circuitry. The invention preceded the start of a new project (IST-LCOS4LCOS, short: L4L) that additionally aims at single panel solutions. In other words, the project adds two additional design requirements compared to the implementations described in paragraph 4.3: one regarding the voltage span, the second regarding frame speed. Increased levels of parallelism in the video data stream only partially address the frame speed requirement. Paragraph 5.2 addresses the issue of AMs with an additional frame memory; it explains a few possibilities of how to include a second frame memory inside the AM. The authors thoughts came across the double DRAM circuit; along with other original circuit ideas a patent was successfully applied for [50]. The following paragraphs introduce the bucket-brigade and double-dram circuits and briefly pixels circuits operating with pulse-width modulation. This last discussion will be the very end-point of this book. However, it is not the end-point of the developments of pixel circuits.

5.2.1 Two-transistor pixels

Consider the circuit below. A second pixel transistor (switch Sw2) controls the connection between the pixel electrode and the storage capacitor. Thus, while Cpixel displays the image data during a frame, Cstorage can be refreshed in the mean time. After switching the CE, Sw2 transfers the new image data to the mirror. The idea is to do this simultaneously for the entire AM, hereby effectively loading a frame at once.
column (data) line row busbar read line
Sw1 Sw2

mirror Cpixel counter electrode

Cstorage AM bulk contact

Figure 5-9: two-transistor pixel circuit CH5-144

Chapter 5: The pixel matrix II smart pixels

CH5 145

There is a catch with this circuit To safeguard the holding ratio, it can be expected that Cpixel is composed of the LC capacitor plus an additional storage much like the pixel circuit described up to here in the text. Note that the figure does not explicitly show this extra capacitor. However, when Sw2 establishes a connection between Cstorage and Cpixel, electric charge (re-)distributes between the two capacitors. Only if the capacitance ratio is high enough, the voltage loss due to the charge redistribution will be negligible the redistribution will hardly affect the signal amplitude. That this is a problem in practice, is clear from the small area available to storage capacitors. Thus, one cannot obtain large capacitor ratios. Current practices yield ratios of about 1:10. The following spice simulation clearly shows how the pixel voltage is no exact copy of the storage voltage, because of capacitive coupling with the gate signal sw2. Neither does the storage voltage follow the column accurately the difference here is about 1/10th as expected from the charge transfer to the pixel and the capacitor ratio used in the simulation model (Cst=0.9pF, Cpix=0.1pF). As a result, typical deviations are 200mV.
200mV

0V SEL>> -200mV V(col)-V(storage) 4.0V

V(storage)-V(pixel)

V(0)

2.0V

0V V(sw1)

0s V(col)

V(sw2)

10ms V(storage) Time

V(pixel)

20ms

Figure 5-10 : two-transistor pixel simulation The GMIN spice parameter affects the computation time and in some cases also the stability of the simulator the default is often good. Here this parameter has been set CH5-145

CH5 146

Chapter 5: The pixel matrix II smart pixels

to 1e-12, i.e. smaller than the default value. This is necessary to avoid an unrealistic quick discharge of Cpixel. The simulation of leakage currents is not included here to make the comparison between the different pixel architectures, only the ideal behavior is of interest. A second catch consists of the fact that this time the pixel capacitor gets its charge from Cstorage, thus not directly from the much larger column capacitor. Because of the limited capacitor ratio, one can expect that the charge from the previous image will influence the net charge after the charge redistribution. I.e. under certain circumstances, the voltage on the pixel electrode will be different depending on the previous image content. The predictable visual effect is image retention really not desirable at all. The impact of this can be minimized with the addition of a reset transistor that systematically discharges Cpixel right before (re-)connecting with Cstorage (see Figure 5-11). The charge redistribution effect, with associated signal loss still remains though.
column (data) line row busbar read line
Sw1 Sw2

reset

Sw3

Cstorage AM bulk contact

Cpixel counter electrode

Figure 5-11: reset transistor to cancel the image retention issue The important funcional contribution from the addition of the second switch, Sw2, is that an image can be displayed while updating the storage capacitors of the active matrix a set of frame data is written into the AMs secondary frame memory, while the primary (pixel) memory directly drives the LC. Unfortunately, the transfer from the secondary to the primary leads to a loss in signal amplitude. To make this circuit still worthwhile, a corrective measure can be taken upfront. I.e. the signal loss could be taken into account when generating the video signals with an amplification of the DACs output, with an adjusted look-up table or with a combination of both. Two issues reduce the effectiveness of this idea. First, the LC capacitance is difficult to estimate as it is voltage dependent and non-linear. Consequently, the amplification factor is variable unless the LC directors are forced into an identical state prior to the refresh. Second, the signal voltage range CH5-146

Chapter 5: The pixel matrix II smart pixels

CH5 147

gets reduced. Indeed, with a 3.3V pixel transistor that can hardly handle the required 2.5V modulation voltage, it will be unable to handle 2.5V x lossCompensatingFactor. Thus larger pixel transistors, capable of passing 5V signals, are required. Because the charge redistribution is the centre of the problem, a smart circuit was devised by Prof. Dr. Ir. A. Van Calster [50]. In its most basic configuration, it counts 3 transistors and two capacitors and is therefore treated next. As can be expected, after this bucket-brigade circuit, a four-transistor circuit will be presented (double DRAM). Finally, even more transistors are used to implement PWM schemes introduced at the end.

5.2.2 The bucket brigade architecture.

Consider the circuit in figure 5-12 below. Here also, the row select line connects Cstorage to the column for updating. The two other transistors are operated in sequence to copy the charge on Cstorage to the sum Cbucket + Cpixel Cbucket (definition of Cbucket). The copying can be done simultaneously for a complete row, a set of rows or even for the entire AM. For a start, one can assume that the bucket and storage capacitors are dimensioned such that Cstorage = Cbucket. The select, copy and pre-charge lines are all inactive before the charge copy operation. To start, the pre-charge line is activated to pre-fill the bucket capacitor to the maximum. The growth of Sw3s Vt limits this maximum. Then, the copy line is activated; i.e. the voltage on Sw2s gate rises. The mirror potential rises as the charge on Cbucket cannot disappear; at the same time Sw2 establishes a connection between Cstorage and Cbucket, once Vgate(Sw2) reaches the threshold Vt(Sw2). Cbucket will discharge into Cstorage until Sw2s Vt has grown so large that no more current can flow. Cstorage is filled to the maximum. At this point, the remaining charge in Cbucket equals the charge stored in Cstorage before the start of the transfer.
column (data) line

Sw1

Sw2

Cbucket

Sw3

row (select) line copy line precharge line Vref

Cstorage AM bulk contact

Cpixel counter electrode

Figure 5-12: the bucket brigade circuit CH5-147

CH5 148

Chapter 5: The pixel matrix II smart pixels

A possible problem is the peak voltage reached by the mirror. A very quick change of Vgate(Sw2) propagates to the mirror electrode. A moderate slope for Vgate(Sw2) allows Cbucket to already discharge, hereby diminishing the peak voltage reached by Vgate(Sw2). A second issue concerns the implementation of Cbucket: it is a floating capacitor and therefore requires the technology to provide for poly-poly capacitors. An extra lithography step and associated masks are required for this. Note that the (W)XGA implementations described in chapter 4 anyway use poly-poly capacitors as they are available in the selected technology other technologies were not readily available to the lab at that time.
200mV

0V

-200mV 4.0V

V(col)-V(pixel)

V(0)

2.0V

SEL>> -0.5V 0s 10ms V(sw1) V(sw2) V(sw2) V(sw3) V(col) V(pixel) V(storage) Time

V(sw1)

20ms

Figure 5-13: simulation of the bucket-brigade circuit Figure 5-13 illustrates simulation results that allow to verify the operation of the above circuit and that also allow to compare with the circuit described in 5.2.2 (the two-transistor pixel circuit). The major advantage of the bucket-brigade circuit is the almost perfect copying of the storage voltage onto the pixel node. With perfect copying, the reader will understand this assumes there is no multiplicative effect from the ratio in total capacitance on the storage and pixel nodes. It is almost perfect as here too, the gate signals propagate through the gate-drain/source parasitic capacitance. Observed deviations are of the order of 10mV and appear to be rather CH5-148

Chapter 5: The pixel matrix II smart pixels

CH5 149

stable, because both the overlap capacitance and the switching of the gate signal are rather stable as well. The difference between the column voltage and the final pixel voltage is approximately 10x smaller as compared with the two-transistor circuit (compare with figure 5-11).

5.2.3 The double dram architecture.

A cost aspect of the bucket brigade circuit the floating bucket capacitor can be a reason to look for an alternative. The double-dram circuit eliminates the need for a floating capacitor and somewhat weakens the issue of charge redistribution from the two-transistor pixel circuit described earlier. So only one poly mask is necessary to build the two storage capacitors; because one of the storages remains connected to the mirror during the entire frame, the holding ratio is safeguarded. To avoid the image retention problem, a fifth reset transistor can be added in a very similar way as described in paragraph 5.2.1.
writeA
Sw1a Sw2a

readA

column (data) line

CstA CstB Cpixel

Sw2b

Sw1b

readB writeB

Figure 5-14: the double dram circuit A new option here is that the data line can be split into two data lines: one to update storageA, the other to update storageB. This allows e.g. to implement a scrolling color scheme combined with CE switching: each pixel simultaneously holds the image data for both the frame polarities. Maybe this can be advantageously combined with a dual rail video amp this is something to be investigated. Toggling of CE can be done at about any moment. Another option is to combine the double dram and bucket brigade circuits to avoid hard-to-control signal losses [50].

CH5-149

CH5 150

Chapter 5: The pixel matrix II smart pixels

5.2.4 PWM for fast gray scales.

Now that the AM can be refreshed at once, another hurdle shows up. Indeed, at high frame rates, the response time of the electro-optical effect can be too slow, much too slow. In particular, the transitions between intermediate gray levels are slower than full on/off transitions. Hence, for minimal LC response time, it can be advantageous to exploit on/off transitions only. The gray level is determined by the fraction of the frame time during which the pixel is turned on. This is what is meant with pulsewidth-modulation (PWM). At some point during a frame, the LC must switch state (on/off either way). That instant is determined by for instance, an analog voltage stored in the pixel. Each pixel contains a comparator circuit that compares the analog value to a ramping signal. The following figure shows a basic PWM circuit, for simplicity only, directly driven by voltage sources.

Figure 5-15: a basic comparator circuit block for PWM As soon as the ramp reaches the (stored) data voltage plus a threshold, the comparator output swaps. In the simulation output shown below, the pixel voltage CH5-150

Chapter 5: The pixel matrix II smart pixels

CH5 151

effectively switches between 0 and 5V within a frame time: the pixel voltage starts with a transition to 5V and, depending on the stored voltage, returns back to 0V after a while. The illustration has been obtained by stepping the stored data voltage. In reality, the Vdata source is replaced by e.g. the double dram circuit described before. Note that the pixel capacitor electrodes are both driven: one by the pwm circuit, the other by the CE (counter-electrode) signal; this is necessary to obtain the voltage span required by the LC.

Figure 5-16 : basic PWM circuit behavior (see figure 5-15) The previously described AM circuits plus a compact comparator at pixel level are sufficient to implement a PWM scheme. Again, the issue is to select a very compact comparator as it adds even more transistors to the pixel circuit. Our patent [50] includes some ideas on how to implement such comparator. Herebey, one more limiting factor has been taken into account: the power consumption of the PWM circuit block. The comparator draws current when it is about to switch (see the simulation of the above basic circuit in figure 5-17). The power consumption of the comparator is of major concern. The reason for this is that the current consumption is multiplied by the number of pixels i.e. a million or two. A current of one micro-amp per comparator means one or two amps for the entire AM when all pixels switch at the same moment. In other words, the current peak will be reached for AM-wide uniform gray levels.

CH5-151

CH5 152
400uA

Chapter 5: The pixel matrix II smart pixels

200uA

SEL>> -40uA 5.0V

Id(Mnpwm2)

2.5V

0V 9ms V(pixel) V(data) 12ms V(CE) V(ramp) Time 16ms 19ms V(rmp_data)

Figure 5-17: basic PWM operation transition current The figure below show a circuit I developed. It implements an improved PWM circuit combined with an in-pixel memory circuit; in this case the double dram. The stored voltage is compared with a ramping reference voltage. The current limiting transistors Mnchop1, Mpchop1 and Mpchop2 efficiently limit the current sinking when the PWM circuit is about to switch; i.e. when the data voltage is just below the ramp signal minus the threshold voltage. The first input transistors (Mnpwm1 and Mppwm1) can be held into a near-transition state where a lot of current flows from the supplying ramp source to the ground. The reason why a nmos chopping transistor is needed (Mnchop1), is that it avoids adirect path from the ramp-supply to ground at all times provided it is only turned on when its pmos counter-part (Mpchop1) is off.

CH5-152

Chapter 5: The pixel matrix II smart pixels

CH5 153

Figure 5-18: pspice circuit of the combination of a double dram pixel memory block, plus a smart current-limited PWM circuit block

CH5-153

CH5 154

Chapter 5: The pixel matrix II smart pixels

Without the current chopping transistors, the power consumption of the comparator reaches about 70 micro-amps in simulations. The current-limiting transistors (figure 5-18) Mpchop1, Mpchop2 and Mnchop1 reduce the power consumption to magnitude orders of 50nA.

Figure 5-19: effects of current limiting on the output curve The behaviour of such circuitry is somewhat particular, because the output will stick to the previous state as long as two conditions are met: a) the current-limiting transistors are off and b) the input voltage is not high enough compared to the ramping signal so that the inverters are out of the transition region. It is as if the power supply is being sampled: The last circuit presented in the text aims at shortening the decision window of the comparator circuit: assume the output is low. As soon as the output starts to rise, one can think about forcing the output to rise further. This is what happens with the circuit in Fig 5-20: the output bypasses the first inverting stage by pulling its output low. Consequently, the PWM output has to rise further. Compared to the previous idea with only current-limiting transistors, one fewer control signal is required. It is an interesting approach, because as the number of control signals rise, less and less area is avaible on the conductor layers. In other words, this idea is attractive. However, detailed analysis during the LCOS4LCOS project reveals that with this circuit not all transistors can have minimal sizes. The area gain resulting from the fewer control signals is mitigated by the required transistor dimensions. As a result this circuit ends up costing more area. During the LCOS4LCOS project this circuit has not been implemented. The testing of this circuit idea hopefully will be done in future.

CH5-154

Chapter 5: The pixel matrix II smart pixels

CH5 155

Figure 5-20: attempt to get shorter decision time than with the comparator circuit of fig 5-18 CH5-155

CH5 156

Chapter 5: The pixel matrix II smart pixels

Nice simulations, so what about the implementations? The silicon-implementations have been done by our partners in the LCOS4LCOS project (in particular, Atmel and Thomson). Several publications are associated with the circuit ideas presented in this chapter; aspects concerning the optical architecture and the performance of some of the smart pixel circuits were/will be presented in: Eurodisplay 02: LCOS4LCOS booth (Thomson) o Rear-Projection Technologies for consumer applications o Polarization analysis on LCOS based projection systems IDW '03 conference: Spice Model for a Dynamic Liquid Crystal Pixel Capacitance (IMEC) SID 04 : Three-dimensional modelling of a projection TV system by dynamic LC simulation and ray tracing ( UCLondon / Thomson) SID06 : report of the results of the LCOS4LCOS project (to be published, co-authored by myself) [] As a final note in this chapter, I would like to re-state my personal contributions: I am the author of two circuit blocks: the double dram and the current-effective PWM scheme presented above. I made several simulations to show the potential advantage of the different circuits. Other ideas, developed by Andr Van Calster, Herbert De Smet and Geert Van Doorselaer are grouped in our patent. I had many fruitful discussions with Herbert, which I am profoundly grateful for. The real implementations of the circuits presented in this chapter have been done during the LCOS4LCOS project. The results of this projects are currently being readied for presentation at SID06.

CH5-156

Conclusion
The work presented in this thesis concerns the design of lcos microdisplay backplanes for projection applications. The aim of the present text is to clarify microdisplay backplane design. Chapter one introduces lcos microdisplays. Chapters two and three discuss design rules it tells how to design. Chapters four and five describe what circuits to implement. Chapters two and four describe limitations encountered; chapters three and five respectively formulate answers to these limitations. The rigorously structured design approach is necessary to be able to demonstrate the lcos concept. The design needs for the stitching approach have been carefully analyzed. This analysis results in an original stitching procedure, which is detailed in chapter three. This procedure gives designers full control over the design process and enables first-time-right designs. An advantageous by-product is that this procedure also helps to organize the reticule sets in a way that allows the production of microdisplays with different formats. For prototyping, this is a cost-effective approach. The filing date of TowerSemiconductors patent about stitching design rules, coincides with the end of the MOSAREL project and hereby demonstrates the originality of the combination of stitching with x-Si processing. The successful implementations described in chapter four are limited by low frame speeds and a non-standard medium voltage requirement. Both limitations affect the cost of the backplane. Chapter five describes what circuit concept is needed for the active matrix to operate at higher frame rates and at standard voltage ranges. The ideas developed are protected by a world-wide patent. Some of the resulting circuits are under evaluation by the LCOS4LCOS consortium. In this thesis little has been said about integration of (a part of) the system electronics onto the microdisplay backplane. Further integration is necessary to reduce the cost of lcos systems. Currently there is a lot of interest for the HDTV format. This format was not implemented by the author; however, all the elements needed for such a design have been described. Battery-powered applications require low power consumption another area to be investigated eventually. I believe I am right to state that I pioneered lcos backplane design the successful designs served as a basis for the research work performed by several lab colleagues and contributed to many publications and to the creation of a spin-off company. My colleagues Geert Van Doorselaer and Dieter Cuypers had the lead for the system electronics, and for the system optics (cell assembly and projection setup) respectively. My responsibility covered the entire silicon design of several backplane ICs. In the technology plane, I developed an original design procedure for stitching. The

implementations during the MOSAREL project have shown how well this approach works. In the circuit plane I implemented the classic AM architecture for different display formats. The successes with these design ultimately led to the creation of the GEMIDIS startup company. I also developed three smart pixel circuit ideas for multi-million pixel arrays these ideas form an important part of the patent we were granted. Some of the ideas have been implemented for single-panel HDTV backplanes and have also been tested during the LCOS4LCOS project. The results hereof will be presented at SID06. Finally I would like to thank again all my lab colleagues and scientific peers for the interesting discussions and fun time during my research. I am also very grateful for the funding of the several projects: national, European and bilateral funding made this research possible as well as the hard work of the project leaders who wrote several very successful project proposals.

Appendices

CH6 - 159

Appendices
This sixth chapter is intended as a collection of useful information. The datasheet section truly deserves the label appendix as chapters 3 and 5 refer to parts of it. The last two sections are more introductory manuals for the use of crucial software blocks. There is no way to make the designs described in appendix AX1 without these tools. This is why the datasheet section and the manuals were grouped together: for an lcos designers reference. The manual sections have a broader scope in that they apply to many full custom design jobs. Although this chapter is the last one, and on first hand the least interesting from a purely scientific point of view, this chapter contains a summary of many, many hours of tedious searching searching for the way to make this software work, how to make this software do what I wanted. I hope the information herein can shelp help other designers concentrate more efficiently on the real design challenges.

6.1 Datasheets of three microdisplay chips

The order in this section roughly sketches the microdisplay design projects order over time. The descriptions are kept at a minimum, to contain the essential information without too much overhead. The intended structure of the datasheets is as follows: a) Brief design project description (aim, result) b) Picture of mask design and chip floorplan c) Pin-out of main circuit and eventual application notes d) Schematic of main circuit (down to transistor level) and eventual simulation results e) Pin-out of test circuits and eventual application notes f) Schematic of test circuits

6.1.1 Feasibility: a 32x18 matrix with integrated column scanner

a) Design project description (goals, results) This design project aims primarily at evaluating the feasibility of (a) micro-display design(s) (1994-1995). Assembly of a micro-display panel is not yet one of the goals. The chip measures 3680m by 2900m and features a simple shift register, a small active matrix (18x32 pixels) with two different pixel layouts, a limited number of externally addressable row and column bus-bars, sub-block test circuits including CH6-159

CH 160

Appendices

a single pixel layout and a scaled version of it. All these elementary blocks allow to check circuit operation and to measure the pixel transistor characteristics. The pitch is 40m in a 2Metal-12V-2.4m CMOS technology of Alcatel Mietec1. The name used to refer to this design is SiLCD. Summary of most significant results: micro-display design is feasible indicative value for attainable pitches proof that a light shield is necessary to keep the pixel-transistors off current at acceptable levels [Paper Johan 1995, Japan]. b) Picture of mask design and chip floorplan

Figure 6-1 : SiLCD chip layout In Figure 6-1, the bondpad abstract cells need replacement by their corresponding layouts. The AMs have openings in the passivation layer for probing; the separate pixel and its scaled version provide a means to measure the pixel characteristics through probing as well. Contacting the ground bondpad through probing is eventually necessary. Figure 6-2 depicts the SiLCD chip floorplan. The figure does not show bonding pads nor the exact signal names refer to the picture above for these.

currently AMIS Oudenaarde, Belgium

CH6-160

Appendices
sync clks row i/p and 2SSR o/p test circuits

CH6 - 161

SCANNER

AM1

AM2

pixel 3x pixel

column o/p

Figure 6-2 : SiLCD chip floorplan c) Pin-out of main circuit and (brief) application notes The supply voltage span is 12V de row inputs are 12V selection pulses. The video switches (CMOS pass transistor pairs) correctly pass a 12V video signal to the columns; the (single) pixel transistors however cannot correctly transfer signals within this voltage range to the pixel electrodes (mirrors), because of the bulk effect. CLK1 reads the sync pulse on a falling edge of CLK1 note that CLK1 and CLK2 should not overlap to avoid race conditions. Relevant bondpads (the main circuit and the test circuits share some bondpads):
Pad Pad name nr. R2 R4 R8 R11 R12 R16 K0 30 29 28 27 26 25 24 23 22 21 20 19 18 17 Short description 3 row (i/p) 5th row (i/p) 9th row (i/p) 12th row (i/p) 13 row (i/p) 17 row (i/p) 1st column AM1 (o/p)
rd

Pad name D4S5 G4 D3S4 G3 VIDEO D2S3 G2

Pad nr. 13 12 11 10 9 8 7 6 5 4 3 2 1 34

Short description drain 4 source 5 gate 4 drain 3 source 4 gate 3 analog video i/p drain 2 source 3 gate 2

G6 16 gate 6 D5S6 15 drain 5 source 6 G5 14 gate 5

K1 K2 K15 K16 K17 K31 D6

2nd column AM1 (o/p) 3rd column AM1 (o/p) last column AM1 (o/p) 1st column AM2 (o/p) 2nd column AM2 (o/p) last column AM2 (o/p) drain 6

CLK2 33 clock 2 (i/p) SYNC\tDFF 32 synchronization i/p GND! 31 ground (0V)

VDDA D1S2 G1 DFFQn DFFQ_ DFFAS CLK1

supply voltage (12V) drain 1 source 2 gate 1 output trap n (inverted, not buffered) input trap n (not buffered) after switch : signal after 1ste switch clock 1 (i/p)

Table 6-1: main circuit bondpads CH6-161

CH 162 c) Schematic of main circuit

Appendices

The main circuit is composed of a scanner, a set of video switches and two small active matrices of the DRAM type. The scanner is a chain of 32 12V shift registers; the output of each register directly drives a CMOS switch. A CMOS switch is composed of a parallel connection of a nmos and pmos transistor. It passes the full 12V range of signals; as opposed to single transistor switches, which can only pass a smaller voltage range. The CMOS switch passes the video signal on to a column. The term column driver refers to the combination of the video switches and scanner. There is no scanner to drive the row electrodes; just a few rows are externally addressable. The other row electrodes are floating this is indeed a (small?) design error. See paragraph f) for more details on the shift register, in particular the explanation on the two-stage shift register.
CLK1,2 sync video R02 R04 R08 C16 C17 C31 R10 R11 C00 C01 C02 R16 8c x 9r 16c x 18r

Figure 6-3 : SiLCD main circuit schematic The difference between the two active matrices AM1 (left) and AM2 (right) is that AM2 has no substrate contacts inside this increases the area available for the storage capacitor. The layout of AM2 does not feature mirroring of the pixel layout as with AM1. Further area gains are possible with mirrored pixel layouts. Note that the layout pitch is 50m; the real dimension is 40m, because of scaling down to 80% of the physical mask dimensions. The single pixel transistor cannot pass 12V signals because of the bulk effect. At 12V source-bulk polarization, the threshold increases to ~4V. With nmos pixel switches, the voltage range of signals that can pass lies between 0 and ~8V; for pmos pixel switches, the range goes from ~4V to CH6-162

C15

Appendices

CH6 - 163

12V. Both AM1 and AM2 use nmos pixel switches. Column drivers designed to drive both types of AM need CMOS video switches; usually only one type is used, causing CMOS video switches to be unnecessary. d) Pin-out of test circuits (bonding pads and separate probe pads) Bondpads for testing: [grayed-out fields are for the main circuitry only]
Pad Pad name nr. R2 R4 R8 R11 R12 R16 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 Short description 3rd row (i/p) 5th row (i/p) 9th row (i/p) 12th row (i/p) 13 row (i/p) 17 row (i/p) Pad name D4S5 G4 D3S4 G3 VIDEO D2S3 Pad nr. 13 12 11 10 9 8 7 6 5 4 3 2 1 Short description drain 4 source 5 gate 4 drain 3 source 4 gate 3 analog video i/p drain 2 source 3

D6 G6 D5S6 G5

K0 K1 K2 K15 K16 K17 K31

1st column AM1 (o/p) 2nd column AM1 (o/p) 3rd column AM1 (o/p) last column AM1 (o/p) 1st column AM2 (o/p) 2nd column AM2 (o/p) last column AM2 (o/p) drain 6 gate 6 drain 5 source 6 gate 5

CLK1 CLK2 SYNC\tDFF GND!

G2 VDDA D1S2 G1 DFFQn DFFQ_ DFFAS

34 33 32 31

gate 2 supply voltage (12V) drain 1 source 2 gate 1 output trap n (inverted, not buffered) input trap n (not buffered) after switch : signal after 1ste switch clock 1 (i/p) clock 2 (i/p) synchronization i/p ground (0V)

Probe pads: [warning: always contact bondpad 31!] The six test pads outside the AM are 108 x 108m (135m in the layout x 0.8 scaling). Arranged in two groups of three, the function is (clockwise) source, drain, and gate. Substrate connects via pad 31! The other test pads are inside the active matrices. The intention is to allow monitoring of pixel electrode voltages. The storage capacitor is this small compared to the input capacitance of a measurement probe, that precise quantitative measurements are not possible. e) Schematics of test circuits (2-stage shift register, 2 pixel layouts, test transistors) Two-stage shift register: The purpose of this circuit (part) is to locate eventual design problems; the number of measurement points is limited because the chip area significantly increases with the number of bondpads. For this layout, the largest number of bondpads fills the periphery, without exploding the chip area.

Table 6-2 : test circuit (bond-) pads

CH6-163

CH 164

Appendices

Figure 6-4 : 2-stage shift register schematic Each shift register must drive a CMOS switch; this requires two complementary control signals. This is why the output is composed of the outputs of two inverters connected in series. Both inverters simultaneously serve the purpose of inversion and of buffering. To meet the 50m layout pitch condition, it is necessary to stack two subsequent registers. If the stacking includes all of the invertors/buffers, then two metal tracks need to be drawn from the top shift register towards the corresponding switch. Placing half of the inverter/buffer below both registers frees the area of one track. This eases the layout job; note that the drawing left does not show the supply (power) lines or clock lines these need area to of course.. The controlling bondpads are: GND!, VDDA, SYNC\tDFF, CLK1, CLK2 ; the measurable nodes are : DFFAS (dot), DFFQ_ (square) and DFFQn (triangle). Two pixel layouts: The first is a copy of the pixel layout in the matrix; the second is a scaled (x3) version of it. However, the scaling cannot be perfect because the thickness of the layers does not scale. The layer thickness defines the size of stray capacitances and for this very first design, no data were available to estimate the size of the parasitic (layout) components. The test pads always have the same size (108m side) and are on a 200m grid. Test transistors: The aim is to determine the optimal pixel transistor dimensions. The test proved that the smallest transistor is the best choice. With smaller pixel transistors, one has more room for the storage capacitor, smaller s/d junction capacitance and a smaller diode area sensitive to photoconduction. The mobile charge mobility in Si is more than large enough for a speedy charging of the storage capacitor. Here too, that single GND! pad must define the substrate bias voltage.

CH6-164

Appendices
Trt 6 w=9m l=9m Trt 5 w=9m l=6m Trt 4 w=9m l=3m Trt 3 w=6m l=9m Trt 2 w=6m l=6m Trt 1 w=6m l=3m

CH6 - 165

6 G D 6

6 G S 5 5 D

5 G S 4 4 D

4 G S 3 3 D

3 G S 2 2 D

2 G S 1 1 D

! D N G

Figure 6-5 : SiLCD test transistors schematic

a) Brief design project description (aim, result) This is the first design really deserving the name micro-display backplane design; the design and re-design cover the period 1995-1996, and is referred to as the REFLEC design. The TFCGroup financed and set up this initiative. The technology chosen for is a combination of a standard 2.5m HBiMOS technology from Alcatel Mietec and a proprietary back end process fully developed within TFCG. The HBiMOS technology offers CMOS 15V transistors amongst others. The design covers a quarter of a wafer. The processing was done via INVOMEC MPW services. Note that the rectangular wafer pieces need lcos back-end processing! Each quarter wafer bears two 160x120 backplanes (n and p- type AM) and a whole string of test chips. The backplane chip dimensions were primarily determined by the loose tolerances required for the in-house cell assembly. In 1995, cell assembly was a very new subject in the TFCGroup. The pixel pitch is an enormous 80m, primarily dictated by the SiLCD projects tolerance requirements and a largely overdimensioned storage capacitor. The result is excellent. Both the back-end process as the control electronics enabled scores of successful demonstrations on many occasions. This early stage success was primordial for boosting the MOSAREL project. b) Pictures of quarter wafer mask layout, quarter wafer and chip floorplans

6.1.2 HBiMOS 1/16th VGA (160x120 display)

CH6-165

CH 166

Appendices

Figure 6-6 : quarter-wafer mask layout

Flip chip test chip Test chips, a N- and p-type AMs Test chips, b

Figure 6-7: quarter-wafer floorplan CH6-166

3x 3bit digital flip chip driver ICs

Align markers

Appendices

CH6 - 167

The backplane chip floorplan shown below is valid for both the n- as the p-type; most driver sub-cells are identical (except the pixel cell and AM border cells). Note that the bidir_SR_upper/lower and row_driver_left/right planes all use the same bidirectional scanner building block. The column driver differs from the row driver because of the extra row_of_switches_upper/lower planes.

N- or p-type AM layouts; row driver polarity swap is hard-wired in left_edge and right_edge

Figure 6-8 : 160x120 backplane chip floorplan c) Pin-out of main circuit (p- and n-type 160x120 microdisplays) The main circuit is composed of two 160x120 microdisplay backplane chips. Note that the 160x120 lcos chips are point-symmetric from an interconnection point of view. This means rotations of 180 degrees do not affect the meaning of the connection pads. This is true even for the CE (on-chip) connection pads. To obtain this symmetry, the design has double pads for some of the signals. Some pads are unique. The pad numbering is linked to the design of the very first silica carrier for these microdisplays. See Figure 6-9 for a picture of the layout of this carrier; the numbering starts with the lower-right pad and goes counter-clockwise. The carrier pads have a pitch of 0.05. The carrier pad numbers correspond with the chip pad numbers.

CH6-167

CH 168

Appendices

The pad names are not all unique to reflect the fact that pad function is independent from eventual 180-degree rotations. The pin-out list starts with the top left pad and ends with the bottom right pad.
Pad nr. 41 42 43 44 45 46 47 48 49 50 51 52 ConPad name Brief description nects [upper left column driver + upper left row driver connections] _SIR left-hand upper col scanner in (shift right) _SOL left-hand upper col scanner out (shift left) upper column driver 0V left supply upper left Video input upper column driver 5V left supply

Figure 6-9 : first version of the backplane carrier layout (5cm x 5cm)

40 VddCol 39 VssCol 33 SHr_l 32 Video

upper scanner shifts to the right when Hi

31 vddaCol GATE_H _ROW__in CLK1 CLK2 69 SH_U_D

upper column driver 15V supply

forces left row driver output to 0 when LO left row scanner input (downward scan direction) left row scanner clock signal1 (read) left row scanner clock signal2 (lock) left row scanner scans downwards when Hi

CH6-168

Appendices
53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 Pad nr. 40 39 38 37 36 35 34 33 32 31 30 29 80 _ROW__OUT left row scanner output for upwards scan direction left row scanner 5V supply 67 vddRow left row scanner 0V supply 66 vssRow 25 Vref 24 vssaCol upper column driver 'analog' ground 0V left row driver 'analog' ground 0V left row driver 'analog' supply 15V 2.5V reference voltage for scanners

CH6 - 169

62 vssaRow 61 vddaRow 20 Vbpl

[lower left column driver + lower left row driver connections] left row driver 15V supply 59 vddaRow vssaRow left row driver 0V 58 23 bulkMat 17 vssCol 56 vref 55 vssRow lower column driver analog 0V supply 2.5V refV for scanners left row driver 0V supply

backplane voltage pin, ~7.5V nominal

AM bulk connection ! 0V for ntype, 15V for ptype AM

54 vddRow row_out 52 shu_D 7 6 5

_row__in vddaCol Video sh_R_L

left row scanner input for upwards scan direction lower column driver analog 15V lower left video input

left row scanner output for downward scan direction left row scanner direction (upwards when Lo)

left row driver 5V supply

4 3

g__high _sil sor colclk1 colclk2 VssCol VddCol

lower column driver enable (enabled when Hi) lower left column scan input when shifting to the right lower left column scan output when shifting to the left lower column driver clock1 (reads on falling edge) lower column driver clock2 (outputs on rising edge) lower left column driver 0V supply lower left column driver 5V supply

lower column scanner direction (to the left when Hi)

ConPad name Brief description nects [upper right column driver + upper right row driver connections] upper right column driver 5V supply 43 VddCol 44 VssCol colclk2 colclk1 sor _sil g__high 45 shR_L upper column driver clock2 (outputs on rising edge) upper column driver clock1 (reads on falling edge) upper right column scan o/p when shifting to the right upper right column scan input when shifting to the left upper column driver enable (enabled when Hi) upper column scanner direction (to the right when Hi) upper column driver analog 15V supply upper right video input upper right column driver 0V supply

46 Video 47 vddaCol 12 shU_D

_row__in

right row scanner input for downward scan direction right row scanner direction (upwards when Hi)

CH6-169

CH 170
28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 row_out 14 vddRow 15 vssRow 16 vref 57 vssaCol

Appendices
right row scanner output for upward scan direction right row driver 5V supply

2.5V refV for scanners

right row driver 0V supply

63 bulkMat 18 vssaRow

AM bulk connection ! 0V for ntype, 15V for ptype AM right row driver 0V supply right row driver 15V supply

upper column driver analog 0V supply

19 vddaRow

[lower right column driver + lower right row driver connections] backplane voltage pin (~7.5V nominal) 60 vbpl vddaRow right row driver 'analog' supply 15V 21 22 vssaRow 64 vssaCol 65 vref 26 vssRow lower column driver 'analog' ground 0V right row scanner 0V supply right row driver 'analog' ground 0V

2.5V reference voltage for scanners right row scanner 5V supply

CLK2 CLK1 ROW_in GATE_H 71 vddaCol 72 Video 73 SHr_l

27 vddRow ROW__OUT 29 SHU_D

right row scanner clock signal2 (lock) right row scanner clock signal1 (read) right row scanner input (upward scan direction) forces right row driver output to 0 when 0 lower column driver 15V supply lower left Video input lower scanner shifts left when Hi lower column driver 5V supply

right row scanner output for downwards scan direction right row scanner scans upwards when Hi

79 VssCol 80 VddCo _SOL _SIR

lower column driver 0V supply

lower right col scanner out (shift right) lower right col scanner in (shift left)

Table 6-3 : 160x120 HBiMOS backplane pin-out d) Schematic of main circuit (down to transistor level) To start, please refer to Figure 6-8 for the chip floorplan. It is valid for both types of AMs (n- or p-type). The n-type seems more interesting because of the high(est) channel mobility. However, this aspect used to be of major concern for TFT technologies, but is nearly of no concern for x-Si lcos. On the other hand, the p-type allows the bulk of the AM to be biased without a forced connection to the rest of the chip (potential effects form substrate noise, substrate currents). In practice, experiments did not allow to link performance differences to the AM type.

CH6-170

Appendices
Row_n nmos chip bulk used within the AMN chip separate AM bulk used within the AMP chip Row_n pmos

CH6 - 171

The difference between the n- and p-type lcos chips is the transistor type used in the AM pixel circuit. The control signals for both chips are pretty much the same; the difference lies in the voltage range of the video signal and in the polarity of the row signal. The range of the video signal has an upper voltage limitation (below the highest supply voltage) with the nmos version and similarly, has a lower voltage limitation (above the lowest supply voltage) with the pmos version. The peripheral electronics must consider this range difference. Both row signal polarities are hard-wired in the layout of the row driver planes. The on-chip connections between the AM rows and the row-driver are different for the AMP and AMN backplanes as such, this is polarity issue is transparent for the peripheral electronics. Referring to Figure 6-8, the planes called upper_edge, lower_edge, left_edge and right_edge are different for the AMP and AMN designs, in addition to the AM itself. Altogether, only the AM (core) cells, the AM border cells and the chip identification cells differ. The rest of the circuitry is identical to both the n- and the p-type versions. There are double row and column drivers. The outputs of each column driver can be set to HiZ state (disconnected); the same however, is not true for the outputs of the row driver. Each row driver can independently select the HiZ state for the entire AM; however, without precautions, one row drivers output can conflict with the other row drivers output. One drawback the circuit suffers from is the over-dimensioned and serial buffering of all the signals. The excess buffering costs layouting time and area (corner cells ul, ur, bl and br for (upper-left, upper-right, etc.), while the serial nature makes the timing of the signals dependent on the scan direction. The peripheral electronics provide a work-around for this; however, for much larger display formats this work-around may be difficult to find. A clock buffer tree is essential for new(er) designs. The corner cells each provide for 20 bonding pads and signal buffering/wiring. Note that the ul and br are identical; the same is true for ur and bl. The row and column drivers use the same 10-stage wide library cells to implement the stack of bidirectional shift register, level shifter and video switch (column driver only). Completion of the design of these cells, implies completion of the design of the row and column drivers. CH6-171

Figure 6-10 : two 'REFLEC' pixel architectures

CH 172
Scan buffers scanner sir sor enable level shifter

Appendices
video switch

shDir

scEna

0 1 2 3

LS

9
to rows or to vid. switches
SL

c1b c2b

sil

sol

The scanner building cell does not buffer the scan direction signal shDir. This is ok, as one does not expect the scan direction switching speed to be relevant. To avoid the spilling of data from one row to the next, the scan enable signal scEna is buffered on the contrary (switches two times per line period). The same buffering mechanism is foreseen for the clock signals c1b and c2b. The scanner inputs are sil and sir corresponding with a shift direction to the right (scan input left hand side) and to the left respectively (scan input right hand side). The naming of the scanner outputs follows a similar format (sol and sor). The output of the bidirectional scanner is gated by an AND gate, before feeding into the bank of level shifters. The output of the level shifters drive the rows of the AM directly (row driver) or drive the bank of video switches (column driver). The video switches simply are 15V CMOS switches. The circuits for the bidirectional shift register element and level shifter are shown below. The shDir signal from fignXXX splits into two opposite signals sh_left and sh_rite for shift left and shift right respectively. The inverters marked with a 2 denote inverters that contain an additional transistor to force the output to VREF (see e.g. pad 16 in Table 6-3), this is done each time the back-to-back inverters could change state. CH6-172

Figure 6-11 : 160x120 row and column driver building block

SL
LS

SL

to cols video

Appendices

CH6 - 173

inL(N)

(N)

c1b
2

inR(N)

sh_left sh_rite

c2b
2

inR(N-1)

inL(N+1)

Figure 6-12 : 160x120 bidirectional scanner element The first stage of the level shifter consists of a passive pull-up inverter. The signal polarity is such that under normal operating conditions (only one row and column addressed at a time), there are only two passive pull-up resistors that drain current from the 15V supplies (one for the rows, one for the columns). The size of the resistance is function of switching speed (capacitive load) and of area (width of the pull-down transistor); the actual value is a compromise between the two.

lsIn

out out

Figure 6-13 : 5 to 15V level shifter schematic

CH6-173

CH 174 e) List of test chips

Nr. of instances 5 2 Name of test instance switcheschip positioningcross Description

Appendices

5 4 5 5 5 5 5 6 4 5 1 3

module_metal3_rules pmosdrampixeltestchip test_conf_whole_nmos _juli95_v3 tinv ldandlschip test_an_shr2 array_pmos array_nmos ssrchip dmosarray itc_hbimos flip2_top

test w/l video switch driven by level shifter lcos back-end process alignment markers (planar/metallization) ; there are more of these alignment markers on the lcos chips themselves test module to evaluate metal3 process (vias, widths, resistance...) pmos pixel array 28x28 test chip 16x16 nmos test active matrix with direct row/column contacts test 15V inverter, test passive pull-up level shifter line decoder test circuit, level shifter test circuit test row driver building block (scanner + row logic) test transistor array (48 different w/l pmos) test transistor array (48 different w/l nmos) test row/column scanner element (scanner + logic) experimental dmos transistor array test chip for flip-chip technology validation 3-bit digital column driver with DAC for flip chip

The other chips on the quarter wafer are test chips. The intention is to experiment novel transistor structures, flip-chip mounting, integrated DACs and to validate the silicon design as well as to validate the in-house and external Si processing. Not all test chips underwent testing, because of the success with the main lcos chips. Some of the test circuits indeed serve the sole purpose of helping with debugging the hardware. A number of test chips occur several times on the quarter wafer and in both the test chip areas to evaluate eventual performance variations, and to provide a sufficient quantity of test material. Because the quarter wafer alone sees the lcos back-end process, the performances of the chips at the edges of the quarter wafer are slightly more at risk. This is another reason to have more than one instance of a test chip. The lcos chips have no circuitry along the border of the quarter wafer. This design is a re-design on a smaller scale (quarter wafer instead of half a wafer) a so-called A1 design, A0 meaning the design is perfect from the first shot on. Table 6-4 lists a brief description of each of the test chips.

Table 6-4 160x120 HBiMOS test chip listing

CH6-174

Appendices

CH6 - 175

a) Brief design project description (aim, result) The name MOSAREL is short for Mono-crystalline Silicon Active Matrix Reflective Light Valve. The European Commission financed Mosarel as Esprit IV project EP-25340. In this text, the focus is on lcos backplane design solely. Many of the figures shown in the following pages result from patient work of a dear colleague. Project Synopsis (from the public report[]): An ultra-high-resolution projection display system based on reflective light valves (LV) with 2560x2048 pixels is developed. The main applications are graphic workstation displays (end-user Barco, Belgium) and head-up displays (HUD) (enduser Sextant, France; now part of Thales Avionique). Other potential application areas are microdisplays for personal viewing applications and virtual reality headsets. Combining a submicron CMOS technology for self-scanning active matrix addressing with a reflective nematic liquid crystal (LC) technology, the LV can be realized in existing ASIC (Alcatel Microelectronics, Belgium; now AMIS) and LC (Thomson-LCD, France ; now Thales LCD) facilities, without major investments. All necessary design and back-end processing (Imec, Belgium) and LC (Stuttgart, Germany) skills are present in the consortium. Fifteen m pixels yield small 38x31 mm Light Valves, a requirement in light, compact, cost-effective optical projection systems. Lcos backplane design plan: The project is extremely ambitious in terms of specifications. The large number of pixels has a tremendous impact on the entire project. The optical system needs to be adapted for the large cell diagonal; the video source and format conversion electronics need a specific design; the spacer-less cell assembly is another major hurdle and finally, the Si design is not standard at all because of the stitching issue. Because of the complexity and amount of hurdles, there are two major design tasks involved. A first task consists of the design of a so-called test vehicle. The intention behind the test vehicle is to elaborate, to fine-tune and to provide a means for validation of all the steps involved. The second task is to design the final demonstrator chip; this chip combines with the results from the test vehicle into the desired GXGA or 5 million pixel light valve. Project result(s): At some point during the project, the first task appeared very hard to finalize. The idea of the test vehicle proved essential to build up experience. The test vehicle cell proved hard to get functional. It never fully functioned, because of some design errors and because of difficulties with the lcos backend steps in Si processing (layers after the second metallization). The difficulty lied in localizing the origin of the problems: ?design or ?technology. The design errors consisted of 1) a missing CH6-175

6.1.3 Mosarel GXGA, SXGA and XGAp

CH 176

Appendices

design rule check and 2) errors stemming from a lack of (software) tools to check the theoretical result of both the stitching and the mix and match patterning approach. The mix and match processing sequence proved to be a bad choice: the processing creates shorts at the boundary between 5X litho areas and 1X areas. Solving this technological problem requires stepper lithography for each portion of the die. As the first step already was over time, the complete redesign of the full test vehicle mask set became impossible. Instead, a VGA resolution mini test vehicle was to provide a means for testing the technology without the mix&match and stitching tricks. This simple design never saw implementation. The design of the final demonstrator resulted in the creation of a reticule set that provides for three different chipsets. The lack of thorough software verification resulted in a number of minor design errors; most importantly, the lcos backend Si processing had major difficulties with chip planarity issues. However, the chip functionality was sufficient to prove the concept. A few cells fitted into a projection system and allowed for a limited series of tests. Lack of investment (time, money and people) led to a premature end of this project. The biggest hurdle with industrial, large scale lcos manufacturing is indeed yield. b) Pictures of mask design and chip floorplans - test vehicle : reticle set, 1X masks and floorplan - demonstrator chipset : reticle set and chip floorplans [demonstrator chipset]

Figure 6-14 : main reticle set CH6-176

Appendices

CH6 - 177

Figure 6-15 : second nwell and metal2 reticles

Figure 6-16 : main reticle floorplan

CH6-177

CH 178

Appendices

Figure 6-17 : floorplan for second nwell and metal2 reticles

Figure 6-18 :GXGA chip layout picture

CH6-178

Appendices

CH6 - 179

Figure 6-19 : GXGA module partitioning

Figure 6-20 : SXGA chip layout picture

CH6-179

CH 180

Appendices

Figure 6-21 : SXGA module partitioning

Figure 6-22 : XGAp chip layout picture

CH6-180

Appendices

CH6 - 181

Figure 6-23 : XGAp module partitioning

Figure 6-24 : mask for 6 GXGA interconnection flexes

CH6-181

CH 182 c) Pin-out of GXGA chip and eventual application notes

Left series (pads at the top of the chip) Nr Name Description 1 VBSD1 (sub)block select data input 2 VBSC1 row (sub)block select clock

Appendices

Right series (pads at the top of the chip) Nr Name Description 1 ANWB 15V analog nwell bias 2 BCKPL backplane contact; shorted to BRGij 3 VSRO1 output of left row shift register 3 DNWB 5V digit nwell bias 4 VGND1 0V digit supply left Rdriver 4 5 VSYN1 left Vsync 5 6 VCLK1 left row clock 6 7 VHLD1 left shift register hold 7 8 VVDD1 5V digit supply left Rdriver 8 9 VGNA1 0V 'analog' supply left Rdriver 9 10 VENA1 enable of left Rdriver 10 11 VVDA1 15V 'analog' supply 11 12 LRTL1 left row test line 1 12 HSYN3 BL3: see block1 (BL1) 13 LRTL2 left row test line 2 13 HCLK3 BL3: " 14 CTST1 column test bar 1 14 HENA3 BL3: " 15 CTST2 column test bar 2 15 HBSC3 BL3: " 16 HSYN1 BL1: Hsync input 16 HBSD3 BL3: " 17 HCLK1 BL1: column driver clock 17 BRG13 BL3: " 18 HENA1 BL1: coldriver enable 18 VID13 BL3: " 19 HBSC1 BL1: (sub)block select clock 19 SRG13 BL3: " 20 HBSD1 BL1: (sub)block select data 20 VID23 BL3: " 21 BRG11 BL1: backplane return ground 1of2 21 SRG23 BL3: " 22 VID11 BL1: video input 1of4 22 VID33 BL3: " 23 SRG11 BL1: substrate return ground 1of3 23 SRG33 BL3: " 24 VID21 BL1: video input 2of4 24 VID43 BL3: " 25 SRG21 BL1: substrate return ground 2of3 25 BRG23 BL3: " 26 VID31 BL1: video input 3of4 26 HVDD3 BL3: " 27 SRG31 BL1: substrate return ground 3of3 27 HVDA3 BL3: " 28 VID41 BL1: video input 4of4 28 HGND3 BL3: " 29 BRG21 BL1: backplane return ground 2of2 29 HGNA3 BL3: " 30 HVDD1 BL1: 5V digit supply 30 HSRO3 BL3: " 31 HVDA1 BL1: 15V analg supply 31 HSYN4 BL4: " 32 HGND1 BL1: 0V digit ground 32 HCLK4 BL4: " 33 HGNA1 BL1: 0V analg ground 33 HENA4 BL4: " 34 HSRO1 BL1: shift register output 34 HBSC4 BL4: " 35 HSYN2 BL2: SEE BL1 DESCR 35 HBSD4 BL4: " 36 HCLK2 BL2: " 36 BRG14 BL4: " 37 HENA2 BL2: " 37 VID14 BL4: " 38 HBSC2 BL2: " 38 SRG14 BL4: " 39 HBSD2 BL2: " 39 VID24 BL4: " 40 BRG12 BL2: " 40 SRG24 BL4: " 41 VID12 BL2: " 41 VID34 BL4: "

CH6-182

Appendices
42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 7 58 59 60 61 62 SRG12 VID22 SRG22 VID32 SRG32 VID42 BRG22 HVDD2 HVDA2 HGND2 HGNA2 HSRO2 BL2: BL2: BL2: BL2: BL2: BL2: BL2: BL2: BL2: BL2: BL2: BL2: " " " " " " " " " " " " 42 43 44 45 46 47 48 49 50 51 52 53 54 SRG34 VID44 BRG24 HVDD4 HVDA4 HGND4 HGNA4 HSRO4 CTST3 CTST4 RRTL2 RRTL1 VVDA2

CH6 - 183
BL4: " BL4: " BL4: " BL4: " BL4: " BL4: " BL4: " BL4: " column test bar 3 column test bar 4 right row test line 2 right row test line 1 15V 'analog' supply right Rdriver enable right Rdriver 0V 'analog' gnd supply right Rdriver 5V 'digit' supply right Rdriver right row driver sr hold right row clock right Vsync 0V 'digit' supply right Rdriver output of right row shift register right (sub)block select clock right (sub)block select data

55 VENA2 56 VGNA2 57 VVDD2 58 59 60 61 VHLD2 VCLK2 VSYN2 VGND2

62 VSRO2

63 SUBST 0V pwell bias contact; substrate ctct 63 VBSC2 64 AMWB 0V AM well bias; shorted to SRGij 64 VBSD2

Timing The circuit design assumes 60 Hz as nominal frame rate; the intention is to make trials with frame rates up to 100 Hz. Nominal operation: 60 Hz: Frame time: 16.67 ms (60 Hz) ; line time: 8.138 s (122.8 kHz) The front porch is selectable between 0 and 200 ns (nominal 100 ns). This value is a rough estimate. The back porch is selectable between 0 and 800 ns (nominal 400 ns). This value is a rough estimate. Column and video clock period = 47.7 ns, or approx. 21 MHz ( because of 16-fold parallelism and a guess of 500ns total porch time) Column clock period as function of the frame rate and display resolution (1s porch time) : o S,GXGA : add 6.25ns * (1-TotalPorchTime/1s) o XGAp : add 5.21ns * (1- TotalPorchTime/1s) CH6-183

Table 6-5 : GXGA pinout

CH 184
100

Appendices

GXGA SXGA XGAP

Period of horizontal clock (ns) with 1s total porch time

80

60

40

20

60

70

80

90

(Hz)

100

Figure 6-25: HCLK period as function of the frame rate Application notes: Because of the photolithographic stitching procedure, the pins numbers go from 1 to 64 for the left connector and again from 1 to 64 for the right connector. There are four identical blocks of 19 pins (L16-L34; L35-L53; R12-R30 and R31 to R49), corresponding with the four column driver segments.
Pin 01: VBSD1 Input : (sub-) block select data input for the left row driver; this is the input of the DFF shift register in. There is one DFF per block of 32 outputs of the row driver. Polarity : A 1 selects the left shift register block; a 0 selects the right block. For timing info: see descriptions of HBSC and HBSD. Pin 02: VBSC1 Input : row (sub-) block select clock for the left row driver; this is the clock of the DFF shift register. Polarity : reading of data occurs on the falling slope. For timing info: see descriptions of HBSC and HBSD. Pin 03: VSRO1 Output : end-output of left row shift register; here you expect to see the VSYN1 pulse after 2048 clock pulses Polarity : same as VSYN1 Pin 04: VGND1 Supply : 0V digital supply left Row driver Pin 05: VSYN1

CH6-184

Appendices

CH6 - 185

Input : left v-sync pulse Polarity : one "1" pulse; after that zeroes, see figure below; rise and fall times < 10ns Pin 06: VCLK1 Input : left row clock Polarity : one "1" pulse; after that zeroes; rise and fall times < 10ns Pin 07: /VHLD1 Input : left shift register hold: forces all outputs of the left row driver to zero Polarity : a voltage of 0V forces all outputs of the row driver to 0V Voltages : 5 volts = 1, 0 volts = 0 Timing : rise and fall times <10ns; the values in the figures below are measured at 50% of the maximum voltage. A change in row voltage occurs either after a falling VCLK edge or after a VHLD edge. For a rising row edge, the delay is 30ns; for a falling row edge, the delay is 66ns. Set VHLD to 1 to de-activate this signal.
>100ns VCLK 0.0V 5.0V >10ns

5.0V

VSYN

0.0V

5.0V selectN 0.0V 5.0V selectN+1 0.0V 5.0V VHLD 0.0V

30ns 12.0V

66ns

RowN 0.0V 12.0V RowN+1 0.0V 135ns

Figure 6-26 : GXGA row-driver timing data (part 1) CH6-185

CH 186

Appendices

Pin 08: VVDD1 Supply : 5V digit supply left row driver Pin 09: VGNA1 Supply : 0V 'analog' supply left row driver Pin 10: /VENA1 Input : block-enable signal of left row driver: This signal is the input of a D-FF shift register (and clocked simultaneously by the same clock signal as VBSD1: VBSC1). The state of the shift register stages determines if the high voltage outputs of the 32output driver sub blocks are enabled. Polarity : A 1 corresponds with a non-enabled output (high Z state) Pin 11: VVDA1 Supply : 15V 'analog' supply left row driver, 15V is absolute maximum and 12V is the nominal value. Pin 12: LRTL1 Output : (left row-driver test line 1): end output of Block Select static shift register. Here you expect to find the VBSD1 signal delayed by 64 clock pulses of VBSC1. Polarity : same as VBSD1 Pin 13: LRTL2 Output : (left row-driver test line 2): end output of Enable static shift register. Here you expect to find /VENA1 signal delayed by 64 clock pulses of VBSC1. [Polarity : same as /VENA1] Pin 14 : CTST1 HVinput : (column test bar 1) : controls a switch connecting column 1 with column 2, column 5 with column 6, column 4n+1 with column 4n+2 for testing purposes. This pin also connects to pin R50 (CTST3) Polarity : a "1" makes the connection. Normally, this signal is "0".

Voltages : 15 volts = "1", 0 volts = "0"

VHLD 0.0V 5.0V

5.0V VCLK 0.0V 30ns RowN 0.0V 12.0V RowN+1 0.0V ~100 ns overlap 12.0V 66ns

Figure 6-27 : MOSAREL timing data (part 2) CH6-186

Appendices

CH6 - 187

Pin 15: CTST2 HVinput : (column test bar 2): controls a switch connecting column 3 with column 4, column 7 with column 8, column 4n+3 with column 4n+4 for testing purposes. This pin also connects to pin R51 (CTST4) Polarity & voltages: see description of pin14, CTST1. Pin 16: HSYN1 Input : Hsync input Polarity : a "1" at the beginning of every line time; then "0" Voltages : 5 volts = "1", 0 volts = "0" Timing : rise and fall times ~3.5ns ; the clock period @ 60Hz framerate and 500ns porch time = 47.7ns.The sync pulse is delayed by +- 15ns with respect to the HCLK pulse train. See picture of simulation input sync and clk waveforms below.

Pin 17: HCLK1 Input : column driver clock Polarity : data reading occurs on the falling slope; keep low to lock data. Timing : rise and fall times ~3.5ns; GXGA period @ 60Hz frame rate and 500ns porch time = 47.7ns. Delay between VIDx1 and HCLK1: assume the four VIDx1 signals are set at the values needed for the first set of four columns. The first set of four column switches are passing the VIDx1 signals to the columns up to +- 63ns after the falling edge of HCLK1.See the table below for temperature dependency : e.g. wait 63.5ns @58C after the first falling edge of HCLK1 before starting to change VIDx1 to the values needed for the second set of four columns. Repeat changing VIDx1 at the same pace as HCLK1 of course.

Figure 6-28 : GXGA column-driver timing data

CH6-187

CH 188
VCLK(falling edge) to new VIDxy : time distance between VCLK and switch-gate signals going down below 0.5V
66 Min. delay in ns 65 64 63 62 61 60 -20 -10 0 10 20 30 40 50 60 70 80 90 10 0 Temperature (degrees Celcius)

Appendices

Series1 Linear

Figure 6-29 : temperature dependence of clock - video switch delay

Gate signal of first block of columns 62.3n s

Start pulse (HSYNC)

HCLK

Start activating the next block of columns

Figure 6-30 : video switch delay simulation at 25 C

Pin 18: HENA1

CH6-188

Appendices
Input

CH6 - 189

: column driver segment enable; if this signal is low, the outputs of the segment connect to the on-chip backplane voltage and to each other. Polarity : "1" = driver enabled; "0" = columns connected to backplane voltage. Set this pin to 5V for normal operation of the column driver. Timing : when switching this signal, a delay of a few frame periods occurs before an eventual change to come into effect. Pin 19: HBSC1 Input : (sub-) block select clock for the first column driver segment; this is the clock of a D-FF static shift register. Polarity : data reading occurs on the falling slope; keep low to lock data. Timing : rise/fall time <1ns, e.g. 100ns cycle time (10Mz); data must not change during falling HBSC edges. Pin 20: HBSD1 Input : (sub)block select data input for the 1st column driver segment; this is the input of a D-FF static shift register. There is one DFF per block of 32 outputs of the column shift register; corresponding with 128 columns. Hence, in every column driver segment, there are 5 DFF outputs in total. Polarity : A 1 selects the bottom shift register; a 0 selects the top one. Timing : 5ns rise/fall time, e.g. 100ns cycle time (10Mz). Data must not change during falling HBSC edges. Hi->Lo transitions must ONLY occur when HBSC is Hi, take sum of rise and fall times as margin. Pins 21, 29: BRG1,2 block1 Supply : backplane return ground, shorted to BCKPL Pins 22, 24, 26, 28: VID1,2,3,4 block 1 Input : video input 1,2,3,4 of4 Pins 23, 25, 27: SRG1,2,3 block 1 Supply : substrate return ground, 0v guard around video input lines Pins 30-33: HVDD1, HVDA1, HGND1, HGNA1 Supply : respectively 5V, 15V, 0V, 0V Pin 34: HSRO1 Output : end-output of dynamic shift register for this column driver segment; here you expect to see the (inverted) HSYN1 pulse after 160 clock pulses. Polarity : inverted with respect to HSYN1 Pins 35-53: see description of pins 16-34 (column driver block 1) Pins 54-62: not connected Pins 63-64, 1-3 (2nd connector): SUBST, AMWB, ANWB, BCKPL, DNWB Supplies : respectively, 0V pwell bias, 0V AM well bias to be shorted with SRGij, not connected, backplane contact to be shorted with BRGij, 5V digital nwell bias. Pins 4-22, 23-49 (2nd connector): column blocks 3 and 4, see corresponding pins 16-34 of block 1 Pins 50-64 (2nd connector): see corresponding pins 1-15 of first connector

CH6-189

CH 190

Appendices

d) Schematic of main circuit (down to transistor level) general architecture of the GXGA backplane, redundancy, enable signals, details of the column driver, details of the row driver General architecture of the GXGA backplane: The driver is integrated around the 2560x2048 active matrix.
Column driver (LV+HV)

Row driver LV Row driver HV

Row driver HV
HGND

Active Matrix
2560 x 2048

Column test

Above the matrix, there is a single column driver, consisting of 4 identical segments each counting 640 outputs. Each of these segments can be disabled, meaning that all 640 outputs are connected to backplane voltage. Every segment receives 4 parallel data lines that are connected to 4 consecutive columns at a time, as shown in Figure 6-32.
VID1-4 BRG SRG1-4 HSYN HCLK HENA HBSC HBSD HVDD HVDA HGNA HSRO

Figure 6-31 : GXGA block schematic

160 bit shift register 2 3 n 160

On the left side of the matrix, there is a complete row driver with 2048 outputs. On the right side, there is another complete row driver with 2048 outputs. These row drivers can be used simultaneously or one of them (or both) can be disabled (meaning that the outputs are put in a hiZ state). The enabling can be done per 32 outputs of every row driver (see below)

Figure 6-32 : schematic of column driver segment

CH6-190

Row driver LV

Appendices

CH6 - 191

Redundancy: ROWS+COLUMNS: If there is a problem in one of the shift registers, a big part of the display will not function. Therefore, the shift registers in row and column drivers are doubled. In other words, a primary and a secondary shift register are put in parallel.
D Q Q en D Q Q en D Q Q en

32 bit SR ... 32 bit SR ...

32 bit SR ... en 32 bit SR ...

32 bit SR ... en 32 bit SR ...

en

32 bit Level Shifter ...

32 bit Level Shifter ...

...

32 bit Level Shifter ...

These shift registers are divided in small units (blocks) of 32 outputs. Every unit can be selected from either the primary or the secondary shift register. This is schematically illustrated in Figure 6-33. In this figure, a third (static) shift register is seen, consisting of D-flipflops, that selects between the primary and the secondary shift register blocks. This Block Select shift register must be initialised before the driver can be used. For this purpose, the (H/V)BS(C/D) Block Select lines are used (see pinout list).

Figure 6-33 : shift register redundancy

Enable signals COLUMNS: The 4 column driver segments can individually be enabled or disabled. In the disabled state, all 640 HV outputs are connected to the backplane voltage. There are 4 HENA signals, one for every segment. The primary purpose of the column disable function is to shut off the display in a proper manner: first, the column driver is disabled; while the row driver remains active, all pixels are automatically brought on backplane voltage; if the backplane voltage is then slowly brought back to zero, the pixel voltages will follow the backplane voltage until it is zero. Then all other signals can be switched off as well.

CH6-191

CH 192
DSROUT 1 HENA DSROUT 2 DSROUT 160 HENA

Appendices

(NOR)

/EN

TO LS 1

TO LS 2

TO LS 160

/EN

Figure 6-34 : enable circuit inside a column driver segment

VVDA

/V E N A,N TO R OW n

/V H LD D S R O U T ,n

ROWS: The High Voltage outputs of the row drivers can be enabled per block of 32 outputs. A non-enabled block puts its HV outputs into high impedance state. In order to decide which blocks are enabled, the VENA signals are used. They are used the same way as the Block Select signals (i.e. the enable bits are stored in a static shift register with one output per block). In fact, the Block Select shift register and the Enable shift register share the same clock (VBSC). Therefore, the VBSD and the VENA bits should always be supplied simultaneously. A second HOLD signal forces all row driver outputs to 0; however, this signal has a lower priority compared to the VENA signals.

Figure 6-35 : row driver enable logic

Detailed description of the column driver The following pages describe in a top-down order the column driver in more detail. The first drawing shows a complete driver segment (640 outputs), which comprises the 5 shift register sub-blocks (32 outputs each), an enable circuit, a level shifter array and 2 switch arrays. Each of these components will be elaborated further in the following drawings.

CH6-192

Appendices
HCLK HBSC HBSD HSYN HENA
HENA BSD IN BSC HCLK

CH6 - 193

BSD OUT OUT

BSD IN

BSC

HCLK

BSD OUT OUT

BSD IN

BSC

HCLK

BSD OUT OUT

Column driver SR sub-block

DSROUT1 2 32

Column driver SR sub-block

DSROUT1 2 32

Column driver SR sub-block

DSROUT1 2 32

HSRO

DSROUT1 2 TOLS1 TOLS1 TOSW1 2 2 2

32 32 32 32

33 33 33 33

34 34

/EN /EN

Enable circuit
64 64

64

129 129 129 129

130 130 130 130

160 160 160 160

HENA /EN /EN /EN OUT

34

VID1-4 BCKPL

/EN OUT

34

162 Level Shifters

64

160x4 switches 640 switches

columns 1-128

columns 129-256

columns 513-640

The second drawing shows the shift register sub-block. It is clear that the state of the BSD static shift register influences the clock generation. If the Q output of the BSD D-FlipFlop is "1", the clock generator for the bottom shift register is working normally, while the clock generator for the top shift register is in the "phi2 low" state, which means that phi2="0" and /phi2="1" all the time. This can be verified on the detailed drawing of the clock generator circuit (see below). This results in the top shift register having high impedant outputs and the bottom shift register working normally.

Figure 6-36 : 640 o/p column driver segment block schematic

CH6-193

CH 194
BSC BSD
D C Q /Q

Appendices
HCLK BSD OUT
PHI2 low HCLK PHI2 low

Clock generator
PHI1 /PHI1PHI2/PHI2

HCLK

Clock generator
PHI1 /PHI1PHI2/PHI2

PHI1 /PHI1PHI2/PHI2

PHI1 /PHI1PHI2/PHI2

PHI1 /PHI1PHI2/PHI2

IN

CD_sr unit cell

IN

OUT

CD_sr unit cell

IN

OUT

CD_sr unit cell

IN

OUT

1
CD_sr unit cell
IN OUT PHI1 /PHI1PHI2/PHI2 IN

2
CD_sr unit cell
OUT PHI1 /PHI1PHI2/PHI2

...
IN

32
CD_sr unit cell
OUT PHI1 /PHI1PHI2/PHI2

OUT

DSROUT1

DSROUT2

DSROUT32

Figure 6-37 : column driver shift register sub-block

/PHI1

/PHI2

IN

OUT

PHI1

PHI2

In these drawings, the internal circuit of the clock generator (6-39) and the shift register unit cell (6-40) is shown. The shift register is a classic dynamic SR.

Figure 6-38 : dynamic SR cell

CH6-194

Appendices

CH6 - 195

phi2low HCLK

/phi2 phi2

phi2low HCLK

/phi1 phi1

Figure 6-39 : clock generator circuit The following drawing shows the level shifters. The first stage is always a resistive load invertor. These circuits use the high voltage power supply (HVDA). Also the /EN signals are level shifted.
/EN TO LS 1 TO LS 2 TO LS 160 /EN

/EN OUT

TO SW 1

TO SW 2

TO SW 160

/EN OUT

The drawing below shows the video switches. Every "TO SW" signal controls 4 switches, connecting the 4 VID signals with 4 consecutive columns. Note that column 1 is connected to VID1, column 2 to VID2, and so on.

Figure 6-40 : 162 level shifters

CH6-195

CH 196
TO SW 1
VID 4 VID 3 VID 2 VID 1

Appendices
TO SW 2 TO SW 160

1 2 3 4 5 6 7 8

640

The last drawing in this section shows the switches that connect the columns with the backplane voltage if the /EN signal is high (HENA low).
/EN BCKPL 1 2 3 4 640

Figure 6-41 : 160 x 4 switches

640

Figure 6-42 : 640 switches Detailed description of the row driver

VCLK VBSC VBSD /VENA VSYN
BSD BSC VCLK /ENA IN BSD OUT /ENA OUT OUT BSD BSC VCLK BSD OUT /ENA OUT OUT BSD BSC VCLK BSD OUT /ENA OUT OUT

RD SR sub-block

/ENA IN

RD SR sub-block

/ENA IN

RD SR sub-block

64

LRTL1 LRTL2 VSRO

/VENA,1 DSROUT1 2

32

/VENA,2DSROUT1 2

32

/VENA,64 DSROUT1 2

32

/VHLD
/VENA,1 DSROUT1 2 /VHLD 32 /VENA,2DSROUT1 2 /VHLD 32 /VENA,64 DSROUT1 2 /VHLD 32

RD o/p logic and LS

TO ROW1 2 32

RD o/p logic and LS

33 34 64

RD o/p logic and LS

2017 2018 2048

rows 1-32

rows 33-64

rows 2017-2048

The row driver is in many ways similar to the column driver, but there are also some important differences, as shown in the schematics below.Figure 6-43 shows the complete row driver (2048 outputs), consisting of 64 shift register sub-blocks and 64 output logic-and-levelshifter blocks with 32 outputs each,. The Figure 6-35 shows the output logic and level shifter sub-block per output. Note that unlike the HENA CH6-196

Figure 6-43 : row driver with 2048 outputs

Appendices

CH6 - 197

signal in the column driver, the /VENA enable signal is not common to all subblocks, and that there is a /VHOLD hold signal that has no equivalent signal in the column driver. The row driver shift register sub-block is almost the same as the column driver shift register sub-block, except that there are 2 static shift registers; one for the Block Select Data (BSD) and one for the Enable data. The building blocks (clock generator and shift register unit cell) are identical to the ones used in the column driver.
BSC BSD /ENA
D C Q /Q D C Q /Q

VCLK BSD OUT /ENA OUT

HCLK PHI2 low HCLK PHI2 low

Clock generator
PHI1 /PHI1 PHI2 /PHI2

Clock generator
PHI1 /PHI1 PHI2 /PHI2

PHI1 /PHI1 PHI2 /PHI2 IN OUT IN

PHI1 /PHI1 PHI2 /PHI2 OUT IN

PHI1 /PHI1 PHI2 /PHI2 OUT

IN

Row driver shift register unit cell

Row driver shift register unit cell

Row driver shift register unit cell

1
Row driver shift register unit cell IN OUT IN

2
Row driver shift register unit cell OUT

...
IN

32
Row driver shift register unit cell OUT

OUT

PHI1 /PHI1 PHI2 /PHI2

PHI1 /PHI1 PHI2 /PHI2

PHI1 /PHI1 PHI2 /PHI2

/VENA,x

DSROUT1

DSROUT2

DSROUT32

Figure 6-44 : row driver shift register sub-block

CH6-197

CH 198 e) Schematic of test vehicle

12 35 36 8 extra 45 64 1 2 3 BP 1 2 2 6 10 16 23 BV 20 41 42 43

Appendices

64

13

16

Figure 6-45 : MOSAREL test vehicle schematic f) Pin-out of test vehicle and eventual application notes
Left connector (from left to right)
Pin function 1 ground (not connected at chip side) 2 row block 16 even 3 4 5 6 7 8 10 11 12 13 14 15 16 17 18 19 9 row block 16 odd row block 15 odd row block 15 even row block 14 even row block 14 odd row block 13 even separate row nr. 2 (replaces 1st row of block 13 odd) row block 12 even row block 12 odd row block 11 even row block 11 odd row block 10 even row block 10 odd row block 9 even row block 9 odd row block 8 even row block 13 odd Pin 1 2 4 6 3 5

Right connector (from left to right)

function column block 10 even backplane voltage column block 11 even column block 12 even column block 13 odd column block 13 even column block 14 odd column block 14 even column block 12 odd column block 11 odd

7 8 10 11 13 12 9

column block 15 odd

14 15 16 17 19

separate column nr. 2 (replaces 1st column of block 16 odd) column block 16 odd column block 16 even column block 17 odd

column block 15 even

18

column block 17 even column block 18 even column block 18 odd

CH6-198

Appendices
20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 row block 8 odd row block 7 odd row block 7 even row block 6 even row block 6 odd row block 5 even row block 5 odd separate row nr. 1 (replaces 1st row of block 5 odd) row block 4 even row block 4 odd row block 3 even row block 3 odd row block 2 even row block 2 odd row block 1 odd 21 23 20 22 column block 19 even column block 20 even BV row block 9 odd column block 20 odd column block 19 odd

CH6 - 199

24 25 26 27

row block 9 even

row block 10 odd

28 30

29

row block 10 even row block 11 even row block 11 odd

31 32 34 36 37 38 39 40 41 42 43 44 45 47 33 35

row block 12 odd row block 12 even row block 13 odd separate row nr. 2

row block 1 even ground (connected to chip ground) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) extra (diffraction tests) column block 1 odd column block 2 odd column block 1 even column block 2 even column block 3 odd column block 3 even column block 4 odd column block 5 odd column block 4 even column block 5 even separate column nr.1 (replaces 1st row of block 13 odd) column block 6 odd column block 7 odd column block 6 even column block 7 even column block 8 odd column block 8 even column block 9 odd column block 9 even

row block 14 odd row block 14 even row block 15 odd row block 16 odd

row block 13 even

row block 15 even row block 16 even

ground (connected to chip ground) column block 20 even column block 20 odd column block 19 even column block 18 even

46

column block 19 odd

48 49 50 52 51 53

column block 18 odd column block 17 even column block 17 odd column block 16 odd column block 16 even separate column nr. 2 (replaces 1st row of block 13 odd)

54 55 57 59 56 58

column block 15 even column block 15 odd column block 14 odd column block 13 odd

column block 14 even column block 13 even

60 61 63 62

column block 12 even column block 12 odd column block 11 odd column block 11 even

column block 10 odd

64

ground (not connected at chip side)

Table 6-6 : MOSAREL test vehicle pin-out

CH6-199

CH 200

Appendices

6.1.4 TMDC XGA

a) Brief design project description (aim, result) A company called tmdc (Taiwan Micro-display Corporation) ordered this design. Actually, this company does the cell assembly and intends to sell lcos panels with the electronics, light sources and optical components. The company also secures access to a top foundry in Taiwan. The silicon technology from UMC provides a CMOS process on 8 wafers that combines 3.3V-0.35m and 18.0V-3m transistors (UMC EHV35 3.3V/18V 2P5M PSUB/POLYCIDE GOX 65/465). The 18V range is necessary considering the continuous illumination, no CE switching architecture employed. The projects target is a 60Hz 1024x768 lcos panel for use in a three-panel projection system. After fine-tuning some of the masks, excellent performance resulted. The threepanel projector served as demo projection system with XGA resolution. Commercial applications and customer demand however, push for higher display resolutions. The control electronics show operation at frame frequencies well above 60Hz this is a way to reduce the visibility of flicker. Safety margins used at design time explain this increase in performance The eight-inch wafers carry more than 50 chips. Each chip measures 22.0mm by 19.4mm; this dimension includes the width of the 600m-wide sawing street. The sawing street contains test-key patterns, align markers In other words, the circuit dimensions are 21.4mm by 18.8mm. The circuits split into test circuitry and the useful backplane circuit. The test circuitry is accessible through independent probe pads, before cell assembly. The backplane circuitry has two row drivers for redundancy. XGA cell bondpads Column driver Right row driver Left row driver

Active matrix (AM)

Test circuits Figure 6-46 : tmdc XGA block diagram CH6-200

Appendices

CH6 - 201

b) Picture of mask design and chip floorplan The test circuits are located in (three of) the four corners of the chip: see Figure 6-46. The upper left corner contains no test circuits for the XGA. Note that the test circuits are not necessary for the operation of the lcos. The corner areas are layouted in such a manner that they can lie in the wafer handling zone, and thus be nonfunctional, hereby increasing the die count per wafer. The backplane circuitry itself contains three large blocks: the active matrix (AM), the column driver and the row driver. The AM counts 1024 by 768 pixels, plus 32 extra rows and columns for easy alignment of the 3 (color) images. The pixel dimension of 17.6m corresponds with a maximal usage of the available reticule area. This is to maximize the light throughput. Therefore, the AM measures 18585.6m by 14080.0m. A 400m wide, square donut surrounds the AM at the mirror level. This AMborder connects to the outside via a separate bondpad (B10). Except the bondpads, the AM and the AMborder, the entire chip is filled with floating squares. The bondpads are located at the top of the chip (reference orientation). The lower left corner of the AM is at 1707.2m from the left chip edge and at 2085.0m from the lower chip edge. The middle of the AM is 575m below the middle of the chip; horizontally, both centers align. Vertical alignment of the centre coordinates of chip and AM is not possible, because of requirements from both the cell assembly process and the wire bonding process. These requirements primarily stem from tolerances on the CE glass positioning, from the CE glass thickness and bonding head size. top
Nr. 1 Nr. 10 Nr. 35

left

scan down RWDOWN=Hi

scan to the left SHRI=Lo

Nr. 43

right

bottom

The bonding pads for the main backplane circuitry are at the top of the chip, in between two alignment markers. These markers are necessary for cell assembly by HANA corp. The pad numbers range from 1 to 43; four of them are not CH6-201

Figure 6-47 : XGA assembly floorplan

CH 202

Appendices

implemented however. The pad size is square, 100m per side; the pad pitch is 300m at least. c) Pin-out of main circuit and eventual application notes Description
3.3V supply

Bond pad descriptions: Name Nr Type

Lgnd Lpsup3 Ldisa B3 B2 B1 Bias

Supply

Supply

0 volt substrate/ground supply

LSOP /RWCLK /HOLD ----Lpsup18

B4 B5 B7 B8 B9 B6

Output Input -NC-NCSupply Input

Disables left row driver when Hi ; blocks /HOLD and /RWCLK for this driver ; transition to Hi only allowed when /RWCLK is Hi Left shift register output ; pulse width is multiple of /RWCLK period Row clock ; RWSYN is read in on rising edge of this signal

Force active row drivers outputs to 0 when Lo ; to avoid row overlap No connection : pad not implemented No connection : pad not implemented 18.0V supply

AMborder Cpsup18 Cpsup3 Cgnd

B12 Supply B14 Bias B13 Supply B15 Output

B11 Supply

B10 Bias

Metal strip around the display matrix ; black level voltage 18.0V supply 3.3V supply

Amstref CSOP CSNC SLD1 VID1 VID2 VID3 VID4 SLD2 SLD3 SLD4 SLD5

0 volt or black level (if this is DC)

0 volt substrate/ground supply

/ACPU

B17 Input B18 Bias B20 Bias B22 Bias B24 Bias B26 Bias B19 Video B21 Video B23 Video B25 Video B27 Input

B16 Input

Column shift register output ; pulse width is multiple of CCLK period Column sync input (width of Lo dip =< CCLK period) Clamps/Pulls all 4 video lines to ACMX when Lo

0V shield for video inputs ; on-chip connection to substrate and SLDi Column voltage for columns 1+4*k , k=0,1,2,3, Identical to SLD1 Identical to SLD1 Identical to SLD1 Ientical to SLD1 Column voltage for columns 2+4*k, k=0,1,2,3, Column voltage for columns 3+4*k, k=0,1,2,3, Column voltage for columns 4+4*k, k=0,1,2,3,

/ACPD SHRI

ACMX CCLK

B28 Supply B30 Input B29 Bias

Clamps/Pulls all 4 video lines to 0 volt when Lo Most positive clamping voltage (=< psup18)

Column shift register shifts to the right when Hi Column clock ; CSNC is read in on falling edge

CH6-202

Appendices
Cgnd Cpsup3 B32 Supply B34 Bias B36 Bias B33 Supply B35 Supply B37 Input B31 Supply 0 volt substrate/ground supply 3.3V supply Black level 18.0V supply 18.0V supply

CH6 - 203

Cpsup18 LiShield Rpsup18

Rwdown RWSYN -----

Active row register shifts down when Hi No connection : pad not implemented No connection : pad not implemented

B38 -NCB39 -NCB41 Bias

Row sync pulse (width less than one /RWCLK period)

RSOP Rdisa Rgnd Rpsup3

B40 Output

B42 Supply B43 Supply

Right shift register output ; pulse width is multiple of /RWCLK period Disable right row driver when Hi ; see B3 3.3V supply 0 volt substrate/ground supply

Application notes: [remarks, qualitative waveforms, timing data] Important remarks: Pad numbering starts at the upper left pad from nr. 1 to nr. 43, although there are only 39 pads implemented: pads marked with a x in the drawing (paragraph 1.1) are not implemented. Signals of type input see pin-out table are all 3.3V CMOS and have +5ns rise and fall times; logic low (=0V) is noted as Lo and logic high (=3.3V) is noted as Hi. HiZ stands short for the high impedance state. There are 3 different supply voltages: logic 3.3V, analog 18V (nominal) and clamping level 11V maximum. Clock signals ideally have 50% duty ratio; this is evident for the column clock signal, for the row clock there is a lower limit to the pulse width. The column driver has 4-fold parallelism: there are 1056/4=264 column clock pulses during each line period. Therefore, the column scanner is shorter than the row scanner (800 stages without parallelism). Depending on the RWDOWN and SHRI signal levels, scanning occurs from top to bottom (RWDOWN=Hi) and from left to right (SHRI=Hi). Internal multiplexers make sure the scanner input and outputs are rerouted to the corresponding pads. Top is at the side of the backplane bondpads. Scan direction is not to change often during operation of the display. There are two row drivers, each driving all of the 800 rows. The only control input with separate pad is the row driver enable. The other row control signals have common pads. Normally one must not enable both drivers simultaneously. When both row drivers are active simultaneously, there is a risk of conflicting row driver outputs. These conflicts can permanently destroy the cells functionality! Especially at startup, one should take care to disable both drivers from the very beginning on (power-up, Ldisa[3] and Rdisa[41] should thus be Hi). Another solution consists of activating the HOLD mode that forces all the row-driver outputs to be Lo. In a CH6-203

Table 6-7 : tmdc XGA main circuit bondpad description list

CH 204

Appendices

later stadium the shift registers should be put into a known state (e.g. all 0), only one driver being enabled at a time. Do not drive several rows at once: the 18-volt supply track widths are not dimensioned for these higher current levels. It therefore is safer first to empty the row drivers shift registers before turning on the 18V supplies: while a driver is active, the 18V drivers are activated as well and current is flowing through the 18V supply tracks. Transistor threshold does not allow video voltages larger than 11V. Qualitative waveforms a) Left or Right Row Driver signals; two consecutive rows are simultaneously active during the transition from one to the next row (/HOLD is Hi).
/RWCLK RWSYN Row nr. 001 Row nr. 002 001 002 003 800

0d 0p 1 2 3 4

ovlp
Row nr. 800

fall rise

R/LSOP

Frame(n)

Frame(n+1)

The disable signal is Lo to allow the activation of the rows and /HOLD is Hi: see next paragraph for the descriptions of L/Rdisa and /HOLD. The important timing parameters are: the minimum time between the edges of the sync pulse and the rising edge of /RWCLK : 1, 2 the delay between this rising /RWCLK edge and the row signal edges : 3, 4 the rise and fall times of the row signals : rise, fall the amount of overlap between two consecutive rows : ovlp b) Effects of /HOLD and L/RDISA signals; the disable function has priority, puts the outputs of the row driver into HiZ state and blocks /RWCLK and /HOLD: (comparison between 3 possible combinations)

Figure 6-48 : qualitative row driver signals

CH6-204

Appendices
Other rows Active row R/Ldisa /HOLD t=0, overlap between successive rows t=0, /HOLD cannot influence driver o/p

CH6 - 205

5
t=0, /HOLD forces driver o/p to be 0 volts

Timing parameters are: row response time to /HOLD signal : 5, 6 row response time to L/Rdisa signal : 7, 8

Figure 6-49 : /HOLD and L/RDISA signal combinations

c) Column driver signals The circuit concept for the column scanner is simpler than for the row driver. There is no hold or disable function. The settling time for the column voltage increases with the voltage difference between the initial and final state. The rise time of the column signals is larger than the fall time, for equal absolute voltage change.
CCLK CSNC 00 1 00 2

11

00 3

26 4

9 10

Sw.block 1 2 3

CSOP Row(k) Row(k+1)

Timing parameters are: the minimum time needed between the edges of the CSNC sync pulse and the falling edge of CCLK : 9, 10

Figure 6-50 : column driver signals

CH6-205

CH 206

Appendices

for every new line, time distance between the second falling CCLK edge and the moment the video data for the second block of 4 columns can be applied to the video inputs VID1,2,3,4 : 11 .

d) Active clamping signals

CCLK /ACPU /ACPD VIDx ACMX

If the active clamping circuit is used, the effective time available for scanning the columns is equal to the line period minus the time needed to clamp the video busbars to either gnd or ACMX. Timing parameter data (from simulations)
Param 0p,d 1 2 3 4 ovlp rise fall 5 6 7 8 9 10 11 12 Relating to /RWCLK Values >120ns >120ns >160ns 107ns 120ns 59ns 80ns 64ns 68ns 72ns Comments Pulse/Dip width minimum ; 50% duty cycle ideal Measured at 50% voltage level Measured between voltage /RWCLK and 17.5V rowN and 0.5V rowN From 0.5V to 17.5V Overlap when _HOLD is Hi

Figure 6-51 : active clamping signals

/RWCLK, RWSYN /RWCLK, RWSYN /RWCLK, rowN /RWCLK, rowN rowN, rowN+1 rowN rowN

/HOLD, all rows /HOLD, all rows L/Rdisa, all rows L/Rdisa, all rows CCLK, CSNC CCLK, CSNC CCLK, VIDx CCLK

t [ /HOLD at 0.3V and row at 0.5V ]

From 17.5V to 0.5V

t [ /HOLD at 3.0V and row at 17.5 V]

>10ns >=30ns >15ns

Measured at 50% voltage level T=0.0C ; for T=100C, add 8ns

CCLK period; 50% duty cycle!

Table 6-8 : simulated timing data (XGA) CH6-206

Appendices d) Schematic of main circuit: AM, column- and row drivers

CH6 - 207

The active matrix: The active matrix counts 1056x800 pixels; each pixel is composed of a single transistor switch and a storage capacitor. The storage capacitor is a poly-poly capacitor in parallel with metal-metal capacitors. These metal-metal capacitors connect to the pixel electrode at one terminal and to several fixed-voltage connections at the other terminal (CE, SB and LS: counter electrode (~Vmax(video)/2.0) , substrate (0V), light shield (e.g. CE voltage). See Figure 6-52 below.
column(k)

row(r)

disabl e
Parasitic cap between pixel electrode and row line

Parasitic cap between row and column busbars

sb Cst Cce Csb Cls sb : substrate ls : light shield

st : storage (main capa) ce : counter-electrode or lc

The column driver: The column driver features 4-fold parallelism to address the AM. Therefore, there are four video lines feeding into quad video switches (blocks of four video switches). A video switch is made of a single nmos transistor and is driven by a 3.3V->18V level shifter. This level shifter also drives the other three switches in the block. A bidirectional scanner (shift register) of 1056/4=264 stages drives the level shifters. A (set of) clock signal(s) controls the scan speed; the load on the clock lines necessitates a clock buffer tree. The clock buffer tree splits into eight equal blocks, making the column driver to split into eight column blocks (CBx, x=1->8). Each CBx drives 132 columns. See below for a drawing. The bidirectional nature of the scanner requires the synchronization signal CSNC to be input at one of the two ends of the scanner (signals RSYN and LSYN). In principle, a 1->2 mux can accomplish this task; the direction signal SHRI controls to which end the sync pulse is input. With a 2->2 mux however, the output of the scanner can be made available for testing through a dedicated pad.

Figure 6-52 : pixel schematic

CH6-207

CH 208
CCK1,2 Rite,Left CBx Rite,Left clk buffers 33 bidir shift regs 33 level shifters 132 video switches 132 columns

Appendices

CCK1,2 Rin/Lout VID1-4 Lin/Rout

VID1-4

Rin/Lout

Lin/Rout

132 columns

VID1-4

VID1-4

The bidirectional nature of the scanner also requires a re-ordering of the clock signals; that is why the RITE and LEFT signals are needed for the generation of the non-overlapping clock signals CCLK1 and CCLK2. The building block that generates the clock signals is called clock separator with this design.
CSN C CSOP RITE LEFT CCLK CCK1 CCK2 LSYN RSYN LSYN clamping VID1-4 SLD1-5 ACMX /ACPU /ACPD

Figure 6-53 : block schematic of one CBx

RSYN

CB1

CB2

CB3

CB8

Col 1-132

Col 133-264

Col 265-396

Col 925-1056

Figure 6-54 : block schematic of the column driver The last sub-block not discussed yet, is the active clamping circuit. This block adds a fixed voltage to the video signals per frame. The video data voltage signal splits into a threshold voltage and a modulating voltage. The peripheral electronics generate the modulating voltage signal outside the chip; it is added to the clamping voltage through an external coupling capacitor. This trick reduces the voltage range of the video op-amp in the peripheral electronics. CH6-208

Appendices

CH6 - 209

The two following pages give details on all the sub-blocks. The clock buffer tree: dir and not-dir determine the shift direction: the order of clock signals must be adjusted to the scan direction. Two tracks distribute CK1 and CK2 over the entire column driver. CK1 and CK2 are buffered and inverted per column block CBx. In below, the portion left of the blue dotted line is the clock separator (one per column driver); the portion to the right are the circuits for buffering, inverting and re-ordering of the clock signals. per CBx, the clock signal path is further split up in two; in total there are 16 buffers for each of the signals 1, 2 and their respective inverted companions.
CCK1 dir CCLK

2
dir dir

1
dir dir

dir

dir

dir

dir

dir

CCK2 x8 (1 per CB)

x2 (2 per CB)

Figure 6-55 : column-clock buffer tree The active clamping circuit: the inputs /ACPU, /ACPD and ACMX are common to the 4 video inputs two nmos transistors can short the video lines to 0V or to ACMX; this last ones voltage range is low enough to be switched with a single trt switch the level shifter contains a 3.3V buffer/inverter: all control signals are 3.3V signals

CH6-209

CH 210
3.3V ACPU VIDx ACPD level shift 3.3V 18V level shift 18V

Appendices

ACMX

Figure 6-56 : active clamping circuit The 2-2 mux: dir and not-dir determine the shift direction the shift direction cannot be determined solely from the CSOP (column scanner output) signal
CSOP CSNC

dir dir

RSYN

bidirectional shift register (264 stages)

LSYN

Figure 6-57 : sync-in and -out plus buffers The bidirectional shift-register: It is composed of two interwoven shift-registers, one for each shift direction. The registers are made of tri-state inverters; the direction signals (dir, /dir) and the clock signals connect to the enable pin of the 3-state inverter. Note that when changing the shift direction, /f2 takes over the function of f1, f2 the function of /f1, etc. The ordering of the tri-state inverters in the layout affects the Si area consumption of this building block. CH6-210

Appendices
2 dir dir 2 1
Rin out Lin

CH6 - 211

1
in enable/ disable

disable/ enable in enable/

Figure 6-58 : a bidirectional shift-register element The level shifter: the level shifter is inverting (odd number of inversions) the 18V buffering is necessary to allow for the rather large capacitive load (the gates of four video switches)

1i to_LS

2i

3i

4i

5i to_4switch

18V

Figure 6-59 : column driver level-shifter The row-driver: The row driver is similar but not equal to the column driver. There are no videoswitches and no clamping circuits in the row driver. There is no parallelism either. However, there are two row driver blocks for redundancy and the adressing of two subsequent rows is not always desirable. Thus, the scanner counts 800 stages, can be disabled (output can be set into the high impedant state, or disconnected from the rows) and features a hold function (force all of the AM into the off-state). The rowdriver splits into 10 yRBx row blocks, each having 80 18V outputs. The y stands for L (left) or R (right).

CH6-211

CH 212
hold, disable

Appendices

hold, disable

CK1,2 down,up

CK1,2 down,up

Din,Uout yRBx

Uin,Dout

Din,Uout

clk/hold buffers 80 bidir shift regs 80 logic blocks 80 level shifters 80 rows Uin,Dout

80 rows

Figure 6-60 : block schematic of the row blocks

/RWCLK /HOLD LDISA Lhold

Lrck1, Lrck2

Rrck1, Rrck2 Rhold

RDISA

RSOP RWSYN LSOP

RSHDO, RSHUP RUSYN RDSYN LDSYN LUSYN LSHDO, LSHUP RwDown

Figure 6-61a : block schematic of the (double) row driver

CH6-212

Appendices

CH6 - 213

RSHDO, RSHUP Rhold Rck1, Rck2

LDISA Lrck1, Lrck2 Lhold LSHDO, LSHUP

LRB01 R001-080 LRB02 R081-160

R001-080 RRB01 R081-160 RRB02

LRB10 R721-800

R721-800 RRB10

Figure 6-62b : block schematic of the (double) row driver Each yRBx contains a bidirectional shift register of 80 stages. The outputs of the shift registers combine with the hold and enable signals to drive a bank of level shifters. The outputs of these level shifters directly connect to the rows : the outputs switch between 0V, 18V and HiZ. As with the column driver, a buffer tree for the clock and hold signals is desirable as well. To limit the number of pads, the signals up-down, hold, sync and rwclk are shared between the left and the right row driver. The enable signals and the outputs of the 2-2 muxs are not shared. Note that the hold and rowclk signals are overruled by the enable signal: the state of a row driver can be frozen by means of the enable signal alone and independently from the other driver. For the clock separator, the 2-2 mux and the details of the bidirectional shift register, please refer to the column driver description. The only difference worth noting is that the clocking speed is much lower; therefore, the row driver speed allows for a design that generates a larger separation between the non-overlapping clock signals. The buffering tree is similar to the one from the column driver, except that at the last inverting buffer, there are five copies per yRBx block (compared to 2 per CBx).

RDISA

RDSYN

RUSYN

LUSYN

LDSYN

CH6-213

CH 214
18V

Appendices

disable

toRow

SRo/p hold

The only big difference is with the logic and level shifter blocks: by separate driving of the output nmos and pmos, one can obtain the HiZ output of the row driver. This necessitates blocking the shift register output and the hold signal. When enabling the driver, the hold signal overrides the shift register output to force the output to 0V. e) List of test circuits and test pad configuration The table shows the list of test circuits, the corners where they are located and the corresponding probe pad numbers.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Name of test circuit Pixel trt 3/3 Pixel trt 3/3 + m4 3.3V inv 56|28 18V inv 36|12 Video switch 6x50/3 HiZ 3st inv 3|1.5 Active 3st inv 3|1.5 Row clock separator Column level shifter 3.3V protection diode (power) 3.3V protection diode (gnd) 18V protection diode (gnd) 2-2 MUX (a) 2-2 MUX (b) Row logic (disabled) Corner LL LL LL LL LL LL LL LL LL LL LL T01, T02, T09, T10 T21, T22, T27, T28 T03, T04, T11, T12 T05, T06, T13, T14 T07, T08, T15, T16 T17, T18, T23, T24 T19, T20, T25, T26 T29, T30, T31, T34, T35 T37, T38, T39, T41, T42 T33, T40 T32, T36 Test bondpad numbers

Figure 6-63 : row driver logic and level shifter

LR LR LR LR

T43, T51 T48, T49, T50, T56, T57, T58 T67, T68, T69, T70, T71, T72 T59, T60, T61, T62, T63, T64,

CH6-214

Appendices
16 17 18 19 20 21 22 23 24 25 26 Row logic (enabled) 2NOR gate 3.3nmos 3.5/0.35 3.3nmos 7.0/0.35 3.3pmos 7.0/0.35 3.3pmos 14/0.35 18nmos 12/3 18nmos 48/3 18pmos 36/3 18pmos 72/3 18pmos 3.4/6 LR UR UR UR UR UR UR UR UR UR LR

CH6 - 215
T65, T66 T44, T45, T46, T47, T52, T53, T54, T55 T73, T74, T75, T76, T77 T78, T79, T80, T81 T84, T85, T88, T89 T82, T83, T86, T87 T94, T95, T100, T101 T92, T93, T98, T99 T90, T91, T96, T97 T102, T103, T108, T109 T104, T105, T110, T111 T106, T107, T112, T113

Table 6-9 : XGA test circuits and corresponding pads

UPPER LEFT
108 102

UPPER RIGHT
109 103 110 104 96 90 76 73 70 67 74 71 68 66 62 57 49 58 50 77 75 72 69 41 37 34 29 23 17 09 01 42 38 35 30 24 18 10 02 39 36 31 25 19 11 03 32 26 20 12 04 33 27 21 13 05 28 22 14 06 15 07 16 08 40 111 105 97 91 112 106 98 92 86 82 113 107 99 93 87 83 100 94 88 84 80 78 101 95 89 85 81 79

63 59 51 43 52 44 53 45 54 46

64 60 55 47

65 61 56 48

LOWER RIGHT

LOWER LEFT

Figure 6-64 : corner floorplan and numbered probe pad configuration

CH6-215

CH6 - 216

Appendices

6.2 Hands-on dracula

The aim is to introduce swiftly the Dracula verification software package. Note that the aim of the text is not completeness, it is rather a guided tour to highlight important points the author came across. Hopefully for the interested reader, this results in shorter learning times and eventually it results in a clearer understanding of layout verification. All the examples assume the so-called 'FLAT' mode. This mode does not take advantage of the casual repetition in the layout. This results in rather large data files (temporary and final), and therefore results most probably in longer computation times ("run times"). On one occasion, a DRC job ran in hierarchical mode. It indeed resulted in shorter run times. However, the job setup time proved to be much longer, partially because of a lack of experience and partially because the layout must be correctly partitioned into two levels. The designer must manually promote some layout cells to so-called 'HCELLs'. Incorrect HCELL choices can lead to much longer run-times. Automatic partitioning of the layout data would probably be an excellent tool to make this check mode more attractive and really" time efficient. Spreading the job over several computers of the network, so-called "distributed processing", has been looked at. However, I could not figure out a correct setup. The other options still have to be tested.

6.2.1 Introduction: general remarks

Dracula is the name of a program that automates the verification of IC layouts. Such programs are essential, because manual verification is very time consuming and does not (always) result in a 100% error free design. An in depth check with verification software will require some time anyway, not only because it also requires a thorough understanding of the software. A complete check always splits in a sequence of verification jobs. A typical example of such a sequence could be: (order does matter) mask generation, design rule checks, electrical rule checks, layout versus schematic comparison, layout parameter extraction/parasitic resistance extraction. Often used abbreviations for these steps are MASKGEN, DRC, ERC, LVS and LPE/PRE. This sequenced approach helps untangling heaps of otherwise confusing error messages. Once all checks are successful, the layout data format needs conversion into a socalled fracture format. Eventually this conversion goes together with a mask bias step1. This step is sometimes referred to as MASKPREP. Usually the foundry itself
1

Note that if dracula is used to generate mask layouts for TCAD (MASKGEN), it is necessary to include the mask bias for correct process simulation.

CH6 - 216

Appendices

CH6 - 217

or by a mask making company performs this step; furthermore proprietary test key patterns, alignment markers and/or process validation modules (PVMs) need to be added to the mask layout at this stage. The dracula software can do this data conversion (mask fracturing). This text will only mention it, though. After each correction, all of the previous checks have to be re-done: layout verification is an iterative process. As designs often contain repeated cells, one expects repeated error messages. As a result, the verification software sometimes produces a real jungle of error messages. The only efficient way is to tackle the first error message and start over. After some iteration, no more error messages show up. However, this does not imply the design is error free! Indeed, this depends on the completeness of the check sequence and correctness of layout data. As an example, checking the maximum width of polygons on a 'via' layer is insufficient to guarantee all 'vias' are squares of exactly X2 m2. A badly positioned text label is another example of how to jeopardize the error sensitivity. The wrong position can generate false errors but can also fail to generate a painfully pertinent open/short circuit error message. Error sensitivity is function of the correctness of both design data and input deck! For each job, the program input consists of design data (layout polygons and/or circuit netlists) and an 'input deck'. The input deck is an ASCII text file containing lists of verification instructions. It is composed of three main blocks: a description block specifying the name of the layout database, search paths of executables, run directory, a list of program options... ; an input layer block translating references to layout layers between the different layout databases (GDSII->Dracula) and defining (eventually derived) conduction layers an operation block containing commands for layer operations, rule checks and netlist extractions. The foundries distribute the input decks; these files are templates that require some additions/modifications. There are (computer) system dependent variables that must concord with the users system. Examples are the search path to the program executables or details of mask definitions. Sub-micron technologies often offer more than five interconnect layers, but not all designs do require all of them. Then the input decks need concordant modifications. Everyday practice has shown input decks assuming an opposite layer tone or that mask X corresponds to some other layer number. Before starting a job, it is mandatory to verify thoroughly the input decks themselves, and to understand the verification commands. One option is to write the input deck completely from scratch, but then there is a liability issue. It therefore is reasonable to put it this way: it absolutely is no 'click and play' or easyto-use, user-friendly software. The next paragraph describes the input deck more in detail for the 'FLAT' run mode. The details of other run modes are not treated (e.g. 'HIER' mode). The examples can serve as templates to create new input decks. The third paragraph explains the creation of run files and the evolution of a dracjob.

CH6 - 217

CH6 - 218

Appendices

6.2.2 The input deck: 'description', 'input layer' and 'operation blocks
A dracula input deck is an ASCII text file that describes one single step (job) of the check sequence. A 'decent' check is composed of several jobs (>3) so several different input decks are needed. Typically, the string of jobs starts with mask generation (MASKGEN) and a design rule check (DRC). The 'design rules' are process (foundry) dependent and defined by the foundry. One could be tempted of checking the mask data prior to the mask generation: less data manipulations results in time gain. Beware this is a fata morgana. Indeed, layer operations can produce unexpected results, eventually leading to fatal (DRC) errors. The safest way is first to generate all the mask information and second to perform a DRC on the complete mask data. With full custom IC design, DRCs are probably the most often executed jobs as they check larger and larger portions of the chip layout.
description input deck input layer operation program options input data info layer definitions conductor def's layer operators bias,logic,selection,maskgen,stitching design rule check (DRC) electrical rule check (ERC) layout versus schematic (LVS)

extraction related ops

layout parameter extraction (LPE)

parasitic resistance extraction (PRE)

The third job is typically an electrical rule check (ERC); say, a consistency check on drawn devices, interconnections and corresponding labels. A fourth job compares the desired/simulated circuit with the circuit extracted from the layout: this is a layout versus schematic comparison (LVS). Further jobs perform extraction of parasitic components (capacitors and resistors): layout parameter extraction and parasitic resistance extraction (LPE/PRE). Eventually, the extracted parasitic components are included in the schematic for a final proof. A last job performs the mask bias and mask fracture steps. Although (a) stitching job(s) is (are) a special case of MASKGEN, it (they) will be treated separately because the data have to be prepared and organized in a specific way see chapter two.

Figure 6-65 : block diagram of paragraph 6.2.2

6.2.2.1 The 'description' block

This first block contains the name of the layout data file and optional program settings. The following example gives an overview of often-needed statements and/or options. Note that it is not the aim to give a detailed description of all CH6 - 218

Appendices

CH6 - 219

possibilities for completeness, please refer to the online manuals (e.g. "openbook" or cadence help menus). The semicolon indicates the start of a comment. Some options are in comments, as only some specific jobs need them. Finally note that the description block must be enclosed within a "*DESCRIPTION", "*END" pair.
*DESCRIPTION ; start of description block ;ABORT-SOFTCHK= YES ; abort on soft connection check violations CHECK-MODE = FLAT ;/HIER/... ; flat/hierarchical dracula checking mode CSEN = NO ; case sensitivity is turned off: to avoid conflicts ; between any of spice, gdsout, cdlout and dracula FLAG-ACUTEANGLE= YES ; check for acute angle polygons FLAG-NON-45 = YES ; check for edges not on multiples of 45 deg FLAG-OFFGRID= YES ; check for polygon coordinates not on grid FLAG-SELFINTERS= YES ; check for polygon self-intersect ; the FLAG- commands help to avoid problems with the mask fracturing ; /GET add2description.txt ; PDRACULA command to read in another file ;GROUND-NODE= GND* ; define ground node labels ;HCELL = shift_reg_elem ; define cell 'shift_reg_elem' as hierarchical cell ;NOT-HCELL = via1 ; exclude this cell from hcell list ;HCELL-FILE = mudis_hc.txt ; name of include file containing hcell defs ;HCELL-RULE = ; hcell rule file (see manuals) INDISK = $HOME/cad/myDesign.gds ; filename of input database ;INMAG/OUTMAG= 5 ;/10/2/3/; in/out magnification factor - MASKPREP KEEPDATA = YES ; (NO) ; store any intermediate data; (do not) ;KEEPDATA = INQUERY ; store for viewing with the "inquery" tool of the layout editor LISTERROR = YES ; show errors in output log ;MIRROR =X ;/Y ; flip layout around horizontal/vertical axis MODE = EXEC NOW ; execution mode ;MODEL = MOS[N],N MOS[PH],PH ; spice transistor model name definitions OUTDISK = $HOME/cad/DRC_uDisplay.gds ; file name of output database OUTPUT-ONE-LAYER=NO ; /YES: ERC error polygons on the same layer / not ;PG = MEBES ; fracture format type ;PGE-COMPACT= FULL ; default PG data compaction ;POWER-NODE= VDD* ; define power node labels PRIMARY = udisplay ; name of top cell in design hierarchy PRINTFILE = dracjob ; log file of dracula run PROGRAM-DIR= /ecad/cadence2000/tools/dracula/bin ; executable path (system) RESOLUTION = 0.001 MICRON ; max. accuracy of input db ROTATION = 90 ;/270/0/180/; rotate layout dracula uses bandscan ; algorithms: best performance with minimal band length

Figure 6-66 : example of a description block (template)

CH6 - 219

CH6 - 220

Appendices

6.2.2.2 The 'input layer' block

The input layer block is enclosed between a statement pair "*INPUT-LAYER", "*END". This block performs two important and delicate tasks: translation of layer names and layer numbers : read-in of layout data definition of (derived) conduction layers and of labeling priorities To start, it is necessary to realize there are differences between the database (dB) formats used to represent the layout data. Three dBs are concerned: Cadence, Dracula and GDSII. Each of the three databases uses a different representation mechanism for both layer and text data. GDSII intervenes here, as it is the format used to transfer designs to the foundry (or mask house). Translation to GDSII format thus occurs at least once. Either (a) text string(s) or (a) number(s) represent a layer. Cadence uses two text strings, known as 'layer-purpose pairs'. Thus, there are two (sub-) strings defining a layer in Cadence. The GDSII standard works with pairs of numbers: layer number and datatype number. With GDSII, the numbers' range is limited to [0-63]. Note there is no absolute relation between a Cadence purpose name and a GDSII datatype number. Finally, Dracula refers to data with single so-called polygon layer names.
Layout database Cadence-Virtuoso Dracula GDSII Ref.type Layer name, purpose name Polygon layer name Layer number, datatype number layer table input layer block

The exchange data between these three databases requires the use of translation tables. One task of the input layer block is to define the translation into the Dracula polygon layer names. Translation between Cadence and GDSII can be done by means of a layer translation table file and the Cadence ''Physical stream In, Physical stream Out" (PIPO) software module. The SYSTEM/SYSIN and SYSOUT statements in the description block control the in- and output formats for Dracula. The default setting for SYSOUT is the same as SYSTEM or SYSIN2. If Dracula inputs a GDSII file, two translation operations are required: one to translate the Cadence dB into a GDSII dB and a second to translate the GDSII dB into the Dracula format. This potentially is a disadvantage, because two translations cost more time than one, and above all, because the probability of translation table errors doubles. Yet it seems to be a good way of working. Of course, it is essential to know precisely the layerNumbers and associated function of the layer. Some layers contain direct mask information, others logic data from which mask data can be derived, yet others contain both. Each foundry has it own layer-numbering method. However, a practical experience has shown
2

Figure 6-67 : summary of layer translations

CH6 - 220

'SYSTEM' and 'SYSIN' are equivalent

Appendices

CH6 - 221

differences in layer numbers between two input decks from the same foundry for the same technology! Thus, in practice, one must not assume there is a fixed relation between a technology and a set of layer numbers. This would be the most logical situation though. Let the reader guess how much time the author (and probably many others) lost in finding this weird little bug Dracula can also read a Cadence dB directly; this seems simpler and safer because there seems to be only one translation. Again, practical experience has shown documentation from a foundry assigning mask numbers 13 and 14 to layers with Cadence names 'LYR' and 'noLYR' (a probable heritage from older versions of the technology or of its documentation, it really has nothing to do with number 13). To make 'sure' wrong interpretation would result, it was not all clear whether the mask type was light working field or dark working field Thus, watch your layers, no matter whether there are one or two layer translations! Another reason to work with GDSII as intermediate format is that transfers of mask data to the mask house are often done by means of GDSII files (personally, I always did). Running checks on the same tape-out files is safer! Additionally, conversion of Cadence format to GDSII allows smashing several layer-purpose pairs onto the same GDSII (layer, datatype) number. The Dracula software can perform such operations too, but it is likely to cost much more time and computing resources. The following layout trick (see Figure 6-70) explains the usefulness of this possibility: connections on a metal1 layer can represent ground, power, signal or floating nodes. Using different purposes in cadence to draw the different connection categories (ground, power, signal or floating) helps avoiding undesirable shorts with full custom design. Each layer-purpose pair can indeed have a distinct visual representation. E.g. blueblue polygons represent metal1 signal connections, blue polygons with a red border represent power nets on metal1, those with a green border are ground nets on metal1, etc. Finally, the GDSII format is independent from Cadence software/dB versions; this is interesting from the point of view of archiving. For the remainder of the text, it is assumed GDSII is always used as intermediate format. Sometimes it can be necessary to combine the results from a previous dracula job with new layout data. The statements IMPORT and EXPORT serve this purpose. They read in, respectively generate files of the form *.DAT in the dracula run directory (the '*' wildcard stands for any valid polygon layer name). A statement like 'IMPORT prevmask' expects the presence of a file 'prevmask.DAT' in the run directory. For reasons of simplicity, it is best to run a job containing 'IMPORT' statements in the same run directory as the run directory of the job containing the corresponding 'EXPORT' statements. To summarize there are three ways dracula can read in data: PLNameA = layerNr datatypeNr. ; reading a GDSII dB file IMPORT = prevMsk ; reading a polygon layer file PLNameB = layerName. purposeName ; reading cadence dB Some of the polygon layers represent conducting layers (conductors). Definition of which polygon layer names represent conductors is done within the CONNECTLAYER statement. Note that the conductor's name(s) eventually represent the result from (a) layer operation(s). Each set of connected conductors is associated with a CH6 - 221

CH6 - 222

Appendices

net name defined either automatically during extraction or by manually placed labels. Manually placed labels are an excellent means to define the net names of important nodes: otherwise, (different) subsequent runs can yield different names for the same physical net. Eventually, the generated names are just meaningless, making it hard to analyze the extracted netlist data afterwards. The TEXT= statement in the input-layer block defines on which layer(s) labels must be placed. Please notice that manual placement of labels is time consuming and only useful to inspect polygon data corresponding to specific nets (ERC), to increase the number of 'initial correspondence pairs (LVS run), to extract parasitics on a specific net (LPE/PRE), to analyze (simulations of) extracted netlists more easily. Because GDSII does not link text objects to polygon objects, it is necessary to have a clear convention on how to interpret the labels. It must be clear with which set of connected conductor polygons a label must be associated. The position of a label with respect to some polygon is essential for correct interpretation; the fact that polygons on different and thus eventually electrically isolated layers can overlap, creates the necessity for a labeling priority rule. Labels are associated with a polygon according to the following rules: the reference coordinates3 of the label must fall within or on the edges of at least one conductor polygon. If not, that specific label is ignored. This is the 1st rule. Watch out when defining wells as conductors! when more than one conductor polygon satisfies the first rule, the layer priority is used to assign the label. The layer priority is implicitly defined in the CONNECT-LAYER statement, unless otherwise specified by the TEXTSEQUENCE statement finally, each label is assigned only once to a polygon; note that a single polygon can be assigned several, hopefully identical labels The last polygon layer name in the 'CONNECT-LAYER' list has the highest priority unless a TEXTSEQUENCE statement overrides this priority ranking. The TEXSEQUENCE statement can prohibit specific conductor layers to be assigned labels. This is very useful with microdisplay backplane layouts, because the uppermost metal (mirror) layer is organized as a matrix of square electrodes covering the entire chip. However, most of the signals run on tracks underneath this top-level conducting layer. Assigning a label to some polygon underneath is very difficult without exclusion of the mirror layer from the priority (by a correct TEXTSEQUENCE statement).
NOT AND POLY POLY POLYRES POLYRES POLYC PYRES ; conductor poly ; poly resistor

Figure 6-68 : double usage of poly layer

Reference coordinates of labels can be viewed in the Cadence 'Virtuoso' layout editor by activating the 'Layer origins' button in the Display Options window.

CH6 - 222

Appendices

CH6 - 223

The names of conductor layers do not necessarily have to be one of the layers present in the layout database. Conductor layers can be so-called derived layers, which result from a combination of layers. Figure 6-68 is an example of a conductor layer resulting from a layer operation: the 'POLY' layer in the GDSII file represents connections unless a special 'poly resistor' layer overlaps it. No (portions of) polygons on the poly layer covered by the POLYRES layer can be conductors. Below are examples of an input layer block and corresponding lTable.
*INPUT-LAYER; start of input layer block imaginary process CHIP = SUBSTRATE 15 ; rectangle enclosing all data + 15 m = chip boundary WELL = 1 ; gds layer number 1 is interpreted as well layer DIFF = 2 ; gds nr. 2 is interpreted as thin oxide regions POLY = 3 ; polysilicon gate material layer NINP = 4 ; n+ implant layer PINP = 5 ; p+ implant layer CTCT = 6 ; contact windows in ILD layer MET1 = 7 ; first metal layer VIA1 = 8 ; contact windows in IMD1 layer '; TEXT = 60 ; layer containing label information TEXT = 61 ATTACH MET1 ; labels on layer 61 attach only to met1 polygons ;EXPORT = WELL ; stores polygon data in a file 'WELL.DAT'(KEEPDATA=YES)' ;IMPORT = WELL ; reading dracula data (file='WELL.DAT') from a previous run ; import & export cmds can be used for 'virtual stitching' CONNECT-LAYER = PWEL NWEL PSD NSD POLY MET1 ; (4 derived layers) TEXTSEQUENCE = POLY MET1 ; consider labels only on poly, met1, *END ; end of input layer block

Figure 6-69 : example of an input layer block

;*** layer translation table *** ; semicolons start comments ; Cadence GDSII ;Lname Lpurpose Lnumber datatypeNr. nwell drawing 1 0 active drawing 2 0 poly drawing 3 0 polyres drawing 3 0 ; merge if no res extraction is done nplus drawing 4 0 pplus drawing 5 0 contact drawing 6 0 metal1 drawing 7 0 ; metal1 layout data on layerNumber 7 metal1 ground 7 0 ; other boundary color for m1 gnd lines metal1 supply 7 0 ; same for VDD lines drawn on metal1 via1 drawing 8 0 metal2 drawing 9 0 ';... text drawing 60 0 ; text layer for labeling ;m1txt drawing 61 0 ; text attaching to m1 only comm drawing 62 0 ; layer used to add comments ;...

Figure 6-70 : example of a layer translation table

CH6 - 223

CH6 - 224

Appendices

6.2.2.3 The operation block

Layer operators (MASKGEN, STITCHING, DRC, MASKPREP) Extraction related commands (DRC, ERC, LVS, LPE/PRE)

Layer operators (MASKGEN, STITCHING, DRC, MASKPREP) Syntax of layer operation statements Mask biasing : e.g. for maskprep Layer AND/NOT/OR logic : e.g. for virtual stitching Maskgen : automatic generation of (additional) mask data Selection operators : e.g. for design rule checks Syntax of layer operation statements General form: <operation><[options]> <layerList> <parameter> <derivedLayer> where:
<OUTPUT pgFileName layerNr> <comments>

: operator name; see the manuals for complete lists : they modify the precise effects of <operation> : list of (polygon) layer(s) (names) operated on : equation/dimension parameter (keyword/number) : name of the derived layer (result) : writes the result to the output in the format defined by the SYSTEM and/or SYSOUT statements in the description block pgFileName : string of maximum 6 characters defining the first part of the name of the output cell layerNr> : two-digit numbers defining the second part of the name of the output cell; the full name thus becomes 'pgFileNamelayerNr' (concatenation). This number also defines the GDSII layerNumber. One topcell contains all output cells. With GDSII as output format, the name of the topcell is given by the concatenation of the string "out" with the name of the input topcell. This input topcell name is defined by the 'PRIMARY' statement in the description block. <comments> : starts with a semicolon and is very useful as explanation for some of the DRCs (DRC command statements are not always intuitive). INQUERY also uses this comment string as description of the error polygons.
<operation> <[options]> <layerList> <parameter> <derivedLayer> <OUTPUT

CH6 - 224

Appendices

CH6 - 225

; defines transistor gates: diffusion and gate AND DIFF POLY gates ; error: contact without metal1: CT and (not M1) - output cellName is NOMETC54 NOT CTCT MET1 badct OUTPUT NOMETC 54 ; merging polygon layers gatePLY and resiPLY into a single allPLY polygon layer OR gatePLY resiPLY allPLY ; oversize via4 by 0.1um in al directions ; the GROW command is similar to, but not identical to SIZE see manual for details SIZE VIA4 BY 0.1 VIA4B ; big via4 ; spacing check : find areas between metal1 polygons narrower than 1.2 microns EXT[H] MTL1 LT 1.2 OUTPUT ERR62 45 ; min. space < 1.2

Figure 6-71 : example of layer operation statements

Mask bias: e.g. for maskprep Usually, the foundry or mask maker performs the mask bias step together with the addition of testkey patterns and alignment markers. The aim of mask biasing can be to anticipate effects from processing. Not all masks necessarily need biasing. A typical example is the compensation of the thermal diffusion of nwells: heat cycles during processing cause the nwell to 'grow'. To obtain nwell contours as the designer wants, an under-sizing (negative bias) of the polygons is required. It is possible that some masks need a positive bias (process/foundry/working field tone dependent).
; simple bias statement SIZE NWELLO BY -1.3 NWELL ; nwell undersized into final nwell

Usually, the biasing occurs just before the mask fracturing (i.e. translating into a format used by mask maker) and this step is transparent for the designers. However, when stitching is involved, it is not straightforward to implement biasing correctly. For LF masks that need biasing, the module separating polygons must be pre-biased to make sure the edges end up right. Be aware of the artifacts resulting from sizing of non-orthogonal edges. These artifacts arise when the amount of over-sizing is greater than or equal to half the smallest of the minimum spacing, width or notch dimensions for that particular mask. The layout rules should be such that the usually rather small bias does not lead to these artifacts. The paragraph on maskgen gives more details.

Figure 6-72 : a basic mask bias operation

CH6 - 225

CH6 - 226

Appendices

Layer logic (and, not, or) : e.g. for stitching The logic operators most often used are 'AND', 'OR' and 'NOT'. The 'NOT' operator computes the overlap of one layer with the complement of the 2nd layer.
; simple layer logic: via1 not covered with metal1 NOT VIA1 MTL1 BV11 OUTPUT DRC6 45 ; via1 without m1

The basic layer operators used with virtual stitching procedures simply are 'OR' and 'AND'. Nothing special, except that the source data must be organized in a specific manner (chapter two). With positive resists, polygon data that represent areas to be etched must be 'ORed' (digitized area tone = light; working field tone = dark). Polygon data that represent areas to remain untouched (digitized area tone = dark) must be 'ANDed'. Intuitive examples are via/contact layers (to be ORed: etched areas), respectively metal layers (to be ANDed: untouched areas).
*input-layer gds01= 1 ; nwell gds02= 2 ; active IMPORT = lyr01 ; previous nwell printed pattern IMPORT = lyr02 ; previous active printed pattern *end *operation ; calc result from consecutive prints OR gds01 lyr01 lout01 OUTPUT NWELL 01 ; nwell- working field = dark, OR AND gds02 lyr02 lout02 OUTPUT ACTIV 02 ; active- working field = light, AND *end

Figure 6-73 : the 'not' layer operator

Figure 6-74 : layer logic - basic stitching operations

Maskgen: automatic generation of (additional) mask data

; super simple generation of pwell mask: excerpt of an operation block NOT CHIP NWELL PWELL ; pwell is whole chip without nwell areas

For many technologies, one can compute some of the mask layouts from the layout data of other masks. Usually, the number of drawing layers is larger than the number of masks. A combination of several layers determines how a mask finally looks like. Remember the example of the merging of several layer-purpose pairs into one GDSII number. It is evident that this provides a tool to simplify the designers' job: although the designer must cope with more layers, the layouts are easier to maintain and the design rules are easier to deal with4. The [Figure 6-76 -> Figure 6-80] figures below explain a typical and weird sequence. Besides generating the complete layout of some masks, it can be necessary to modify the layout of a specific mask. An example is the addition of so-called dummy patterns to increase the fill factor of a layer (to homogenize the ICs' (breaking down of complexity, an excellent example is the mask generation of the active mask in the I2T100 process from AMIS-Oudenaarde)
4

Figure 6-75 : a simple pwell mask generation

CH6 - 226

Appendices

CH6 - 227

planarity) typical for processes using CMP. The fill factor must be uniform on a scale smaller than the integration distance of the CMP process. The next few lines explain the apparently bizarre sequence "not, size CD, size +CD, and, not". The AND" operation is needed to cut undesirable artifacts outside the original dummy pattern. The last NOT removes artifacts that grew into areas forbidden for dummy generation. The designer indeed desires three things simultaneously: a dummy exclusion area augmented by oversized m1 polygons, no artifacts from sizing operations and no features below a given minimum critical dimension (CD).
; dummy fill deck ; add description block here *INPUT-LAYER M1L = 10 ; m1 layout data M1NOD = 30 ; m1 dummy exclusion IMPORT = PAT1 ; m1 dummy pattern covers all of the chip, ; the file "PAT1.DAT" must exist in rundir *END ; of input-layer block *OPERATION SIZE M1L BY 2.0 NODUM1 ; dummies >2 away from existing m1 OR NODUM1 M1NOD NODUM ; add to m1 dummy exclusion areas NOT PAT1 NODUM M1DUM ; OUTPUT M1D 10 ; raw M1 dums SIZE M1DUM BY -CD M1DUMS ; eliminate features <2CD ('SMALL') SIZE M1DUMS BY CD M1DUMB ; get remaining pgs back to original ; size ('BIG') AND M1DUMB M1DUM M1DLBF ; limit to original m1 dummy pattern NOT M1DLBF NODUM M1DUMF ; M1DUMF metal 1 dummy 'FINAL' OR M1L M1DUMF M1 OUTPUT M1MASK 10 ; merge&output new m1 mask *END ; of dummy fill operation block

Figure 6-76 : dummy pattern generation to homogenize the metal1 fill factor

Be very careful with size operations: 'size' moves edges; and all (portions of) edges falling completely inside a polygon disappear. A sequence of size +X followed by a size X will fill narrow gaps (notches) having a dimension smaller than or equal to 2X. This can be a wanted effect. However, with edges not along multiples of 90 degrees, surprising patterns can result. The next drawings show how the opposite sequence, size X, size +X, can also yield surprising and undesirable results!

CH6 - 227

CH6 - 228

Appendices

+X

-X

Figure 6-77 : +x, -x fills narrow 'notches', ok

+X

-X

Figure 6-78 : +x, -x results in 'undergrow; DRC min notch width should be > 2x

-X

+X

Figure 6-79 : -x,+x eliminates too thin (portions of) patterns, ok

-X

+X

Figure 6-80 : -x, +x results in 'overgrow'; solved by 'anding' with the source

CH6 - 228

Appendices

CH6 - 229

Selection operators: e.g. for design rule checks Often used commands are: 'AREA', 'ENC', 'EXT', 'SELECT', 'WIDTH'. For the detailed description of each operator and all their options, see the manuals. It is good practice to compare the topological rules documentation from the foundry with its implementation. It is also useful to add comments at the end of DRC output sequences as these can be read with the INQUERY tools: when viewing the results of a verification run with INQUERY, error polygons are highlighted and described by either the verification statement or by the comment if present. As a designer might not know the dracula semantics by heart, and as some selection operations can be lengthy and referring to meaningless intermediate polygon layer names, comments are most welcome for swift analysis and correction of the errors.
SELECT WIDTH EXT[H] ENC[TOE] AREA NWELL INSIDE HV HNWEL ; define high voltage nwell DIFF LT 3.0 OUTPUT Er09 45 ; min width of mosfet = 3u VI1 LT 2.0 OUTPUT Er10 45 ; via1 spacing < 2u VI1 M1 LT 1.0 OUTPUT Er11 45 ; m1 ovlp via1 < 1u METAL1 RANGE 0 0.3 SMT1 OUTPUT Er12 45 ; m1 polygon too small

Figure 6-81 : examples of selection operators

CH6 - 229

CH6 - 230 Extraction related commands (DRC, ERC, LVS, LPE/PRE) node dependent design rules establishing connectivity device extraction typical ERC commands typical LVS commands Node dependent design rules
EXT[H] EXT[N] EXT[NR'] AND NWELL LT 2.0 OUTPUT Er13 45 ; absolute minimum spacing is 2DBU ; the 'N' option bans equipotential wells from following check NWELL LT 4.0 OUTPUT Er14 45 ; 4DBU (disconnected nwells only) NWELL LT 7.0 TCNWL ; 7DBU (disconnected nwells) AND... TCNWL HNWELL ERHNW OUTPUT Er15 45 ; being Hvoltage nwell = error : HV wells must be connected

Appendices

The reason why DRC still shows up in this paragraph is the fact that some design rules depend on netlist information. An often recurring example is the minimum separation distance between two nwells (assume p-doped wafers). When the two nwells are always at the same potential (connected nwells), this distance can be shorter than when the wells can be at arbitrarily different potentials. In the example, EXT[H] checks NWELL polygons for an absolute minimum spacing of 2 microns (assuming micron is the database unit, DBU) and outputs the violation area. The next EXT[N] checks unconnected nwells for a minimum spacing of 4 dB units. The third EXT[NR] outputs unconnected nwells less than 7u apart to the TCNWL polygon layer. TCNWL polygons in a high voltage region generate an error cell Er1545. Thus, unconnected, high voltage nwell spacing must be 7 dB units at least; unconnected, low voltage nwell spacing is 4 dB units; otherwise, the minimum spacing is 2u, irrespective of the voltage region. These checks require Dracula to interpret the NWELL layer as a conducting layer. Establishing connectivity Connectivity is established through the CONNECT,SCONNECT and STAMP commands. The first two commands propagate a node name trough a set of conductor polygons that make up a connection. The CONNECT-LAYER statement in the input layer block defines the polygon layer names that represent conductors. STAMP forces the node name from one polygon layer onto a second. With CONNECT, conductor polygons are assigned the same node name whenever they simultaneously overlap the same contact or via. It is essential to note that the node names pass in both directions here either from a lower to an upper or from an upper to a lower conductor, depending on which one the software comes across first. With SCONNECT (soft connect) node names pass in one direction only. Simultaneous overlap of a contact/via by both layers is again required. Often the layers representing the bulk (pwell, nwell ) are defined as soft connect layers to be CH6 - 230

Figure 6-82 : examples of node dependent rules

Appendices

CH6 - 231

sure no supply tracks are fed via/through the bulk (SOFTCHK statement with ERC). The STAMP statement copies (overwrites) the node name of one polygon onto a second, overlapping polygon (it does not propagate node names towards the rest of connected polygons!). The connectivity to be established is function of the subsequent verification checks. For the previous paragraph (node dependent DRC), it is enough to define only and but the ways nwells can be electrically connected to each other. E.g. it would be enough to include nwell, metal1 and metal2 in the connect layer statement (e.g. for a double metal process). Of course, this would be far from sufficient to extract full netlists from layouts that include poly conductors/resistors, n(p)plus cross-unders, or to perform soft connection checks.
; excerpt showing connectivity commands ; *INPUT-LAYER block CONNECT-LAYER = NWL NND MT1 MT2 ; *OPERATION block AND NWL NPL NND ; nwell and nplus = nwell nplus diffusion AND NWL NND NCO ; nwell nplus diffusion = nwell contact ; need diff. polygon layer name for contact AND NWL HV HNW ; high voltage nwell (definition needed for HV ; device extraction) ; CONNECT MT2 MT1 BY VIA ; via connects metal1 and metal2 polygons CONNECT MT1 NND BY CNT ; cnt (contacts) connects nwell nplus with metal1 SCONNECT NWL NND BY NCO ; NND connects to nwell in 1 direction only STAMP HNW BY NWL ; HNW is a subset of NWL that needs node info ; for HV device extraction ; *END

Figure 6-83 : connectivity examples

Device extraction Dracula recognizes devices (mosfets, bipolars, diodes, capacitors, resistors) through the ELEMENT statements. Extraction of predefined device types is easy. It is possible to detect more than one sub-type (low voltage nmos, high voltage nmos) and to make the association with an appropriate model name. A device specific layer makes the recognition easy; terminal specific layers define the device terminals. The device/terminal specific layers are often derived layers; sometimes, dedicated device marker layers are used. In such cases, the designer is required to mark the devices in the layout with the appropriate marker layer! With these explanations, the syntax of the ELEMENT statements in the example should be easy to understand.

CH6 - 231

CH6 - 232

Appendices

;ELEMENT MOS[modelCode] deviceLayer gateLayer sourceDrain bulk ELEMENT MOS[N] NACT3V POLY NSDR PWEL ; 3.3V nmos ELEMENT MOS[P] PACT3V POLY PSDR NWEL ; 3.3V pmos ELEMENT MOS[PH] PACTHV POLY HPSD HNWL ; HV pmos ; ELEMENT RES[PR] POLYRES POLY ; extract poly resistor PARAMETER RES 25 ; square resistance value for poly resistor ;LPESELECT[S] RES GE 200 OUTPUT LAYNET ; output res larger than 200Ohms only

Extraction of device parameter values is not necessary for ERC, so scaling parameters do not need definition. For LVS, LPE and PRE it is necessary to provide the software with values for e.g. area capacitance, square resistanceetc. The PARAMETER statements following the ELEMENT statement do this. Electrical rule checks ERC MULTILAB SAMELAB SOFTCHK NDCOUNT ELCOUNT ERCs verify the electrical integrity of the circuit layout. The aim of an ERC check is to eliminate errors that could heavily disturb the normal course of subsequent LVS runs. It is an extra step that avoids the loss of time resulting from repetitive, unsuccessful LVS jobs. Definition of the conducting (derived) layers, how they connect to each other and component extraction statements form the preparation to the real ERC checks. Signals running across large sub-circuit blocks are 'global signals'. Examples of these global signals definitely are the power supplies, main clock signals, etc. ERCs check the integrity of the global signals by checking the consistency of the given set of labels: it looks for shorts and open circuits. The MULTILAB (shorts) and SAMELAB (opens) statements check for these respectively. A slightly more complex check on power signals is the so-called soft connection check (SOFTCHK statement). The SOFTCHK statement outputs polygons that do not carry a label and that are soft connected to another. This check is to ensure proper power supply distribution. Occasionally, the designer assigns two different node names to a polygon. Occasionally, the designer assigns the same node name to two unconnected polygons. An ERC run detects these cases with the MULTILAB and SAMELAB verification commands. MULTILAB outputs polygons that are assigned more than one label; whereas SAMELAB outputs polygons that are assigned the same label and at the same time are unconnected. An ERC also verifies the integrity of components connections (transistors, diodes, resistors and capacitors). E.g., a floating transistor gate (only connected to itself) is CH6 - 232

Figure 6-84 : device extraction example

Appendices

CH6 - 233

most probably an error. ELCOUNT and NDCOUNT count the number of devices connected to a net, respectively the number of nets connecting to a device. Layout versus schematics LVS LVS jobs compare between the intended circuit (from simulations) and the circuit as defined by the mask layout. The presence of devices is checked against, as well as device dimensions like the width and length of mosfets, etc. Because Dracula performs the comparison, it is necessary to translate the circuit description into a compatible format. The LOGLVS command does this translation: it generates the 'LVSLOGIC.DAT' file from e.g. a CDL input netlist. It is not always necessary for the translation to include all devices present in the circuit netlist. The CDL netlist can contain optional statements that steer the comparison process: sometimes it is useful to neglect a specific device type or is it acceptable to allow swapping of input pins of logic gates. Most of these controls are available when creating the CDL netlist in Cadence. Layout parameter & parasitic resistance extraction LPE/PRE This is probably the last of the verification jobs en allows to extract parasitic components significant enough to eventually disturb or at least influence the chip's operation. In other words a more precise spice netlist is extracted that can be used for electrical simulation of the layout. Typically, large chips contain long interconnection wires and thus larger parasitics can be expected with and. A good designer anticipates the influences from parasitics by over-dimensioning components without excess though. More precise computation of the impact is only possible once the layout is finished. And hopefully, no major modifications are needed as a result from this final computation Parasitics emerge from the moment a connection wire is drawn: coupling capacitances emerge as well as resistors. With LPE/PRE a minimum threshold can be imposed to reduce excessive computing overhead from parasitics with negligible impact. PRE defines more devices compared to LPE; especially parasitic capacitors and parasitic resistances are looked at.

6.2.3 Evolution of a dracula job: a UNIX command sequence

This paragraph describes practical 'software' details on how to execute one type of dracula runs. Most probably there are other ways to run verification tasks. First there is a description of the main executables: function, command line name and output filenames. The second part presents an example of a UNIX command sequence that attempts to mention all the useful steps for a verification run. The dracula preprocessor ("PDRACULA") checks the syntax of the input deck file. This program can be run standalone in some terminal window. Some jobs like LVS also require a netlist file as input. A program called "LOGLVS" translates this netlist into a dracula specific circuit description format. The translation must be finished CH6 - 233

CH6 - 234

Appendices

before starting the preprocessor, because the availability of the circuit description is checked. LOGLVS generates a file "LVSLOGIC.DAT" referenced in the description block. After checking the syntax and availability of data files, the PDRACULA preprocessor generates script files needed for the execution of the job. By the way, note the file "PDRACoutput" is a status report of the PDRACULA process. PDRACULA sometimes halts a job
schematic layout Export CDL (netlist) LOGLVS temp drac.data

Export Stream (gds file)

PDRACUL DRACRUN jxsub.com A

Figure 7-85 : schematic operation of a dracula job

The script called "jxsub.com runs the job in background. Watching the file "dracjob.log" allows for continuous monitoring of the status of the job; changing the log file-name is possible within the description block. Observing the job's progress continuously, can be easily done with the UNIX command "tail f dracjob.log | grep STAGE". This command returns the stage reached by the process. The file "jxrun.int" contains the total number of STAGES in the job. A job's error/logging/warning messages are stored in the files "dracjob.???". 'sum', 'erc', 'err', 'drc', etc. eventually replace the ???-wildcard. Error polygons generated by Dracula can be output in both GDSII and Cadence formats; however, all of the Dracula generated data can be viewed on top of the layout data by using the INQUERY software module. It avoids lengthy read-in of large GDSII dBs and subsequent placing of error cells on top of the original data. Furthermore, different colors correspond with different error messages, making it much easier to analyze the results. INQUERY is called from within the layout window (Virtuoso>Tools menu), and can only work if the KEEPDATA variable is set to the correct value in the description block. Next is an example of a (template of a) UNIX command sequence. For the example only, square brackets indicate optional steps in the command sequence; round brackets "{","}" contain explanations or comments. The comments (round brackets {,} included) must not be part of the UNIX commands.

CH6 - 234

Appendices

CH6 - 235

[ {START: check search path settings in the .login and/or .cshrc files for correct terminal settings in UNIX: at least "/ecad/cds_2000/tools/dracula/bin" must be part of the search path (anno 2001) to be able to run the Dracula software} [vi $HOME/.login ;] [vi $HOME/.cshrc] ] {end of terminal settings check} [ CIW~>Export CDL (cdlout) {generate a CDL netlist: needed for some of the LVS/LPE/PRE jobs} {e.g. cellN == 'udisplay', viewN== 'schematic', o/p fileN == 'udisplay_netlist'} {wait until CDLOUT finishes} vi si.log {check the log file; CDL == Component Description Language} cp microdisplay_netlist $HOME/DRACULA/thisRun/netlist {copy the netlist file to the dracula run dir} LOGLVS << ! /cir <CR> /con <CR> /sum <CR> /x ! <CR> {Compile the network description into Dracula format: 2 files are generated, namely "LVSLOGIC.DAT" and "TREECELL.DAT"} ] { end of eventual CDLout translation and netlist compilation, LVS case } [ CIW~>Export Stream (gdsout) { e.g. input can systematically be done through a GDSII file format } {wait until GDSOUT terminates} vi PIPO.LOG {check gdsout log file (PIPO == Physical-In,Physical-Out) ; the list of translated layers, warning and error messages must be checked!} ] {end of translation into GDSII format} [ PDRACULA {preprocessor execution} /G $HOME/DRACULA/DRACDECKS/thisRun.txt {get inputDeck} /F {finish} [PDRACULA << ! /G $HOME/DRACULA/DRACDECKS/thisRun.txt /F ! {alternative for use with script files; the text between the exclamation marks is input for the preprocessor} ] vi PDRACoutput {check preprocessor output file} ] {end of preprocessor job} [ cat jxrun.int {view total number of STAGES in the job}] jxsub.com {start verification job} tail f dracjob.log | grep STAGE {print the STAGE number Dracula reached; the name of the log file depends on the PRINTFILE statement in the description block; if the stage number does not change in time, then either: the machine is slow (heavy processing load, can be due to large dB size) or a license is still in use by the/another user. Note that it can happen that the Dracula license daemon crashes; in this case the user must kill the Dracula processes still running and must re-start the job. } {wait until the Dracula job terminates} vi dracjob.sum {analyze the summary of run results; .log for logfile} [vi dracjob.erc or .lvs {analyze the results from eventual ERC or LVS run} ] [vi SPICE01.DAT {analyze the extracted netlist} ] [Virtuoso Layout~>INQUERY {interactive analysis of dracula output} ] ] {end of verification job, end of cmd sequence}

Figure 6-86 : UNIX verification command sequence

CH6 - 235

CH6 - 236

Appendices

6.3 The skill interpreter

The aim is to give an idea of how skill code was used for the designs described in this book.

6.3.1 A brief introduction

Usefulness and how it works
[implementation of algorithms] [lists as inputs, note on versions] Lisp/C-like code interpreter The skill language has a syntax similar to C and works with lists. It is essential to know that the elementary datastructure is a list item. The language sequentially processes series of lists (hence interpreter); these lists can consist of an instruction name followed by input data. In simple terms, the design data are organized as objects; each object is a list of fields. Each field has a name and points to any kind of datastructure : a list of strings, a number, an object With objects pointing to other objects, one has the basis for hierarchical data structures exactly what is needed by IC designs. The design data in cadence are organized as lists that contain series of objects, i.e. data with associated behavior (call-back functions). The datafields in these objects can be accessed with the '~>' operator : object235~>field1 returns the value or object stored in field1 of object235. Customization of cadence sub-menus The file "./menus/layEdit.menus" in the user's home directory allows for customization of the pull-down menus in the Virtuoso layout editor. To add a submenu in e.g. the layout editor's "Options" menu, simply add a previously defined menu item "myMenuItem" to the list in "lecOptionsMenu". A menu item must be a list of three objects its name, a string defining the text that appears in the pulldown menu and a string containing the name of the associated action(s). E.g. The "cadenceRunDir/.cdsinit" initialization file can contain instructions that customize the keyboard short cuts (so-called bindkey definitions) and mouse strokes. The use of keyboard short cuts can significantly reduce the access times to menu commands.

Hands-on skill code

For detailed information of course I refer to the manuals; a first study should clarify the most essential notions: literals, datatypes, object identification (ID) and operators, procedures.

CH6 - 236

Appendices

CH6 - 237

The list presented below is meant to shorten the time needed to start writing some code. It is intended to help anyone who needs to write code for the first time; of course, this does not help with programming style and certainly does not provide a full documentation of the language. load(), list(), case(), while(), procedure(), strcat(), geGetSelectedSet() geGetWindowCellView(), printf(), sprintf(),csh(), shell(), infile(), outfile(), close(), fscanf(), if(), foreach(), car(), cdr(), cadr(), caar(), caddr(), cadar(), cadadr(), tconc(), hiCreateAppForm(), hiCreateLabel(), hiCreateSeparatorField(), hiCreateMenuItem(), hiCreateStringField() hiDisplayForm(),hiFormCancel( hiCreateRadioField(), hiCreatePUlldownMenu(), hiCreateBBoxField(), hiCreateStringField(), hiCreateFloatField(), hiCreateButton(), hiCreateIntField(), hiSetFormToDefaults() hiGetWindowList(), hiCreateCyclicField(hiGetAppType(), hiDeleteBannerMenu(), hiInsertBannerMenu(), dbOpenCellView(), dbCreatePolygon(), dbCreateLabel(), dbOpenCellViewByType(), dbCreatePolygon(), dbCreateParamInst(), dbSave(), dbClose(), dbCreateInst() , getWorkingDir()

CH6 - 237

CH6 - 238

Appendices

6.3.2 Function and/or structure and/or source code of some useful routines
6.3.2.1 Automating labels for LVS
It is obvious that for layouts with many repetitive structures (arrays), it can be an exhausting task to label nets manually. Instead, it is very easy to create a label that has a parameterized name and a parameterized placement coordinate. The following example shows how to label the 768 outputs of a XGA row driver cell. Make sure the active window (the last layout window the user accessed) contains the layout to label and that the layout is saved. Then load the following file called 'fileName' type 'load("fileName")' in the input line of the CIW.
; *** fileName = 20020411_XGA013_RD_levelShift_opLabels.ilu ; get the object ID of the active window it returns a cellView ID ; cv is the ID of the cellView to operate on cv=geGetWindowCellView()

; initialize counter the rows will be named row<1> to row<768> i=1 ; start loop; at each cycle, create a new string and place the label 11.9u further up while(i<=768 ; variable 'devar' holds the label name ; sprintf prints "row<"+the value of counter+">" into the string variable sprintf(devar "row<%d>" i) ; create the label, the y coordinate is at 400, the x-coord is at -35.95 for 1st label ; labels for LVS are on layer-purpose m1txt-drawing ; the label is centered both in x and y around the x,y coordinate, orientation is 0deg ; the font is 'stick' and the font size is 0.5 dbCreateLabel(cv list("m1txt" "drawing") -35.95+(i-1)*11.9:400.0 devar "centerCenter" "R0" "stick" .5 ) ; end dbCreateLabel ; increment counter i++ ) ; end while and end of file

After execution, it is good practice to verify the modifications in the layout bugs do show up from time to time. Thus, avoid immediately saving the cellview, as saving makes it harder to undo the modifications. To undo, click 'Discard edits' in the layout editor's menu 'Design' (does not work if the modified layout has been saved).

Figure 6-87: a routine for quick labeling of large designs

CH6 - 238

Appendices

CH6 - 239

6.3.2.2 Collecting the list of instance coordinates in a layout

The next file shows how to build a text file containing the placement coordinates of a given instance in the currently open layout. The instance name and the output file name are hard-coded in the file.
; fileName = "Xiaohong.ilu" ; 3mar99 vdsteen ; To get list of placement coordinates of given instance in current layout view ; type 'load("Xiaohong.ilu")' to execute this file. ; ; last layout window selected, is automatically operated on; be careful! ; -> FILL IN INSTANCE NAME (case sensitive, no %$#@... characters) ; OUTPUT file is named 'cList.txt' and is located in the cadence runDir NameOfInstance="bondpad_v1" NameOfOutputFile="cList.txt" ; get ID of current cellview (graphics editor Get Window CellView) CID=geGetWindowCellView() ; get a list of IDs of all instance headers in the current cellview (CID) ; an instance header is a list of parameters of a cell that was placed x times in CID AllHeaders=CID~>instHeaders ; find the desired header (cellName equal to NameOfInstance), build instance list foreach( header AllHeaders if( header~>cellName==NameOfInstance WantedInst=header~>instances)) ; print the xy coordinate pairs 4 each instance from desired header to the output OPF=outfile(NameOfOutputFile) fprintf(OPF ";-------------------------------------------------------------\n") fprintf(OPF "; cList for instance %s in design %s\n" NameOfInstance CID~>cellName) fprintf(OPF ";-------------------------------------------------------------\n") foreach( instnce WantedInst fprintf(OPF "%L\n" instnce~>xy) printf("%L\n" instnce~>xy) ) fprintf(OPF ";-------------------------------------------------------------\n") close(OPF) ;write file to disk ; print message to the CIW window printf("Output in file %s\n" NameOfOutputFile) ; end of file

The lithography of stitched layouts requires precise location of the different circuit blocks (instances of the different modules) on the reticule, on the die and on the wafer. Usually, the reference coordinate (0.0, 0.0) is at the lower left corner of the reticule layout and in the middle for both the die and wafer layouts. All these data together determine where to flash the different modules on the wafer.

Figure 6-88 : routine to list instance coordinates

CH6 - 239

CH6 - 240

Appendices

6.3.2.3 Automatic generation of 'regular' polygons: fan-out from pitch1 to pitch2

Fan-out of a bus of connections can cost a lot of area if it is done in a 'manhattan' style (orthogonal connections). The use of 45-degree angles when drawing connections is more area efficient, but it can cost a lot of time if it is done manually on a very wide signal bus. Therefore, a routine is described that allows implementing connections between a signal bus with pitch1 to one with pitch2.
; eight point polygon to implement fanout from pitch1 to pitch2 ; width2 is larger (4.4) than width1 (3.0) (the track to connect to) ; the leftmost pair of tracks have equal middle coordinates. ; get db id number of current (opened) cellview cid=geGetWindowCellView() x1r=-6 ; middle coordinates of rightmost track of pitch 1 y1r=0.0 x2r=-70.2 ; middle coordinates of rightmost track of pitch2 y2r=125 i=0 ; reset counter to 0 while(i<642 ;create 1284/2 tracks (one half) x0=x1r-1.5 ; define the eigth coordinate points y0=y1r (x5,y5) (x4,y4) x1=x0+3.0 y1=y0 x2=x1 y2=y1+15+1.5 x3=x2r+1.5 (x6,y6) (x3,y3) y3=y2+(x2-x3) x4=x3 (x2,y2) y4=y2r x5=x2r-1.5 (x7,y7) y5=y2r x6=x5 y6=y7+(x7-x6) x7=x0 y7=y2-1.5 (x0,y0) (x1,y1) ; draw this polygon dbCreatePolygon(cid list("MT2" "drawing") list( x0:y0 x1:y1 x2:y2 x3:y3 x4:y4 x5:y5 x6:y6 x7:y7 )) x1r=x1r-11.9 ;move to the next track pair x2r=x2r-11.8 i++ ) ;end of while and file

Figure 6-89 : fan-out routine

CH6 - 240

Appendices

CH6 - 241

6.3.2.4 Automatic generation of wafer and die floorplans

Computing the maximum nr. of chips per wafer and associated layouts Automatic generation of reticle/die/wafer layouts for stitching and collecting the list of print centre coordinates The aim of these routines is to be able to easily provide the foundry with print centre coordinate data. This kind of data is essential for stitching jobs. The die layout will determine the wafer layout. Combined with the reticle module location data, this allows to provide the necessary coordinate offsets for the wafer stepper. Additionally, a first impression of the resulting die layout can eventually help to debug the layout.

The essence of these routines can be summarized as follows. The translation of layout data into the GDSII format is a common operation. Maintaining the translation table files, archiving them and archiving GDSII data is best done by using some conventions. Most essential is a file naming convention, so that it is easy to find out what the file is/was intended for. The routines provide a file naming mechanism and easy access to fixed/standard locations for these files. The naming convention results in rather long filenames. Therefore the second utility of the routines is to automatically copy the full filenames into the input fields of the 'gds streamOut form or to automatically setup input and output parameters for verification jobs. Altogether, it allows for a quick setup of GDSII translation jobs and of verification jobs.The file naming convention is: yyyymmdd_techname_intention_jobtype.filetype, where YYYYMMDD is the date in 'year month day' concatenated format TECHNAME is a name referring to the technology and/or project INTENTION is the name of the source data or a comment depending on the JOBTYPE subgroup: o gds output files: libraryName_cellName_viewName o technology file: libraryName_comment o layer trans. table: comment o dracula i/p deck: comment o unknown: comment JOBTYPE is one out of five subgroups: o gds output files: GDS, DRO, ERO, LVO, GEO, STO for GDS translation, DRC, ERC, LVS, MASKGEN and stitching o/p o technology files: TFD o layer translation table: LTB o dracula i/p deck: DRC, ERC, LVS, LPE, GEN, ST1, ST2 o unknown: ??? CH6 - 241

6.3.2.5 Routines for easier setup of verification jobs, GDSII translation and maintenance

CH6 - 242

Appendices

FILETYPE is the file extension depending on the JOBTYPE subgroup: o gds output files: gds o technology file: tf o layer trans. table: ltable o dracula i/p deck: drac o unknown: ???

Besides the file naming, I formalized the file location (directory name): ./Projects/ThisProjectDir/ o CDLOUTS/ o DRAC_FILES/ o DRAWINGS/ o GDSOUTS/ o LAYERTAB_FILES/ o SIMULATIONS/ o SKILLS/ o TAPEOUT/ o TECHFILES/

CH6 - 242

References

References - 243

References
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 A. G. Fisher, T. P. Brody, andW. S. Escott, Design of a liquid crystal color TV panel, presented at the IEEE Conf. Display Devices, Oct. 1972 T. P. Brody, J. A. Asars, and G. D. Dixon, A 6x6 in 20 lines-perinch liquid-crystal display panel, IEEE Trans. Electron Devices, vol. ED-20, pp. 9951001, Nov. 1973 F. C. Luo, W. A. Hester, and T. P. Brody, Alphanumeric and video performance of a 6"x6" 30 lines/inch TFT-LC display panel, in Proc. SID, pp. 9494, 1978 P.G. Barten, "Contrast sensitivity of the human eye and its effects on image quality", SPIE press G. Wyszecki, W.S. Stiles, "Color science: concepts and methods, quantitative data and formula", John Wiley & Sons
"Electronic Display Technology: Advanced Materials and Emerging Markets", Technical Insights Inc., 2000

C. Chinnock, "Microdisplay report", Insight Media, published monthly C. Chinnock, "Projection monthly", Insight Media, published monthly A. Kitai, (Mc Master University, Ontario, Canada), cited by The New York Times, Aug 8, 2002 H. Kawamoto, "The history of liquid crystal displays", Journal of the IEEE, Vol. 90, No. 4, 2002 B.Verweire, R. Defever, "Limitation of resolution of lcos based projection displays by diffraction effects", Proc. IDRC '99, pp. 489492, 1999 H.Barna et al, "Liquid crystal alignment degradation induced by photo-irradiation", SID Digest, Vol. 32, pp. 980-, 2001 C-H. Wen et al., "Photostability of liquid crystals and alignment layers", Journal of the SID, Vol. 13, no.9, pp. 805-811, 2005

M. Bone, "Front-projection optical-system design for reflective lcos technology", Journal of the SID, Vol.9, pp. 227-, 2001 M.L. Jepsen et al., "0.9 SXGA Liquid Crystal on Silicon Panel with 450 Hz. field rate", Microdisplay Conference 2001, Westminster (Colorado), pp. 106-, 2001 C. Heine, S. Linz-Dittrich, "Color management system for 3-panel lcos projectors", SID Digest, Vol. 32, pp. 1189-, 2001 C.W. McLaughlin, "Progress in Projection and Displays", Proc. of the IEEE, Vol. 90, No. 4, 2002

Large-Area

References - 243

References - 244 18

References

19

20

21

22 23 24 25

D. Cuypers, G. Van Doorselaer, J. Van den Steen, H. De Smet, and A. Van Calster, Assembly of an XGA 0.9" LCoS display using inorganic alignment layers for VAN LC, in Proceedings of Eurodisplay 2002, (Nice, France), pp. 551-554, October 2002. D. Cuypers, G. Van Doorselaer, J. Van den Steen, H. De Smet, and A. Van Calster, A projection system using vertically aligned nematic liquid crystal on silicon panels, in Proceedings of IDW 2002, (Hiroshima, Japan), pp. 45-48, December 2002. D. Cuypers, H. De Smet, G. Van Doorselaer, J. Van den Steen, and A. Van Calster, Measurement methodology for vertically aligned nematic reflective displays, in Proceedings of SPIE, Vol. 5002, (Santa Clara), pp. 62-72, January 2003. A. Van Calster and D. Cuypers, Reflective vertically aligned nematic liquid crystal microdisplays for projection applications (invited paper), in Proc. of the 2000 international symposium on Electronic Imaging Vol. 3954: Prction Displays 2000: Sixth in a Series, (San Jose, USA), pp. 112-119, SPIE, January 2000. D. Cuypers, H. De Smet, and A. Van Calster, Fringe-field induced disclinations in van lcos panels, in Proceedings of 11th Internation Display Workshops (IDW), (Niigata, Japan), December 2004. MOSAREL public report, Esprit-IV project EP-25340, 2001 M.T. Pfeiffer, "LC modes for lcos", SID seminar, 2001 G. Van Doorselaer, B. Dobbelaere, M. Vrana, X. Xie, N. Carchon, J. Van den Steen, J. Vanfleteren, H. De Smet, D. Cuypers, and A. Van Calster, A paper-white chip-based display MCM package for portable IT products, in Proc. of the IMAPS'98, (San Diego, USA), pp. 543-547, IMAPS, November 1998. J. Van den Steen, N. Carchon, G. Van Doorselaer, C. De Backere, J. De Baets, H. De Smet, J. De Vos, J. Lernout, J. Vanfleteren, and A. Van Calster, Technology and circuit aspects of reflective PNLC microdisplays, in Proc. of the 1997 International Display Research Conference, (Toronto, Canada), pp. 195-198, SID, September 1997. G. Van Doorselaer, N. Carchon, J. Van den Steen, D. Cuypers, J. Vanfleteren, H. De Smet, and A. Van Calster, A reflective polymer dispersed information display made on CMOS, in Proc. of the ICPS, (Antwerpen, Belgium), pp. 222-225, September 1998. R.L. Melcher et al., "Design and fabrication of a prototype projection data monitor with high information content", IBM J Res Dev, Vol. 42, no. 3/4, pp 321-, 1998 C. De Backere, M. Vermandel, and V. C. A, Light shielding with liquid crystal on silicon displays, in Proc. of the 1998 International Display Research Conference, (Seoul, Korea), pp. 415-418, SID, September 1998.

26

27

28 29

References - 244

References 30 31 32 33

References - 245

34

35 36

37

38

39 40 41 42

D. Cohen et al., "Stitching design rules for forming interconnect layers", US. Patent 6225013, 2001 H. De Smet, J. Van den Steen, N. Carchon, D. Cuypers, C. De Backere, M. Vermandel, A. Van Calster, A. De Caussemaeker, A. Witvrouw, H. Ziad, K. Baert, P. Colson, G. Schols, and M. Tack, The design and fabrication of a 2560x2048 pixel microdisplay chip, in Proc. of the 19th International Display Research Conference (EuroDisplay99), (Berlin, Germany), pp. 493-496, SID, September 1999. J. Van den Steen, H. De Smet, G. Van Doorselaer, A. Van Calster, and Colson, Cost effective reticule design for very high resolution Si backplane prototypes, in Proc. of IDW, (Sendai, Japan), pp. 235-238, ITE, December 1999. Y. Zhu et al., "Simulation of Pixel Voltage Error for a-Si TFT LCD Regarding the Change in LC Pixel Capacitance", IEEE Trans. Electron Dev., Vol. 48, no. 2, 2001 H. De Smet, J. Van den Steen, and D. Cuypers, Spice model for a dynamic liquid crystal pixel capacitance, in Proceedings of the 10th international Display Workshops (IDW), (Fukuoka,Japan), pp. 5356, December 2003. H. De Smet, J. Van den Steen, and D. Cuypers, Electrical model of a liquid crystal pixel with dynamic, voltage history-dependent capacitance value, Liquid Crystals, vol. 31, no. 5, pp. 705-711, 2004. J. De Baets, A. Van Calster, J. Van Den Steen, G. Van Doorselaer, D. Wojciechowski, G. Schols, and J. Witters, Design of an x-Si active matrix for high resolution reflective displays, in Proc. of the 15th Int. Display Research Conf., (Hamamatsu, Japan), pp. 477479, SID, October 1995. B.Lee, C. Park, S. Kim, M. Jeon, J.Heo, D. Sagong, J. Kim, J. Souk, " Reducing gray-level response to one frame : dynamic capacitance compensation", SID 2001 Digest, pp1260-1263, 2001 K. Sekiya, H. Nakamura, "Overdrive method for TN-mode LCDs Recursive system with capacitance prediction", SID Digest, Vol. 32, pp. 114-117, 2001 K. Sekiya et al., "Overdrive for compensating color-shift on field sequential color TFT-LCDs", SID Digest, Vol. 35, pp. 408-, 2004 K. Sekiya et al., "Liquid crystal display device", Patent no. US20020175907, 2002

G. Meynants et al., "A 35mm 13.89 million pixel CMOS active pixel image sensor", The PICS Conference, Rochester NY, pp. 58-61, 2003 Shacham et al., "Method of forming reticle from larger size reticle information", US. Patent 6194105, 2001

References - 245

References - 246 43

References

44 45 46 47

48

49

50 51 52 53 54 55 56 57

K.F. Chiang, S. Wu, S. Chen, "Fringing field effect of the lcos devices", Jpn. J. Appl. Phys., Vol. 41, pp4577-4585, 2002 F. Bruyneel et al., "Comparison of reflective PNLCDs and reflective single-polarizer Heilmeier guest-host displays", Journal of the SID, Vol. 9, pp. 313-, 2001 H. De Smet, D. Cuypers, A. Van Calster, J. Van den Steen, and G. Van Doorselaer, Design, fabrication and evaluation of a highperformance XGA VAN-LCOS microdisplay, Displays, vol. 23, no. 3, pp. 89-98, 2002. G. Van Doorselaer, D. Cuypers, H. De Smet, J. Van den Steen, A. Van Calster, K.-S. Ten, and L.-Y. Tseng, A XGA VAN-LCoS projector, in Proceedings of Eurodisplay 2002, (Nice, France), pp. 205-208, October 2002. D. Cuypers, J. Van den Steen, G. Van Doorselaer, H. De Smet, and A. Van Calster, WXGA LCOS projection panel with vertically aligned nematic LC, in Proceedings of the 10th international Display Workshops (IDW), (Fukuoka,Japan), pp. 1541-1544, December 2003. H. De Smet, J. Van den Steen, G. Van Doorselaer, A. Van Calster, "Refresh pixel circuit for active matrix", patent no. WO2004001715, 2002 G.G. Gilette et al., "Display device drive circuit", Patent no. US4,766,430, 1988 A. Dupont et al., "Actuating circuit for a liquid crystal display", Patent no. US5206632, 1993 S.C. Tan, X.W. Sun, "Generic design of silicon backplane for lcos microdisplays", SID Digest, Vol. 33, pp. 200-, 2002 S.E. Shields, AC activated liquid crystal display cell employing dual switching devices, US Patent 4870396, 1989 D. McKnight, Display system having electrode modulation to alter a state of an eletro0optic layer, US Patent 6329971, 2001 K.M. Johnson, D.J. McKnight, I. Underwood, Smart spatial light modulators using liquid crystals on silicon, IEEE J. Quantum Electron., Vol. 29, no. 2, pp. 699-714, Feb. 1993 T.N. Blalock et al., True color 1024x768 microdisplay with analog in-pixel pulsewidth modulation and retinal averaging offset correction, IEEE J. of Solid-State Circuits, vol. 36, pp. 838-845, May 2001

J. Van den Steen, G. Van Doorselaer, D. Cuypers, H. De Smet, A. Van Calster, F. Chu, and L. Tseng, A 0.9 XGA LCoS backplane for projection applications, in Proceedings of the Microdisplay 2001 conference, (Westminster, Colorado), pp. 87-90, August 2001. H. Aoki, "Timing measurement and simulation of a TFT-LCD panel using pixel macro models", SID Digest, Vol. 30, pp. 524-, 1999

References - 246

References 58 59

References - 247

60 61

S. Lee et al, Novel frame buffer pixel circuits for liquid-crystal-onsilicon microdisplays, IEEE J. of Solid-State Circuits, vol. 39, pp.132-139, Jan 2004 A. OHara, J.R. Hannah, I. Underwood, D.G. Vass, and R.J. Holwill, Mirror quality and efficiency improvements of reflective spatial light modulators by the use of dielectric coatings and chemicalmechanical polishing, Applied Optics, vol. 32, no.28, Oct 1993 Dieter Cuypers, Vertically aligned nematic liquid crystal microdisplays for projection applications, Doctoral Thesis, Faculty of Applied Sciences, University of Gent, 2005, ISBN 90-8575-025-X T. Borel, H. De Smet, J. Van den Steen et al., Report on the results of theLCOS4LCOS project, SID06, to be published

References - 247

References - 248

References

Further reading
1 2 3 4 5 6 7 8 9 D. Plus et al., "System for applying brightness signals to a display device and comparator therefore", Patent no. US5,170,155, 1992 A. Van Calster, Microdisplays for portable IT products and projection applications, in Proc. of the fifth international display workshops, (Kobe, Japan), pp. 135-138, IDW, December 1998. E.G.Colgan and M.Uda, On-chip metallization layers for reflective light valves, IBM J Res Develop 42, pp. 339-345, 1998 F.E.Doany, R.N.Singh, A.E.Rosenbluth, et al, Projection display throughput: Efficiency of optical transmission and light-source collection, IBM J Res Develop 42, pp. 387-399, 1998 K.H.Yang and M.Lu, Nematic LC modes and LC phase gratings for reflective spatial light modulators, IBM J Res Develop 42, pp. 401410, 1998 D. McKnight, "Display system having electrode modulation to alter a state of an electro-optic layer", US patent no. 5920298, 1999 E.H.Stupp, M.S.Brennesholtz, "Projection Displays", Wiley-SID Series in Display Technology, 1999

10

11 12

13

F.A. Perner, "Liquid crystal display with pixel circuits with memories", Patent no. EP0965976, 1999 H. De Smet and A. Van Calster, Microdisplays, in Proc. of the 11th Symposium and Seminar on Displays and Display Technology (VISU 99), (Grenoble, France), Le Club Visu (SID France), January 1999. H. De Smet, Liquid crystal displays on monocrystalline silicon, in Proc. of the 11th Symposium and Seminar on Displays and Display Technology (VISU 99), (Grenoble, France), Le Club Visu (SID France), January 1999. H. Pauwels and H. De Smet, Addressing of liquid crystal displays, in Proc. of the 6th International Conference on Mixed Design, (Krakow, Poland), pp. 21-26, June 1999. A. Van Calster and D. Cuypers, Reflective vertically aligned nematic liquid crystal microdisplays for projection applications (invited paper), in Proc. of the 2000 international symposium on Electronic Imaging Vol. 3954: Prction Displays 2000: Sixth in a Series, (San Jose, USA), pp. 112-119, SPIE, January 2000. A. Van Calster, Analog addressed microdisplays (invited), in Proceedings of the 198th Meeting of the Electrochemical Society, (Phoenix, Arizona), pp. 295-302, October 2000.

References - 248

References 14 15 16 17

References - 249

A. Van Calster, Miniaturized reflective displays (invited), in Proceedings of the 7th International Display Workshops (IDW'00), (Kobe, Japan), pp. 179-182, November 2000. A.J. Blanksby, M.J. Loinaz, "Performance analysis of a color CMOS photogate image sensor", IEEE Trans. Electron Dev., Vol. 47, no.1, 2000

18

19 20 21 22

23 24 25 26 27 28 29

D. Cuypers, Vertically aligned nematic liquid crystal displays for projection purposes, in Proc. of the Spring '00 meeting of the SIDME-Chapter meeting, (Balzers (Liechtenstein)), SID, March 2000. F. Bruyneel, H. De Smet, A. Van Calster, and J. Egelhaaf, Measurement and evaluation of the applicability of reflective displays for direct applications, in Proceedings of the 7th International Display Workshops (IDW'00), (Kobe, Japan), pp. 4548, November 2000. F. Bruyneel, H. De Smet, J. Vanfleteren, and A. Van Calster, Characteristics of PNLC in reflective displays, in Proceedings of the 20th International Display Research Conference (IDRC), (Palm Beach, Florida), pp. 217-220, September 2000. G.W. Jones, O.F. Prache, "Field sequential reflective liquid crystal display with frame buffer", patent no. WO0014713, 2000 K. Neyts et al., "Electrical Modeling of Interface Roughness in Thin Film Electroluminescent Devices", IEEE Trans. Electron Dev., Vol. 47, no. 2, 2000 S. Wu, G. Xu, "Cell gap and twist angle determinations of a reflective liquid crystal display", IEEE Trans. Electron Dev., Vol. 47, no. 12, 2000 A. Van Calster, Microdisplays: A cost-effective technology for highresolution displays (invited), in Proceedings of the conference on Displays and Vacuum Electronics, (Garmisch-Partenkirchen, Germany), pp. 115-119, May 2001. B. Maximus, "Projection Displays", SID seminar M-11, 2001 C. Chinnock, "Microdisplay overview", SID seminar, 2001 F. Bruyneel et al., "Comparison of reflective PNLCDs and reflective single-polarizer Heilmeier guest-host displays", Journal of the SID, Vol. 9, pp. 313-, 2001 F. Okumura, "Active matrix liquid crystal displays", SID Seminar, 2001

F. Bruyneel, H. De Smet, J. Vanfleteren, and A. Van Calster, Cell gap optimization and alignment effects in reflective PDLC microdisplays, Liquid Crystals, vol. 28, no. 8, pp. 1245-1252, 2001. G. P. Crawford, "Liquid crystal technology", SID seminar, 2001 I. Underwood, "Silicon technology and VLSI desing for microdisplays", SID seminar, 2001 References - 249

References - 250 30 31 32 33 34

References

35 36 37 38 39 40 41 42 43

J. Chen, M.G. Robinson, G. Sharp, "Two-panel architecture for reflective lcd projector", SID Digest, Vol. 32, pp. 1084-, 2001 J.Gandhi et al., "Experimental Comparison of Contrast Ratio Vs. F/# For Various Reflective lcos Modes", SID Digest, Vol. 32, pp. 326-, 2001 L.D. Silverstein, "Color in electronic displays", SID seminar, 2001 S. Pathak, J.E. Payne, "Drive circuit for liquid crystal display cell", Patent no. WO0135384, 2001 F. Bruyneel, H. De Smet, and A. Van Calster, Reduction of the switching time of polymer dispersed liquid crystal using field oriented addressing, IEEE Electron Device Letters, vol. 23, no. 7, pp. 401-403, 2002. M. Lu, "Nematic lc technology for Si wafer-based reflective spatial light modulators", Journal of the SID, Vol. 10, pp. 37-, 2002

S. Lee et al., "A lcos microdisplay driver with frame buffering pixels", SID Digest, Vol. , pp. 202-, 2002 M.L. Jepsen, "Rear Projection TV:Single Panel LCOS Lumen Throughput", SID Digest, Vol. 34 pp. 122-125, 2003

D. Cuypers, H. De Smet, and A. Van Calster, Fringe-field induced disclinations in van lcos panels, in Proceedings of 11th Internation Display Workshops (IDW), (Niigata, Japan), December 2004. J.A.Shimizu et al., "Contrast improvements for scrolling color lcos", J. Soc. Inf. Dis., Vol.12, pp. 469-, 2004 D. Cuypers, H. De Smet, A. Van Calster, "Fringe field effects in microdisplays", SID Digest, Vol. 36, pp. 1298-, 2005 J.E. Anderson et al., "Fast VAN lcos microdisplay", SID Digest, Vol. 36, pp. 1366-, 2005

M.L. Jepsen, "Why Analog Silicon May Be Best For LCOS Digital TV", SID Digest, Vol. 36, pp. 1358-1361, 2005 S. Hashimoto et al., "SXRD (Silicon x-tal reflective display); a new display device for projection displays", SID Digest, Vol. 36, pp. 1362-, 2005 ---ooo the end ooo---

References - 250