ee Chapter 11 == Layout of Analog and Mixed ! Analog-Digital Circuits | Franco Maloberti Department of Electronics, University of Pavia - 27100 Pavia, Waly 1. Introduction In many of the chapters of this book, theoretical considerations of the design of analog circuits, data converters and mixed analog-cigital systems have been discussed. The design flow consists of the following steps after theoretical analysis: logic, electrical simulation and verification and finally the translation | of results into the physical description (and physical verification) of the system its layout. It is well known that designing a layout is a tedious and error prone task. Thus, in order to avoid trivial work and to reduce the possibility of (human) error, convienient CAD tools for automatic, error-free layout generation have been introduced. For the design of digital circuits, where an extremely large number of transistors is very common, the use of such CAD tools is mandatory. However, the principles on which these tools are based do not coincide with the design strategies that should be used for analog or mixed analog-digital systems. For digital circuits compactness and speed are the main criteria driving clectrical and physical implementation; to-guarantee, analog, circuit accuracy, signal-to-noise ratios (SNR) and bandwidth (for operation with small signals) are important design issues. They must be.carefully taken into account in the layout design phase. To date, economic and technical aspects (simplicity and better codificability) | have favoured the development of CAD tools oriented towards dealing with digital design problems. However, because of the increasing importance of integrated mixed analog-digital systems, a new generation of CAD tools will be made available soon. They will be capable of addressing the design problems (including, of course, layout) involved with entire signal processing systems. It is important for designers to be capable of critically analysing automatically generated solutions, in order to identify any weak points and suggest suitable modifications. Moreover, critical analysis of the results favours the technical growth of the designer himself. The purpose of this chapter is to provide the basic background necessary in order to carry out the functions of the designer mentioned above. anJ ea Layout of Anatog and Mixed Aualog-Digital Circuits 2. Differences Between Layout and Circuit Silicon technology (and its limitations) has been extensively covered in another chapter of this book. Here, we shall only recall those results which are relevant to the correct design of circuit layouts. In particular we will analyse why a designed layout and its corresponding real circuit are different. There are several reasons for this. Among them, the most relevant are: - lateral diffusion | etching under protection boundary dependent etching three-dimensional effects | and, obviously, the errors and limitations associated with mask production and mask alignment. The diffusion of doping atoms is intrinsically a three-dimensional effect. Thus, the edge of a metallurgical junction does not proceed only in the direction perpendicular to the silicon surface but also laterally under the mask-defined protection. If the depth of the junction is x the lateral diffusion is in the range 0.6-0.8 x. (Fig. 1.a) [1]. This effect must be predicted and taken into account at the circuit design level. However, fluctuations in lateral diffusion contribute to a certain amount of random error. In order to be sure that unwanted patterns are completely eliminated, materials are over-etched and the etching also proceeds slightly undermask-defined protection even if unisotropic (plasma assisted) etching is used. It turns out that the resulting pattern is smaller than the one defined by the layout (Fig. 1.b) / (undercut effect). Moreover, the undercut effect is boundary dependent: etching is less active when small strips of material must be removed (Fig. 1.). So, not only undercut but also its boundary dependence must be taken into account. The surface of an integrated circuit, in particular when small linewidths are used, is not flat. The selective oxidation of silicon determines a silicon consumption and, at the same time, the growth of an oxide layer whose thickness is larger than that of the consumed silicon. Selective growth of chemical vapox deposed layers also contribute to non-flat structures. So it results (Fig. 2) that the geometrical dimensions (L1) of patterns grown on the top of non-flat surfaces are different from those of their designed layout (L).Franco Maloberti 43 SiO, protection Mask 06-0.8x Etched layer Elched layer a} by e Fig. 1:4) Lateral diffusion; b) etching under the protection; ) boundary dependent etching. ‘The above described effects are not very substantial for digital systems; by contrast, they may have a significant impact on the accuracy of analog circuits and must be avoided or compensated. Fig, 2: Error in the pattem size due to tridimensional effects 3. Absolute and Relative Accuracies The global performance of an analog circuit is strongly dependent on the absolute and the relative (matching) accuracy of its basic components. In turn, these accuracies depend on the relevant properties of materials and on the geometries of the components. The absolute accuracy of the properties of materials can be controlled only at the technological level; by contrast, relative inaccuracies, due to gradients and local variations, which are also controllable at the technological level, can be compensated with suitable layout techniques, ‘The factors affecting the static performances of MOS transistors, capacitors and resistors are summarized in Tables 1, 2 and 3, [2]. The parameters are influenced in process controlled and layout controlled effects. The relative importance can be deduced by their typical absolute and relative accuracies.SL 344 Layout of Analog and Mixed Analog-Digital Circuits ‘Table 1: Parameters relevant for transistor performance, TYPICAL ABSOLUTE [ MATCHING PARAMETER | SYMBOL VALUE accuracy | ACCURACY ‘Threshold Ven osv o2v | Smv at wom dist mobility (n) Mn 700 am? /V see 15% 1% at 100 dist. mobility (p) Hp 200 am? /V see 15% 1% at 100 dist. oxide capacit. 6% ‘Table 2: Parameters relevant for capacitor performance. TYPICAL | ABSOLUTE | MATCHING VALUE ACCURACY | ACCURACY PARAMETER | SYMBOL Diel-constant | ex (5102) 38 2% 002% oxide thickness Tox “an We DAB% at 100m cist poly oxide os Dan FCG Toon aise thick. length - 03 um 0.04 um at 20 waa Ww - 03 um 0.04 um at 208 Table 3: Parameters felevant for resistor performance. assoLuTe | MATCHING PARAMETER ACCURACY ACCURACY, Giff. resistivity Poly resistivityFranco Maloberti 45; Since the various parameters depend on different technological steps, they can be assumed to be statistically independent. It results that the inaccuracy affecting the considered element can be evaluated by summing up quadratically the various error contributions. From the tables it follows that geometrical parameters (W and L) exibit a 10% absolute inaccuracy if their value is 3 um and a 1% mismatch if their value is 4 um. In the case in which other parameters contribute with lower error or mismatch, it is necessary to use dimensions larger than the ones indicated, in order to have inaccuracies not dominated by geometrical parameters. From the tables it also appears that the mismatch is, at least in some of the parameters, dominated by gradient effects: as the distance of matched elements, increases the matching accuracy worsens. In this case it is necessary to compensate gradient effects with interdigitized and common centroid layout arrangements. These arrangements will be considered in a following section. 4, Layout of MOS Transistors Typically MOS transistors in analog circuits have a relatively large W/L aspect ratio (in the range 10-100). In special cases (for example, transistors in the output stage of a line driver) the aspect ratio W/L can be as high as 1000 or more. The layout of such elements must face two main problems: the large aspect ratio of the resulting physical structure and the series resistance at the source and at the drain. It is quite common to use non-straight gates for digital application, in order to optimize the aspect ratio, as shown in Fig. 3. Given that the matching of orthogonal elements is very poor and the contribution to the gate length of elements at the corners is not well defined this is not tised-much in analog precision applications. An additional big disadvantage of the layout in Fig. 3 is that the drain and the source connections are taken only in two localized points. Since the active area has a specific resistance that is as large as something like 10 a per square (@/2), and since the source and the drain span several squares, the actual equivalent circuit is not that of a single transistor but the more complex configuration shown in Fig. 3 where many elementary transistors are interconnected to one another through drain and source resistances. The voltages at the source and the drain of the elementary transistor are not the same. Even for a relatively low current (100 1A) a voltage drop of several mV can result across the parasitic resistor of the chain. This drop voltage is equivalent to an input-referred offset which, for precision applications, cannot be neglected.36 Layout of Analog and Mixed Analog-Digital Circuits Active Area 00 Metal =m Poly = Contact Drain Gate Source Source Gate Drain Fig. 3: Layout of a digital MOS transistor and its equivalent circuit © Active Area 3 Metal = Poly Contact| Fig. 4: Layout of an analog MOS transistor. This drawback is overcome by using straight transistors with multiple source and drain contacts (Fig. 4). The structure is again equivalent to distributed elementary transistors connected in parallel; however, the metal line on the top of the drain and source contacts the underneath area at many points, thus shorting the drain and source diffused resistances. In some styles of layout only one big contact ranging over the entire source or drain area is used instead of multi-contacts placed at the minimum distance allowed by the design rules. However, even if the two solutions are circuitally equivalent, the approach shown in Fig, 4 is usually preferred because it guarantees a better circuit reliability. Many contacts, with minimum spacing, lead to a reduced curvature of the surface of the metal, thus reducing the risk of micro-fractures (potential sources of failure) in the body of the metal connections (Fig. 5). In the case of transistors with a large W/L aspect ratio a wide straight transistor is not very manageable. To get a more suitable shape it is common to split such a transistors into a number of parallel equal parts arranged in a stack (3). Fig. 6 shows an example of layout of a wide transistor split into eight equal parts. All the sources and drains of each element are connected in parallel by suitable metal connections. The gates are also interconnected by using polysilicon. It should be observed that some of the drain and source connections serve twoFranco Maloberti a7 different elements. This corresponds to a reduction of the silicon area and to an associated reduction of the parasitic capacitances of the two junctions source- substrate and drain-substrate. Metal lines run outside the active area of the transistor in order to ensure the best contact of source and drain. As a second- order effect, it should be noted that while the inner gates of the stack see the same boundary (two gates at the two sides), the gates at the end elements have a different periphery: unmatched undercut will determine slightly different lengths. This minor mismatch is not very important if it concerns elements of a stand-alone big transistor but, it should be taken into account when different transistors, which could be placed on the same stack, must match one another. Metal Oxide Fig. 5: Metal profile with multi-contacts and only one contact. & Aative Area ‘Poly ca Metal mi Contact 7 Gus wide transistor. The layout structure in Fig. 6 that contains only one transistor can be generalized by putting different transistors or parts of them on the same stack ‘A typical situation would be a differential matched pair shown in Fig. 7. Two transistors, MI and M2, which are made up of a differential pair, are split in five equal parts and are arranged into an interdigitized fashion. The connection to the drains (DM1 and DM2) is taken with metal combs placed at the two sides of the stack. The common sources (SM1-SM2) are contacted with a 2 aR SEE348. Layout of Analog and Mixed Analog-Digital Circuits serpentine metal path. It should be noted that crossings (with resulting parasitic resistances) with poly underpass are strictly avoided. DM1 I we fal 2 i oun it ia Aalive Area [a [al a Poly | © Metal oem Contact Fig. 7: Layout of interdigitized stacked differential pair. 5. Layout of Resistors Integrated resistors can be obtained by the suitable use of resistive layers [4]. The specific resistance of a typical resistive layer can range from-a few tens of 2/2 (ohm per square) to Ka/; hence, in integrated resistors the number of squares, and thus the L/W ratio, is often quite large. For this, itis common to make the layout with a serpentine arrangement. As in the case of transistors, the main practical problems concern the estimation of the contribution of the corners of the serpentine and the control of absolute and matching accuracies, For resistors where a limited accuracy is permitted it is enough to assume that the contribution of a corner element is equivalent to one-half square; for precise realizations this approximation is not suitable and it is suggested to use rounded comers (Fig. 8 a) or straight elements interconnected by metal. In this case controlling and taking into account the contact resistances resistive layer- metal is very important. ‘The boundary-dependent undercut effect determines any mismatch. It can be avoided by the use of dummy elements placed around the resistor's layout (Fig. 8a, 8b). Gradients in the properties of the materials also can create problems. They are usually compensated for with interdigitized structures [5]. Fig. 8b shows a possible implemetation of the technique: equal elements of two resistors R, and R, that should match are interleaved and connected in series by metal (low resistive) connections. The centroids of the two structures are closeFranco Maloberti 34g to each other so that the mismatch due to gradients is limited in such a small distance. As already mentioned, it is important to carefully design the end points of the resistive elements. If the current is disturbed from its laminar flow a localized resistance at the end points can result, which can be as high as 1 square of material. In Fig. 8 examples of bad and good end points are shown along with a very rough estimation of the localized resistances (measured in squares) [6]. Dummy on) Dummy Dummy Fig. 8: Examples of layout of resistances and their terminations.350 Layout of Analog and Mixed Analog-Digital Circuits (a) (b) Fig, 9: Two resistors unmatched (a) and matched (b) with respect to thermal gradients. When designing the layout of accurate resistors it is also worth while remembering that their value is temperature-dependent. If a power device is present on the chip, its power dissipation determines temperature gradients. In this case matched resistors should be arranged with their centroids placed symmetrical with respect to the power devices (Fig. 9) With resistors of high values (hundreds of Ka or Ma), layers with high specific resistance must be used; typically they-are the well or pinched-wells. The use of such a layers determines a remarkable interaction between the body of the resistance and the surroundings (coupling, noise, etc... a8 will be discussed later). In order to limit these negative effects it is recommended to put a substrate bias ring around the resistor and, with serpentine shapes, to use substrate bias between the parallel strips. A better insulation between the strips is therefore ensured (Fig. 10). An additional point to be taken into account for precise resistances is the pressure dependence of the resistive layer used. With plastic packages a fluid material is injected into a mould and its successive solidification gives rise to a big stress on the silicon. It can be as high as several hundreds of atmospheres (a value that is close to the fracture limit of silicon). Such a high pressure gives rise to an unpredictable resistor variation. Moreover, for monocrystalline layers realized in <100> silicon the effect is also unisotropic: the pressure induces variations of the electrical properties in a different way in the two orthogonal directions; the unisotropy has a minimum in the 45°direction. Because of this minimum, 45° is often the preferred orientation of resistors designed with monocristalline layers in <100> silicon.Franco Maloberti 351 Well Substrate bias Fig. 10: Layout of resistor realized by well diffusion. 6, Layout of Capacitors Capacitors are achieved in MOS technology by using diffusion, polysilicon or metal as plates and silicon oxide or polysilicon oxide as dielectric. The structures correspond to a parallel-plate element with capacitance given by: C=£08e¢4 a £0 is, as known, the absolute dielectric. constant. Relative dielectric constant (e,), the area of the plates (A) and the oxide thickness (t,,) are the main parameters controlling the capacitor value. However, second-order effects must be taken into account when precise capacitors have to be designed. These effects are related to fabrication inaccuracies and to the fringing effects at the boundary of the structures. Fabrication inaccuracies can give rise to errors in oxide thickness. When this error corresponds to a gradient, its first-order effect is cancelled by matched elements that are arranged in a common centroid fashion. Fig. 11 shows the layout of two capacitors (C, and C,) that need to be equal. The two capacitors are split into eight equal parts that are connected in parallel. This arrangement guarantees that a gradient in the oxide thickness either in the x or y direction does not affect the capacitor matching {7]. Another significant limititation which is due to fabrication steps is the undercut effect. A capacitor is a parallel-plate structure with the upper one normally smaller than the lower one. The area of the smaller plate is assumed to be that of the capacitor. However, because of the undercut effect the actual352 Layout of Analog and Mixed Analog-Digital Circuits area is smaller than the designed one. If the plate is a rectangle with designed sides a and b, with an undercut x, the actual area A’ is smaller than the designed A A'=A-2x(a +b) = A-xP @ POLY 1 Fig. 11: Layout of two matched capacitors (common centroid structure) The resulting reduction is proportional to the perimeters of the plate. It follows that in order to get the same proportional reduction in matched capacitors they should have the same area-perimeter ratio. This condition can easily be fulfilled in matched capacitances whose ratio must be a rational number. Equal “unit” capacitances connected in parailel can do the job; equal elements, of course have the same area-perimeter ratio. For non-integer ratios, the following strategy is usually used: a number of unit capacitors are connected in parallel with the addition of a rectangular element, which value ranges from 1 to 2 unit capacitances (Fig. 12). Some technologies allow use of a contact poly 2-metal even in the thin oxide region where poly I overlaps poly 2. In this case the layout is more efficient, as shown in Fig, 12 b.petra Franco Maloberti 353 uyiTy CAPACITOR Pouy2 WELL POLY) Pov () Fig. 12: Layout of a non-integer multiple of unity capacitance; a) poly interconnection ofthe top plates; b) metal interconnection of the top plates. In order to limit the boundary-dependent errors, rounded or 45° corners are used. When the contact on the thin oxide area is not allowed by the technology (there is the risk of faliure), plate endings for exiting the thin oxide area must be used. Furthermore these poly endings should match. For precise applications even the contribution of parasitics of the metal lines used for interconnections is important. The capacitors should be designed in such a way that the parasitic capacitances are minimized and matched. It is not possible to concentrate the above considerations, illustrated in Fig. 13, into standard design rules. However, the awareness of the multiplicity of these practical problems will stimulate one to think about an optimum layout solution.354 Layout of Analog and Mixed Analog-Digital Circuits a) Common centroid 'b) Dummy poly? strips .©) Contact on top of thick oxide d) Matched poly2 terminals 8) Matched metal ines ) Protective well ) Well multiple biased Fig. 13: Two matched capacitors laid out accordingly to some analog tricks. Finally, it is worth mentioning a fundamental limitation on the capacitor accuracy. This comes from fringing effects. Equation (1) is valid under the assumption that the electric field between the two plates of the capacitor is uniform. In reality, at the boundary of the plates the electric field is not uniform and its fringing causes an intrinsic error in the use of equation (1). Since the fringing depends not only on the voltage of the two plates but also on the voltage of nearby conductors, its effect cannot be quantified. However, this contribuion to capacitor inaccuracy is proportional to the ratios t,,/W and t,,/L. If these ratios are smaller than 1/500 the effect becomes negligible compared to those resulting from fabrication-associated errors. 7. Stacked Layout for Analog Cells ‘The advantages of using stacks of transistors have already been discussed. They involve reducing the parasitic capacitance of the source-substrate and drain substrate junctions and have the advantage of saving area because they allow the sharing of the same contact space (for the source or drain). Moreover, the use of elements with the same orientation improves the matching and also reduces the effects of physical parameter gradients by allowing the use of interdigitized structures. This section considers a generalization of the stacked technique applied to analog, building blocks. It is evident that the most efficient situation is the design of a stack with elements that have the same width, considering that the length of the stacked transistors is not very important. To get an idea of the topological degrees of freedom, it is worthwhile noting that, if a transistor is made of an even numberFranco Maloberti 355 of parts, its stacked representation will have its source connection (or the drain) on both sides of the stack; by contrast, when an odd number of elementary components are used, the source is on one side of the stack and the drain is on the other. Thus, depending on the number of parts in which a transistor is, divided both source and drain or same terminal can be made available at the ending of the stack. If different transistors have sources or drains connected to the same electrical node they can be combined in the same stack in order to share the area corresponding to the common node. For this it is necessary that elements with equal width, whether they are single transistors or part of them, are combined into a single stack. Advantages in active area and junction capacitance reduction will result. Fig. 14 shows an example where four transistors (part of a telescopic folded-cascode op-amp) are designed on the same stack. Since the width of the transistors is an integer multiple of 40 um, they are split into a number of parts 40 ym wide. The two transistors M, and M, are also interdigitized in order to get good matching. The sources of My and M, partially share the same contact area (node 1 and 2) with the drains of M, and M, Fig. 14: Schematic and stacked layout of four n-channel transistors. When designing an analog cell, simply changing the dimensions of a few transistors can lead to a very critical situation, whereas by contrast many other elements can, within limits, have a range of dimensions without significantly affecting the performance of the circuit. In order to achieve a well-designed layout, it is often worth changing the dimension of non-critical transistors (layout oriented design technique) in such a way that the dimensions fit the requirements of the stacked arrangement. The layout of an analog cell can be organized as the interconnection of superimposed or side by side stacks; all the stacks should be made of elements which have equal width and a topological symmetry that corresponds to the electrical symmetry [8] a356 Layout of Analog and Mixed Analog-Digital Circuits Two examples of a fully stacked layout of a two-stage op-amp and of a folded cascode operational amplifier are given in Fig. 15 and Fig. 16. The arrangement of the transistors is given in the associated schematics. It is interesting to observe that the layout drawn with the suggested approach is compact and regular with a close correspondence between the electrical and physical symmetries. Moreover, all the transistors have the same orientation, ensuring good matching, For the two stage op-amp the compensation capacitor occupies all the empty areas. Only one level of metal is used in the interconnections. If the technology makes more than one level of metal available the extra layers can be used for interconnecting at system level. 8. Digital Noise Coupling In analog circuits that are integrated with digital sections the problem of digital noise coupling and the techniques for its avoidance are very important issues. The sources of digital noise coupling are: = capacitive couplings = couplings through the power supply = couplings through the substrate Capacitive coupling, is mainly originated by analog, lines which are routed parallel to the digital buses. Even if the transversal coupling of metal lines (or metal and poly lines) is very weak its effect becomes substantial when two paths are running parallel for a long distance. This situation is typical when reference voltages and clock signals must be delivered to a distant analog block, or, when for shielding purposes the analog Vjyp (or Veg) is used to bias a shielding ring that is placed close to another shielding ring which in turn is biased by a digital power supply. In such cases the noise protection given by the shielding can be Gliminated by the capacitive coupling between the analog and the digital lines. “Another source of capacitive coupling is crossing. Generally digital signals are run as a bus with a lot of lines, It follows therefore that a crossing can often mean interaction with a lot of noisy sources. This kind of coupling is critical when the analog line is connected to the virtual ground of an operational amplifier. The collected noise is directly integrated into the feedback element and is made available at a low impedance node.Franco Maloberti 357 — Me) | M6} | Me} Ma | M4) Mz} | ta | M7} M7} || bal © Fig, 15: a) Schematic of a two stages op-amp; b) transistor's arrangement; ¢) stacked layout.358 Layout of Analog and Mixed Analog-Digital Circuits (a) SEEDER Malan eee MCT (b) i Fig, 16: a) Schematic of a folded cascode op-amp; b) Transistor’s arrangement; c) Stacked layout.Franco Maloberti 359 In switched capacitor circuits, crossing between analog lines and clocks is often unavoidable. A typical layout of complementary switches is shown in Fig. 17 [9]. It should be noted that the metal used to interconnect the n-channel and the p- channel transistor must cross the digital phase driving the gate of the transistors that are placed in the well. tierce ea mS Se i Ft Fig, 17: a) Layout of a minimum area cmplementary toggle switch; b) layout for large W/L ratio. The reduction of noise injection due to capacitive coupling often is achieved from a simple awareness of the problem. A suitable placement and routing allows us to limit its effect. When parallel analog and digital lines (which should be kept as short as possible) are really necessary, they can be decoupled by using an appropiately large distance in between and, if necessary, by putting a grounded line (horizontal shielding) between the two critical paths. Another important source of digital noise are power supply connections. An integrated circuit usually employs the same pins for both the analog and the digital biasing. Moreover the circuit is connected to the external world by a pad, a wire bonding and a pin. After the pad, a common analog and digital biasing line runs for a while before there is a definite splitting of the analog biasing network and the digital biasing network. This situation is described by the equivalent circuit shown in Fig. 18. In general, the current in the analog part of the circuit is constant or slightly varying in time; by contrast, the current flowing in the digital section is made up of sharp pulses, almost synchronous360 Layout of Analog and Mixed Analog-Digital Circuits with the clock, The current pulses are necessary to charge or discharge ‘he parasitic capacitance of the nodes driven from one Jogic state to the Fomplementary one. The amplitude of the pulses is dominated by the switching, Sf digital output drivers, since they are required to actuate externa! loads that are typically a capacitance of 100 pF (in series with the bonding inductance). ‘The bias current (analog and digital) determines a voltage drop aV due to the eetanee of the connection, and more importantly to the inductance of the bonding. Lerrrivanel Digital section __» Analog — section 1 Ht i i R, av: Pin Fig. 18: Equivalent circuit of the bias connection in a mixed A/D circuit AV=R gy + Ralraog te Gt 8 ‘Ac.a rule of thumb, the bjpical inductarice of a bonding connection is around PS mm. Thus, for usual bondings, it can range from 9 to 10 nile Moreover, 1 uiifmional contribution of capacitance and inductance of frame and pin Ue dd also be taken into accOunt. ‘The resulting inductance is such that the Shop voltage can be as large as tens or hundreds of m\/ Table 4 shows typical Hook ing loads for duat-in-line (DIL) and chip-carrier (CC) 40-pin packages. ; ‘Table 4; Parasitic inductances and capacitances in packages. ; Fement | COOL) | HOM) cco) HCO : [pin 1 pin 40 | 25pF | 15 nH 1.0 pF 3 nH i | pins, pind6 LS pF | nH 1.3 pF 6 nH } [pin 10,pin3i| 07 pF | 4ni 1.0 pF 3 nk | pin 20;pin 21 | 25 pF [ 15 nH 1.0 pF 3 nHaaa Franco Maloberti 361 The drop voltage expressed by (3) (usually referred to as ground or Vip bouncing) gives rise to an effect equivalent to a power supply noise. This noise has high frequency components; in the analog section it must be rejected by ptoper power supply rejection ratios (PSRR). Unfortunately PSRR is quite poor at high frequencies. Therefore it is very important to reduce the generation of the power supply noise. In general this is achieved by suitable bonding and biasing strategies that refer to the following general rules: - Firstly, it is advisable to keep as separate as possible the analog from the digital biasing networks. They should merge only very close to the pad = Secondly, when possible, it is recommended to use separate pads for the analog and digital section with relative bonding (eventually multiple-ones) to the same pin (Fig. 19). In such a case the inductance of the analog bonding acts as a filtering element. Pad Pin Li Digital NOMI Digital section Le R, digital 1 Tet OMA = Analog section Pad R, ‘analog Analog Fig. 19: Equivalent circuit of bias connection with separate bonding, - Thirdly, when extra pins ate available, separate pins for the analog and the digital bias should be used. Such a complete separation in the biasing of the two sections gives evident advantages in terms of noise limitation, even if special care must be paid to substrate biasing. A delay in the substrate biasing (because of a delayed biasing of the analog or digital supply) can determine latch-up. - Fourthly, it is always suggested to choose for bias connections the pins that are in the middle of the frame. These result in the minimum parasitic inductance. Another important source of noise is the coupling through the substrate. Output digital drivers very often employ transistors with a huge W/L aspect ratio since they must actuate large external capacitances. The drain diffused areas are consequently large and big capacitive couplings with the substrate results (because of the drain-substrate junction). When the driver switches from one logic state to the complementary state the output node exhibits a fast excursion with a resulting capacitive current, Moreover, during switching, high362 Layout of Analog and Mixed Analog-Digital Circuits current flow can result in impact ionization in the drain area (hot carrier effects) with a resulting substrate current. The relatively high specific resistance of the substrate establishes a considerable amount of noise voltage even for an extremely low avalanche current. The noise coming from the substrate can be limited either by reducing the noise injection or by using shielding strategies. In turn, the noise injected can be reduced by limiting the couplings with the substrate and by special care in the design of digital output drivers. They, in particular, should control the derivative of the output current in order to limit the inductance-dependent drop voltage component. CIRCUIT Fig. 20: Example of layout with well protection strategy. Another method of reducing the substrate noise is to use suitable shields for intercepting the substrate noise and draining it towards non-critical nodes. Shielding can be achieved by plates or by rings. A typical plate shield is a well diffusion put under a capacitor array or under a critical metal line. In order to have real shielding (and non-noise collecting) the plate must be carefully biased and connected to a quiet, low-impedance node. Shielding can even be achieved by using the circuit itself or the wells inside which part of the circuit is integrated. For example, in a digital output driver the layout can be arranged in such a way that results in self-shielding: the transistor inside the well is designed surrounding its complementary transistor (Fig, 20). Shielding rings are typically well rings or substrate-bias rings. Substrate bias offers a low impedance path for the noise currents that are going around the substrate. Wells create a surface barrier along the path of those noisy currents.Franco Maloberti 563 Very often a substrate bias ring and a concentric well placed on the side of the analog circuit are used in order to create a double protection for the analog sections. A very much debated point concerns which voltage is better to bias the shielding rings: analog or digital Vpp or Vg. The use of an analog bias determines a corruption of quict lines (if a non-negligible current must be collected); by contrast, the use of a digital bias reduces the efficiency of the shielding. The decision is often decided on by the specific situation. However very often the digital biases are preferred. 9. Floor Planning of Mixed Analog-Digital Blocks A typical mixed analog-digital block typically contains in the analog section an input signal conditioner, a continuous time or a switched capacitor filter and, eventually, data converters. These analog blocks are made up by active analog cells (operational amplifiers or comparators), passive components (resistors or capacitors) and switches. Moreover, very often a specific digital logic constitutes an essential part of the block (for example, the generator of the disoverlapped phases). The design of the layout of active and passive components must be oriented by analog system requirement. For example, if operational amplifiers must be placed side to side, it is worthwhile to use only one bias block. In this case, the biasing lines in the layout of the op-amp should cross the op-amp in a fashion that the connection is automatically established when the cells are placed side to side (Fig. 21). More in general, the input-output connection in the cell must be located in the proper side, in order to minimize the path and crossing of the inter-block routing. Thus, before designing the layout of the components, it is necessary to define the floor plan of the analog block. The general of reference are: = put the analog critical components as far as possible from the digital elements = make the connections to the critical nodes as short as possible ~ avoid crossing between the analog biasing lines and digital busses Fig. 22 shows the typical floorplan of a switched capacitor filter. It can be noted that the switches that are the components closer to the digital world are placed on one side of the layout. The operational amplifiers that, by contrast, are the more critical analog elements, are placed on the other side of the layout. The capacitors that are in the middle are usually protected by a well shielding. The crossing of analog bias and signal lines and the digital bus bringing in the switches command is strictly avoided.364 Layout of Analog and Mixed Analog-Digital Circuits VDD VSS Fig, 21; Path of bias lines and Vpp - Vsg for basic analog cells. Phases Switches Prot. Ring Capacitor Array Bias cell & Op-AMPS Fig. 22: Typical floorplan of an SC filter. For a fully differential structure, the floor plan shown in Fig. 23 is normally used. The arrangement utilized for a single ended circuit is made symmetrical around the operational amplifiers. The switches are now at the two sides of the floor plan and the digital busses used to drive them are never crossing and far away from the analog ones.Franco Maloberti 365, Digital bus Analogue bus Fig. 23: Typical floorplan of a fully differential SC filter. At a more complex level, when many analog and digital sections must be arranged on the same chip, guidelines similar to the ones already mentioned for floor planning of switched capacitor circuits should be used.In addition, it is important to introduce a well-defined physical separation between the analog and the digital circuitry with suitable protections and decoupling, as discussed in Section 8 A special care must be put in the power supply distribution: networks as separated as possible, for the analog and digital section, must be used. These recommendations are resumed in Fig. 24 where a. possible floorplan of a mixed analog-digital circuit with protection and power supply distribution is shown.366 Layout of Analog and Mixed Analog.Di Analog o DO _ | Guard Rings Section ——— vop yoo Analog Digital vss ‘LI| vss Digital O Section Output Buffers OO) Fig. 24: Typical foorplan of mixed analog digital chip. 10. Concluding Remarks The layout of analog and mixed analog/digital circuits is an important issue for integrated systems. While digital layouts are realized with automatic tools, the physical description of analog blocks is still mainly manual and, generally speaking, requires a specific expertise.. The realization of mixed analog/digital circuit must combine results obtained from automatic tools with the results of manual activity. New computer aids for this kind of task will become available in the near future; however, it is necessary to have the knowledge to critically analyze and interactively optimize the solutions proposed by existing and coming CAD tools. 11. References I) SM. Sze (Ed.), VLSI Technology, McGraw-Hill, New York, 1983. Austria Mikro Systeme: AMS Design Rules CMOS Manuals. K.C. Hsieh, P.R. Gray, D. Senderowicz, D.G. Messerschmid, "A low noise chopper stabilized differential switched-capacitor filtering technique,” IEEE I. Solid-State Circuits, SC-16, 708-715, 1981 SHEFranco Maloberti 367 4] PR. Gray, RG. Mayer, Analysis and Design of Analog Integrated Circuits, Join Wiley & Sons, New York. {I DJ. Allstot, W.C. Black, “Technological design considerations for monolithic MOS switched capacitor filtering systems,” Proceedings of the IEEE, 967-986, 1983. [ A.B. Grebene, Bipolar and MOS Analog Integrated Circuit Design, John Wiley & Sons, New York, 1984. ( J-L. McCreary, P.R. Gray, "All-MOS charge redistribution analog-to-digital conversion techniques,” IEEE J. Solid-State Circuits, SC-10, 371-379, 1975. Bl U. Gatti, F. Maloberti, V. Liberali, "Full stacked layout of analogue cells,” Proc. IEEE Int. Symp. on Circ. and Syst., 1123-1126, 1989. BIR. Gregorian, G. Temes: Analog MOS Integrated Circuits, John Wiley & Sons, New York, 1986,