CN118604585B - A behavioral sensor based on ALU - Google Patents

Abstract

The invention belongs to the field of electric digital data processing, and provides an ALU-based behavior sensor which provides more accurate behavior data for an artificial intelligent model for SoC chip aging prediction. The invention comprises the following steps: the system comprises a pressure generating module, an ALU group, a state control machine, a sampling counting module and a memory, wherein the ALU group comprises a plurality of ALU combination logic chains, each ALU combination logic chain is configured into a pressure sensor in a pressure mode and a ring oscillator in a test mode, and is used for simulating the behavior state of a tested circuit; the pressure generating module provides pressure input for the pressure sensor, and the sampling counting module samples the vibration frequency information of the ring oscillator and stores the vibration frequency information in the memory; the state control machine is used for controlling the ALU group and the sampling counting module. The invention builds the ring oscillator by taking the ALU as a main body, can describe the behavior of the tested circuit more accurately, and has the advantages of low invasiveness and stronger monitoring capability.

Description

A behavioral sensor based on ALU

Technical Field

The present invention belongs to the field of electrical digital data processing, relates to an aging sensor applied to SoC chip aging prediction, and specifically provides a behavior sensor based on an ALU (Arithmetic and Logic Unit).

Background Art

As the complexity of SoC chips continues to increase, complex semiconductor processes and harsh working environments will accelerate the aging of SoC chips. Therefore, aging detection and aging prediction of SoC chips have become particularly critical. SoC chips mainly include components such as processors, on-chip interconnects, storage and peripherals. Among them, the processor is the core, responsible for the control and calculation functions of the entire SoC chip, making the processor the component with the largest load, highest temperature and fastest aging in the SoC chip; further, the core part of the processor is the computing unit composed of ALU. Therefore, the core of aging prediction of SoC chips lies in the aging estimation of ALU in the processor.

The main aging effects of processors include hot carrier injection (HCI), negative bias temperature instability (NBTI) and time-dependent dielectric breakdown (TDDB). Among them, NBTI and HCI have the greatest impact on processor aging. With the improvement of semiconductor technology, NBTI has gradually become dominant. The NBTI effect is affected by many parameters, mainly including: voltage, temperature and signal probability (SP). Therefore, in the SoC chip aging prediction method based on artificial intelligence model, aging sensors such as voltage sensors, temperature sensors and behavior sensors are needed to collect data such as voltage, temperature and manufacturing process deviation to complete model training and prediction. Behavioral sensors are generally composed of a ring oscillator, which is composed of a buffer. The aging process of the circuit under test is simulated to characterize the behavior of the circuit under test in its manufacturing process and working environment. Although the aging sensor composed of a ring oscillator has a high degree of accuracy for most circuits, for complex SoC chips, especially for the computing circuits in the processor, the accuracy of the aging sensor still needs to be improved.

Summary of the invention

The purpose of the present invention is to provide an ALU-based behavior sensor to provide more accurate behavior data for an artificial intelligence model used for SoC chip aging prediction, thereby enabling the artificial intelligence model to more accurately predict SoC chip aging.

To achieve the above object, the technical solution adopted by the present invention is:

A behavior sensor based on ALU includes: a pressure generating module, an ALU group, a state control machine, a sampling and counting module and a memory; wherein the ALU group includes: a plurality of ALU combination logic chains, each of which is configured as a pressure sensor in a pressure mode and as a ring oscillator in a test mode, so as to simulate the behavior state of a circuit under test; the pressure generating module provides pressure input for the pressure sensor, and the pressure input of each ALU combination logic chain is different; the sampling and counting module samples the vibration frequency information of the ring oscillator and stores it in the memory; the state control machine is used to control the ALU group and the sampling and counting module.

Further, the ALU group includes: N ALU combination logic chains, N≥10; the N ALU combination logic chains are numbered in sequence, each ALU combination logic chain includes: a multiplexer and an ALU, the ALU combination logic chain switches between a stress mode and a test mode under the control of a mode switching signal, and the mode switching signal is generated by a control state machine;

In the pressure mode, the two input ends of the ALU select the pressure input generated by the input pressure generation module through the multiplexer, the control end of the ALU is randomly assigned a function code by the control state machine, and the output end of the ALU is output to the sampling counting module;

In the test mode, the control end of the ALU is fixed with a function code by the control state machine. One input end of the ALU selects a high-level input through a multiplexer, and the other input end of the ALU selects a feedback input through a multiplexer. The last bit of the feedback input is set to the inverted result of the last bit of the output, and the remaining bits are all low levels. The output end of the ALU is output to the sampling counting module, and the high-level input and feedback input are provided by the control state machine.

Furthermore, the pressure generation module is composed of a linear feedback shift register, which generates a random number and combines it with a high level and a low level to form a pressure input in a preset ratio. By adjusting the preset ratio, the signal probability of the pressure input is dynamically adjustable, thereby providing different pressure inputs for each ALU combinational logic chain.

Furthermore, the sampling and counting module includes: a register and a counter. In the test mode, the register uses the output of the ALU as a clock signal, and records the number of output flips of the register through the counter as the oscillation frequency information of the ring oscillator; after the sampling is completed, the counter is cleared and the number of the ALU combinational logic chain and the count value are saved in the memory.

Furthermore, the control cycle of the state control machine includes N sub-cycles, which sample N ALU combinational logic chains in sequence to obtain count values; in each sub-cycle, the state control machine controls the ALU combinational logic chain to execute the stress mode and the test mode in sequence, and controls the sampling counting module to complete the sampling in the test mode.

Based on the above technical solution, the beneficial effects of the present invention are:

The present invention provides an ALU-based behavior sensor, which uses the ALU as the main body to build a ring oscillator, and can more accurately describe the behavior of a circuit under test. Without affecting the function of the circuit under test, the behavior of the circuit under test under the manufacturing process is described, and more abundant and accurate behavior data is provided for an artificial intelligence model, so that the artificial intelligence model can more accurately predict the aging of the circuit under test. Compared with existing behavior sensors, the ALU-based behavior sensor provided by the present invention is less invasive to SoC chips, and has stronger behavior monitoring capabilities and higher accuracy for SoC chips. In addition, the present invention provides more information, which is beneficial to artificial intelligence model training.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the structure of an ALU-based behavior sensor provided by the present invention.

FIG. 2 is a schematic diagram of the structure of the ALU in the behavior sensor provided by the present invention.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and beneficial effects of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments.

The present embodiment provides an ALU-based behavior sensor, which is manufactured synchronously with the circuit under test on the same chip, and characterizes the deviation of the circuit under test under its manufacturing process by simulating the aging process of the circuit under test; the behavior sensor is shown in FIG1, and includes: a pressure generation module, an ALU group, a state control machine, a sampling and counting module and a memory; wherein the ALU group is the main part of the behavior sensor, including: a plurality of ALU combination logic chains, each ALU combination logic chain is configured as a pressure sensor in a pressure mode and as a ring oscillator in a test mode, so as to simulate the behavior state of the circuit under test; the pressure generation module is used to simulate the generation of dynamic pressure, and provide pressure input for the pressure sensor, and the pressure input of each ALU combination logic chain is different; the sampling and counting module is used to sample the vibration frequency information of the ring oscillator and store it in the memory; the state control machine is used to control the ALU group and the sampling and counting module.

Further, the ALU group includes: N ALU combination logic chains, N ≥ 10; the N ALU combination logic chains are numbered in sequence, and each ALU combination logic chain is shown in FIG2, including: a multiplexer (MUX) and an ALU, the ALU combination logic chain has a stress mode and a test mode, the multiplexer selects the input of the ALU under the control of a mode switching signal (Test Enable) to achieve mode switching, and the mode switching signal is generated by a control state machine; the ALU has two input terminals, an output terminal and a control terminal, and in the stress mode, the ALU is configured as a pressure sensor, and the two input terminals of the ALU select the pressure input (Stress Input) generated by the input pressure generating module through the multiplexer, and the control terminal of the ALU is randomly assigned a function code by the control state machine, and the ALU obtains an output after completing the operation (ALU Output); In test mode, the ALU is configured as a ring oscillator, the control end of the ALU is fixed by the control state machine with a function code, one input end of the ALU selects a high-level input (each bit is high) through a multiplexer, and the other input end of the ALU selects a feedback input through a multiplexer. The last bit of the feedback input is set to the inverted result of the last bit of the output (ALU Output), and the remaining bits are low. The high-level input and feedback input are provided by the control state machine.

Furthermore, the pressure generation module provides pressure input for each ALU combinational logic chain, and the pressure input for each ALU combinational logic chain is different; the pressure in the aging process specifically refers to the signal probability (SP), that is, the duty cycle of the signal. The pressure generation module is composed of a linear feedback shift register (LFSR). In order to simulate the change of pressure of the ALU in the processor during actual operation, this embodiment generates pressure input by generating random numbers through the linear feedback shift register (LFSR). However, the signal probability (SP) of the generated random number is approximately 50% during the long-term aging process. Therefore, this embodiment realizes dynamic adjustment of the signal probability (SP) by controlling the ratio of the high level (1), the low level (0) and the random number, thereby providing different pressure inputs for each ALU combinational logic chain to ensure data richness;

In this embodiment, a linear feedback shift register (LFSR) is exemplarily given to generate a signal probability (SP) of a pressure input, as shown in Table 1. It can be seen from the table that by controlling the ratio of a high level (1), a low level (0) and a random number, the signal probability (SP) can be dynamically adjusted; and the higher the ratio accuracy, the higher the adjustment accuracy of the corresponding input signal probability (SP) is.

Table 1

LFSR ID Time ratio SP 0 0: Random number = 7: 1 0.0625 1 0: Random number = 5: 3 0.1875 2 0: Random number = 3:5 0.3125 3 0: random number = 1: 7 0.4375 4 All random numbers 0.5000 5 1: Random number = 1:7 0.5625 6 1: Random number = 3:5 0.6875 7 1: Random number = 5:3 0.8125 8 1: Random number = 7:1 0.9375

Furthermore, the sampling and counting module includes: a register and a counter. In the test mode, the register uses the output of the ALU as a clock signal. When the output of the ALU changes, the output of the register flips once. The counter records the number of times the output of the register flips, and completes the sampling of the oscillation frequency information of the ring oscillator configured by the ALU. After the sampling is completed, the counter is cleared and the number of the ALU combination logic chain and the count value are saved in the memory (DPRAM); the start and end of the sampling are controlled by the state control machine, and the counter is cleared by the state control machine;

Furthermore, the state control machine is used to control the ALU group and the sampling counting module. The control cycle of the state control machine includes N sub-cycles, and the N ALU combinational logic chains are sampled in sequence to obtain counting values. In each sub-cycle, the state control machine controls the ALU combinational logic chain to execute the stress mode and the test mode in sequence, and controls the sampling counting module to complete the sampling in the test mode. In addition, for each ALU combinational logic chain, a test can be performed once in the initial state.

In practical applications, the above-mentioned behavior sensor is used to monitor the behavior of the SoC chip. In a typical application scenario, a number of the above-mentioned behavior sensors are arranged around the SoC chip under test and manufactured synchronously on the same chip, which can minimize the sensor measurement deviation caused by the chip-to-chip error during the manufacturing process; in addition, the SoC chip under test is also arranged with a temperature sensor, a voltage sensor, etc., and the data of each sensor is transmitted through a data transmission network and sent to the host computer through an interface for processing. In the host computer, the oscillation frequency of the ring oscillator configured in each ALU combination logic chain can be calculated according to the count value stored in the behavior sensor, which is specifically expressed as: f=1/t, t=T/X, f is the oscillation frequency of the ring oscillator, t is the oscillation period of the ring oscillator, T is the sampling duration, and X is the count value of the counter; further, the delay value of each ALU combination logic chain can be calculated according to the oscillation frequency, which is specifically expressed as: tp = t/2. Since one oscillation cycle requires the ALU combination logic chain to flip once, the time of half a cycle is the delay value of the combination logic chain. Moreover, the corresponding pressure value (signal probability, SP) can be obtained a priori according to the number of the ALU combinational logic chain stored in the behavior sensor, thereby obtaining the comparison relationship between the pressure value and the delay value. On this basis, during the training of the artificial intelligence model, the pressure value, voltage value, and temperature value are used as the input of the model, and the delay value is used as the output of the model to form a training sample, thereby completing the model training; according to the trained artificial intelligence model, the real-time measurement values of the temperature, voltage, and pressure of the SoC chip under test are input into the model, and the model outputs the delay prediction value of the SoC chip under test, and the real-time aging degree of the SoC chip under test is estimated through the delay prediction value.

The above description is only a specific implementation mode of the present invention. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other alternative features that are equivalent or have similar purposes; all the disclosed features, or all the steps in the methods or processes, except for mutually exclusive features and/or steps, can be combined in any way.

Claims (3)

1. A behavior sensor based on ALU, characterized in that it includes: a pressure generating module, an ALU group, a state control machine, a sampling and counting module and a memory; the ALU group includes: a plurality of ALU combination logic chains, each of which is configured as a pressure sensor in a pressure mode and as a ring oscillator in a test mode, so as to simulate the behavior state of a circuit under test; the pressure generating module provides a pressure input for the pressure sensor, and the pressure input of each ALU combination logic chain is different; the sampling and counting module samples the vibration frequency information of the ring oscillator and stores it in the memory; the state control machine is used to control the ALU group and the sampling and counting module; The ALU group includes: N ALU combination logic chains, N≥10; the N ALU combination logic chains are numbered in sequence, each ALU combination logic chain includes: a multiplexer and an ALU, the ALU combination logic chain switches between a stress mode and a test mode under the control of a mode switching signal, and the mode switching signal is generated by a control state machine; In the pressure mode, the two input ends of the ALU select the pressure input generated by the input pressure generation module through the multiplexer, the control end of the ALU is randomly assigned a function code by the control state machine, and the output end of the ALU is output to the sampling counting module; In the test mode, the control end of the ALU is fixed with a function code by the control state machine, one input end of the ALU is selected to input a high-level input through a multiplexer, and the other input end of the ALU is selected to input a feedback input through a multiplexer, the last bit of the feedback input is set to the inverted result of the last bit of the output, and the remaining bits are all low levels, and the output end of the ALU is output to the sampling counting module, and the high-level input and feedback input are provided by the control state machine; The pressure generation module is composed of a linear feedback shift register. The linear feedback shift register generates a random number and combines it with the high level and the low level to form a pressure input in a preset ratio. By adjusting the preset ratio, the signal probability of the pressure input can be dynamically adjusted to provide different pressure inputs for each ALU combinational logic chain. 2. According to claim 1, the ALU-based behavior sensor is characterized in that the sampling and counting module includes: a register and a counter. In the test mode, the register uses the output of the ALU as a clock signal, and records the number of output flips of the register through the counter as the oscillation frequency information of the ring oscillator; after the sampling is completed, the counter is cleared and the number of the ALU combinational logic chain and the count value are saved in the memory. 3. According to claim 1, the ALU-based behavioral sensor is characterized in that the control cycle of the state controller includes N sub-cycles, and N ALU combinational logic chains are sampled in sequence to obtain count values; in each sub-cycle, the state controller controls the ALU combinational logic chain to execute the pressure mode and the test mode in sequence, and controls the sampling counting module to complete the sampling in the test mode.