Aggressive scaling of CMOS technology in sub-90nm nodes has created huge challenges. Variations due to fundamental physical limits, such as random dopants fluctuation (RDF) and line edge roughness (LER) are increasing significantly with technology scaling. In addition, manufacturing tolerances in process technology are not scaling at the same pace as transistor's channel length due to process control limitations (e.g., sub-wavelength lithography). Therefore, within-die process variations worsen with successive technology generations. These variations have a strong impact on the maximum clock frequency and leakage power for any digital circuit, and can also result in functional yield losses in variation-sensitive digital circuits (such as SRAM). Moreover, in nanometer technologies, digital circuits show an increased sensitivity to process variations due to low-voltage operation requirements, which are aggravated by the strong demand for lower power consumption and cost while achieving higher performance and density. It is therefore not surprising that the International Technology Roadmap for Semiconductors (ITRS) lists variability as one of the most challenging obstacles for IC design in nanometer regime. To facilitate variation-tolerant design, we study the impact of random variations on the delay variability of a logic gate and derive simple and scalable statistical models to evaluate delay variations in the presence of within-die variations. This work provides new design insight and highlights the importance of accounting for the effect of input slew on delay variations, especially at lower supply voltages. The derived models are simple, scalable, bias dependent and only require the knowledge of easily measurable parameters. This makes them useful in early design exploration, circuit/architecture optimization as well as technology prediction (especially in low-power and low-voltage operation). The derived models are verified using Monte Carlo SPICE simulations using industrial 90nm technology. Random variations in nanometer technologies are considered one of the largest design considerations. This is especially true for SRAM, due to the large variations in bitcell characteristics. Typically, SRAM bitcells have the smallest device sizes on a chip. Therefore, they show the largest sensitivity to different sources of variations. With the drastic increase in memory densities, lower supply voltages and higher variations, statistical simulation methodologies become imperative to estimate memory yield and optimize performance and power. In this research, we present a methodology for statistical simulation of SRAM read access yield, which is tightly related to SRAM performance and power consumption. The proposed flow accounts for the impact of bitcell read current variation, sense amplifier offset distribution, timing window variation and leakage variation on functional yield. The methodology overcomes the pessimism existing in conventional worst-case design techniques that are used in SRAM design. The proposed statistical yield estimation methodology allows early yield prediction in the design cycle, which can be used to trade off performance and power requirements for SRAM. The methodology is verified using measured silicon yield data from a 1Mb memory fabricated in an industrial 45nm technology. Embedded SRAM dominates modern SoCs and there is a strong demand for SRAM with lower power consumption while achieving high performance and high density. However, in the presence of large process variations, SRAMs are expected to consume larger power to ensure correct read operation and meet yield targets. We propose a new architecture that significantly reduces array switching power for SRAM. The proposed architecture combines built-in self-test (BIST) and digitally controlled delay elements to reduce the wordline pulse width for memories while ensuring correct read operation; hence, reducing switching power. A new statistical simulation flow was developed to evaluate the power savings for the proposed architecture. Monte Carlo simulations using a 1Mb SRAM macro from an industrial 45nm technology was used to examine the power reduction achieved by the system. The proposed architecture can reduce the array switching power significantly and shows large power saving - especially as the chip level memory density increases. For a 48Mb memory density, a 27% reduction in array switching power can be achieved for a read access yield target of 95%. In addition, the proposed system can provide larger power saving as process variations increase, which makes it a very attractive solution for 45nm and below technologies. In addition to its impact on bitcell read current, the increase of local variations in nanometer technologies strongly affect SRAM cell stability. In this research, we propose a novel single supply voltage read assist technique to improve SRAM static noise margin (SNM). The proposed technique allows precharging different parts of the bitlines to VDD and GND and uses charge sharing to precisely control the bitline voltage, which improves the bitcell stability. In addition to improving SNM, the proposed technique also reduces memory access time. Moreover, it only requires one supply voltage, hence, eliminates the need of large area voltage shifters. The proposed technique has been implemented in the design of a 512kb memory fabricated in 45nm technology. Results show improvements in SNM and read operation window which confirms the effectiveness and robustness of this technique.

Aggressive scaling of CMOS technology in sub-90nm nodes has created huge challenges. Variations due to fundamental physical limits, such as random dopants fluctuation (RDF) and line edge roughness (LER) are increasing significantly with technology scaling. In addition, manufacturing tolerances in process technology are not scaling at the same pace as transistor's channel length due to process control limitations (e.g., sub-wavelength lithography). Therefore, within-die process variations worsen with successive technology generations. These variations have a strong impact on the maximum clock frequency and leakage power for any digital circuit, and can also result in functional yield losses in variation-sensitive digital circuits (such as SRAM). Moreover, in nanometer technologies, digital circuits show an increased sensitivity to process variations due to low-voltage operation requirements, which are aggravated by the strong demand for lower power consumption and cost while achieving higher performance and density. It is therefore not surprising that the International Technology Roadmap for Semiconductors (ITRS) lists variability as one of the most challenging obstacles for IC design in nanometer regime. To facilitate variation-tolerant design, we study the impact of random variations on the delay variability of a logic gate and derive simple and scalable statistical models to evaluate delay variations in the presence of within-die variations. This work provides new design insight and highlights the importance of accounting for the effect of input slew on delay variations, especially at lower supply voltages. The derived models are simple, scalable, bias dependent and only require the knowledge of easily measurable parameters. This makes them useful in early design exploration, circuit/architecture optimization as well as technology prediction (especially in low-power and low-voltage operation). The derived models are verified using Monte Carlo SPICE simulations using industrial 90nm technology. Random variations in nanometer technologies are considered one of the largest design considerations. This is especially true for SRAM, due to the large variations in bitcell characteristics. Typically, SRAM bitcells have the smallest device sizes on a chip. Therefore, they show the largest sensitivity to different sources of variations. With the drastic increase in memory densities, lower supply voltages and higher variations, statistical simulation methodologies become imperative to estimate memory yield and optimize performance and power. In this research, we present a methodology for statistical simulation of SRAM read access yield, which is tightly related to SRAM performance and power consumption. The proposed flow accounts for the impact of bitcell read current variation, sense amplifier offset distribution, timing window variation and leakage variation on functional yield. The methodology overcomes the pessimism existing in conventional worst-case design techniques that are used in SRAM design. The proposed statistical yield estimation methodology allows early yield prediction in the design cycle, which can be used to trade off performance and power requirements for SRAM. The methodology is verified using measured silicon yield data from a 1Mb memory fabricated in an industrial 45nm technology. Embedded SRAM dominates modern SoCs and there is a strong demand for SRAM with lower power consumption while achieving high performance and high density. However, in the presence of large process variations, SRAMs are expected to consume larger power to ensure correct read operation and meet yield targets. We propose a new architecture that significantly reduces array switching power for SRAM. The proposed architecture combines built-in self-test (BIST) and digitally controlled delay elements to reduce the wordline pulse width for memories while ensuring correct read operation; hence, reducing switching power. A new statistical simulation flow was developed to evaluate the power savings for the proposed architecture. Monte Carlo simulations using a 1Mb SRAM macro from an industrial 45nm technology was used to examine the power reduction achieved by the system. The proposed architecture can reduce the array switching power significantly and shows large power saving - especially as the chip level memory density increases. For a 48Mb memory density, a 27% reduction in array switching power can be achieved for a read access yield target of 95%. In addition, the proposed system can provide larger power saving as process variations increase, which makes it a very attractive solution for 45nm and below technologies. In addition to its impact on bitcell read current, the increase of local variations in nanometer technologies strongly affect SRAM cell stability. In this research, we propose a novel single supply voltage read assist technique to improve SRAM static noise margin (SNM). The proposed technique allows precharging different parts of the bitlines to VDD and GND and uses charge sharing to precisely control the bitline voltage, which improves the bitcell stability. In addition to improving SNM, the proposed technique also reduces memory access time. Moreover, it only requires one supply voltage, hence, eliminates the need of large area voltage shifters. The proposed technique has been implemented in the design of a 512kb memory fabricated in 45nm technology. Results show improvements in SNM and read operation window which confirms the effectiveness and robustness of this technique.

With advancements in technology, the dimensions of transistors are scaling down. It leads to shrinkage in the size of memory bitcells, increasing its sensitivity to process variations introduced during manufacturing. Failure of a single bitcell can cause the failure of an entire memory; hence careful statistical analysis is essential in estimating the highest reliable performance of the bitcell before using them in memory design. With high repetitiveness of bitcell, the traditional method of Monte Carlo simulation would require along time for accurate estimation of rare failure events. A more practical approach is importance sampling where more samples are collected from the failure region. Even though importance sampling is much faster than Monte Carlo simulations, it is still fairly time-consuming as it demands an iterative search making it impractical for large simulation sets. This thesis proposes two machine learning models that can be used in estimating the performance of a bitcell. The first model predicts the time taken by the bitcell for read or write operation. The second model predicts the minimum voltage required in maintaining the bitcell stability. The models were trained using the K-nearest neighbors algorithm and Gaussian process regression. Three sparse approximations were implemented in the time prediction model as a bigger dataset was available. The obtained results show that the models trained using Gaussian process regression were able to provide promising results.