SRAM Design And Yield Analysis Based On Mixture Importance Sampling

Static random access memory plays a critical role in memory due to its high speed, robust reliability and low power consumption. It is widely used in SOCs. With the development of CMOS process, the performance of SRAM is continuously improved, and costs continue to decrease. As the technology scales down to 100 nm and below, the random fluctuations of MOS threshold voltage begin to limit the exaltation of SRAM yield seriously. Because of those, variation-aware design has become a hot spot in the design of SRAM in recent years, especially in the full-custom design of SRAM. The traditional analytical method of PVT already can not satisfy the need of high yield of SRAM, and using the Monte Carlo method analyze process variation’s effects on the stability and performance of an SRAM has been widely accepted.This dissertation analyses the process impact on SRAM stability. Based on the stringent requirement of SRAM cell failure rate in high-density SRAM, the thesis discusses the disadvantages of traditional Monte Carlo analytical method, such as slow convergence, long simulation time and so on. To solve this problem, the thesis realizes a Fast Monte Carlo analytical method based on Mixture Importance Sampling, which greatly improved simulation speed. The proposed Fast Monte Carlo analytical method is applied to a 32 M SRAM test chip design based on a 40 nm CMOS process. After the stability analysis of SRAM cells, the most stable one is used to the design of the test chip. Simulation results indicate that the SRAM test chip has a robust stability and keeps high yield under the influence of process variations.