OpenOpt: Open-source SRAM optimizer cuts area by 73.6% with 61x faster simulation

OpenOpt is a new open-source framework for SRAM design optimization that jointly optimizes both architecture parameters and transistor sizing using equivalent circuit models. Unlike traditional methods that treat these aspects separately, OpenOpt models inactive SRAM cells as equivalent RC loads and static power sources, dramatically accelerating simulation by up to 61.4x while keeping read/write delay errors below 0.22% and power errors under 1.68%. This co-optimization approach, built on FreePDK45, reveals complementary gains from both architecture selection and transistor sizing, validated through ablation experiments.

The framework integrates seven optimization algorithms, including simulated annealing (SA), particle swarm optimization (PSO), Bayesian optimization variants, and multi-objective evolutionary algorithms. Among them, MOEA/D achieves the highest Figure of Merit (8.2721), resulting in a 6.2% improvement in static noise margin (SNM), a 73.6% reduction in area, and a 42.3% reduction in peak power. OpenOpt is publicly available on GitHub, enabling chip designers and researchers to rapidly explore the large design space of SRAMs for better performance, area, and power trade-offs.