Benchmarking optimization algorithms for BSIM-CMG compact model parameter extraction

Accurate and efficient compact model parameter extraction is critical for integrated circuit design using advanced transistor architectures. In this work, we benchmark four widely used optimization algorithms– Differential Evolution (DE), Particle Swarm Optimization (PSO) and Bayesian Optimization (BO)–for BSIM-CMG parameter extraction in nanosheet FETs. Each optimizer is evaluated in both conventional (non-dynamic) and dynamic configurations, where the latter incorporates staged hyperparameter adaptation to balance exploration and exploitation. Benchmarking demonstrates that dynamic algorithms consistently outperform their non-dynamic counterparts, with DE-Dynamic exhibiting the highest consistency and lowest variance across multiple runs. This benchmarking study identifies the most suitable optimization strategies for compact model parameter calibration and underscore the significant advantages of dynamic hyperparameter adaptation, offering a more reliable and efficient approach to automated parameter extraction for complex compact models.