SMAS: A Generalized and Efficient Framework for Computationally Expensive Electronic Design Optimization Problems

Abstract

Many electronic design automation (EDA) problems encounter computationally expensive simulations, making simulation-based optimization impractical for many popular synthesis methods. Not only are they computationally expensive, but some EDA problems also have dozens of design variables, tight constraints, and discrete landscapes. Few available computational intelligence (CI) methods can solve them effectively and efficiently. This chapter introduces a surrogate model-aware evolutionary search (SMAS) framework, which is able to use much fewer expensive exact evaluations with comparable or better solution quality. SMAS-based methods for mm-wave integrated circuit synthesis and network-on-chip parameter design optimization are proposed and are tested on several practical problems. Experimental results show that the developed EDA methods can obtain highly optimized designs within practical time limitations.