AI-Driven Logic Synthesis and Optimization in Modern Electronic Design Automation: From Learned Heuristics to Generative Circuit Design

Abstract

AbstractThe exponential growth in integrated circuit (IC) complexity, combined with tight power, performance and area (PPA) limits, has strained traditional Electronic Design Automation (EDA) techniques. Classical logic synthesis relies on hand-engineered heuristics whose runtime and solution quality struggle to scale to modern designs. This article reviews how artificial intelligence, specifically Graph Neural Networks (GNNs), Reinforcement Learning (RL) and generative transformer models, is being applied to logic gate synthesis and optimization. This article extends the author's earlier study and organizes the field into predictors, selectors, operators and assistants, compare AI-driven and heuristic flows, survey key applications, and discuss open problems in functional correctness, generalization and scalability. We propose a hybrid propose-and-verify architecture in which learned components suggest transformations and classical formal engines guarantee equivalence, together with an evaluation protocol for fair, reproducible comparison. This is a review and perspective; no new experimental results are reportedKeywords—Electronic Design Automation, Logic Synthesis, And-Inverter Graph, Graph Neural Networks, Reinforcement Learning, Generative Circuit Models, PPA Optimization..