Auto-CEC: Combinational Equivalence Checking via Intelligent Sweeping Engine Selection

Combinational Equivalence Checking (CEC) is essential for circuit verification, but traditional heuristic-based sweeping engine selection often results in inefficiencies. To address this, we enhance the previous Hybrid-CEC method by integrating proposed BDD sweeping, which significantly improves performance for circuits with high XOR chain density, achieving a 13.27× speed-up over Kissat solver. To address the lack of effective guidance for solvers in different verification scenarios, we propose the Auto-CEC framework. This intelligent framework leverages DeepGate2 embeddings and a convolutional neural network (CNN)-based classifier to dynamically predict the most suitable sweeping engine. Experimental results on industrial benchmarks demonstrate that Auto-CEC effectively balances accuracy and efficiency, achieving speed-ups of 6.02× to 11.29× over existing approaches.