Identifying Good Rules for Efficient SAT Encodings of Single-Constant Multiplication Using Machine Learning

Researchers propose a neuro-symbolic method to speed up SAT encodings for single-constant multiplication in hardware optimization.

A graph neural network predicts promising decomposition rules, then symbolic search prunes unlikely choices.

On unseen 17–32 bit constants, encoding time dropped by 10x–100x and memory use fell by over 97%.

The approach also sharply reduced branching while keeping addition counts near optimal, improving scalability for synthesis.