A Scalable and Nearly Uniform Generator of SAT Witnesses

TL;DR

This paper introduces a scalable method for generating SAT witnesses uniformly or nearly uniformly, with theoretical guarantees and practical effectiveness for large constraint sets, improving over prior heuristic and theoretical approaches.

Contribution

It presents a new scalable algorithm for uniform SAT witness generation that combines theoretical guarantees with practical efficiency.

Findings

The method achieves near-uniform distribution of SAT witnesses.

It performs efficiently on large constraint sets.

The approach outperforms previous heuristic and theoretical methods.

Abstract

Functional verification constitutes one of the most challenging tasks in the development of modern hardware systems, and simulation-based verification techniques dominate the functional verification landscape. A dominant paradigm in simulation-based verification is directed random testing, where a model of the system is simulated with a set of random test stimuli that are uniformly or near-uniformly distributed over the space of all stimuli satisfying a given set of constraints. Uniform or near-uniform generation of solutions for large constraint sets is therefore a problem of theoretical and practical interest. For Boolean constraints, prior work offered heuristic approaches with no guarantee of performance, and theoretical approaches with proven guarantees, but poor performance in practice. We offer here a new approach with theoretical performance guarantees and demonstrate its practical utility on large constraint sets.