Neural Network Co-Processor for Scalable SAT Solving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current SAT solvers face scalability issues when dealing with systems that have thousands of variables and millions of constraints, limiting their ability to practically solve Boolean satisfiability problems.
Innovation Solution
The development of digital hardware circuits and co-processors equipped with neural network circuitries for unsupervised learning, which include binary input interfaces, state machine memory modules, neuron circuits, and neural network acceleration modules to iteratively update weights and guide the search for optimal solutions in the solution space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional SAT solvers and algorithms are used, then they can solve simple Boolean satisfiability problems, but they scale poorly and cannot practically solve problems with thousands of variables and millions of constraints
Solution Approach 1:
The patent replaces traditional algorithmic SAT solvers with a neural network-based system. The neural network learns patterns from training data and directly predicts satisfying assignments, substituting the mechanical step-by-step solving process with a learned model that scales better to large problems.
Solution Approach 2:
The patent transforms the discrete Boolean problem into a continuous learning problem by using neural network weights and probabilities. The binary variables are handled through softmax outputs and probability-based sampling, changing the parameter space from discrete to continuous for optimization purposes.
2Adaptability or versatility
If general mathematical solutions are applied to SAT problems, then they provide theoretical frameworks, but they are not known to solve all possible types of SAT problems
Solution Approach 1:
The neural network is trained on problem instances and learns to solve similar problems independently. Once trained, the model can handle various SAT problem types without requiring manual algorithm selection or configuration, achieving self-service adaptability across different problem domains.
Solution Approach 2:
The patent creates a universal solver that can handle multiple types of SAT problems through a single neural network architecture. The model is designed to process different problem formulations and constraints types, providing multi-functional capability while maintaining consistent solution quality.
Data Source
AI summary
An exemplary co-processor system and method includes a solver circuit with a neural network configured for unsupervised learning, the solver circuit including: a binary input interface configured to receive binary inputs corresponding to variables and clauses for a Boolean problem; a state machine memory module; neuron circuits coupled to the state machine memory module; a neural network machine memory module having arrays of weights corresponding to nodes in a neural network; a state machine circuit operably coupled to state machine memory module, the binary input interface, and the neural network memory module, where the state machine circuit is configured to (i) compute a score from the Boolean states of the clauses (ii) determine a plurality of learning probabilities to generate a plurality of weights; and (iii) provide the plurality of weights to NN memory module, where the weights of the state machine memory module are iteratively updated through unsupervised learning.