Abstract:Nature-inspired computation is receiving increasing attention. Various Ising machine implementations have recently been proven to be effective in solving numerous combinatorial optimization problems including maximum cut, low density parity check (LDPC) decoding, and Boolean satisfiability (SAT) problems. In this paper, a novel method is presented to solve SAT or MAX-SAT problems with a CMOS circuit implementation. The technique solves a SAT problem by mapping the SAT variables onto quantized capacitor voltages generated by an array of nodes that interact through a network of coupling units. The nodal interaction is achieved through coupling currents produced by the coupling units, which charge or discharge capacitor voltages, implementing a gradient descent along the SAT problem's cost function to minimize the number of unsatisfied clauses. The system also incorporates a unique low-complexity perturbation scheme to avoid settling in local minima, greatly enhancing the performance of the system. The simulation results demonstrate that the proposed SKI-SAT is a high-performance and low-energy alternative that surpasses existing solvers by significant margins, achieving more than 10 times faster solution and 300 times less power.
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve
This paper aims to propose a new method using CMOS circuits to solve Boolean satisfiability (SAT) or maximum satisfiability (MAX-SAT) problems. Specifically, the paper introduces a system called SKI-SAT, which maps SAT variables to quantized capacitor voltages generated by an array of nodes and achieves interactions between nodes through a coupling unit network. This system implements gradient descent on the cost function of the SAT problem to minimize the number of unsatisfied clauses. Additionally, the system introduces a unique low-complexity perturbation scheme to avoid local minima, significantly enhancing the system's performance.
### Main Contributions
1. **High Efficiency and Low Power Consumption**: Simulation results show that the proposed SKI-SAT system outperforms existing solvers, with solution speeds more than 10 times faster and power consumption reduced by 300 times.
2. **Gradient Descent and Perturbation Mechanism**: Through the gradient descent algorithm and a unique perturbation mechanism, the system effectively avoids local minima and finds the global optimal solution.
3. **CMOS Compatibility**: The system design is compatible with standard CMOS processes, easy to scale, and cost-effective, capable of accommodating hundreds or even thousands of clauses.
4. **Support for Higher-Order Polynomials**: Unlike existing methods, SKI-SAT is not limited by polynomial terms and can handle polynomials of any order, improving the system's versatility and applicability.
### Background and Motivation
- **Importance of the SAT Problem**: The SAT problem is a core issue in computer science, widely used in electronic design automation, hardware verification, and cryptanalysis. Despite its simple definition, the SAT problem is NP-complete and typically requires exponential time to solve.
- **Limitations of Existing Methods**: Although existing SAT solvers perform well in some cases, they still face challenges when dealing with large-scale and high-complexity problems. Recently, various Ising machine implementations have been used to solve quadratic unconstrained binary optimization (QUBO) problems, but many NP optimization problems involve higher-order polynomials, making them difficult to map directly to the Ising model.
- **Nature-Inspired Computing**: Nature-inspired computing models, such as Ising machines, map problems to energy spaces and use gradient descent methods to gradually approach the optimal solution with the help of random perturbations. However, these methods still have limitations when dealing with higher-order polynomial problems.
### System Design and Implementation
- **Theoretical Analysis**: By converting the SAT problem in CNF form into a cost function and deriving the gradient vector of the cost function, the SKI-SAT circuit topology is constructed.
- **Circuit Implementation**: The system consists of N nodes, each containing a capacitor, an input for charging or discharging, and a comparator. The comparator compares the capacitor voltage with a threshold to generate a binary output variable. Interactions between nodes are achieved through a coupling unit network.
- **Perturbation Mechanism**: A unique perturbation mechanism is introduced, which helps the system escape local minima and find the global optimal solution by disconnecting the influence of satisfied clauses on node states at specific times.
### Simulation Results
- **Circuit-Level Simulation**: Circuit-level simulations were conducted using the Cadence Virtuoso environment and the TSMC 65nm process design kit. The simulation results show that the SKI-SAT system can find the global minimum within 40 nanoseconds, with an average current consumption of 31.4 mA and power consumption of 37.68 mW.
- **Statistical Performance**: Statistical analysis of numerous runs using a MATLAB behavioral model verified the effectiveness and reliability of the SKI-SAT system.
In summary, this paper proposes an efficient CMOS circuit implementation method for solving SAT and MAX-SAT problems, featuring high performance, low power consumption, and good scalability.