An Efficient Simulated Oscillator-Based Ising Machine on FPGAs

Xingjian Gao,Tingting Zhang,Jie Han,Bailiang Liu

DOI: https://doi.org/10.1109/NANO61778.2024.10628658

2024-07-08

Abstract:Ising model-based computing has emerged as an efficient approach for solving combinatorial optimization problems. In particular, oscillator-based Ising machines (OIMs) have shown promising performance in such metrics as solution quality, time-to-solution, and energy-to-solution. Existing designs, however, require customized chips and use sparsely connected topologies with limited precision for the spin coupling coefficients. In this paper, a simulated oscillator-based Ising machine (SOIM) is designed with a fully connected topology and high coefficient precision for an efficient implementation on field-programmable gate arrays (FPGAs). To this end, an FPGA-oriented model is first developed to describe the underlying mechanism of an oscillator-based Ising machine based on revised differential equations. To save hardware, periodic functions that are expensive to implement are replaced by piece-wise linear functions. Moreover, Gaussian noise is omitted in the system to further save hardware. An OIM simulation algorithm is then proposed to solve the new differential equations using the Euler integration method. From experiments on solving 800-node max-cut problems, the results reach a level of 99% of the best-known values. SOIMs of different sizes are then developed and synthesized on a Zynq UltraScale+ board. Compared with state-of-the-art FPGA-based Ising machines, the SOIM is expected to utilize fewer hardware resources to efficiently solve complex combinatorial optimization problems by leveraging a high coefficient precision and a fully connected topology.

Engineering,Computer Science,Physics