An Enhanced 1440 Coupled CMOS Oscillator Network to Solve Combinatorial Optimization Problems
Markus Graber, Klaus Hofmann · 2023
Today’s world demands more and more computation power combined with increased energy efficiency. Specialized hardware architectures are developed to enable high-performance computing. NP-hard combinatorial optimization problems such as the maximum cut or traveling salesman are especially costly to solve. The usage of analog coupled oscillators to solve such combinatorial optimization problems is a recently emerging approach. Those so-called Oscillator-based Ising Machines (OIMs) exploit the coupling between oscillators for computation. In this work, we present experimental results of an integrated OIM to demonstrate the capabilities. Together with an analysis of the scalability, the strengths, and weaknesses are discussed. We propose a new improved topology, which is optimized towards a versatile OIM solver. Our 4.6mm2prototype implemented in a 28nm node offers 1440 oscillators with 11724 configurable couplers.