A Coupled Oscillator Network to Solve Combinatorial Optimization Problems with Over 95% Accuracy
Markus Graber, Klaus Hofmann · 2023
Solving optimization problems is becoming increasingly important for many applications in today's world. Various software solvers and digital hardware accelerators are designed to speed up the computation while decreasing energy consumption. However, NP-complete problems like the maximum-cut are still inefficient to solve using traditional computing methods. The usage of specialized analog coupled oscillator networks is an upcoming fast and energy-efficient option. In order to compete with digital counterparts, a similar solution accuracy at significantly increased speed is required. However, reaching a very good solution with such analog-oriented principles is still a major challenge. The system proposed here focuses on a best possible accuracy and speed, while area and power consumption are of minor priority. Experimental benchmarking shows that the computed solutions reach 95% accuracy and more of commercial software, while the actual computation takes just 714 nanoseconds.