All-to-All Connected Ising Machines with Free-Space Optical Interconnects

Combinatorial optimization problems are ubiquitous in several fields, including logistics, machine learning, and protein folding, to name a few. Such optimization problems require substantial computational resources and time, escalating exponentially with the problem size, thus necessitating the development of innovative, effective computational techniques. The main hurdle with the current computational approaches is inefficiency in dealing with vast, complicated problem sets. Traditional algorithms tend to be linear, and their scalability frequently falls short for complex problems. Current hardware systems too are either specialized, costly, energy-consuming, or incompatible with existing technologies, creating a pressing need for a less costly, more efficient solution.

Technology Description

The technology is an analog computing system that functions through coupled nonlinear oscillators to solve intricate combinatorial optimization problems. This system operates using the weighted Ising model and comprises a fully connected LC oscillator network. The network utilizes minimal-cost electronic components and is compatible with time-honored integrated circuit technologies. Each LC oscillator, or node, within the network can be linked to every other node in the array either through a multiply and accumulate crossbar array or optical interconnects. What sets this technology apart is its exceptional performance in terms of obtaining solutions to randomized MAX-CUT problem sets with high accuracy rates; 98% for binary weights and 84% for five-bit weight resolutions. In larger networks, the time to solution scales directly with oscillator frequency, indicating that sizeable coupled oscillator networks could solve computationally rigorous problems more rapidly and efficiently than traditional algorithms.