Novel Ising model using dimension-control for high-speed solver for Ising machines

Abstract

Rapid optimization and recognition problem solving is needed in a future IoT society. Two novel Ising models suitable for hardware implementation for Ising machines are proposed. The Ising model is expected to solve optimization and recognition problems extremely fast by mapping the parameters of the problem onto spins and interactions, and searching for the ground state. This method is different from von-Neumann architecture which is poor at solving these problems. However, in actual problems in IT, complex interactions, such as all interactions between spins, cannot be incorporated into the LSI. The two proposed models can consider all interactions equivalently by assessing only the interactions of adjacent spins. The successful operation of these proposed models was confirmed for the traveling-salesman-problem and support vector machine as an important way to solve problems for the future IoT. The convergent energy was reduced to about 4% of the initial and compares favorably with the ideal model with all interactions. Also, a recognition rate of about 90% was obtained, which is the same as that of the ideal model. Furthermore, VLSI block diagrams for both models are presented.