A novel Ising model processing achieving all interactions only by adjacent spins for a high-speed solver for versatile Ising machines

摘要

To implement the Ising model on hardware and create "things" with optimization capability for a future IoT society, two methods for solving an optimization and recognition problem by using a new general-purpose Ising model suitable for LSI chip implementation are proposed. An ordinary model cannot solve many optimization problems, by searching for the ground-state, because the number of interactions is not enough for mapping the problems to the model. This ordinary Ising model considers only interactions of adjacent spins. Actually, to solve an optimization problem, interactions between all spins must be considered. However, all interactions are impossible to implement in an LSI chip. Therefore, to consider all interactions equivalently by adjacent spins only, two models are proposed. By using these model that implements all interactions, it is possible to obtain good candidates for optimal solutions for the "traveling salesman problem" and the "support vector machine". Moreover, since the features of original Ising model with all interactions is known as a general-purpose model, this result means that the proposed model can be used to obtain the optimal solution candidates for most optimization problems. Further, LSI block diagrams of the two models are presented.