Spintronics probabilistic computer: Realization of proposals by Feynman and Hinton using a spintronics device

Abstract

In 1981, Feynman proposed a computer that operates probabilistically at the hardware level, along with the concept of quantum computers. Boltzmann machine learning, proposed by Hinton et al., the 2024 Nobel laureates in Physics, also models a magnetic system consisting of probabilistically fluctuating spins. As the increasing power consumption associated with the widespread use of artificial intelligence becomes a pressing challenge, this paper discusses spintronics probabilistic computers that naturally realize the proposals of Feynman and Hinton, enabling energy-efficient artificial intelligence computation. Proof-of-concept demonstrations, including combinatorial optimization, machine learning, and quantum simulation, as well as the development of superparamagnetic tunnel junction devices for enhancing computing performance, are described.