2020年第67回応用物理学会春季学術講演会

講演情報

一般セッション(口頭講演)

10 スピントロニクス・マグネティクス » 10.3 スピンデバイス・磁気メモリ・ストレージ技術

[13a-A501-1~8] 10.3 スピンデバイス・磁気メモリ・ストレージ技術

2020年3月13日(金) 10:00 〜 12:00 A501 (6-501)

山本 竜也(産総研)

11:00 〜 11:15

[13a-A501-5] Solving Integer Factorization with Stochastic Magnetic Tunnel Junctions and a Quantum Adiabatic Algorithm

〇(DC)William Andrew Borders1、Ahmed Zeeshan Pervaiz2、Shunsuke Fukami1,3,4,5,6、Kerem Yunus Camsari2、Supriyo Datta2、Hideo Ohno1,3,4,5,6 (1.RIEC Tohoku Univ.、2.School of Electrical and Computer Engineering Purdue Univ.、3.CIES Tohoku Univ.、4.CSRN Tohoku Univ.、5.CSIS Tohoku Univ.、6.WPI-AIMR Tohoku Univ.)

キーワード:Magnetic Tunnel Junction, Probabilistic Computing, Optimization Problems

Optimization problems are a class of computationally complex problems that conventional computers struggle to solve. This struggle stems from the fact that optimization problems search for the most efficient solution out of a multitude of possible answers. Quantum computing shows potential for solving these problems, but is currently limited by technological barriers including decoherence, the number of implementable many-body interactions, and a current requirement of cryogenic temperatures for operation.
In this work, we demonstrate the use of a probabilistic computer to solve integer factorization, which can be cast as an optimization problem [1]. We develop an algorithm initially from the field of adiabatic quantum computing, but instead connect bits electrically, which allow us to circumvent some of the above challenges quantum computing faces. We implement these bits as probabilistic bits, or p-bits, which are composed of stochastic magnetic tunnel junctions (s-MTJs) fluctuating in time between 0 and 1 at room temperature. We first describe the characterization of s-MTJs where we modify a typical magneto-resistive random-access memory (MRAM) stack structure. We then connect the s-MTJ with CMOS circuit components to create a p-bit where the probability for the output to be either 0 or 1 is controlled by the input. By connecting eight p-bits we then demonstrate integer factorization up to 945. These result show that probabilistic computing with p-bits can provide a classical analog to quantum computing and offer a scalable hardware for solving optimization problems.
A portion of this work was supported by ImPACT Program of CSTI, JSPS KAKENHI grant numbers 17H06093 and 19J12206, Cooperative Research Projects of RIEC, and ASCENT, one of six centers in JUMP, an SRC program sponsored by DARPA.
[1] William A. Borders et al., Nature 573, 390 (2019).