2021年第82回応用物理学会秋季学術講演会

講演情報

一般セッション(口頭講演)

FS フォーカストセッション「AIエレクトロニクス」 » FS.1 フォーカストセッション「AIエレクトロニクス」

[13a-S101-1~10] FS.1 フォーカストセッション「AIエレクトロニクス」

2021年9月13日(月) 09:00 〜 11:45 S101 (口頭)

佐藤 昇男(NTT)、江崎 瑞仙(東芝)

11:30 〜 11:45

[13a-S101-10] Colored Noise Induced Synchronization of Ultra-low Power Consumption Spintronics Stochastic Neurons

Zhiqiang Liao1、Kaijie Ma1、Siyi Tang1、Md Shamim Sarker1、Hiroyasu Yamahara1、Hitoshi Tabata1 (1.Univ. of Tokyo)

キーワード:Superparamagnetic tunnel junctions, Spintronics stochastic neurons, Synchronization

Superparamagnetic tunnel junctions (SMTJs) are nanodevices that can oscillate without driving signals. Similar to a human brain, when external information-carrying signals control this stochastic characteristic, SMTJs can process the neuromorphic computing tasks. Thus, it makes them be considered as a promising candidate for physical AI. To control the randomness of the SMTJs, a noise-induced phase-locking method was proposed to control the synchronization of SMTJs with ultra-low power consumption. However, the studied noise is confined to Gaussian-distributed white noise.
Compared with white noise, natural noises generally have “color”. For example, the fluctuation of device performance and the Brownian motion of electrons can cause different colored noise, so SMTJs will inevitably be affected in practical applications. Therefore, it is necessary to investigate the effect of colored noise on SMTJs' subthreshold synchronization.
This work numerically studied the SMTJs based on the experimental data. Under the subthreshold driving voltage, the influence of colored noise on SMTJs is analyzed quantitatively. The results show that red and pink noise can suppress the subthreshold synchronization of SMTJs, while violet and blue noise can phase-lock SMTJs into a subthreshold signal with lower energy than white noise.