2019年第66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)

31 フォーカストセッション「AIエレクトロニクス」 » 31.1 フォーカストセッション「AIエレクトロニクス」

[12p-W933-1~11] 31.1 フォーカストセッション「AIエレクトロニクス」

2019年3月12日(火) 13:15 〜 16:00 W933 (W933)

葛西 誠也(北大)

15:15 〜 15:30

[12p-W933-9] Two-bit input binary task with reservoir computing using nanomagnet array

Kazuki Tsujimoto1、Hikaru Nomura1、Taishi Furuta1、Yuki Kuwabiraki1、Ryoichi Nakatani1、Yoshishige Suzuki1 (1.Osaka Univ.)

キーワード:reservoir computing, macro-spin simulation

Neural network has made great achievements in recent years. However, a general neural network implemented on a CPU/GPU has a problem of power consumption. To reduce the power consumption, reservoir computing which use physical phenomenon for numerical calculation has been reported. Recently we demonstrated reservoir computing with nanomagnet array reservoir. In this reservoir computing, static magnetization of the nanomagnets in the reservoir are used as node state. Each node state are connected via dipole field of the nanomagnets. This reservoir can perform single-bit input binary tasks. However, more complex tasks such as image recognition require not a single-bit input but a multi-bit input.
In this presentation, we demonstrate reservoir computing with two-bit binary task using a nanomagnet array as a reservoir. We performed macro-spin simulations. As a results, an output matrix of this reservoir can be trained to perform two-bit binary input tasks with delay up to one.