[18p-P06-1~11] FS Focused Session "AI Electronics" (Poster)

[18p-P06-8] Simulation for parallel processing on magneto-otpcial diffractive deep neural network

〇(D)Hotaka Sakaguchi¹, Hiroyuki Awano², Hirofumi Nonaka³, Fatima Zahra Chafi¹, Takayuki Ishibashi¹ (1.Nagaoka Univ. of Tech., 2.Toyota Tech. Inst., 3.Aichi Inst. of Tech.)

Keywords:

optical computing,magneto-optical effect,Neuromorphic computing

To solve the enormous power consumption problem of deep learning, we have developed a magneto-optical diffraction deep neural network (MO-D²NN) that utilizes the magneto-optical effect of magnetic materials. In the previous report, we reported on the development of an online learning technique using the Monte Carlo method, but it is necessary to reduce the learning time. Therefore, in this study, we investigated accelerating the computation by setting up multiple magnetic domain patterns in a single diffraction layer and performing parallel computation. In the presentation, we will report the results of learning simulation of spatial parallel operations using MO-D²NN.