Integrated-circuit implementation of chaotic Boltzmann machines for deep learning hardware

2016 Fiscal Year Final Research Report

• Project/Area Number: 15K12110
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Soft computing
• Research Institution: Kyushu Institute of Technology
• Principal Investigator: Morie Takashi 九州工業大学, 大学院生命体工学研究科, 教授 (20294530)
• Research Collaborator: YAMAGUCHI Masatoshi 九州工業大学, 大学院生命体工学研究科 ; KATO Takashi 九州工業大学, 大学院生命体工学研究科 ; WANG Quan 九州工業大学, 大学院生命体工学研究科 ; UENOHARA Seiji 九州工業大学, 大学院生命体工学研究科 ; KAWASHIMA Ichiro 九州工業大学, 大学院生命体工学研究科
• Project Period (FY): 2015-04-01 – 2017-03-31
• Keywords: ソフトコンピューティング / ニューラルネットワーク / 電子デバイス・機器 / 集積回路 / 深層学習 / 機械学習

Outline of Final Research Achievements

Boltzmann machines are an important stochastic model in machine learning, and a chaotic Boltzmann machine (CBM) model has been proposed using deterministic chaos. We have implemented this model by both analog and digital integrated circuits (ICs), evaluated their performance, and developed the circuit design technology that can be applied to deep-learning hardware. We have implemented an exponential function used in the CBM model by the characteristics of the subthreshold region of a MOSFET in analog ICs, and by bit shift operation in digital ICs, and verified the operation of CBMs using real IC chips. We have designed a 100-neuron CBM in a digital IC, applied it to max-cut problems, and verified the output of the optimum solutions.

Free Research Field

脳型集積システム