Presentation 2023-03-01
Programmable Binary Hyperdimensional Computing Accelerator for Low Power Devices
Yuya Isaka, Nau Sakaguchi, Michiko Inoue, Michihiro Shintani,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Hyperdimensional computing (HDC) can perform various cognitive tasks efficiently by mapping data to hyperdimensional vectors consisting of thousands to tens of thousands of dimensions. On the other hand, since the main operations of HDC, Bind, Permutation and Bound, require several cycles in the computing unit, it is not necessarily efficient to perform HDC on CPU platform. In this paper, we propose a programmable accelerator specialized for HDC. Our accelerator can execute various tasks at high speed and with low power consumption by cooperating with the CPU. Furthermore, by making it possible to freely select three operations per cycle, our accelerator enables to support any HDC encoding method. Through evaluation experiments with the ARM-v7 processor, we show that the above operations can be accelerated by a maximum of 169 times. We also confirmed that our accelerator can improve the energy-delay product up to 13,469 times in the training of handwritten character recognition task.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Hyperdimensional Computing / Hardware accelerator / Domain-specific architecture
Paper # VLD2022-76,HWS2022-47
Date of Issue 2023-02-22 (VLD, HWS)

Conference Information
Committee HWS / VLD
Conference Date 2023/3/1(4days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Makoto Nagata(Kobe Univ.) / Minako Ikeda(NTT)
Vice Chair Yuichi Hayashi(NAIST) / Daisuke Suzuki(Mitsubishi Electric) / Shigetoshi Nakatake(Univ. of Kitakyushu)
Secretary Yuichi Hayashi(Sony Semiconductor Solutions) / Daisuke Suzuki(NAIST) / Shigetoshi Nakatake(NBS)
Assistant / Takuma Nishimoto(Hitachi)

Paper Information
Registration To Technical Committee on Hardware Security / Technical Committee on VLSI Design Technologies
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Programmable Binary Hyperdimensional Computing Accelerator for Low Power Devices
Sub Title (in English)
Keyword(1) Hyperdimensional Computing
Keyword(2) Hardware accelerator
Keyword(3) Domain-specific architecture
1st Author's Name Yuya Isaka
1st Author's Affiliation Nara Institute of Science and Technology(NAIST)
2nd Author's Name Nau Sakaguchi
2nd Author's Affiliation San Jose State University(SJSU)
3rd Author's Name Michiko Inoue
3rd Author's Affiliation Nara Institute of Science and Technology(NAIST)
4th Author's Name Michihiro Shintani
4th Author's Affiliation Kyoto Institute of Technology(KIT)
Date 2023-03-01
Paper # VLD2022-76,HWS2022-47
Volume (vol) vol.122
Number (no) VLD-402,HWS-403
Page pp.pp.19-24(VLD), pp.19-24(HWS),
#Pages 6
Date of Issue 2023-02-22 (VLD, HWS)