講演抄録/キーワード |
講演名 |
2023-03-02 14:15
[記念講演]DependableHD: A Hyperdimensional Learning Framework for Edge-oriented Voltage-scaled Circuits [Memorial lecture] ○Dehua Liang(Osaka Univ.)・Hiromitsu Awano(Kyoto Univ.)・Noriyuki Miura・Jun Shiomi(Osaka Univ.) VLD2022-93 HWS2022-64 |
抄録 |
(和) |
Voltage scaling is a promising approach for energy efficiency but also brings challenges to guaranteeing stable circuit operation. To tackle such issues, we propose DependableHD, a learning framework based on HyperDimensional Computing, which supports the systems to tolerate bit-level memory failure in low voltage with high robustness. DependableHD introduces the concept of margin enhancement and utilizes noise injection to improve the robustness. Under 10% memory error, DependableHD exhibits 1.22% accuracy loss on average, which achieves an 11.2x improvement. It supports the systems to reduce the supply voltage from 400mV to 300mV, which provides 50.41% energy reduction while maintaining accuracy. |
(英) |
Voltage scaling is a promising approach for energy efficiency but also brings challenges to guaranteeing stable circuit operation. To tackle such issues, we propose DependableHD, a learning framework based on HyperDimensional Computing, which supports the systems to tolerate bit-level memory failure in low voltage with high robustness. DependableHD introduces the concept of margin enhancement and utilizes noise injection to improve the robustness. Under 10% memory error, DependableHD exhibits 1.22% accuracy loss on average, which achieves an 11.2x improvement. It supports the systems to reduce the supply voltage from 400mV to 300mV, which provides 50.41% energy reduction while maintaining accuracy. |
キーワード |
(和) |
HyperDimensional Computing / Voltage Scaling / Robustness / Energy Efficiency / / / / |
(英) |
HyperDimensional Computing / Voltage Scaling / Robustness / Energy Efficiency / / / / |
文献情報 |
信学技報, vol. 122, no. 402, VLD2022-93, pp. 111-111, 2023年3月. |
資料番号 |
VLD2022-93 |
発行日 |
2023-02-22 (VLD, HWS) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
VLD2022-93 HWS2022-64 |
|