Approximate Computing Platform Maximizing Computating Efficiency of Near-Threshold Circuits

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K19768
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 60040:Computer system-related
• Research Institution: Osaka University (2021-2023); Kyoto University (2020)
• Principal Investigator: Shiomi Jun 大阪大学, 大学院情報科学研究科, 准教授 (40809795)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: 計算機システム / 省エネルギー / 低消費電力化 / ニアスレッショルドコンピューティング / 近似コンピューティング

Outline of Final Research Achievements

This project developed a high-efficiency computing platform combining Near-Threshold Computing (NTC), where integrated circuits operate at less than half of the nominal supply voltage, and Approximate Computing (AC), which can tolerate small computational errors. However, NTC has a drawback that it can cause partial operation failures in the target circuit. This project developed NTC circuit design methods that absorb this drawback with AC. Specifically, this project was divided into brain-inspired computer design, low-voltage on-chip memory design, and dynamic voltage optimization methods. This project results in three publications in international journals, three presentations at international conferences, two presentations at domestic conferences, and three awards. The research concept of combining NTC and AC has been successfully demonstrated. Its impact is thus very significant.

Free Research Field

集積回路設計

Academic Significance and Societal Importance of the Research Achievements

あらゆるヒトやモノにIoT (Internet of Things)デバイスが組み込まれ、多種多様な情報を絶え間なく通信する次世代情報化社会が到来している。リアルタイム性、バッテリ駆動時間、CO2排出量の問題から、自動運転・ファクトリオートメーションなどの新サービスを支える集積回路の省エネルギー化・高性能化が喫緊の課題となっている。同時に、AC技術適用先の典型例である人工知能技術が急速に発展している。この社会トレンドをおさえ、NTCの欠点である動作技術をAC技術で克服し、NTC回路の利点である省エネルギー動作を享受できた意義は大きい。