ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing

In this paper, we present a low-power anomaly detection integrated circuit (ADIC) based on a one-class classifier (OCC) neural network. The ADIC achieves low-power operation through a combination of (a) careful choice of algorithm for online learning and (b) approximate computing techniques to lo...

Full description

Saved in:
Bibliographic Details
Main Authors: Kar, Bapi, Gopalakrishnan, Pradeep Kumar, Bose, Sumon Kumar, Roy, Mohendra, Basu, Arindam
Other Authors: School of Electrical and Electronic Engineering
Format: Article
Language:English
Published: 2022
Subjects:
Online Access:https://hdl.handle.net/10356/160503
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
id sg-ntu-dr.10356-160503
record_format dspace
spelling sg-ntu-dr.10356-1605032022-07-25T08:26:45Z ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing Kar, Bapi Gopalakrishnan, Pradeep Kumar Bose, Sumon Kumar Roy, Mohendra Basu, Arindam School of Electrical and Electronic Engineering Delta-NTU Corporate Laboratory for Cyber Physical Systems Engineering::Electrical and electronic engineering Anomaly Detection Approximate Computing In this paper, we present a low-power anomaly detection integrated circuit (ADIC) based on a one-class classifier (OCC) neural network. The ADIC achieves low-power operation through a combination of (a) careful choice of algorithm for online learning and (b) approximate computing techniques to lower average energy. In particular, online pseudoinverse update method (OPIUM) is used to train a randomized neural network for quick and resource efficient learning. An additional 42% energy saving can be achieved when a lighter version of OPIUM method is used for training with the same number of data samples lead to no significant compromise on the quality of inference. Instead of a single classifier with large number of neurons, an ensemble of K base learner approach is chosen to reduce learning memory by a factor of K. This also enables approximate computing by dynamically varying the neural network size based on anomaly detection. Fabricated in 65nm CMOS, the ADIC has K = 7 Base Learners (BL) with 32 neurons in each BL and dissipates 11.87pJ/OP and 3.35pJ/OP during learning and inference respectively at Vdd = 0.75V when all 7 BLs are enabled. Further, evaluated on the NASA bearing dataset, approximately 80% of the chip can be shut down for 99% of the lifetime leading to an energy efficiency of 0.48pJ/OP, an 18.5 times reduction over full-precision computing running at Vdd = 1.2V throughout the lifetime. National Research Foundation (NRF) This work was supported in part by Delta Electronics Inc. and in part by the National Research Foundation Singapore under the Corporate Laboratory@UniversityScheme. 2022-07-25T08:26:45Z 2022-07-25T08:26:45Z 2020 Journal Article Kar, B., Gopalakrishnan, P. K., Bose, S. K., Roy, M. & Basu, A. (2020). ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing. IEEE Transactions On Very Large Scale Integration (VLSI) Systems, 28(12), 2518-2529. https://dx.doi.org/10.1109/TVLSI.2020.3016939 1063-8210 https://hdl.handle.net/10356/160503 10.1109/TVLSI.2020.3016939 2-s2.0-85097353599 12 28 2518 2529 en IEEE Transactions on Very Large Scale Integration (VLSI) Systems © 2020 IEEE. All rights reserved.
institution Nanyang Technological University
building NTU Library
continent Asia
country Singapore
Singapore
content_provider NTU Library
collection DR-NTU
language English
topic Engineering::Electrical and electronic engineering
Anomaly Detection
Approximate Computing
spellingShingle Engineering::Electrical and electronic engineering
Anomaly Detection
Approximate Computing
Kar, Bapi
Gopalakrishnan, Pradeep Kumar
Bose, Sumon Kumar
Roy, Mohendra
Basu, Arindam
ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
description In this paper, we present a low-power anomaly detection integrated circuit (ADIC) based on a one-class classifier (OCC) neural network. The ADIC achieves low-power operation through a combination of (a) careful choice of algorithm for online learning and (b) approximate computing techniques to lower average energy. In particular, online pseudoinverse update method (OPIUM) is used to train a randomized neural network for quick and resource efficient learning. An additional 42% energy saving can be achieved when a lighter version of OPIUM method is used for training with the same number of data samples lead to no significant compromise on the quality of inference. Instead of a single classifier with large number of neurons, an ensemble of K base learner approach is chosen to reduce learning memory by a factor of K. This also enables approximate computing by dynamically varying the neural network size based on anomaly detection. Fabricated in 65nm CMOS, the ADIC has K = 7 Base Learners (BL) with 32 neurons in each BL and dissipates 11.87pJ/OP and 3.35pJ/OP during learning and inference respectively at Vdd = 0.75V when all 7 BLs are enabled. Further, evaluated on the NASA bearing dataset, approximately 80% of the chip can be shut down for 99% of the lifetime leading to an energy efficiency of 0.48pJ/OP, an 18.5 times reduction over full-precision computing running at Vdd = 1.2V throughout the lifetime.
author2 School of Electrical and Electronic Engineering
author_facet School of Electrical and Electronic Engineering
Kar, Bapi
Gopalakrishnan, Pradeep Kumar
Bose, Sumon Kumar
Roy, Mohendra
Basu, Arindam
format Article
author Kar, Bapi
Gopalakrishnan, Pradeep Kumar
Bose, Sumon Kumar
Roy, Mohendra
Basu, Arindam
author_sort Kar, Bapi
title ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
title_short ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
title_full ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
title_fullStr ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
title_full_unstemmed ADIC: anomaly detection integrated circuit in 65-nm CMOS utilizing approximate computing
title_sort adic: anomaly detection integrated circuit in 65-nm cmos utilizing approximate computing
publishDate 2022
url https://hdl.handle.net/10356/160503
_version_ 1739837397537914880