EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing

This paper proposed an energy-efficient reconfigurable accelerator for keyword spotting (EERA-KWS) based on binary weight network (BWN) and fabricated in 28-nm CMOS technology. This keyword spotting system consists of two parts: the feature extraction based on melscale frequency cepstral coefficient...

Descripción completa

Guardado en:

Detalles Bibliográficos
Autores principales:	Bo Liu, Zhen Wang, Hu Fan, Jing Yang, Wentao Zhu, Lepeng Huang, Yu Gong, Wei Ge, Longxing Shi
Formato:	article
Lenguaje:	EN
Publicado:	IEEE 2019
Materias:	Keyword spotting binary weight network approximate computing Electrical engineering. Electronics. Nuclear engineering TK1-9971
Acceso en línea:	https://doaj.org/article/08b9754fd8cb4ab6905f47947b201669
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

id	oai:doaj.org-article:08b9754fd8cb4ab6905f47947b201669
record_format	dspace
spelling	oai:doaj.org-article:08b9754fd8cb4ab6905f47947b2016692021-11-20T00:00:31ZEERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing2169-353610.1109/ACCESS.2019.2924340https://doaj.org/article/08b9754fd8cb4ab6905f47947b2016692019-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/8743369/https://doaj.org/toc/2169-3536This paper proposed an energy-efficient reconfigurable accelerator for keyword spotting (EERA-KWS) based on binary weight network (BWN) and fabricated in 28-nm CMOS technology. This keyword spotting system consists of two parts: the feature extraction based on melscale frequency cepstral coefficients (MFCC) and the keywords classification based on a BWN model, which is trained through the Google’s Speech Commands database and deployed on our custom. To reduce the power consumption while maintaining the system recognition accuracy, we first optimize the MFCC implementation with approximate computing techniques, including Pre-emphasis coefficient transformation, rectangular Mel filtering, Framing and FFT optimization. Then, we propose a precision self-adaptive reconfigurable accelerator with digital-analog mixed approximate computing units to process the BWN efficiently. Based on the SNR prediction of background noise and post-detection of network output confidence, the BWN accelerator data path can be dynamically and adaptively reconfigured as 4, 8, or 16 bits. For the BWN accelerator, we proposed a time-delay based addition unit to process bit-wise approximate computing for the convolution layers and fully connected layers, and a LUT based unit for the activation layers. Implemented under TSMC 28 nm HPC+ process technology, the estimated power is <inline-formula> <tex-math notation="LaTeX">$77.8~\mu \text{W}~\sim ~115.9\mu \text{W}$ </tex-math></inline-formula>, the energy efficiency can achieve 163 TOPS/W, which is over <inline-formula> <tex-math notation="LaTeX">$1.8\times $ </tex-math></inline-formula> better than the state-of-the-art architecture.Bo LiuZhen WangHu FanJing YangBo LiuWentao ZhuLepeng HuangYu GongWei GeLongxing ShiIEEEarticleKeyword spottingbinary weight networkapproximate computingElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 7, Pp 82453-82465 (2019)
institution	DOAJ
collection	DOAJ
language	EN
topic	Keyword spotting binary weight network approximate computing Electrical engineering. Electronics. Nuclear engineering TK1-9971
spellingShingle	Keyword spotting binary weight network approximate computing Electrical engineering. Electronics. Nuclear engineering TK1-9971 Bo Liu Zhen Wang Hu Fan Jing Yang Bo Liu Wentao Zhu Lepeng Huang Yu Gong Wei Ge Longxing Shi EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
description	This paper proposed an energy-efficient reconfigurable accelerator for keyword spotting (EERA-KWS) based on binary weight network (BWN) and fabricated in 28-nm CMOS technology. This keyword spotting system consists of two parts: the feature extraction based on melscale frequency cepstral coefficients (MFCC) and the keywords classification based on a BWN model, which is trained through the Google’s Speech Commands database and deployed on our custom. To reduce the power consumption while maintaining the system recognition accuracy, we first optimize the MFCC implementation with approximate computing techniques, including Pre-emphasis coefficient transformation, rectangular Mel filtering, Framing and FFT optimization. Then, we propose a precision self-adaptive reconfigurable accelerator with digital-analog mixed approximate computing units to process the BWN efficiently. Based on the SNR prediction of background noise and post-detection of network output confidence, the BWN accelerator data path can be dynamically and adaptively reconfigured as 4, 8, or 16 bits. For the BWN accelerator, we proposed a time-delay based addition unit to process bit-wise approximate computing for the convolution layers and fully connected layers, and a LUT based unit for the activation layers. Implemented under TSMC 28 nm HPC+ process technology, the estimated power is <inline-formula> <tex-math notation="LaTeX">$77.8~\mu \text{W}~\sim ~115.9\mu \text{W}$ </tex-math></inline-formula>, the energy efficiency can achieve 163 TOPS/W, which is over <inline-formula> <tex-math notation="LaTeX">$1.8\times $ </tex-math></inline-formula> better than the state-of-the-art architecture.
format	article
author	Bo Liu Zhen Wang Hu Fan Jing Yang Bo Liu Wentao Zhu Lepeng Huang Yu Gong Wei Ge Longxing Shi
author_facet	Bo Liu Zhen Wang Hu Fan Jing Yang Bo Liu Wentao Zhu Lepeng Huang Yu Gong Wei Ge Longxing Shi
author_sort	Bo Liu
title	EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
title_short	EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
title_full	EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
title_fullStr	EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
title_full_unstemmed	EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing
title_sort	eera-kws: a 163 tops/w always-on keyword spotting accelerator in 28nm cmos using binary weight network and precision self-adaptive approximate computing
publisher	IEEE
publishDate	2019
url	https://doaj.org/article/08b9754fd8cb4ab6905f47947b201669
work_keys_str_mv	AT boliu eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT zhenwang eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT hufan eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT jingyang eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT boliu eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT wentaozhu eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT lepenghuang eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT yugong eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT weige eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing AT longxingshi eerakwsa163topswalwaysonkeywordspottingacceleratorin28nmcmosusingbinaryweightnetworkandprecisionselfadaptiveapproximatecomputing
_version_	1718419817234432000

EERA-KWS: A 163 TOPS/W Always-on Keyword Spotting Accelerator in 28nm CMOS Using Binary Weight Network and Precision Self-Adaptive Approximate Computing

Ejemplares similares