Sub-mW Keyword Spotting on an MCU: Analog Binary Feature Extraction and Binary Neural Networks

被引:14
作者
Cerutti, Gianmarco [1 ]
Cavigelli, Lukas [2 ]
Andri, Renzo [2 ]
Magno, Michele [3 ]
Farella, Elisabetta [1 ]
Benini, Luca [3 ,4 ]
机构
[1] ICT Irst Fdn Bruno Kessler, I-38123 Trento, Italy
[2] Huawei Technol, Zurich Res Ctr, Comp Syst Lab, CH-8050 Zurich, Switzerland
[3] Swiss Fed Inst Technol, Dept Elect Engn & Informat Technol, CH-8092 Zurich, Switzerland
[4] Univ Bologna, Dipanimento Ingn Energia Elettr & Informaz, I-40126 Bologna, Italy
关键词
Feature extraction; Neural networks; Quantization (signal); Task analysis; Power demand; Computational modeling; Memory management; Keyword spotting; quantization; binary neural networks; deep learning; feature extraction; SENSOR;
D O I
10.1109/TCSI.2022.3142525
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Keyword spotting (KWS) is a crucial function enabling the interaction with the many ubiquitous smart devices in our surroundings, either activating them through wake-word or directly as a human-computer interface. For many applications, KWS is the entry point for our interactions with the device and, thus, an always-on workload. Many smart devices are mobile and their battery lifetime is heavily impacted by continuously running services. KWS and similar always-on services are thus the focus when optimizing the overall power consumption. This work addresses KWS energy-efficiency on low-cost microcontroller units (MCUs). We combine analog binary feature extraction with binary neural networks. By replacing the digital preprocessing with the proposed analog front-end, we show that the energy required for data acquisition and preprocessing can be reduced by 29x, cutting its share from a dominating 85% to a mere 16% of the overall energy consumption for our reference KWS application. Experimental evaluations on the Speech Commands Dataset show that the proposed system outperforms state-of-the-art accuracy and energy efficiency, respectively, by 1% and 4.3x on a 10-class dataset while providing a compelling accuracy-energy trade-off including a 2% accuracy drop for a 71x energy reduction.
引用
收藏
页码:2002 / 2012
页数:11
相关论文
共 51 条
[1]  
Alioto M., 2017, Enabling the Internet of Things: From Integrated Circuits to Integrated Systems
[2]   An optimized recurrent unit for ultra-low-power keyword spotting [J].
Amoh, Justice ;
Odame, Kofi M. .
Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 2019, 3 (02)
[3]  
[Anonymous], 2018, GREENWAVES TECHNOLOG
[4]  
[Anonymous], 2017, ARXIV170909888
[5]   Fair and Comprehensive Benchmarking of Machine Learning Processing Chips [J].
Burr, Geoffrey W. ;
Lim, SukHwan ;
Murmann, Boris ;
Venkatesan, Rangharajan ;
Verhelst, Marian .
IEEE DESIGN & TEST, 2022, 39 (03) :18-27
[6]   Compact Recurrent Neural Networks for Acoustic Event Detection on Low-Energy Low-Complexity Platforms [J].
Cerutti, Gianmarco ;
Prasad, Rahul ;
Brutti, Alessio ;
Farella, Elisabetta .
IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, 2020, 14 (04) :654-664
[7]  
Cerutti Gianmarco, 2020, P ACMIEEE INT S LOW, P19
[8]   A 2.5 μW KWS Engine With Pruned LSTM and Embedded MFCC for IoT Applications [J].
Chong, Y. S. ;
Goh, W. L. ;
Nambiar, V. P. ;
Do, A. T. .
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS, 2022, 69 (03) :1662-1666
[9]  
Delbruck T, 2010, IEEE INT SYMP CIRC S, P2015, DOI 10.1109/ISCAS.2010.5537160
[10]   Cloud Computing: Issues and Challenges [J].
Dillon, Tharam ;
Wu, Chen ;
Chang, Elizabeth .
2010 24TH IEEE INTERNATIONAL CONFERENCE ON ADVANCED INFORMATION NETWORKING AND APPLICATIONS (AINA), 2010, :27-33