A 510nW 0.41V Low-Memory Low-Computation Keyword-Spotting Chip Using Serial FFT-Based MFCC and Binarized Depthwise Separable Convolutional Neural Network in 28nm CMOS

📋 论文概要

本文提出了一款功耗仅510nW、电压0.41V的关键词唤醒芯片，采用串行FFT的MFCC特征提取和二值化深度可分离卷积神经网络，解决了传统KWS芯片功耗过高（>5µW）的问题，实现了超低功耗的始终在线语音接口。

💡 主要创新点

• 首次实现510nW级别的关键词唤醒芯片，功耗比之前工作降低一个数量级。
• 采用串行FFT的MFCC和二值化深度可分离卷积，大幅降低内存和计算量。

🏷 关键词

📄 原文摘要

is a strong requirement for always-on speech interfaces in wearable and mobile devices, such as Voice Activity Detection (VAD) and Keyword Spotting (KWS) [1-5]. A KWS system is used to detect specific wake-up words by speakers and has to be always on. Previous ASICs for KWS lack energyefficient implementations having power <5µW. For example, deep neural network (DNN)-based KWS [1] has a large on-chip weight memory of 270KB and consumes 288µW. A binarized convolutional neural network (CNN) used 52KB of SRAM,141µW wakeup power at 2.5MHz, 0.57V [2]. An LSTM-based SoC used 105KB of SRAM and reduced power to 16.11µW for KWS with 90.8% accuracy on the Google Speech Command Dataset (GSCD) [3]. Laika reduced power to 5µW [4], not including the Mel Frequency Cepstrum Coefficient (MFCC) circuit.

👥 作者与机构

Weiwei Shan1, Minhao Yang2, Jiaming Xu1, Yicheng Lu1, Shuai Zhang1,

Tao Wang1, Jun Yang1, Longxing Shi1, Mingoo Seok3 Southeast University, Jiangsu, China EPFL, Neuchâtel, Switzerland 3 Columbia University, New York, NY 1 2 Ultra-low power