← 返回 JSSC 论文列表
📄 下载 JSSC 原文 PDF
JSSC 2021第1期Memory28nm

A 510-nW Wake-Up Keyword-Spotting Chip Using Serial-FFT-Based MFCC and Binarized

一款基于串行FFT和二进制神经网络的超低功耗唤醒关键词识别芯片
28nm CMOS, 0.41V, 40kHz, 0.51μW, 0.23mm²
关键词识别超低功耗串行FFT二进制神经网络近阈值电压
采用串行FFT的梅尔频率倒谱系数特征提取电路
小型化二进制深度可分离卷积神经网络分类器
帧级增量计算技术与近阈值电压操作
Abstract
We propose a sub- µW always- ON keyword spot- ting ( µKWS) chip for audio wake-up systems. It is mainly composed of a neural network (NN) and a feature extraction (FE) circuit. For significantly reducing the memory footprint and computational load, four techniques are used to achieve ultra- low-power consumption: 1) a serial-FFT-based Mel-frequency cepstrum coefficient circuit is designed for FE, instead of the common parallel FFT. 2) A small-sized binarized depthwise separable convolutional NN (D