← 返回 JSSC 论文列表JSSC 2023第11期Digital Circuits65nm
A 47-nW V oice Activity Detector V AD Featuring a Short-Time CNN Feature Extract
一款基于短时CNN和RNN的低功耗语音活动检测器,采用非易失性电容ROM存储权重。
65nm CMOS, 47nW, 0.022mm², 94%/91% hit rate
语音活动检测器低功耗卷积神经网络循环神经网络非易失性存储
▸创新点1:短时卷积神经网络(ST-CNN)是一种方法创新,通过缩短特征提取窗口显著降低了信号泄漏和检测延迟,同时减少了面积和功耗预算,实现了94%/91%的高检测准确率。
▸创新点2:基于RNN的分类器是一种方法创新,通过优化网络结构将VAD参数减少至仅45个,大幅降低了计算复杂度和存储需求,同时保持了高检测性能。
▸创新点3:非易失性电容ROM(CAP-ROM)是一种电路创新,取代了传统易失性存储器,消除了权重预加载过程,减少了功耗(47 nW)和面积(0.022 mm²),提升了系统能效。
▸创新点4:系统创新体现在VAD的鲁棒性设计,在0.9至1.3 V电源电压和0至60°C温度范围内,检测准确率无显著下降,适用于各种边缘设备环境。
Abstract
This article reports an area-and-power-efficient
voice activity detector (V AD) for voice-control edge devices.
It innovates a short-time convolutional neural network (ST-CNN)
and a recurrent neural network (RNN)-based classifier. Such a
classifier shortens the extraction window of the ST-CNN while
reducing its signal leakage, detection latency, and area and
power budgets. The RNN also aids in parameter reduction of
the V AD to only 45. We also propose the non-volatile capacitor-
ROM (CAP-ROM) a