14.1 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.
Weiwei ShanMinhao YangJiaming XuYicheng LuShuai ZhangTao WangJun YangLongxing ShiMingoo SeokPublished in: ISSCC (2020)
Keyphrases
- low memory
- keyword spotting
- convolutional neural network
- speech recognition
- cmos technology
- analog vlsi
- nm technology
- high speed
- silicon on insulator
- metal oxide semiconductor
- low power
- power consumption
- speech processing
- image sensor
- face detection
- memory requirements
- hidden markov models
- speech signal
- document images
- neural network
- pattern recognition
- power dissipation
- printed documents
- handwritten documents
- machine learning
- feature extraction
- input image
- frequency domain
- probabilistic model
- automatic speech recognition
- feature vectors
- floating point
- ibm power processor
- multiscale
- language model