A 3.8μW 10-Keyword Noise-Robust Keyword Spotting Processor Using Symmetric Compressed Ternary-Weight Neural Networks

A ternary-weight neural network (TWN) inspired keyword spotting (KWS) processor is proposed to support complicated and variable application scenarios. To achieve high-precision recognition of 10 keywords under 5dB Clean wide range background noises, a convolution consists 4 layers fully connected layers, with modified sparsity-controllable Truncated Gaussian Approximation based training used. End end optimization composed three techniques are utilized: 1) the stage-by-stage bit-width selection algorithm optimize hardware overhead FFT; 2) lossy compressed TWN symmetric kernel (SKT) dedicated internal data reuse computation flow; 3) error inter-compensation approximate addition tree reduce marginal accuracy loss. Fabricated in an industrial 22-nm CMOS process, realizes up real-time 11 noise types, 90.6%@clean 85.4%@5dB. It consumes average power 3.8 μW at 250KHz normalized energy efficiency 2.79× higher than state-of-the-art.