A $5.2\mu \mathrm{W}$, 2-to-8-Channel Scalable, Speaker-Tracking Microphone Array Featuring a CNN-Defined AFE

This paper describes a smart microphone array with speaker-tracking capability to enhance the output voice signal-to-noise ratio (SNR). The speaker verification (SV) module identifies the target speaker in both near and far fields and intelligently adjusts the beamforming direction-of-arrival (DoA) and the analog front-end (AFE) gain for a better output voice SNR. The system chip in 28nm process is bonded with a capacitive micro-electro-mechanical system (MEMS) microphone, and it integrates an in-sensor-computing channel that can arbitrarily scale between 2 to 8, to realize a small form-factor DoA. With also our energy-efficient delay and sum (DAS) and SV circuit techniques, the system chip consumes only <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$5.2\mu \mathrm{W}$</tex>.