SynSense has taped out its Xylo Audio 3 ultra-low-power audio processing platform built on an neuromorphic AI inference core.
Xylo Audio 3 is built on TSMC’s 40nm CMOS low power logic process for real-time neuromorphic AI audio signal processing capabilities with lower die cost.
The company has added more flexibility in network configuration coupled with lower-power operation for edge AI for near-sensor auditory applications, for keyword recognition, environmental sound monitoring, industrial vibration monitoring and more, empowering a wide range of edge intelligence applications.
“We expect engineering samples of Xylo Audio to be available at the end of 2023, with commercial volume sales to follow,” said Dylan Muir, Vice President of Global Research Operations at SynSense.
The tapeout marks a milestone for the commercialization of SynSense’s neuromorphic AI audio processing technology using a temporally sparse, quantized neural networks, also known as a spiking neural network, or SNN.
The highly efficient SNN temporal processing units in Xylo reduce the energy consumption for complex temporal signal analysis tasks by a factor of ten says SynSense, a spin out of hardware developed at the Institute of Neuroinformatics of the University of Zurich and ETH Zurich.
The chip integrates a configurable audio front-end (AFE) which accurately converts audio inputs into spikes, with excellent compatibility for a range of analog and digital MEMS microphones.