SynSense advances neuromorphic audio processing
31/10/2023
Xylo Audio 3 is based on the Taiwan Semiconductor Manufacturing Company (TSMC)’s 40 nm CMOS logic low-power process and is able to deliver real-time ultra-low-power audio signal processing capabilities while reducing chip costs.
This tape-out is being described as a milestone for the commercialisation of SynSense’s neuromorphic audio processing technology.
“With Xylo Audio we are expanding the sensory processing capabilities of our neuromorphic hardware to real-time audio and vibration monitoring. This tape-out is an important step for our Xylo family devices, incorporating advances in low-power design as well as improved audio interfacing, and readies Xylo Audio for volume production,” said Dylan Muir, Vice President of Global Research Operations at SynSense.
Based on temporally sparse, quantised neural networks (spiking neural networks (SNNs)), Xylo Audio 3 is an ultra-low-power always-on artificial intelligence (AI) chip for auditory processing. The highly efficient spiking neural network temporal processing units in Xylo can reduce the energy consumption for complex temporal signal analysis tasks by at least ten
times.
Xylo Audio 3 integrates a configurable audio front-end (AFE) that accurately converts audio inputs into spikes, providing compatibility for a range of analogue and digital micro-electromechanical systems (MEMS) microphones.
The latest enhancements to the Xylo Audio family feature more flexibility in network configuration coupled with lower-power operation. Xylo Audio 3 enables edge artificial intelligence for near-sensor auditory applications, keyword recognition, environmental sound monitoring, industrial vibration monitoring and more, empowering a wide range of edge intelligence applications.
“We are providing Xylo Audio to many commercial and research partners, who are developing applications ranging from in-home audio event monitoring to underwater acoustic signature detection. Xylo enables our partners to build sophisticated audio-driven intelligence into their battery-powered edge devices,” added Dylan.
In addition to its chip design, Xylo Audio 3’s use of the Taiwan Semiconductor Manufacturing Company 40 nm process contributes to cost reduction and ensures the stability of tape-out and production.
SynSense has already launched development kits based on Xylo Audio and the company is actively supporting customers in application development. Dylan added: “We expect engineering samples of Xylo Audio to be available at the end of 2023, with commercial volume sales to follow.”
SynSense (formerly aiCTX) was established in 2017 and is a world leader in supplying neuromorphic intelligence and application solutions. The company focuses on the commercialisation of neuromorphic intelligence, based on over 20 years of experience at the University of Zurich and ETH Zürich. Centring on edge computing applications, SynSense provides ultra-low-power consumption, ultra-low latency inference ASICs and IP blocks, as well as full-stack application development services and is claimed to be the only global neuromorphic technology company that involves both sensing and computing.
