POLYN, a fabless semiconductor company supplying application-specific Neuromorphic Analogue Signal Processing (NASP) technology and products, has developed NeuroVoice, a Neuromorphic Front End (NFE) chip that can perform on-device Voice Activity Recognition.
SkyWater is a US-based semiconductor manufacturer and a DMEA-accredited Category 1A Trusted Foundry.
In always-on, at-microphone use cases, NeuroVoice completely offloads neural network processing for voice detection without the use of a microcontroller (MCU) or Digital Signal Processor (DSP). This enables an extremely energy-efficient AI-empowered sensor node, freeing up the MCU for other general-purpose computation.
Working at 30µW in active mode, the solution is designed to wake up any other system in the IoT device upon voice detection.
The NeuroVoice chip supports various system configurations through standard interfaces for MCU and simple pin control, which assists customers in developing new solutions, and could be integrated either like a system-in-package (SIP) or module with minimal design complexity.
“This is an important milestone for our technology, to guarantee our first product and the continuous innovation of analogue neuromorphic IP, expanding our NFE product lines,” said Aleksandr Timofeev, co-founder and CEO of POLYN Technology. “The first production chip using SkyWater’s S90LN platform will showcase the possibilities for the next generation of AI-enabled sensors – in a range of industries – for extracting useful information at ultra-low-power and inference latency.”
According to Ross Miller SVP, Strategic Marketing at SkyWater, “The need for efficient, low-power distributed data processing at the edge is huge and growing, and POLYN’s solution addresses this need. Our Advanced Technology Services (ATS) business model allowed us to merge POLYN’s proprietary structures with our S90LN technology platform to build a hybrid technology solution.”
