POLYN Technology signs first customer for NASP Tiny AI chips

1 min read

POLYN Technology, a fabless semiconductor company providing Neuromorphic Analogue Signal Processing (NASP) Tiny AI chips, has signed its first contract for NeuroVoice, its latest product for voice processing.

NeuroVoice is an AI chip that performs on-chip Voice Activity Recognition and industry-unique voice extraction from any noisy background. It also offers additional voice features such as Wake Word Detection and Keyword Spotting. All voice features can be used in any combination or separately, based on the customer’s requirements.

The contract is with a home appliance company will see POLYN deliver a solution with a neural network that recognises voice commands offline, eliminating any need to access the cloud and will ensure that end users’ privacy won’t be violated.

The NASP-based NeuroVoice has an analogue neuromorphic design, which performs maximum MAC operations in analogue. The NASP framework and tools convert a neural network into a silicon chip implemented in 65nm CMOS and BEOL.

“The contract demonstrates market demand for a new generation of voice processing products,” said Aleksandr Timofeev, CEO of POLYN. “It represents the customer’s choice of NeuroVoice as the lowest power consumption chip available in the market.  The simplicity of its integration was also an important selection factor for this customer.”

NeuroVoice will be integrated right after the sensor to provide a binary output of voice detection to wake up other functions. In the first stage of the project, POLYN will provide ultra-low-power Voice Activity Recognition with a very small die size of 1.5x1.5mm for always-on environment monitoring for Voice Detection. The final product will contain all voice features in 3.5x3.5mm die size, delivering a total power consumption of only 150µW for the whole inference.

“We anticipate that this contract will result in several millions of dollars of revenue for POLYN over the next two years,” Timofeev indicated. “This is just the first of many customers we expect to take advantage of NeuroVoice.”