
Always-on neural network for sensor fusion
AONDevices is a fabless chip company founded in 2018 by a group of industry veterans with expertise in audio processing and AI.
The AON1100 can detect multiple voice commands as well as sound events, such as a baby crying or a car backfiring using a single microphone. The same chip can also detect motion patterns such as walking or falling.
AON1100 features two neural network cores, AONVoice and AONSens. The AONVoice delivers 90 percent accuracy in 0dB SNR conditions. Including front-end processing and all required memory, AONVoice core consumes less than 150microwatts when implemented in 40nm ULP silicon during continuous speech.
“We are engaged with several tier-1 customers in key markets. With engineering samples and development kits now available, we can support OEM customers looking for complete solutions that don’t sacrifice privacy or trade accuracy for power,” said Mouna Elkhatib, CEO of AONDevices, in a statement. “In addition, we continue to enable key silicon partners and OEMs with IP core licenses.”
AONDevices is a partner of the tinyML Foundation and embraces the vision set forth by the organization.
Engineering samples of the AON1100 chip are available now. AONVoice and AONSens cores are also available for licensing as IP cores for integration in an SoC to high-volume customers.
Related links and articles:
News articles:
Machine learning improves ultrasound buttons
Alif benchmarks ARM Ethos-55 neural processing unit
Edge Impulse raises $34m for TinyML tool
Plumerai develops embedded AI accelerator IP core for FPGAs
