Jan 07, 2024 10:00 ET | Hackster.io
With a RISC-V core, two NPU coprocessors, and a DSP, this tiny chip draws under 260μW while performing on-device inference.
On-device machine learning (ML) specialist AONDevices is showing off its latest chip, the AON1120, at the Consumer Electronics Show (CES) 2024 this week — combining a general-purpose RISC-V core with two neural processing units (NPU) to deliver a sub-260μW power draw at full load for audio classification tasks.
“The AON1120 chip is a testament to our commitment to cutting-edge, super low-power edge AI [Artificial Intelligence],” claims AONDevices chief executive officer Mouna Elkhatib of his company’s latest creation. “It enables vast improvements in voice, sound, and multi-sensor fusion applications, crucial for developers and businesses aiming for always-on functionality.”
The AON1120 is a system-on-chip design which features a single general-purpose processor core built atop the free and open-source RISC-V instruction set architecture (ISA) along with a specialized digital signal processor (DSP) and two neural processing unit (NPU) coprocessors designed specifically for machine learning workloads. All told, the company claims, the chip can run in “listening mode” for audio classification tasks with a power draw below 80μW — and draws less than 260μW when “at 100% processing of various AI tasks.”
In addition to voice command recognition with a claimed 90 per cent detection rate with a single microphone and 40 adaptable output classifications, the part also offers biometric, gesture, and motion recognition, plus sensor fusion capabilities and 16 general-purpose input/output (GPIU) pins. These, the company claims, make it suitable for use in a variety of use-cases including health and wellness monitoring, sleep analysis, voice-activated remote control systems, driver authentication, and acoustic monitoring of industrial equipment.
To support adoption of the new chip, AOMDevices is shipping an evaluation board which combines the edge AI processor with a TDK T5838 microphone, USB controller, and USB power, plus dedicated measurement points for core and input/output (IO) power draw. Each kit comes with what the company calls a “complete AI tool suite” including dataset augmentation, training, and inference optimization capabilities.
This evaluation kit is to be demonstrated at the Consumer Electronics Show (CES) 2024 in Las Vegas this week, the company has confirmed, with interested parties invited to contact AONDevices for more information; pricing and availability have not yet been publicly disclosed.
