AONDevices’ AON1100 and AON1120 chipsets provide designers with always-on, real-time AI-driven voice, sound, and sensor-fusion data recognition at extremely low power consumption, the lowest on the market at time of release at <260 µW when processing and <80 µW when listening – this benchmark assumes a single-microphone, 12-keyword voice-recognition workload with >90% accuracy.
The AI models are trained, augmented, and optimised using AONDevices’ AONx360 software platform and are deployed directly to the hardware so that no cloud-based processing or specialist algorithm/firmware development is required. This enables fast time to market and high-privacy AI for applications that include home automation, wearables, automotive HMIs, industrial IoT, entertainment, and more. The high level of performance can even distinguish between different user voices.
AONDevices’ hardware solution comes fully integrated with a dual-NPU RISC-V architecture where each neural network core can be individually configured to suit the workloads required by the user’s application. The more advanced AON1120 device comes in a compact 40QFN package with 16 GPIOs for sensor interfacing and peripheral control based on the analysis of incoming data.
Currently available workloads for AONDevices’ high-efficiency AI algorithms include:
- Multi-wake-word detection
- Voice command recognition
- Acoustic Event Detection (AED)
- Gesture and activity detection/classification (falling, walking, etc.)
In contrast to some competing technologies, AONDevices’ solutions also boast:
- Adaptive voice activity detection with context awareness
- High accuracy in high-background-noise and high-reverb environments
- Speaker-independent or user-locked functionality
- Robustness to variabilities from standard conditions
- High levels of scalability with multiple concurrent command detection
AONDevices produces a simple evaluation kit to test the functionality of their chips. This includes an on-board microphone, a USB port for power supply and programming, headers for I/O and core power measurements, and 4x LEDs for visual confirmation of different classifications.
Detailed performance specifications are available after initial communication, so fill out the form below if you would like to evaluate this technology for use in a commercial project.
