Products
Applications
Solutions
The POLYN Tiny AI Engine is a neuromorphic processing solution designed for ultra-low-power, sensor-driven edge applications. Unlike conventional AI accelerators that rely on digital compute, this chip implements neural networks directly in analogue circuitry—delivering real-time inference at microwatt power levels. That makes it ideal for wearable, automotive, and IoT products that need embedded intelligence but can’t afford the energy cost of cloud or even local digital inference.
By translating a trained neural network into a neuromorphic hardware core, the POLYN Tiny AI Engine operates with deterministic latency and massive energy efficiency. One flagship use case is in smart tyres, where the chip processes accelerometer data inside the tyre to determine road friction and condition—all on a battery designed to last five years. This would be impossible using traditional digital processors at similar size and power levels.
Voice activity detection is another strong application. The POLYN Tiny AI Engine can detect speech using as little as 30 microwatts of power, with offline command recognition and speaker ID also supported. It enables always-on voice interfaces in hearing aids, remote controls, and industrial headsets without requiring cloud connectivity or complex digital SoCs. Gesture recognition, too, is handled with sub-millisecond response and near-zero idle power draw.
Engineers start by training a neural network in a standard digital environment. That model is then compiled into a neuromorphic structure and optimised using POLYN’s proprietary toolchain. The resulting analogue logic is produced using standard 40–55nm processes, ensuring affordability and compatibility with conventional packaging and assembly lines. The POLYN Tiny AI Engine is available as a stand-alone chip or via evaluation platforms to accelerate design.
For designers of edge AI systems with severe power, latency, and form factor constraints, the POLYN Tiny AI Engine offers a viable, production-ready path to embedded intelligence. It sidesteps many of the limitations of digital inference architectures, especially in always-on environments where power consumption must be measured in microwatts—not milliwatts.