ASYGN says ColibryNPU sets MLPerf Tiny energy record

ASYGN says its ColibryNPU microcontroller posted the lowest energy use in the MLPerf Tiny v1.4 Visual Wake Words benchmark, a result the company says could expand battery-powered AI for vision, wearables, IoT and medical devices. The Grenoble-based chip designer says the part is available now in evaluation form.

Why it matters: - ASYGN is targeting always-on AI at the edge, where power draw can block use cases in compact devices. - The ColibryNPU result points to sub-milliwatt AI vision processing, which could extend battery life in IoT, wearables, medical devices and toys. - At one inference per second, ASYGN says a system using ColibryNPU could run for more than three years on a 220 mAh CR2032 coin cell.

What happened: - ASYGN announced results from its ColibryNPU microcontroller on the MLPerf Tiny v1.4 benchmark. - The company says the chip recorded 22.2 microjoules per inference on the Visual Wake Words task. - ASYGN says the benchmark results were published July 7, 2026. - ColibryNPU is a 32-bit RISC-V microcontroller with a neural accelerator. - The company said the device is available now in evaluation form.

The details: - ColibryNPU is designed for TinyML and low-power sensing workloads, including video, audio and environmental signals. - The chip uses a near-memory computing architecture that keeps memory and compute blocks closely connected. - ASYGN says the microcontroller can process video in real time at under 1 milliwatt and at several frames per second. - The design includes 8 dedicated macro-blocks. - Those blocks perform up to 2 multiply-accumulate operations per clock cycle. - ASYGN says peak performance reaches 9.6 GOPS. - The chip includes a dedicated image signal processor for video handling. - ASYGN says the MLPerf Tiny benchmark provides a standardized way to compare inference latency and energy use. - The company linked to the official benchmark page for the v1.4 results: the official benchmark page. - ASYGN also provided a contact link for more information: more information.

Between the lines: - The benchmark result gives ASYGN a concrete claim in a crowded low-power AI market, where energy efficiency is often the main differentiator. - The use of MLPerf Tiny matters because it offers a common test framework, making the comparison easier to verify than vendor-only demonstrations. - ASYGN is positioning the chip for applications that need local inference without frequent charging or large batteries.

What's next: - ASYGN says ColibryNPU is now available for evaluation, which suggests the company is moving toward broader developer testing and potential design wins. - The company is likely to use the benchmark result to court customers building edge AI systems that need long battery life and small form factors.

The bottom line: - ASYGN is pitching ColibryNPU as a rare combination of benchmark-leading efficiency and practical edge-AI capability for ultra-low-power devices.