heliaAOT

Blazing Fast Neural Inferencing

heliaAOT is an ahead-of-time compiler that converts LiteRT models into highly optimized, standalone C inference modules—custom-tailored for Ambiq’s ultra-low-power SoCs. It produces compact, efficient, and human-readable C code with zero runtime overhead, enabling lightning-fast, power-efficient AI at the edge.

Design Resources

heliaAOT Highlights

Zero Guesswork in Memory Allocation

Automatic tensor memory planning eliminates over-allocation and strips away unused code for lean, efficient deployment.

Up to 10x Smaller Code Size

Dramatically reduces flash footprint on Apollo SoCs compared to standard TensorFlow Lite for Microcontrollers.

Deep Optimization & Customization

Fine-tune inference pipelines at the operator, subgraph, or full graph level with advanced techniques like layer fusion, tensor reordering, and intelligent memory placement.

Seamless Integration

Easily integrates as a Zephyr RTOS module or as a plug-in to Ambiq’s neuralSPOT AI Development Kit (ADK) for streamlined workflow.

Uncompromising Performance

Get HeliaRT-level inference speed in a tiny package—HeliaAOT slashes memory footprint by up to 2.6× on MLPerf Tiny models.

Design Resources

HeliaAOT Developer Hub

Explore step-by-step guides, hands-on examples, and performance benchmarks to kickstart your HeliaAOT integration

Ambiq AI Hub

Dive into neuralSPOT AI SDK & toolkit and our heartKIT, sleepKIT, and soundKIT ADKs—everything you need in one place to build, deploy, and optimize AI on Ambiq devices