heliaCORE™ is Ambiq’s optimized neural-network kernel layer for Ambiq SPOT SoCs, enabling developers to quickly bring AI products to market on resource-constrained edge devices with efficient AI kernels and extensive operator acceleration. It supports validation on about 12 times more operator instances than MLPerf Tiny, covering more real-world AI workloads.
This benchmark demonstrates heliaCORE’s performance on Apollo510 (Cortex-M55) through the scalar reference, DSP, and MVE execution paths. Since all three paths operate on the same SoC at identical clock speeds, the results offer a straightforward apples-to-apples comparison of the performance improvements from each ISA optimization path.
AI performance depends on more than model architecture. Efficient kernel implementations help accelerate common AI operations, improve resource utilization, and reduce software integration complexity. For edge devices operating within strict memory and power constraints, optimized AI kernels help developers achieve more predictable performance while accelerating deployment of production-ready AI applications.
Read More
Wearable Devices
Audio and Voice Processing
Industrial Monitoring
Consumer Edge AI Products
heliaCORE Developer Hub – Explore step-by-step guides, hands-on examples, and performance benchmarks to kickstart heliaCORE integration.
Ambiq AI Hub – Dive into neuralSPOT® AI SDK & toolkit and our heartKIT, sleepKIT, and soundKIT ADKs
heliaCORE Developer Hub – Explore step-by-step guides, hands-on examples, and performance benchmarks to kickstart heliaCORE integration.
Ambiq AI Hub – Dive into neuralSPOT® AI SDK & toolkit and our heartKIT, sleepKIT, and soundKIT ADKs
The leak detection industry is evolving. For years, monitoring systems focused primarily on collecting and transmitting data. Sensors…
The dual listing on NYSE and SGX broadens investor access across Asia while reinforcing Ambiq’s long-term commitment to one of the…
Existing tools like Arm’s CMSIS-NN provide a solid industry-standard foundation, but they weren’t built for Ambiq’s specific silicon or the AI workloads Ambiq customers actually ship. heliaCORE™ builds on top of CMSIS-NN rather than replacing it, preserving the familiar APIs and documentation engineers already know, while adding Apollo-specific tuning and coverage that generic tools lack. What devices can use heliaRT™?
Every new edge AI project on Ambiq silicon previously required teams to manually assemble, integrate, and optimize the software stack from scratch, often taking months before any real product work could begin. heliaCORE™ delivers that foundation pre-integrated and pre-optimized, so teams can start building their product immediately rather than rebuilding infrastructure they’ve already built.
MLPerf Tiny is the industry’s standard benchmark for edge AI, it tests 5 models and 80 total operations. Ambiq validated heliaCORE™ against a field-representative suite of 40+ models and 963 operator instances, roughly 12 times the operational volume of the industry benchmark. The gap reflects the difference between what a benchmark tests and what a real product can run.
A product optimized only against standard benchmarks can still perform poorly in the real world if those benchmarks don’t reflect production complexity. Ambiq’s internal profiling found that operations surrounding the main computations, padding, activations, reductions, and reshapes, were consuming significant processing time in real workloads even after headline kernels were fully optimized. heliaCORE™ addresses the full pipeline, not just the parts that benchmarks measure.
heliaCORE™ is the foundation, an optimized kernel layer on which everything else is built. heliaRT™ is the runtime path that executes AI inference on-device using heliaCORE underneath. heliaAOT™ is the ahead-of-time compiler path, pre-baking efficiency into the code before it ships — also powered by heliaCORE. Think of heliaCORE as the engine, and heliaRT and heliaAOT as two different ways to drive it.
Yes. heliaCORE™ includes both MVE acceleration for current-generation Apollo SoCs and DSP acceleration for previous-generation devices. Neither generation is left behind.
Both. heliaCORE™ is built on Arm’s CMSIS-NN — an open, industry-standard ecosystem — and fully preserves its APIs, documentation, and license terms. The Apollo-specific kernel tuning, extended operator coverage, and platform integration are Ambiq additions layered on top of that open foundation.
heliaCORE is available now to Ambiq customers and partners. Documentation and access are available through Ambiq’s GitHub page and at www.ambiq.com/AI/heliaCORE.