Accelerating Edge AI with Ambiq Micro’s heliaCORE Platform

Ambiq Micro has launched a new software platform called heliaCORE. It aims to speed up AI deployment on small, resource-limited devices. These devices run on Ambiq’s Apollo system-on-chips (SoCs).

heliaCORE is built to handle AI workloads efficiently. It combines optimized AI kernels, operator acceleration, and validation tools. This setup improves performance and cuts down on development time.

The platform is built on the Arm CMSIS-NN framework. This foundation lets heliaCORE accelerate neural network kernels and key inference operations. These include activations, padding, reductions, reshaping, and data movement.

Broad Operator Support and Validation

heliaCORE supports over 200 accelerated operators. These operators are optimized specifically for Apollo devices. It also supports MVE and DSP acceleration across current and previous Apollo SoC generations.

The platform was tested on more than 40 AI models. These models covered 53 operator types, 247 unique operators, and 963 operator instances. To put this in perspective, heliaCORE handles roughly 12 times more operator instances than MLPerf Tiny benchmarks.

Part of a Growing AI Ecosystem

heliaCORE is part of Ambiq’s HELIA Edge AI ecosystem. It joins two other platforms: heliaRT and heliaAOT. Together, they provide a comprehensive software stack for edge AI development.

The software is available through multiple integration flows. These include CMake, CMSIS-Pack, Zephyr, and neuralSPOT-X. This variety helps developers use heliaCORE in different project setups.

Dr. Adam Page, Head of AI at Ambiq, said, “heliaCORE is a foundational building block of the HELIA ecosystem. It provides the software infrastructure developers need to deploy edge AI faster and with greater confidence.”

Edge AI devices need fast and efficient software to handle AI tasks on the spot. By improving AI kernel speed and operator support, heliaCORE helps meet these demands. It makes deploying AI models on small chips easier and more reliable.

With heliaCORE, developers can push AI to devices that have limited power and memory. This can open doors to new applications in wearables, smart sensors, and other connected gadgets. The platform’s extensive validation gives confidence that AI models will run smoothly on Apollo SoCs.

Overall, heliaCORE marks a leap forward in edge AI software. It combines powerful acceleration with broad compatibility. This can help bring smarter AI to more devices, closer to where data is generated.