Edge AI hits limits in speed, memory, and security. This M.2 solution can boost AI, protect firmware, and manage power in small spaces.
Edge AI applications face limits in performance, memory, and power while running models like large language models (LLMs) and vision transformer networks. Many solutions are too large, use too much power, or lack security for deployment in industrial, retail, healthcare, or public safety environments.
The Metis M.2 Max from Axelera AI addresses these challenges by delivering PCIe-card-level AI performance in an M.2 form factor. It doubles memory bandwidth, supports up to 16 GB of memory, and manages heat, allowing AI inference in constrained spaces. It is available for standard and extended operating temperatures, making it suitable for different environments.
It solves performance bottlenecks for AI models. Built on Axelera’s Metis AIPU, the M.2 Max boosts convolutional neural network (CNN) performance by 33% and doubles token throughput for LLMs and vision-language models (VLMs), all within a 6.5 W power envelope.
Edge deployments are also vulnerable to firmware attacks and unauthorized modifications. Metis M.2 Max solves this with firmware integrity protection via a Root of Trust, enabling secure boot and firmware updates, ensuring only authenticated firmware runs on devices.
Physical limits are another challenge in edge AI systems. The 2280 M.2 module is available as a board or with a heatsink, reducing card height by 27%. An onboard power probe allows automatic performance adjustments in power- or thermal-limited deployments.
Developers face complexity integrating AI accelerators into projects. Metis M.2 Max solves this with the Voyager SDK, which unlocks the potential of Metis AIPUs and simplifies deployment of both proprietary and standard AI models.
“We continue to set the price-performance ratio benchmark for the AI accelerator market,” said Fabrizio del Maffeo, CEO of Axelera AI. “Our goal is to make it possible for our customers to deploy transformative edge AI applications at scale. Metis M.2 Max and our original Metis M.2 card enable customers to choose the solution that best fits their requirements – from computer vision to large language models to the latest of transformer models.”
