Twitter
Facebook
LinkedIn
Reddit
Pinterest
Email

F&S Elektronik Systeme has introduced an M.2 AI accelerator based on the NXP Ara-240, designed to offload inference workloads from embedded systems. It targets edge applications requiring low-latency processing, including machine vision, multimodal inference, and real-time analytics.

The accelerator delivers up to 40 TOPS of AI performance through the Ara-240 architecture, which integrates multiple neural network processors and vision processing units.

The Ara240 architecture has also appeared in other recent edge AI modules, including Gateworks’ GW16168 M.2 accelerator and Forlinx’s FAI-ARA240-M, both targeting similar embedded inference workloads.


Ara240 DNPU runtime integration w/ host processors(click image to enlarge)

The module adopts a standard M.2 Key-M 2280 form factor, enabling integration into existing systems via PCIe without requiring a full hardware redesign.

It supports PCIe Gen3 and Gen4 interfaces with up to four lanes for high-throughput data transfer between the host processor and the accelerator.


Ara240 machine learning deployment workflow(click image to enlarge)

Onboard memory includes up to 16 GB of LPDDR4, providing capacity for large models and high-bandwidth data processing.

The design also incorporates secure boot and root-of-trust features, supporting deployment in security-sensitive environments.


FS-M2-A block diagram w/ Ara240 & PCIe interface(click image to enlarge)

The module is designed as a companion processor, enabling heterogeneous computing configurations when paired with host platforms such as SMARC modules based on NXP i.MX processors.

In this setup, the accelerator handles AI workloads while the host processor manages system control and application logic.


FS-M2-A(click image to enlarge)

Software support includes the NXP eIQ AI development environment along with compatibility for frameworks such as TensorFlow, PyTorch, and ONNX.

Additional tools for model optimization, including quantization and compilation, are provided to support deployment workflows.

Specifications listed for the FS-M2-A include:

Processor:

NXP Ara-240 Edge AI processor

AI Performance:

Memory:

Form Factor:

Interface:

PCIe Gen3/Gen4 (up to x4 lanes)

Power Supply:

Power Consumption:

Dimensions:

Operating Temperature:

-40°C to +85°C (depending on variant)

Further Information

F&S lists the module at approximately €360 (excluding tax), while DigiKey lists it at around $498 for single-unit quantities.

LinuxGizmos related posts: