Geniatech AIM-M-K and AIM-B2 integrate Ara240 for local AI inference
Apr 28, 2026 — by Giorgio Mendoza 274 views
Geniatech has shared information about the AIM-M-K and AIM-B2 AI accelerator modules based on the NXP Ara240 NPU. Both designs target edge inference workloads, offering up to 40 TOPS of INT8 performance for applications such as computer vision, transformer models, and generative AI.
The AIM-M-K adopts a standard M.2 2280 (M-key) form factor and connects over a PCIe Gen4 x4 interface, allowing it to be integrated into existing x86 and Arm-based systems without requiring changes to the base hardware.
— ADVERTISEMENT —
The module is available with up to 16 GB of LPDDR4 memory for handling larger models and high-throughput workloads.

AIM-M-K top view
The AIM-B2 uses a smaller 40 × 55 mm form factor with a board-to-board PCIe Gen4 interface, targeting embedded designs where direct integration onto a carrier board is required.
It supports a range of host platforms, including systems based on NXP, NVIDIA, Qualcomm, and Xilinx processors.

AIM-M-K bottom view
Both modules are built around the Ara240 NPU and support concurrent execution of multiple AI workloads.
The platform is intended to handle models such as ResNet50, YOLO-based detection networks, and generative AI workloads including Stable Diffusion and LLaMA 2, with local inference used to reduce latency and avoid cloud dependency.

AIM-B2 bottom view
Performance figures listed on the product page include ResNet50 inference at approximately 2 ms latency and Stable Diffusion 1.4 generation in the range of 7 to 10 seconds per image, depending on configuration.
The AIM-M-K operates from a 3.3 V supply and is listed with typical power consumption around 3 W, a TDP of 12 W, and peak operation up to 8 W. The AIM-B2 is specified at around 12 W typical power and supports an operating temperature range of 0°C to 70°C for embedded deployments.

AIM-B2 top view
Software support includes compatibility with frameworks such as TensorFlow, PyTorch, ONNX, and Caffe, with additional support for MXNet on the AIM-B2. Additionally, both modules support Linux and Windows operating systems.
Further Information
The AIM-M-K M.2 module is listed at $645, although it is not currently available for retail purchase. Geniatech indicates that samples can be requested for prototype development. Pricing and availability for the AIM-B2 module have not been disclosed.
Please comment here...