Geniatech AIM AI Accelerator Modules — Hailo, NXP Ara-240, MemryX & DEEPX

Specifications

Overview

Geniatech's AIM family is a set of drop-in AI accelerator modules for industrial edge designs, spanning four different NPU architectures in matching M.2 and board-to-board footprints. The AIM-M-H10 and AIM-M-H8 / AIM-B-H8 use Hailo-10 and Hailo-8 silicon — 40 TOPS INT4 for generative AI and 26 TOPS INT8 at 2.5W respectively. The AIM-M-K and AIM-B-K carry NXP's Ara-240 (Kinara Ara-2) accelerator at up to 40 eTOPS with 16GB of LPDDR4. The AIM-M-MX brings MemryX MX3, the AIM-M-DX brings DEEPX, and the AIM-M-R28 uses Rockchip's RK1828 for generative AI. All are offered in industrial temperature grades with Linux and Windows support and the standard TensorFlow / PyTorch / ONNX / Keras toolchains.

Key Benefits

Architecture choice in one footprint: Hailo, NXP Ara-240, MemryX and DEEPX can be evaluated and swapped inside the same M.2 socket or board-to-board carrier, so a product can be re-targeted at a different NPU without a board respin. Power-scaled options: the Hailo-8 26 TOPS module runs at 2.5W TDP, which suits sealed fanless enclosures where a discrete GPU cannot be thermally accommodated. Generative AI at the module level: the AIM-M-H10 and AIM-M-R28 target LLM and VLM inference rather than classification-only vision pipelines. Industrial qualification: -40°C to +85°C on the Hailo-8 modules, B2B connector variants with a 4-lane PCIe Gen3 interface for custom carrier boards, and multi-stream / multi-model concurrent execution.

Applications

Industrial machine vision and inspection, smart cameras and NVR analytics, robotics and AMR perception, retail and traffic analytics, medical imaging inference, and adding an NPU to an existing x86 or ARM industrial computer via M.2.

Related Products