AI accelerator modules
Expansion-tier M.2 and board-to-board AI coprocessor modules — follow-on programs once the edge host or box platform is in place, not the first export conversation.
6 representative models — browse below or search by model. Carrier compatibility, SDK, and export MOQ confirmed on inquiry.
M.2 & board-to-board AI accelerator modules
Expansion-tier programs — add dedicated NPU inference to existing edge boxes, industrial PCs, and custom carriers after the host platform is defined. Pairs with our edge box lines above.
AIM-M-K
M.2 2280 AI accelerator with NXP Ara-240 NPU — up to 40 TOPS INT8 for transformers, vision, and generative AI on x86 or Arm edge hosts.
- 40 TOPS INT8 · NXP Ara-240 NPU
- M.2 2280 M-key · PCIe Gen4 x4
- Up to 16 GB LPDDR4 on-module memory
- TensorFlow, PyTorch, ONNX · Linux & Windows
- Typical 3 W · TDP 12 W · 0°C to 50°C operating
- On-module memory: Up to 16 GB LPDDR4 (8 GB optional)
- Host processors: x86, Arm
- Power: Typical 3 W · full 8 W · TDP 12 W
- Dimensions: 22 × 80 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
AIM-B-K
Board-to-board AI accelerator with Kinara Ara-2 NPU — up to 40 TOPS INT8 for embedded carrier integration and generative edge workloads.
- 40 TOPS INT8 · Kinara Ara-2 NPU
- Board-to-board · PCIe Gen4 x4
- Up to 16 GB LPDDR4(X) on-module memory
- LLM, YOLOv8, Stable Diffusion class workloads
- Typical 12 W · 0°C to 70°C operating
- Form factor: Board-to-board module
- Host interface: PCIe Gen4 x4 (B2B)
- Host processors: NXP, NVIDIA, Qualcomm, Xilinx
- AI frameworks: TensorFlow, PyTorch, ONNX, Caffe, MXNet
- Operating systems: Linux, Windows
- Dimensions: 40 × 55 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
AIM-B-H8
Compact Hailo-8 board-to-board AI module — 26 TOPS INT8 at ultra-low power for multi-stream vision on custom embedded carriers.
- 26 TOPS INT8 · Hailo-8 NPU
- Board-to-board · PCIe Gen3 x4
- Multi-stream & multi-model concurrency
- Industrial -40°C to +85°C operating range
- Typical 5 W · 2.5 W TDP class efficiency
- Form factor: Board-to-board module
- Host interface: PCIe Gen3 x4 (B2B)
- Host processors: NXP, Renesas, TI, Rockchip, Socionext, Xilinx
- AI frameworks: TensorFlow, TFLite, Keras, PyTorch, ONNX
- Operating systems: Linux, Windows
- Dimensions: 40 × 55 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
AIM-M-H10
Hailo-10 M.2 AI accelerator card — up to 40 TOPS with on-module memory for edge generative AI, LLMs, VLMs, and vision on AI PCs and edge hosts.
- 40 TOPS · Hailo-10 NPU
- M.2 M-key · PCIe Gen4 x4
- Up to 16 GB LPDDR4/4X on-module memory
- Generative AI — LLMs, VLMs, diffusion models
- Typical 12 W · 0°C to 70°C operating
- AI performance: Up to 40 TOPS
- Form factor: M.2 2280 (M-key)
- Host processors: NXP, NVIDIA, Qualcomm, Xilinx
- AI frameworks: TensorFlow, PyTorch, TorchScript, ONNX, Caffe, MXNet
- Operating systems: Linux, Windows
- Dimensions: 22 × 80 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
AIM-M-MX
MemryX MX3 M.2 AI accelerator — four MX3 cores with up to 24 TFLOPS for dense, low-latency computer-vision inference on edge hosts.
- 4× MemryX MX3 · up to 24 TFLOPS @ 1 GHz
- M.2 2280 M-key · PCIe Gen3 x2 or x4
- Industrial -40°C to +85°C operating range
- ONNX, PyTorch, TensorFlow, Keras, TFLite
- Average max 10 W · peak 14 W (M.2 limit)
- Host processors: Arm, x86, RISC-V
- Operating systems: Ubuntu 22.04/24.04, Windows 64-bit, Android 64-bit
- Dimensions: 22 × 80 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
AIM-M-DX
DEEPX M.2 AI accelerator — 25 TOPS INT8 at 2–5 W for robotics, smart cameras, and factory edge vision on x86 or Arm hosts.
- 25 TOPS INT8 · DEEPX NPU
- M.2 2280 M-key · PCIe Gen3 x4
- 4 GB LPDDR5 + 1 Gb NAND on-module
- TensorFlow, PyTorch, ONNX via DXNN SDK
- 2 W min · 5 W max · -25°C to +85°C
- On-module memory: 4 GB LPDDR5 · 1 Gb NAND flash
- Host processors: x86, Arm
- AI frameworks: TensorFlow, TFLite, ONNX, Keras, PyTorch (DXNN compiler)
- Operating systems: Windows 10/11 64-bit · Ubuntu 20.04/22.04 LTS
- Dimensions: 22 × 80 × 4.1 mm
- Supplier program: Geniatech
- MOQ: Confirm on inquiry
Expansion program — pair with edge box SKUs above or confirm host carrier before module-only quotes.
Priority export program — start with edge box and fanless appliance SKUs; accelerator modules are expansion-tier follow-on programs. Share workload, environment, certification, and MOQ for early matching.