Models & Pricing
NXP Ara240 16GB M.2 Module
- 40 eTOPS with flexible precision
- 16GB onboard LPDDR4
- PCIe Gen4 x4 (4-lane)
- Under 2W typical, 12W TDP
- M.2 2280 M-Key
- Secure boot + encrypted memory
Geniatech AIM-M2 16GB
- Kinara Ara-2 NPU, 40 TOPS
- 16GB RAM (4/8GB options)
- PCIe Gen4 x4, 3.3V supply
- Heatsink cooling
- Under 2W typical
- ~$188 direct sample pricing
Forlinx FAI-ARA240-M
- NXP Ara240 DNPU
- 40 eTOPS mixed precision
- Works with i.MX 8M Plus / i.MX 95 hosts
- TensorFlow, PyTorch, ONNX support
- M.2 form factor
- Board-to-board AIM-B2 sibling
Kinara KM-2 / KU-2 Reference
- KM-2 M.2 module reference design
- KU-2 USB 3.0 module
- 2GB RAM guidance for classic AI
- 8GB RAM guidance for generative AI
- Vendor reference hardware
- Linux and Windows drivers
Specifications
NPU
NXP Ara240 (Kinara Ara-2) discrete neural processing unit — multiple NPU + VPU cores
AI Performance
40 eTOPS (INT4 / INT8 / mixed precision)
Onboard Memory
16GB LPDDR4 (4GB / 8GB options)
Host Interface
PCIe Gen4 x4
Power Consumption
Under 2W typical (computer vision); 12W TDP
Form Factor
M.2 2280 M-Key (22 × 80 mm)
Security
Secure boot and encrypted memory access
Supply Voltage
3.3V
Operating Temperature
0 to 50°C; storage –40 to 85°C
AI Frameworks
TensorFlow, PyTorch, TorchScript, ONNX, Caffe, MXNet
Supported Models
Llama 2.0, YOLOv8, Stable Diffusion 1.4, ResNet-50, MobileNetV1 SSD
Benchmarks
Stable Diffusion 1.4 ~7 s/image; Llama-7B ~12 output tok/s; ResNet-50 2 ms latency
Drivers
Linux and Windows
Verified Hosts
NXP, NVIDIA, Qualcomm, AMD Xilinx platforms
Module Variants
M.2 2280 (AIM-M2 / FAI-ARA240-M), board-to-board (AIM-B2), USB 3.0 (KU-2)
Positioning
Generative AI and transformer inference at a lower cost point than rival M.2 NPUs
Typical Use
AI assistants, copilot appliances, gaming, smart retail, physical security, factory automation
Silicon Note
Ara-2 delivers an 8× performance increase over the first-generation Kinara Ara
Overview
The NXP Ara240 — originally developed by Kinara as the Ara-2 — is a discrete neural processing unit aimed squarely at generative AI and transformer workloads at the edge. It pairs multiple NPU cores with dedicated vision processing units to reach 40 eTOPS of mixed-precision (INT4/INT8) throughput, while keeping typical power under 2W in computer-vision workloads and capping TDP at 12W.
The value proposition is cost per usable token or image. Where most M.2 AI accelerators of this class were designed for CNN-era vision models, the Ara-2 was architected for attention-based networks. Independent coverage highlights it running Stable Diffusion 1.4 at roughly seven seconds per image and Llama-7B at around 12 output tokens per second, with ResNet-50 at 2 ms latency — at a module price that has been quoted near $188 for the 16GB configuration, materially below competing generative-capable NPUs.
Integration is deliberately conventional. The module is an M.2 2280 M-Key card using PCIe Gen4 x4 over a 3.3V supply, with secure boot and encrypted memory access built in. Drivers exist for both Linux and Windows, and the part has been validated on NXP i.MX, NVIDIA, Qualcomm and AMD Xilinx host platforms. For industrial customers the natural pairing is an NXP i.MX 8M Plus or i.MX 95 board; Forlinx sells exactly that combination as the FAI-ARA240-M. Geniatech's AIM-M2 is the widely available 16GB variant, with a board-to-board AIM-B2 sibling for customers embedding the NPU directly on their own carrier.
The trade-off to note is thermal: the module is specified for 0–50°C operation rather than the –40°C industrial range some competitors offer, so it suits indoor and climate-controlled edge deployments, AI appliances and on-premise inference boxes rather than outdoor or vehicle-mounted installations.
Target Applications
On-premise generative AI appliances, AI assistants and copilot boxes, smart retail analytics, physical security and video analytics, factory automation inspection, gaming and kiosk intelligence, and any embedded platform that needs local Llama-2 or Stable Diffusion inference without a GPU.
Related Products
Request a Quote — NXP Ara240 M.2 Module
QS Compute — global B2B supply for NXP Ara240 / Kinara Ara-2 modules and partner variants (Geniatech AIM-M2, Forlinx FAI-ARA240-M). Sample quantities and volume allocation available.
Get Your Quote →