Edge AI coprocessor adopts M.2 form factor
Based on BrainChip’s Akida neuromorphic engine, the AKD1500 edge AI coprocessor is now available in a compact M.2 2230 (22×30 mm) form factor. The card features a B+M key edge connector for use with Raspberry Pi 5 and compatible host systems, enabling fanless, plug-and-play AI acceleration without redesigning existing power or cooling systems.
Operating from a 3.3-V supply, the AKD1500 M.2 card connects through a two-lane PCIe Gen2 host interface. The coprocessor integrates 32 neural processing units (NPUs) and delivers up to 800 effective GOPS (INT4) for edge AI inference. It also includes 1 MB of dual-port on-chip memory and consumes a typical 250 mW at 400 MHz.
The event-based digital architecture delivers ultra-low-power acceleration for standard neural network models and supports on-device learning, enabling applications to adapt without a cloud connection or full model retraining. Models are developed and optimized using BrainChip’s MetaTF software flow with TensorFlow/Keras and PyTorch front ends before deployment to the card.
The AKD1500 M.2 card with a B+M key connector is priced at $129 and is available for purchase directly from BrainChip.
The post Edge AI coprocessor adopts M.2 form factor appeared first on EDN.
