BrainChip now offers its AKD1500 neuromorphic edge AI co-processor across three different development platforms, each targeting a different integration point in the Linux and embedded hardware ecosystem. The most interesting for single-board computer enthusiasts is an M.2 2230 B+M Key card that slots directly into a Raspberry Pi 5 via an M.2 HAT, bringing 800 GOPS of neural network inference at a typical draw of just 250 mW. A standard PCIe 2.0 x1 card covers desktop and server hosts, while the tiny BrainBoard 1500, developed with Neuromorphyx, fits an Arduino Nicla-compatible 22.86 x 22.86 mm (0.9 x 0.9 inch) footprint and communicates over SPI or QSPI.
The AKD1500 itself is a fanless, passively cooled chip fabricated on a 22 nm FD-SOI process, with an Akida Neuron Fabric clocked between 5 and 400 MHz and 1 MB of on-chip memory. Its defining trick is on-chip adaptive learning, meaning models can be updated at the edge without a cloud round-trip. The BrainBoard 1500 pushes that further into microcontroller territory, pairing the AKD1500 with 4 MB of flash for model deployment, an integrated two-channel power monitor for real-time inference profiling, and sleep power consumption down to 37 uW when the co-processor is off.
On the software side, all three platforms run on Linux hosts (Debian and Ubuntu), with open-source kernel drivers on GitHub covering mainline kernels from 5.4 up through 6.9. Getting the most out of the PCIe-connected M.2 and PCIe cards requires contiguous memory allocation (CMA) support enabled in the host kernel to handle larger models, a setting the driver's own documentation notes isn't always on by default. BrainChip's MetaTF development environment handles model conversion, quantization, compilation, and deployment using standard TensorFlow and Keras workflows, and includes a model zoo of pre-trained networks plus tools for converting CNNs to spiking neural networks. The BrainBoard 1500 follows a different path, relying on an open-source Arduino library from Neuromorphyx with example sketches for Arduino Nicla Vision and Nicla Sense ME boards. BrainChip also launched its Symphony Community Akida Bundle in August 2026, an open-source software package that lets developers run Akida neuromorphic processors alongside existing compute. Independent of BrainChip's own stack, a community project called rustChip has reimplemented the driver and inference pipeline in pure Rust under an AGPL-3.0 license, removing the dependency on Python, the C++ SDK, and MetaTF; its author submitted AKD1500 device ID support as a pull request to BrainChip's own driver repository in February 2026.
The AKD1500 M.2 development card sells for $129 (€119), the BrainBoard 1500 module for $99 (€91), and the PCIe development card for $149 (€137). A five-pack of bare AKD1500 chips is available for $199.99 (€184). All are available now through BrainChip's online shop.
