Akida Pico FPGA — Cloud Examples

Runnable examples for deploying neural-network models to BrainChip Akida Pico — an ultra-low-power, event-based neural IP for always-on 1-D sensing — on the Akida Pico FPGA cloud platform (a single-NP Pico IP implemented on a Xilinx FPGA and accessed through JupyterLab).

Each example takes a trained model the full path to on-device inference:

build → convert to a streaming (stateful) form → quantize to int8 → convert with Akida's MetaTF toolchain → map onto the Pico → measure and run on hardware.

The Akida software toolchain — MetaTF (akida, cnn2snn, quantizeml, akida_models) — and full API documentation is here: https://doc.brainchipinc.com/index.html

Examples

Example	Task	What it demonstrates
Keyword Spotting	12-class speech-command recognition (audio, 16 kHz)	Streaming SSM keyword spotting on Pico — data → stateful conversion → quantization → Akida mapping → latency/throughput/power → streaming inference.
Bearing Fault Detection	Multi-label vibration fault detection (accelerometer, 42 kHz)	The same pipeline on a 1-D vibration stream — real-time multi-label fault detection, hardware metrics, a float-vs-Akida comparison, and an accuracy-vs-decision-latency study.

Each example folder has its own README.md with the details.

Setup

The Akida Cloud host already has the Pico FPGA attached and conda/Python available.

1. Install dependencies:

   conda install -c conda-forge jupyterlab ffmpeg
   pip install -r requirements.txt

   (ffmpeg is used by tensorflow_datasets to prepare the Speech Commands dataset for the Keyword Spotting example.)

2. Launch JupyterLab:

   ./start-jupyterlab.sh

3. Open an example — e.g. examples/kws/kws_sc12.ipynb or examples/fault_detection/fault_detection_inference.ipynb — and run all cells in order.

The Pico device

The examples run on a real Akida Pico device. Confirm it is visible before running:

akida devices            # or:  python -c "import akida; print(akida.devices())"

You should see one device. Hardware and platform details are in examples/Akida_Cloud_Specs.md.