AkidaTag technical specifications
AkidaTag is a compact, battery-powered edge-AI device built on BrainChip’s Akida™ neuromorphic processing engine. It runs neural networks on the AKD1500 AI co-processor, listens through two on-board microphones, senses motion, and talks to the BrainChip Connect app over Bluetooth Low Energy. Models are built with BrainChip’s MetaTF™ flow and loaded from the app, and the device learns new keywords on-device, with no cloud connection. AkidaTag is made for developers who want to evaluate event-based AI on a wearable-sized, always-on device.
This page is the buyer-facing summary. The engineering reference, with pinouts, the power tree and the full connector detail, is the datasheet; the system view is the block diagram.
| Document status | |
|---|---|
| Hardware described | AkidaTag hardware revision 2 |
| Firmware referenced | The AkidaTag firmware in this repository; the current build is the latest release |
Key features
Section titled “Key features”- BrainChip AKD1500 Akida neuromorphic AI co-processor with on-device learning
- Nordic nRF5340 dual-core Arm Cortex-M33 host with Bluetooth Low Energy
- 16 MB of model storage for the AI processor and 16 MB of firmware and file storage for the host
- Two digital MEMS microphones for always-on audio
- Six-axis accelerometer and gyroscope
- SPI camera header for vision experiments
- Single-cell Li-ion battery support with USB-C charging, fuel gauge and on/off switch
- On-board current sensing of the AI and system rails for power measurements
- Firmware update over Bluetooth or over the USB-C cable
- Companion BrainChip Connect app to load models, run the demonstrations and start on-device learning
Applications
Section titled “Applications”- Keyword spotting and voice control with on-device personalisation
- Vibration and motion classification, anomaly detection
- Low-power vision with an attached SPI camera
- Evaluation of Akida event-based inference on a battery-powered device
Specifications
Section titled “Specifications”| AI co-processor | BrainChip AKD1500: Akida neuron fabric, event-based neuromorphic architecture, 22 nm FD-SOI, 7 mm x 7 mm package |
| AI on-chip memory | 1 MB |
| AI clock | Up to 400 MHz core clock from an on-board 25 MHz crystal; 400 MHz default in firmware |
| On-device learning | Yes; keyword learning exposed over Bluetooth |
| Host to AI interface | SPI, 8 MHz by default, up to 32 MHz |
| Host processor | Nordic nRF5340: 128 MHz Arm Cortex-M33 application core with 1 MB flash and 512 KB RAM, plus a 64 MHz Arm Cortex-M33 network core for Bluetooth |
| Operating system | Zephyr RTOS on the nRF Connect SDK, MCUboot bootloader |
| Model storage | 16 MB SPI NOR flash dedicated to the AI processor |
| Firmware storage | 16 MB SPI NOR flash for firmware updates and a LittleFS file system, in addition to the host’s 1 MB internal flash |
| Microphones | Two digital PDM MEMS microphones; the firmware captures audio at 16 kHz, 16-bit, and the keyword spotting demonstration uses one of the two |
| Motion sensor | STMicroelectronics ISM330DHCX six-axis accelerometer and gyroscope; firmware defaults 208 Hz, plus or minus 8 g and plus or minus 500 degrees per second |
| Camera | SPI camera header (10-pin, 1.27 mm pitch) with switched 3.3 V supply; the firmware supports an ArduCam Mega camera at 96 x 96 and 128 x 128 pixels |
| Battery monitoring | Fuel gauge reporting state of charge to the app; charger status reporting |
| Power monitoring | On-board current sensing of the 1.8 V system rail and the 0.8 V AI core rail, readable from the firmware console |
| Bluetooth | Bluetooth Low Energy peripheral, device name AkidaTag; LE Secure Connections only with bonding; firmware update over Bluetooth |
| Radio | nRF5340 radio, specified by Nordic for -40 to +3 dBm configurable transmit power and -98 dBm sensitivity at 1 Mbps; on-board 2.4 GHz chip antenna |
| USB | USB-C: 5 V charging input and a USB-to-serial console at 115200 baud; firmware update over the cable with nothing but a computer |
| Debug | 10-pin 1.27 mm SWD header |
| Battery | Not included. Fits a single-cell rechargeable Li-ion or Li-Po cell, 3.0 to 4.2 V, on a JST XH 2-pin connector |
| Battery life | Depends on the cell fitted |
| Charging | On-board linear charger with power path, 536 mA nominal fast charge from USB-C 5 V; the device runs while charging; charge time depends on the cell fitted |
| Power switch | On/off slide switch |
| Power consumption | Power consumption figures will be added in a later revision of this document |
| Indicators | One RGB LED and one white LED under firmware control |
| Controls | One user button; the on/off slide switch |
| Board dimensions | 27.7 mm x 39.5 mm, 1.0 mm thick, six-layer PCB, chamfered corners |
| Enclosure | Ships in its enclosure |
| What is in the box | The AkidaTag board in its enclosure. No battery, USB-C cable or camera is included |
Software
Section titled “Software”Models for AkidaTag are built with BrainChip’s MetaTF flow from TensorFlow/Keras or PyTorch, packaged with the model workflow in this repository, and sent to the device by the BrainChip Connect app over Bluetooth. The firmware in this repository runs the demonstrations: keyword spotting with on-device edge learning, motion sensing, and camera capture from the console. Firmware updates arrive over Bluetooth (MCUmgr) or over the USB-C cable from a computer.
| Companion app | BrainChip Connect for Android 13 or later, in pre-registration on Google Play, and coming soon to the iOS App Store |
| Firmware | This repository, under the Apache License 2.0; the current build is the latest release |
| Models | Every firmware release attaches the keyword spotting model packages akidatag-kws-model.zip and akidatag-kws-edge-learning-model.zip |
| Demonstrations | Keyword spotting with on-device edge learning; motion sensing; camera capture |
Resources
Section titled “Resources”- AkidaTag documentation
- BrainChip Connect documentation
- AkidaTag firmware releases
- AkidaTag on the BrainChip Developer Hub
- Developer Hub sign-up
- BrainChip Connect on Google Play (pre-registration)
- AKD1500 product brief
- Help: the BrainChip Discord for questions and GitHub issues for bugs. Security problems are reported privately through the repository’s Security tab (“Report a vulnerability”)
Sources
Section titled “Sources”Every value on this page is taken from the AkidaTag revision 2 design files, the AkidaTag firmware in this repository, the AKD1500 Product Brief V2.4, the Nordic nRF5340 product specification, the Texas Instruments BQ25185 datasheet, the Google Play listing or BrainChip’s product decisions of September 2026 on the enclosure, the box contents and the app platforms. The datasheet lists the source next to each value.