Edge AI Box

$1,495.00 Sold out

The Akida Edge Al Box is offered as a reference design that allows buyers to utilize pre-installed machine learning models or to develop models in their environment.

Contact sales@brainchip.com to white-label the Akida Edge Al Box as a stand-alone system for Al acceleration applications.

What's in the box

Akida M.2 2280 board (AKD1000 + flash + PMIC)
1m flat-flex PCIe interconnect cable
Standoffs and M2.5 hardware (×4)
Quick-start card · printed pinout reference
MetaTF SDK 3.4.1 license key (printed)

Specifications

Processor
Architecture: Akida 1.0 neuromorphic
NPU Count: 1024
Performance: 1.0 TOPS
Quantization: 1 / 2 / 4 / 8 bit
Power
Typical: 30 mW (KWS workload)
Maximum: 100 mW
Wake Latency: less than 200 µs
Interface
Host: PCIe Gen3 x4
AUX: I2C 400 kHz · PDM 3.072 MHz
Reset / Wake: Active-low GPIO
Form Factor: M.2 2280 M-key
Environmental
Operating Temp: -40°C to +85°C
Storage Temp: -55°C to +125°C
Humidity: 5% to 95% non-condensing
Shock / Vibration: MIL-STD-810G method 514.6

Designed for

Keyword spotting: 30 mW · always-on
Visual wake-word: Sub-5 mW idle
Anomaly detection: On-chip learning
Vital-sign sensing: Radar + Akida

Quick Start

Unbox and power on: Connect the 12VDC adapter and press the power button. LED turns green when ready.
Install MetaTF SDK: Run pip install akida on your connected host machine to install the SDK and dependencies.
Run your first model: Launch any of the included example notebooks and deploy a model to the Akida accelerator over PCIe.

Great for low-power AI deployment

Finally a compact device that can run our neural network models without needing a GPU server. Power efficiency is noticeably better than our previous solution. The BrainChip team was also responsive when we had setup questions.

Solid hardware for edge inference

The Edge AI Box delivers on its promise. We tested it for anomaly detection on a factory floor and latency was well within our requirements. Build quality is excellent and the documentation made integration manageable. Would be great to see more SDK examples but overall a strong product.

Impressive edge AI performance

We integrated the Edge AI Box into our computer vision pipeline and were genuinely impressed by the performance. The Akida neuromorphic processor handles real-time inference at the edge with very low power consumption. Setup was straightforward and the compact form factor fits perfectly into our deployment environment. Highly recommend for anyone looking to run AI workloads without cloud dependency.