In the world of Edge AI, there’s always been a tradeoff: high intelligence, always-on capability, or long battery life. If you wanted a device to listen for a voice command or monitor something 24/7, you usually had to accept that the battery would be drained just waiting for a command.

Enter Akida Pico.

Built on BrainChip’s proprietary event-based processing platform, Akida Pico is an ultra-low-power co-processor (or standalone core) designed to give small devices “eyes and ears” without the power-hungry baggage of a traditional CPU tasked with the nuisance of stand-by mode.

Why “Event-Based” is a Game Changer

Traditional AI processors are like a light that stays on all night, even when the room is empty. They’re constantly crunching numbers regardless of whether anything significant is happening.

Akida Pico uses event-based processing, which mimics the human brain. It only “fires” when it detects a relevant change in data (an “event”). If nothing is happening, it consumes almost zero power. This allows it to operate in the microwatt (uW) to milliwatt (mW) range, making it the leanest NPU core in the industry.

Unlike most chips that need a heavy-duty host CPU, this core can operate entirely standalone, pulling just microwatts to milliwatts of power. Pico stays lean by using “power islands” to make sure its standby mode doesn’t incur the leakage of the whole system.

Real-World Use Cases

Akida Pico isn’t just a spec sheet; it’s designed for specific, high-impact “extreme edge” applications.

Wake-up Systems

In many designs, Akida Pico acts as a low-power filter. Instead of the main CPU staying awake to listen for a keyword or detect motion, Pico remains on duty. When it identifies a “qualified event,” like a specific voice command, it sends a low-power interrupt to “wake up” the main MCU: perfect for applications like smart appliances, voice assistants, and wearables.

Healthcare

Imagine a wearable that monitors heart health or detects the early onset of a seizure. Because Pico is purely digital and ultra-efficient, it can perform medical anomaly detection locally on the device without ever needing to send data to the cloud. Patients can enjoy prolonged battery life with a monitor that only alerts a doctor when a specific event is detected.

Industrial Predictive Maintenance

In a factory, thousands of motors hum 24/7. Identifying a failing part early can prevent emergency repairs or lost productivity. Akida Pico can be integrated into remote sensors to perform industrial anomaly detection, analyzing vibration patterns or thermal spikes in real-time. An industrial sensor can run for years, only “reporting in” if it hears a problem.

Developer-Friendly: No New Languages Required

Akida Pico is easy test, train, and deploy with BrainChips Meta TF development tool. MetaTF includes a processor IP simulator for model execution, as well as support for Akida hardware like the AKD1000 reference SoC and Akida 2 FPGA platform. Inspired by the Keras API, MetaTF provides a high-level Python API for neural networks. This API facilitates early evaluation, design, final tuning, and productization of neural network models.

1. Native Support: Works directly with TensorFlow/Keras and PyTorch.
2. Low-Code/No-Code: For those who aren’t AI experts, BrainChip offers turnkey tools to deploy optimized models quickly.

The Bottom Line

Akida Pico is about making the Internet of Things intelligent without the need to tether devices to a charging cable. With its ultra-low power neuromorphic technology at the edge, Pico is proving that you don’t need a massive power budget to have a massive impact.

Ready to check out Akida Pico?

Join us for a live webinar to see Pico in action on Akida FPGA in the Cloud: execute models, assess accuracy, and benchmark performance all from your desktop with no hardware required.