BrainChip announced that its Akida Pico, described as the industry’s lowest-power AI acceleration co-processor, is now immediately available for evaluation through the Akida FPGA Cloud platform.
The remote access service enables system designers to benchmark and validate neuromorphic models on live hardware without requiring local physical boards. The company states this approach reduces development cycles for intelligent wearables and sensors.
Akida Pico is an ultra-low-power AI co-processor that operates in the micro-watt (μW) to milli-watt (mW) power range. It is designed for applications in speech and audio processing as well as medical vital sign monitoring.
The co-processor supports standalone operation, where it independently handles audio and vital signs processing with minimal power for always-on medical and voice systems. It can also function as a co-processor, offloading AI tasks from a microcontroller to improve efficiency and reduce energy use.
For medical vital sign monitoring, Akida Pico enables wearable devices to track health data in real time and detect anomalies.
In speech and audio processing, it supports voice-activated technology, allowing smart assistants and hearing aids to respond to commands in real time while preserving battery life in consumer electronics and smart home devices.
The system is also optimized for industrial anomaly detection, providing continuous, real-time monitoring of time-series data to identify outliers for predictive maintenance in Industrial IoT applications, such as motor vibrations or sensor patterns.
Akida Pico is built to accelerate limited, use-case-specific neural networks with a power profile of less than one milliwatt. It targets always-on monitoring scenarios including voice and audio wake detection, keyword spotting, noise reduction; presence detection in intelligent doorbells and wearables that activate larger processors only on relevant events; real-time healthcare monitoring of medical vital signs in portable devices; and industrial vibration analysis for preventive maintenance.
Through the Akida FPGA Cloud, designers can upload models created with frameworks such as TensorFlow/Keras and PyTorch using the MetaTF software flow, then run them on remotely hosted Akida IP. The platform supports performance verification by measuring latency and power consumption before silicon commitment, offers configurations from one to six neural nodes, and provides rapid prototyping in a secure, browser-based Jupyter Labs environment.
Sean Hehir, CEO of BrainChip, stated that providing remote access to Akida IP via FPGAs removes traditional hardware barriers for engineers. He noted that developers can prove AI concepts in minutes rather than months, whether for high-performance robotics or sub-milliwatt wearables.
BrainChip will host a webinar demonstrating Akida Pico on the Akida FPGA in the Cloud on February 24, 2026, at 8 a.m. PST. Registration details are available at https://brainchip.com/akida-pico-webinar/.
