BrainChip’s Akida™ sets a new standard in efficiency, achieving very high performance with an extremely low-power budget and a negligible thermal footprint. Akida’s performance per microwatt can be orders of magnitude better than the current solutions on the market. Designed for sustainable AI, Akida performs complex functions at the device level, delivering a real-time, instant response.
With more efficient inference on the device, Akida can substantially reduce the need for cloud compute on edge AI applications. On-chip learning allows for customization on-device and doesn’t need additional, expensive, retraining of models on the cloud, ultimately freeing up the cloud for more complex cognitive tasks.
Akida’s efficient performance helps minimize or eliminate raw data that must be sent to the cloud for Edge AI applications thereby reducing exposure of sensitive data on the internet while increasing security. Advanced on-device learning allows for instant customization, keeping changes on-device, private, and secure as no learning data is stored anywhere.
The most performant edge AI architecture, Akida IP is also easy to implement and evaluate. MetaTF software provides a models zoo, performance simulation, and CNN model conversion. The Akida1000 reference chip is fully functional and enables working system evaluation. Our development systems (PCIe boards, Shuttle PCs, and Rasberry Pi) complement Brainchip’s IP and reference SoC to enable the easy design of intelligent endpoints.