The Akida IP platform is a revolutionary neural processor inspired by the workings of the human brain to achieve unparalleled cognitive capabilities and energy efficiency. This self-contained neural processor utilizes a scalable architecture that can be configured from 1 to 128 nodes, each capable of supporting 128 MAC operations. It allows for the execution of complex neural network operations with minimal power and latency, making it ideal for edge AI applications in vision, audio, and sensor fusion.
The Akida IP supports multiple data formats including 4-, 2-, and 1-bit weights and activations, enabling the seamless execution of various neural networks across multiple layers. Its convolutional and fully-connected neural processors can perform multi-layered executions independently of a host CPU, enhancing flexibility in diverse applications. Additionally, its event-based hardware acceleration significantly reduces computation and communication loads, preserving host CPU resources and optimizing overall system efficiency.
Silicon-proven, the Akida platform provides a cost-effective and secure solution due to its on-chip learning capabilities, supporting one-shot and few-shot learning methods. By maintaining sensitive data on-chip, the system offers improved security and privacy. Its extensive configurability ensures adaptability for post-silicon applications, making Akida an intelligent and scalable choice for developers. It is especially suited for implementations that require real-time processing and sophisticated AI functionalities at the edge.
