The Origin E2 is a versatile, power- and area-optimized neural processing unit (NPU) designed to enhance AI performance in smartphones, edge nodes, and consumer devices. This NPU supports a broad range of AI networks such as RNNs, LSTMs, CNNs, DNNs, and others, ensuring minimal latency while optimizing for power and area efficiency.
Origin E2 is notable for its adaptable architecture, which facilitates seamless parallel execution across multiple neural network layers, thus maximizing resource utilization and providing deterministic performance. With performance capabilities scalable from 1 to 20 TOPS, the Origin E2 maintains excellent efficiency up to 18 TOPS per Watt, reflecting its superior design strategy over traditional layer-based solutions.
This NPU's software stack supports prevalent frameworks like TensorFlow and ONNX, equipped with features such as mixed precision quantization and multi-job APIs. It’s particularly suitable for applications that require efficient processing of video, audio, and text-based neural networks, offering leading-edge performance in power-constrained environments.
