The Origin E2 NPU is engineered for power-sensitive devices requiring on-device AI processing. Suitable for smartphones and edge nodes, it supports various AI models and balances performance with low power consumption. By utilizing Expedera's packet-based architecture, the E2 achieves high efficiency and effective resource utilization, facilitating parallel execution across multiple layers to enhance computational throughput.
The E2 is highly adaptable, supporting a broad performance range from 1 to 20 TOPS, making it well-suited for running neural networks like RNN, LSTM, CNN, and DNN without the need for hardware-specific modifications. This ensures seamless integration for customers looking to leverage their existing trained models.
Equipped with a full TVM-based software stack, the Origin E2 simplifies network deployment while offering excellent customization capabilities to meet specific application requirements. Its design emphasizes field-proven reliability with its inclusion in over 10 million consumer devices globally.
