Low-Power Tech

1. Integrated computing architecture for self-research, storage and calculation

Foothold end side AIHardware scenario, independent research and development of storage and calculationComputing ArchitectureTo break the performance bottleneck of the separation of storage and computing, to achieve efficient collaboration between data storage and computing, significantly reduce power consumption, improve computing power utilization, and adapt to the rapid deployment and operation requirements of lightweight end-side models.

2、NPU/DSP/MCUheterogeneous collaboration

The heterogeneous architecture design of NPU (neural network processing unit), DSP (digital signal processor) and MCU (micro control unit) is adopted. Each unit performs its own duties and is linked efficiently. NPU specializes in AI reasoning, DSP optimizes audio and video processing, MCU is responsible for system control, and improves the multi-task processing efficiency and response speed of end-side equipment in all directions.

3Efficient listening Visual perception algorithm

Self-developed adaptive end-side hardware audio Visual perception algorithm, taking into account recognition accuracy and operational efficiency, can quickly achieve voice wake-up, audio noise reduction, image recognition, scene perception and other core functions, lightweight design adaptation end-side resource constraints, to ensure stable perception and accurate output in complex environments.