SG2000 is a high-performance, low-power chip designed for various product fields such as edge intelligent surveillance IP cameras, local facial recognition attendance machines, and smart home devices. It integrates H.264/H.265 video compression and decoding and ISP capabilities. It supports various image enhancement and correction algorithms like HDR wide dynamic range, 3D noise reduction, defogging, and lens distortion correction, providing customers with professional-grade video image quality.

The chip also integrates an in-house TPU, delivering approximately 0.5TOPS of computing power under INT8 operations. The specially designed TPU scheduling engine efficiently provides high-bandwidth data flow for tensor processing unit cores. It also offers users a powerful deep learning model compiler and software SDK development kit. Mainstream deep learning frameworks such as Caffe, Pytorch, ONNX, MXNet, and TensorFlow (Lite) can be easily ported to this platform.