CVU Graphs Library

The Computer Vision Unit (CVU) is a high-performance hardware accelerator block based on the Synopsys® ARC® EV74 Vision Processor within the SiMa.ai MLSoC™ designed specifically for computer vision (CV) and image preprocessing tasks.

Purpose of the CVU

The primary purpose of the CVU is to offload complex vision operations — such as color space conversion, image resizing, and dequantization — from the ARM CPU cores and the Machine Learning Accelerator (MLA). By handling these tasks in dedicated hardware, the CVU:

• Reduces Latency: Accelerates data preparation for AI models, ensuring that the MLA receives input data in the correct format as quickly as possible.

• Improves Efficiency: Performs vision-centric computations more power-efficiently than a general-purpose CPU.

• Optimizes Resource Allocation: Frees up the ARM cores for application-level logic and the MLA for high-throughput neural network execution.

This section contains information about the graphs available for the CVU, how to program the CVU, and frequently asked questions.