Artificial Intelligence Application in the Field of Functional Verification

Abstract

The rising interest in Artificial Intelligence and the increasing time invested in functional verification processes are driving the demand for AI solutions in this field. Functional verification is the process of verifying that the Register Transfer Layer (RTL) implementation behaves according to the specifications provided. This is performed using a hardware verification language (HVL) such as SystemVerilog combined with the Universal Verification Methodology (UVM). Reading, identifying the key elements from multiple documentations, creating the verification plan, building the verification environment, implementing the tests defined, and achieving 100% coverage are usually the steps performed in order to complete the verification process. The verification process is considered finalized when functional coverage is at 100%. There are multiple ideas on how the process can be aided by AI, such as underlining the essential information from documentation, which would help in understanding faster how the Register Transfer Layer implementation works, thus vastly reducing time. In this paper, to greatly reduce the time spent on functional verification, two Convolutional Neural Network (CNN) architectures are implemented to properly classify the information across different documents; both approaches have significant and promising results. The database used for this classification task was created by the researchers using different documentations available.