Efficient approaches for functional validation of soc designs using high-level specifications

Prabhat Mishra,Mingsong Chen

2010-01-01

Abstract:Increasing complexity coupled with time-to-market pressure create a critical need to raise the abstraction level for System-on-Chip (SoC) designs. Functional validation is widely acknowledged as a major bottleneck due to lack of automated techniques and limited reuse of validation efforts between abstraction levels. Simulation is the most widely used form of validation using random or constrained-random tests. Directed tests are very promising for simulation since only fewer directed tests are required compared to billions of random tests to achieve a coverage goal. Currently, directed test generation is performed manually which is time-consuming and error-prone. This dissertation presents a novel top-down methodology for automatically generating directed tests from high-level specifications and reuse them across different abstraction levels. The objective is to reduce the overall functional validation effort. My research has four major contributions: i) it proposes a method that can extract formal models from high-level SoC specifications; ii) it presents an approach that can automatically derive properties based on fault models; iii) it proposes efficient clustering, learning and decomposition techniques to reduce the directed test generation time; and iv) it provides validation refinement approaches to enable reuse of the system-level validation efforts for low-level implementation validation as well as to check the consistency between different abstraction layers. Our experimental results using both software and hardware benchmarks demonstrate that the proposed approaches can significantly reduce the overall validation effort.