Fast Search-Based RTL Test Generation Using Control-Flow Path Guidance

In this paper, we propose a functional test generation method for Register Transfer Level (RTL) designs by systematically performing highly efficient radial searches starting from heuristically selected system states. During test generation, we track the executed path with a lightweight instrumentation and classify tests that have enabled different or unique executions. Our approach heuristically targets least explored paths, unlike methods that simply try to maximize branches reached. We obtain an overall high branch coverage upon directing the testing effort along the exploration around such paths. The unique paths executed are represented in the form of a Search Tree. A path-oriented selection metric which distinguishes tests that cover least explored paths selects the best next starting state from the Search Tree. Our work is engineered with the goal of being able to produce tests for hard-to-reach states using a lean search-based approach, rather than expensive formal or semi-formal constraint-based methods. Experimental results over benchmarks from ITC '99 and IWLS-2005 show that our method rapidly generates short tests that achieve high coverages.