Practical code coverage for Verilog

The ability to detect sections of code that have not been executed in a program or design has always been desired by hardware designers, since unexecuted code is clearly untested code. Unfortunately, there have been virtually no practical mechanisms for doing so. Given the combinatorial number of input possibilities for even a moderate design, exhaustive coverage obviously exceeds the capabilities of simulators. However, this does not mean that there cannot be useful heuristics, short of full coverage, that provide valuable feedback to hardware designers. This paper discusses a number of such practical approaches explored in a joint research project between Sun Microsystems and Chronologic Simulation. In addition to discussing the various intermediate approaches explored, the paper also discusses one approach that appears both practical and useful and its utility in uncovering untested sections in several Sun designs.