This is the final part in the experts' roundtable on verification. In part 1, the experts talked about the state of verification and the progress that has been made. In part 2, the subject broadened to the increasing number of domains where verification has to be performed, including power and software. Part 3 discussed abstraction and the integration of IP.
Taking part in this discussion are Gary Smith, chief analyst at Gary Smith EDA; Paul Martin, senior manager for debug, trace and performance modeling at ARM; Rajeev Ranjan, CTO at Jasper Design Automation; Oz Levia, vice president of marketing and business development at Jasper; Harry Foster, chief verification scientist at Mentor Graphics Corp.; and Viresh Paruthi, senior technical staff member at IBM.
EE Times: We recently had the first version of the Accellera Unified Coverage Interoperability Standard released. This tried to codify where we are with the metrics we have. As we start thinking about future challenges, we are going to need some new coverage metrics, and we have to work out how it all fits together.
Rajeev Ranjan: UCIS has been a long effort that started in 2006. This standard is a first step. It is an API that enables disparate tools to interact -- a framework to talk to each other. I agree that we need new ways to measure what verification has been done. That space is still wide open. We are providing ways to measure metrics in this space. For example, when you use formal verification, you need to measure the sanity of the formal testbench before applying it to verification.
Harry Foster: We are looking at ways of extending coverage beyond the traditional functional coverage metrics, into the system level, particularly in a whole coherent network subsystem such as an AMBA 5-based system. You can't write a functional coverage model. It's a different way of thinking, so we're moving up to the system level, because that is where the problem is moving.
Viresh Paruthi: I second that notion and the idea that we are lacking in terms of coverage goals that span the system. The lower levels do a pretty good job with established best-practices. But at the systems level, we really need to go off and think of coverage goals, and that's where you can stretch your brain to a large extent.
Paul Martin: That's where UCIS starts to help, although it's not solving the problem yet. The issue is you have multiple verification platforms, and you want to converge their results. You want to combine the results from formal and those from the IP provider and merge them into your overall coverage metrics. These things haven't been solved yet.
Foster: You're right. It was a long process and baby steps to where we are today, but I'm optimistic that, now that it's done, it will open up opportunities for innovation. We do have new ways for merging and knowledge about what makes sense to merge.
Oz Levia: Second your point, Paul. There needs to be a drive from the user community to convince the vendors to start some meaningful integration between the databases and the semantics of the coverage, because otherwise everyone's an island, and when everyone's an island...
Martin: You end up duplicating efforts...
Foster: Look how much effort it was in the committee to define the simple semantics of even code coverage. It was a huge problem even coming to agreement.