Which Verification Engine?

Semiconductor Engineering sat down to discuss the state of verification with Jean-Marie Brunet, senior director of marketing for emulation at Mentor, a Siemens Business; Frank Schirrmeister, senior group director for product management at Cadence; Dave Kelf, vice president of marketing at OneSpin Solutions; Adnan Hamid, CEO of Breker; and Sundari Mitra, CEO of NetSpeed Systems. What follows are excerpts of that conversation. To view part 1, click here. Part 2 is here.

SE: Where does formal fit into the verification picture, particularly in regard to areas such as cache coherence and the Portable Stimulus?

Kelf: You can process large amounts of data in a formal engine more easily than you can move the design. Then you can analyze cache coherency with a specific capability that allows you to state the properties or specifics with regard to the Portable Stimulus tool. So it becomes much more straightforward. It’s analyzing that data, and having a platform to do that, which is the next frontier in verification. It’s not the performance of the engines. That’s not an issue. You have to have performance, of course. But it’s how you handle the data in the cloud or locally.

Schirrmeister: Formal is not the only smart engine. The distinction is you need smart tools on top of the four engines. Formal is one data contributor. You merge coverage today from formal, emulation, simulation and prototyping into a verification environment. The smarts is on top of it. To me, formal is one of the core engines. The distinction is whether it’s static or dynamic. You have different static and dynamic engines, and all of those contribute data. You have smart generation of test data and then the analysis on top of all of it. That’s where you want to apply machine learning. You want this thing running 24 x 7 to say you haven’t looked at this piece over here.

Hamid: We’ve figured out how to get the design and the tests to move across all of these engines. We have not gotten all of the verification environments to move completely. And one thing we’re seeing today is that as we accelerate the ability to run tests, we are running out of tests that need to be run. Portable Stimulus provides a way of generating more tests. Also, with Portable Stimulus you don’t have to move the verification environment. We can allow the test tools to translate from Portable Stimulus to each of these environments. There are good reasons why there are different environments. And we can do all kinds of layers on top of them to make them look the same, but that’s just adding overhead. It’s the same for all of the analysis. With Portable Stimulus you can just say, ‘Here are the tests you care about. Just go cover all of them.’ So we’re getting direct feedback from the test generators that they did all of these things, and here’s what went wrong. It’s interesting that we’re looking at machine learning, but some of the issues we’re dealing with today will be dealt with by Portable Stimulus.

Mitra: Formal is extremely ‘smart.’ It’s the only thing that helps us find deadlocks. I can simulate forever and not find the deadlock. I can emulate a lot. So everything complements each other. You can’t do away with anything. In the end it all comes down to analyzing what comes out of the verification environments. Is it a false positive or a real bug? How do you verify that? It’s still up to the engineers and architects to come up with the correct method of doing things and employing all of these tools to do what they know how to do best.

Brunet: I don’t think that, as an industry, we are smarter about the debug.

Schirrmeister: From the formal side we generate the testbenches that get executed and confirmed.

Mitra: Each tool has its own benefits.

SE: Where does the cloud fit into this?

Mitra: We probably have one or two tiny servers in the office. Everything we have done from the beginning is hosted. We have never invested in our own infrastructure. If you’re building massive systems, you can imagine that at the simulator level you have hundreds of licenses and every possible tool you can get.

Schirrmeister: It needs to make sense from a business model perspective, and business models are changing.

Brunet: What’s the level of comfort with the cloud?

Mitra: We are not yet distributing in the cloud, but is it far off? No.

Schirrmeister: There are two elements to the cloud here. One is smaller companies and smaller designs, which normally wouldn’t consider emulation. They couldn’t afford the investment in hardware over three years, so they use it for a specific project.

Mitra: That’s what we have done. We can’t afford an emulation box.

Schirrmeister: But there’s another piece to it, which is peak use. That’s what the big guys are doing. They have an emergency situation where they need many more cycles over a short period of time.

Kelf: That happens on formal, as well. But with emulation, you get these hilly graphs. With formal, it’s sharp spikes. The pay-per-use model works very well for that.

SE: This becomes important for Portable Stimulus, as well, right?

Hamid: Yes, it plays right into that.

Schirrmeister: Nirvana for Portable Stimulus is when you have a huge load of test cases, and in an ideal world you would not know where they execute. That could be done in the cloud, and we have customers looking at how to get the test cases that need to be verified into the cloud. This could vary by the time of day, too, because maybe there is a cheaper option if you don’t use non-peak capacity. We are looking at a different way of arranging schedules for this.

Related Stories
Which Verification Engine?
Part 1: No single tool does everything, but do all verification tools have to come from the same vendor?
Is Verification Falling Behind?
It’s becoming harder for tools and methodologies to keep up with increasing design complexity. How to prevent your design from being compromised.
Tech Talk: Verification
How verification will change as chips are used in machine learning, medical, 5G, and what impact advanced packaging will have on all of this.