Standard Evolution

I recently had the opportunity to sit down with Lu Dai, chairman of Accellera Systems Initiative and senior director of engineering for Qualcomm.

SE: I have noticed a change in the way that Accellera operates these days. In the past, standards were driven by the EDA companies, but recently we have seen a lot more end-user company involvement and they are the companies driving new standards.

Dai: You are right. Accellera started back in the EDA history with Verilog and VHDL. In recent years, we are beginning to see standards that are purely user driven. We have things like UVM-AMS, which are coming from DVCon and SystemC Evolution Day. Most of the people who were asking for it are not from the EDA side of Accellera. The other healthy situation can be seen in the Functional Safety Working Group. The big drivers behind it are the EDA guys, but that’s because they have been working with the user companies. The user companies start by saying they wanted some tools to help with this type of work. EDA companies start to show the tools and the users complained that they don’t work together and so don’t satisfy their needs. This drives the EDA companies to the standards table.

SE: Why is functional safety special?

Dai: One of the special things about functional safety is the user want to have traceability. This means traceability for verification all the way back to the specification. Actually, it is not just verification: it can extend from manufacturing all the way back to the specification. But traceability includes the tool chains in between, that it’s certifiable and traceable. If everybody is using a different format in a different approach, then yes, I can still trace it, but it’s very manual, and there is a lot of overhead.

SE: Several standards have been criticized as being the least common denominator of what the EDA companies know how to support. How will that change?

Dai: We definitely have seen that in the past. One EDA company has a certain approach, and another EDA company may have a slightly different approach, and the user complains about the lack of cross-platform reusability. They want them to work together to push out a common standard. More recently, I see a little bit less of that approach. Users often have in-house solutions already. It’s hard to maintain and they want that solution to be taken over by somebody. So that’s the other approach. The first approach suffers because the user cannot express the need clearly. Does the EDA guy understand their needs properly? For business reasons will often try and tweak existing solutions to match the needs. But if the user company has some in-house solution, that could be a very good approach. That is why we are now seeing companies more willing to offer up their solutions. This approach is not entirely new. Think back to Vera – that came from a commercial company. Today, we are seeing this with UVM-AMS. There have been quite a few donations from user companies. This actually fits the background of AMS pretty well, because there were a lot of user company who complained about the lack of a good solution. They were all hacking something on their own. So now they are sharing their hacks. And they’re trying to make it work.

The Functional Safety Working Group is talking about a kind of spec reuse, from the top all the way down to the leaf node. The features you put into the spec, sometimes indicates what you actually intend to do. There are concerns about leaking their secret sauce. So, there are EDA company donations likely from this one as well, which probably will be safer. That is because it covers a superset, and so a user company will not be seen as pushing certain features into it. They’re worried that by pushing features, they may have leaked information.

SE: In the past, the International Technology Roadmap for Semiconductor (ITRS) laid out a timeline for when certain technologies were going to be needed. This enabled research and development to start and ensure products became available at the right time. Do we need that again perhaps for EDA? For example, the security working group is not tackling differential power safety because technologies do not exist to help with that. Should we place that on the timeline for future needs?

Dai: The people behind these standards were defining the standard based on their need, and not based on what’s available. Some companies have come along a little faster than the others. It will be interesting to see what kind of donations they’re going to do. The security working group is a particularly sensitive working groups, because it’s such a new area. Yet it is projected to be a high growth area. There are also companies who are among the leaders in the industry and yet they are not part of any standards body. They could have contributed to the industry, but they choose to keep their secret sauce.

Companies who are not completely vertical are still open to the standardization idea because they have to borrow the knowledge from somebody. If you completely go vertical, you can do everything in-house, and then they have a little bit less incentive for pushing a standard solution other than from tool chain support.

SE: More companies appear to be going vertical.

Dai: Right. It is following a trend where, historically, you see company going vertical and after a while, doesn’t work out and they stop, they spin off various parts, and then another 10 or 20 years later, they go back to being vertical again. They keep on going back and forth. Going vertical has a lot of negatives. They certainly take on a lot more risk. We are not restricted to certain geographical locations where people are more inclined to do things one way or the other. We continue to reach out to more companies, more people, and hopefully through this, we get a critical mass and influence of the people who are not going that way. If you look at those companies going vertical and not participating in standards work, they actually do use the standards work. It’s not that they do everything in house, they actually look at the standards.

SE: Everyone is talking about AI these days. Will Accellera get involved in this area?

Dai: There have been discussions about training sets. If I start to make my data shareable, then what is my competitive advantage? The big companies have tons and tons of historical data. When they deploy AI they are at an advantage. A smaller company doesn’t have access to much data, so your training set is much smaller and much narrower. The big company does not want to share the data with other people. The same will be true at the national level. One of my colleagues pointed out that China has more data than anyone else because the government essentially has control of all of the data. Compare that to the US. Each company may have a training set but because of privacy reasons, they don’t want to share the data. The attitude is one of, why I would want to share this data with my competitor because he’s going use my data to improve his product.