Problems Ahead for EDA

Experts at the table, part 3: Participants discuss if startups should be being tackling shift left, Agile, embedded software, IP, ESL or something else.


Semiconductor Engineering sat down with , chief technology officer at Carbon Design Systems; , chief executive officer for ; Randy Smith, vice president of marketing for Sonics and Michel Courtoy, vice president of marketing and business development for Kilopass Technology. In part one, the experts talked about innovation in the EDA industry, the breakdown of the funding model and the areas in which EDA needs to focus and grow. Part two examined some of the other difficulties associated with starting new EDA companies. What follows are excerpts of that conversation.


SE: We hear a lot about shift left today. Is this an entrenching of existing tools and methods or a new opportunity?

Smith: Shift left is not enough. We started an effort that we called Agile IC development. We are not trying to sell anything; we are only trying to instigate a discussion. It is intended to make people rethink the values of what they are trying to accomplish with design. We are not meaning that they should use Agile software for development, or the development models, but the concepts are something that people should think about. Shift left is trying to squish things closer together. It says take the existing tools and try and make them work better together without changing the flow much. We need to rethink how we do design if we are going to help it get faster.

Neifert: The people who try and fit it into the ways that they have been doing things are the people who have all the problems. Those who start by saying I want to do it differently and start with a blank sheet are likely to be more successful.

Davidmann: Ten years ago, that was the promise of SystemC. Let’s start again and do it better, more abstract. In principle everything was right, the reality is somewhat different. It was the implementation that was wrong. Raising abstraction and improving productivity was absolutely right, not just make the tools a little better or closer together.

Smith: If you are willing to start over, it is easier to see the holes. Today people are trying to make it work with what they have. There is one guy in the chip design process who does not have an executable language – the chip architect. Everybody else has a language. What does he have? A spreadsheet and he is writing specifications and tables for system requirements but it doesn’t go anywhere.

Neifert: You will have a difficult time selling to one person in a company.

Smith: Yes, it may require a different business model compared to a typical tools model. Maybe it is site license, but if you add enough value, then it will happen.

SE: That was the original argument made against investment in ESL

Davidmann: But they did attempt to go there. It was about trying to build a better methodology for system design but I don’t know if it was philosophically flawed or just bad implementation. It didn’t manage to deliver enough benefit for something like 50 man years of development.

Courtoy: And it has to fit into the business model of large companies. The big three have a business that is working, they are good companies, they are making money. Why should they change? We should not look there for a complete change. It needs to start from a clean sheet.

Neifert: But nothing starts from a clean sheet anymore.

Davidmann: Every day I see things in Google search – alerts. There is a new cell phone out there with a slightly different architectural twist – every day. They are all ARM-based. People can build these quickly.

Smith: There was nothing on the iPhone 4 that helped the iPhone 5 design because the chip design cycle is more than a year. The architectural decisions are from two generations back. This is the learning time necessary to do those chips. Shift left is the right intention when it says we have to get cycles down to six months, but we don’t have companies doing that for these complex designs. Not even close. That is where we need to get to and be able to learn faster. It is a disruptive change to go from a 12 to 15 month cycle down to 6 months.

SE: For the , people are saying they need a one or two month design cycle and that requires a different way of thinking.

Davidmann: And it could be using a much more programmable fabric, but it is quite small. Getting a product to market quickly – what would it take to do that? It may need a whole infrastructure around it. We have to think about the whole system and not just the chip. There are more software engineers per product than hardware engineers. So today, EDA is selling to the smaller group.

Courtoy: The reality is that the software engineers need more help.

Davidmann: Agreed. They have a much bigger challenge. They are still in the days of CAD. There is no formality to it. If they see it work once they ship it. Sure they have some static analysis tools, but it is primitive.

Courtoy: From my experience working with EDA companies, there is a constant discussion about if they should get involved in the software market. Synopsys made the jump, but the argument is that there is no money to be made in that market.

Neifert: It is difficult to establish an ROI. With a chip, you can say if you mess up, you have to create new masks and replace chips. For software, it’s just a new revision.

Davidmann: This is something the EDA companies need to do. Educate them that there is a better way to do things. I watched the Cadence EDA 360 strategy, Mentor Graphics’ Embedded Software Division and now Synopsys. The EDA industry has to go after software. I see a light which I hope with grow. I hope EDA and the EDA Consortium can start talking a lot more about the software that is close to hardware – not the apps. The stuff that semiconductor companies have to worry about. They have to have OSs running and ready for their customers to work on it immediately. Look at the trend of multicore design companies and most companies in this space died because of the software.

Smith: I look at ARMs acquisition of Duolog, and it is clear that ARM is getting into the tools business. They have a few tools related to the chip assembly process, such as register and memory maps. While they tried it before, they had the wrong DNA at that time.

Neifert: We have done quite well with it thank you. Fundamentally the problem needed to be solved differently from the way they were doing it. It needed to be freed from a single IP ecosystem. We were able to do that.

Davidmann: If you are using an ARM and using their new tools, does that mean they become a one stop shop for sub-system design? Imagination Technologies has been very successful with this in a graphics environment and ARM wants to grow into that market.

Smith: There are ways to do that. Fill out your IP portfolio with other things to plug in around it. Second, a tools environment that allows you to stitch things together, measure performance and customize it and then the software development environment that has to support this as well. 80% of phone chips are using those cores and I have heard that Tensilica has surpassed MIPS Technologies in terms of number of cores now. But ARM is the only one with high enough numbers to afford that investment so that they can get the necessary return.

Davidmann: The evolution of ARM providing more tools is to grow its business and make it easier to use its technology. If it is hard to assemble the chip and they are not finding that the EDA industry can help them, then they have to go out and do it themselves. A lot of semiconductor companies are building their own tools again. Apple is an example. They clearly have some fantastic technology. Some semiconductor companies have realized that they need to do things differently and they don’t want to go to the EDA vendors because they don’t want it being used by their competitors.

SE: Are we degenerating back to the days of internal CAD departments again?

Courtoy: This is probably a good thing because it makes the point that hardware is viewed as a differentiator. At Apple ten years ago, there was no hardware, it was all standard parts. They have now changed that equation and that is good for EDA because it is growing the customer base. Google is getting into the market, Amazon Lab 126, Facebook – lots of opportunities.

Leave a Reply

(Note: This name will be displayed publicly)