Editor’s Note. In the initial March edition of this commentary, Lauro Rizatti, Marketing VP of EVE, was quoted as saying that there had been a total of 30M euros invested in his company. The correct figure is 13M euros.
Across the CAD industry (mechanical, electrical, architectural) it is routine for vendors to claim that their software products were designed by engineers for end user engineers. There are varying amounts of truth about this. But most engineers inside of CAD firms are not designing products, even if they were educated to be and actually worked as a practicing engineer out in industry. But I found an example in Lynx, a new product from Synopsys. The genesis of this product was linked to their turnkey design service offering.
I had an opportunity to talk about Lynx and its genesis with John Koeter, VP of Marketing for the Solutions Group for Synopsys.
Would you give us a brief biography?
I graduated in 1986 with a BSEE from Cornell. I have worked in the industry since then. I worked for 11 years at Texas Instruments, primarily in their ASIC division, holding a variety of different jobs. Then I transferred over to AMD in their embedded processor division where I worked for a year and a half or so. Then about 10 years ago I transitioned into Synopsys where I have had different roles including managing the turnkey organization here in our services business and also sales. Lastly, I am now in marketing for IP among other things Lynx and flows in general.
What is the Solutions Group?
That’s a great question. The Solutions Group basically covers IP, services and flows. Those are my three primary areas of responsibility.
What has been the reaction to Lynx?
As you know the Lynx Design System was introduced on March 16th at SNUG at San Jose and was featured the following week at SNUG in Israel. We have gotten very good feedback, good coverage in general and a lot of interest from our customers and from the field. So far things are pretty good. It really seems to have come to market at the right time
Why now? What are the issues that Lynx is trying to address?
It is not a surprise or a mystery to you that semiconductor revenue is way down. They are really struggling right now in terms of having the bottom drop out in the September or October timeframe. Another piece of data which is interesting is that the percentage of tapeouts that occur on schedule. We collect data from all the Synopsys user groups worldwide. I just looked at the numbers from San Jose in 2009 which say exactly the same thing. It has not changed for the last three or four years that I have been looking at it personally. Essentially 70% of tapeouts are late relative to their schedule. 40% are two or more months late. There are lots of economic models out there that show you just how drastic an impact there is by being two or three months late. It is a very, very significant issue. At a time when revenues are down, it is very tough in the world of semiconductors, where they have to do a lot more with less. Complexity is going way up causing schedules to slip. The reason why complexity is up at advanced nodes is new problems that pop up and old problems get even harder to solve. Things like functional verification gets harder because the chip is bigger, because you have to integrate and you have to verify more different areas. But also low power is another area that is very significant with some of the more advanced power management that people are doing at 65nm and 45nm. Not only does it affect your flows and your tools but it also affects the physical implementation. So there are some significant changes in the way you architect your design. According to our SNUG data, for the first time every power and managing power is the number one design challenge that customers are facing. At the same time gate counts continue to go up, nearly 30% over the last couple of years. Things just keep getting harder for designers.
Will you give us an overview of Lynx?
When we introduced the product we call Lynx Design System, we really tired to take a holistic look at the problems that our customers are having in the RTL-to-GDSII flow and how to solve those problems in one system versus solving them in individual tools. Of course, Synopsys continues to do solve problems and continues to innovate at the tool level. When we talk about Lynx, we have four major components to talk about. Part of Lynx is this production flow which has been proven down to 32nm. We have a customer using Lynx at 32nm. We have not taped out at 32nm yet but we have tapeouts at 40nm and a bunch at 50nm and 65nm. So really the heart of Lynx is a production RTL-to-GDSII flow. At the top is something we call the Management Cockpit. The job of the Management Cockpit is to automatically collect 50 different metrics as the flow is running and put them in an SQL database. What the Management Cockpit does is look at those metrics and enables you to track them over time. It is also a web-based utility so you can access it anywhere you can access the web. We also have a Runtime Manager where we automate the setup of the flow: configuring the flow, executing the flow with parallelism and then debugging the results. The last part is what we call the Foundry-Ready System. That basically makes sure that the inputs getting into the flow are very, very clean because as you know people have many different types of IP in their designs. These days seven or eight different types of IP is not unusual. It is not at all unusual for IP to not have consistent view. We have some facilities to check the input so that when you are in Place & Route, it won’t crash, when you find out that there is a missing view. Similarly we have been working with the foundries on the output side to make sure that when you tape out your designs, not only is the manufacturing clean and tends to support high yield but also it will be functionally correct. This is an overview of Lynx.
One of the things we did when we were looking at Lynx is that we did a survey of several hundred customers and asked them “What are the biggest problems you are facing now in the RTL-to-GDSII domain or just in general in implementing your designs?” The number one challenge was hitting power followed by timing. These were the most difficult things to do. You also had the challenges of physical design complexity, schedule predictability, importing IP, design productivity, cost pressures and dealing with the fact that design teams are almost always geographically disperse. We took all of these things into account when we designed Lynx. I am not going to say that we have addressed every single one and solved every one of those but we did take all of those things that designers were telling us that they are struggling with and tried to build features into Lynx to address them. Some of the things we did were building in an advanced methodology, having more transparency and so forth.
The Management Cockpit is a web-based utility. As you are going through the production flow, it tracks 50 different metrics and stores them in an SQL database. These metrics are common things like area, timing, utilization, how much time you are spending in each flow step, .. All of those things are captured automatically while the flow is running. By the way, you can have multiple revisions of the design in that data base. So you can be doing a bunch of what-if scenarios and reporting on what the different results look like. There are built-in reports. You pick the variable you are interested in and it generates the reports.
An example of a graphics report would be the total power of the design over time versus the target. Over the span of a week or so you would typically see the power coming down to where it meets the target and then bounces around the target as the design team is optimizing for other variables.