Experts At The Table: IP Business Models
System-Level Design sat down to talk about what makes a company successful in the IP market with John Koeter, vice president of marketing for IP and systems at Synopsys; Art Swift, vice president of marketing and business development at MIPS Technologies; Vishal Kapoor, vice president of product management for SoC realization at Cadence; and Charlie Cheng, CEO of Kilopass. What follows are excerpts of that discussion.
SLD: Third-party IP is being used increasingly in complex designs, but does the model work? Are people getting paid for all their efforts in creating IP?
Swift: From our perspective it’s going very well. There is a strong trend for people looking to reduce time to market. If you can buy pre-packaged stuff that’s pre-proven—in our case that’s the CPU, software stack and complementary IP around it—then you’re going to speed your time to market pretty dramatically. It’s all about time to market. The other thing that’s going on is an increase in complexity. We’re moving from single-core to many-core designs. We have customers in production with up to 32 cores and customers in development with up to 200 cores. At the same time, we have an explosion of customers in Greater China who are looking for pre-packaged solutions. In some cases they have a less-sophisticated design team but they’re trying to do very complex devices.
Koeter: We’re seeing a significant growth in our IP business, as well. This year it will be roughly $250 million. It’s a big part of our business at Synopsys. Dataquest predicts the IP market will be a $2 billion market over the next four to five years. What we’re seeing on our side is more and more outsourcing. That’s driven by the consumer market, convergence, and time to market issues. Our philosophy is that you’re buying schedules and risk reduction. If you’d have quality IP you’re not providing customers what they need.
Cheng: The semiconductor industry this year will be about $280 billion. That means the IP business is less than two-thirds of 1%. That includes both licensing fees and royalty fees. That’s low for the value that’s being delivered. So is the business thriving? The good thing is it’s $2 billion. The bad thing is it’s only $2 billion. The question is where should it be. Is it simply outsourcing manpower, or is it more than that?
Kapoor: In the 1990s, we started seeing an increase of embedded processors. In 2000 we started seeing the first outsourcing with the interfaces. The reason people did it was that it was standardized, so you lowered the risk. It was localized, so if you messed up it was only a small portion of the chip. This decade you’ll see memory and the fabric from companies like Sonics and Arteris. The challenge there is it is not localized. It affects the entire SoC, so people will be more prudent in what they do. But the real advantage is that it lowers risk—even over cost and schedule. Quality is a big factor in all of this because the industry has a ton of poor-quality providers, which is one of the reasons that Synopsys has done so well. For companies to do well in this business that have to have quality, value, and they also need a sufficient level of differentiation. On the interface side I’m not sure you can provide that, but you certainly can provide that with the memory and the processor.
SLD: Is the cost of running an IP business increasing? It now has to be characterized in much more detail than in the past.
Kapoor: It is, but the cost of the SoC is going up, as well. Complexity is going up everywhere. You used to be able to start an IP business with six people in a garage and come up with the next controller. The demands are going up. I don’t know if it’s disproportionate.
Koeter: I don’t think it’s disproportionate, but the costs are going up. If you look at 28nm, we have to do two layouts for every piece of IP on the periphery because of directional effects and layouts. You have to run Monte Carlo simulations extensively. When you go under 1 volt there are issues like IR droop. In the processor area, people are looking for bigger and bigger solutions with software stacks.
Swift: It’s not the cost of developing the IP that’s going up. We have 44% operating income this year. But the costs are going up on the ecosystem side. Customers expect a complete solution that includes the hardware, the software stack, all of it pre-integrated and pre-tested. This isn’t all bad, though. With all of this you can command a higher revenue and ASP (average selling price) if you can license those stacks and provide time to market for people.
SLD: Is the cost of entry into this market now too high for small companies, particularly at advanced nodes, because of the rising complexity?
Kapoor: It definitely is getting too complex and expensive. To get any part of the design done is getting more complex. Getting the PHY side done is extremely complex. And on the quality side, customers are demanding that this is silicon-proven. You need to run it at least through test silicon.
Koeter: Protocols are getting much more complex, too. Our best estimate is that USB 3.0 is about 20 times more complex from a verification standpoint than USB 2.0. It’s the same thing with PCI Express. It’s highly parameterizable and configurable, so verifying it is extremely complex. If you look at PCI Express 2.0 vs. PCI Express 3.0, the number of pages in the specs has doubled. There are almost 1,000 pages now. Designing IP that’s compliant is a challenge.
SLD: Do you now also have to provide details about what IP can be next to, in terms of noise and power? Is that part of the IP?
Koeter: It’s absolutely part of the IP. We have to assume that people are going to be integrating our IP on complex SoCs, and they may or may not be experts in that particular domain. We have to build in a fundamental amount of performance margins. There aren’t margins in terms of megahertz. They’re margins in terms of the specification. You may have to add an extra 50 millivolts of tolerance so you can afford noise on the SoC and still have working silicon. We do design reviews for all of our standard IP, we do signal integrity services.
Cheng: As an industry average we spend 2% more than a typical technology company, and yet on average we make less profit. The industry has fostered this incessant requirement to reduce cost for the customer. We deliver more value to the GDP. But we are extending more of our R&D dollars to keep pace. The reason we do this is that these kinds of extra checks can make the difference between who ends up helping the customer and who can’t. This industry is growing. The software industry, the online media industry and most other industries are going through consolidation. In contrast, the semiconductor industry is consistently going to deliver 3% to 7% growth.
Swift: It’s not that all of these challenges don’t exist, but in the move toward many-core devices there are all these system-level design challenges now embodied in the SoC. It’s an underserved market in terms of the tools available to the designer. That includes everything from how you power down a core to how you bring it up, how you bring it in and out of a coherency domain, and how you run things at different clock frequencies.
SLD: What you’re talking about is context?
Swift: Exactly. It’s the fabrics, the analysis of the bandwidth required for the different subsystems. Those challenges are much higher priority. It’s not to say that proximity issues and libraries aren’t there. But in terms of the grand challenges for the industry, it’s from the system-level design side.
Kapoor: It’s both. There’s the cost of implementation at the lower geometries, so the associated risk is very high. Ten years ago you delivered a reference software stack to show this is how it would work. Now for any piece of IP you’ve got to have that software stack delivered in a working model rather than a reference model. The value is being extracted on the system side. For us to get some of that value we’ve got to provide more. We can’t just say, ‘Here’s that one piece of PHY at 20nm that’s available off the shelf. There are integration and customization services along with that. If you’re not going to do that—which is the extra 2%–you’re not going to survive. The same trend toward consolidation that’s going on in the semiconductor industry will happen in the IP business.
Tags: Cadence, Kilopass, MIPS Technologies, Synopsys