Experts At The Table: The Business Of IP

Second of three parts: Choosing the right IP; which IP vendors do best; integration challenges; make vs. buy; what it takes to effectively compare IP; differentiating with standard IP.

popularity

Low-Power/High-Performance Engineering sat down to discuss IP supply chain issues with Jim Hogan, an independent VC; Jack Brown, senior vice president at Sonics; Mike Gianfagna, vice president of marketing at Atrenta; Paul Hollingworth, vice president of strategic accounts at eSilicon, and Warren Savage, CEO of IPextreme. What follows are excerpts of that conversation.

LPHP: There’s so much IP available these days that it’s hard to choose which is the best. How do you navigate all of this?
Gianfagna: The market will always flow to the best. You want as many choices as possible so that the good ones succeed and the bad ones fail. Not every piece of IP is going to be good. You need a lot.
Savage: There is a natural bias in the market, though. You’re more likely to buy from your current IP suppliers than a new IP supplier. It’s harder for new companies to get started because you’re an unknown. That’s why it takes a long time for IP companies to take hold in the market. It takes a long time for royalties to kick in, but it also takes a long time for people to know who you are and what your advantages are.
Browne: It is a long time because your credibility is based upon your customers’ success, and the transactions you did last year they might not tape out until next year.
Hogan: Or this can be like Sony Ericsson. They can’t build a new phone because they can’t get the Qualcomm chip because of a foundry shutdown. So they’ve missed a whole cycle and they don’t ship that phone. If you look at how many cores are in an SoC, in 1990 there was an ARM controller, audio, some video. ARM and Intel pulled the applications processor inside and a whole bunch of other application processors sprouted up. Today we have control and we have a whole bunch of apps, and the apps are specialized processors that talk to the software. It’s only limited by the imagination of the people who build the system apps. As long as they keep on building stuff and coming up with innovative ideas and killer apps that people want, the hardware has to respond in performance, cost and power. That’s what I’m betting on.
Gianfagna: The innovation will continue. There will always be a market for new ideas and new apps and new IP. But putting it together is a real bear. It takes too long. It’s too expensive. And it’s too error prone. That sounds like an opportunity to make it more reliable and predictable to integrate IP more quickly. If you can do that it will boost the whole industry.
Hollingworth: From the integrator point of view, one of the trends that also goes along with that is that five years ago someone doing a custom chip would say, ‘Here’s what we’re planning to design.’ They would do all the front-end design, the layout, the physical design. They would put down an IP block and a subsystem containing a microprocessor and a memory subsystem and they would get to the stuff that is really their differentiator. Now, more often than not, they want a complete processor subsystem and they don’t want to design it because they’re not adding value. And then they put their key intellectual property into a couple blocks. And they ask us to assemble it. It becomes more important to stitch stuff together quickly and more easily.
Gianfagna: And that’s easier said than done. There’s a lot of work necessary to make it easier and more reliable.
Hollingworth: Especially if you want it to run really quickly and save power.
Browne: So how do we make the customer more successful? Each of us has a different strategy. IPextreme takes IP that other people have designed and makes it plug and play with standards. eSilicon takes all these permutations and identifies which are the ones that really give you differentiation. And after they pick the IP we’re saying that we can hook it up. Each of us has a unique view of the value that we’re bringing to that. That’s what keeps the IP industry healthy.
Hogan: It’s been proven that big companies have a hard time innovating on this. There’s a robust cottage industry in IP. There’s always an engineer with a unique idea like ESD protection. Over time innovation occurs, and when it occurs at a high rate and magnitude it’s disruptive. Every business model and supplier dies if they’re tied to that old business model and can’t change because they’re a victim of their own success.
Hollingworth: There is a clear graying of the industry. That’s an interesting issue because if all the new graduates are doing social networking apps, what is going to happen to our industry when the current population moves on?
Hogan: Let’s use the analog design as a comparison. All of them have been around since the 1970s and even studied physics in college. What happens when that generation retires? The answer is that people find a different way of doing it. Today the guys building apps don’t even think about this stuff. They look at the reference manual from Apple and away they go. The good news is there are young kids building apps every day. It doesn’t matter if most of them are successful if there are a billion of them.

LPHP: Let’s shift gears here. The whole idea behind IP is make vs. buy, but it doesn’t always perform as you expected it to. So how clean is this business model?
Browne: That’s a typical problem. My customers used to do what we do. They ask, ‘Do you have these features? And can you explain how these features work?’ My application engineers are just teaching them how these features work. But that’s okay, because we have 10 features they will need in the next design. They can’t afford to build their own interconnect anymore or have their own flow. We’re coming in with IP, Atrenta’s tools, we’re qualified in TSMC’s process, we have a Synopsys flow and a UVM flow. But they still have to hook us up to their proprietary SystemC infrastructure. That’s the bigger challenge in IP. It’s what you need to do to get the next chip done. Re-making IP you can buy from someone else usually isn’t the right answer.
Gianfagna: The reason why people buy is that it’s cheaper, for sure, but it’s also more reliable and more robust than you can do internally. If you sell it to 100 people it’s probably better than two people internally. But here’s the problem with the IP market: There’s no vocabulary. There’s no standard language we can use that says how good is a piece of IP. What does it measure up to? We need to come up with a standard vocabulary and a standard language that defines the quality of delivered IP. The only thing we have to differentiate quality today is whether it works. That’s a really painful way to figure it out. There needs to be more industry collaboration to address this weakness in the current market.
Hogan: Standards favor the startup. They don’t have to spend their time coming up with the infrastructure and the interfaces. When you don’t have standards people buy brand, because that gives them some assurance of quality and reliability. What’s required are standards and education, which will build a robust IP community and an economy.
Gianfagna: This is a tremendous opportunity. I’m surprised there is not more traction around it and more effort being applied.
Hollingworth: If you look at the number of design starts, that number is going down. But the value of the market continues to go up. That says the value per design is increasing. The reason anyone uses custom silicon these days isn’t for fun, because everyone knows it costs a fortune to do all the design work, not to mention all the mask costs to get these things out. We don’t see the make versus buy issue so much these days. For most companies it’s a no-brainer. If it exists and it’s proven quality, people will buy it. There are issues about up front costs. The only place we see this working differently is in memory. If you look at a custom SoC today, about 50% of the area is memory, and that’s growing. People use the memory compilers from companies like Synopsys or ARM, but while they do a good job they’re not optimized. One thing we see is that companies are optimizing the instances as they come out of the compiler. You might be able to improve performance by 30%, improve leakage by 30%, improve area by 10%. That’s the only counterintuitive trend. The rest is all make vs. buy.
Savage: If you can buy off the shelf, then you do that. But a lot of those guys have a lot of competitive pressures. They need different functionality from their competitors, who are buying IP off the shelf, as well. Design teams are focusing on designing IP for new standards where there is no IP company out there yet. And that IP has to be inherently re-usable. And then, just on the technology side, there are plenty of companies out there that have specialized SerDes because nobody in the commercial IP market makes anything in the class of what’s shipping in their products. That’s a very specialized technology. For high-volume networking and server boxes, that’s not traditional IP in those boxes. It’s differentiated IP that comes from IDMs.
Browne: We all look at this problem differently. I can show you that if you put a block of memory the size of an A9 on a chip then you’ll get more system performance than you would putting a third or fourth A9 on there. But they don’t want to get involved in designing internal DRAM. External DRAM is a commodity, so you can’t compete and get the value back. We’re looking at the problem differently, though. Where one of us sees a problem, another sees an opportunity. That’s part of the innovation. Which one of those two options is right.
Gianfagna: What are you really getting in terms of IP quality? We’ve tested about 15 pieces of commercial IP with our IP Kit and we’ve found errors in that IP that would have caused substantial delays in design or substantial chip re-spins. Running the tools and finding the problems isn’t that hard, so why don’t we do more of that.