More feature stories by year:
Return to: 2014 Feature Stories
CLIENT: IMAGINATION TECHNOLOGIES
May 28, 2014: Semiconductor Engineering
Imagination Technologies' CEO talks about why his company bought MIPS, the difference between ARM and Imagination, and how the fabless industry will need to refocus.
SE: What concerns you looking out at the semiconductor industry?
Yassaie: There are big changes in the industry over the last two years, such as the verticalization in the mobile space. Every semiconductor company wanted to play in mobile. It was just a fact of life that some of these guys would give up. That was a major change, and it was something we could see coming. Our focus is diverse enough that we came out of it in one piece, but there was definitely some turbulence. It's like getting into a boat knowing that it's going to get rough and you're as prepared as you could be. But we've gone through that and come out the other end.
SE: What do you mean by verticalization of the mobile market?
Yassaie: If you go back six or seven years there were 20 mobile manufacturers. With the advent of Google Android, we've essentially ended up with two platforms. If you look at where the unit shipments are it's Apple, Samsung, Chinese companies, and some smaller guys. If you're a chip provider, you're then faced with, 'Either I win this mega design win or I have nothing, after investing $300 million to $400 million.' That's why a lot of companies decided it's too hard. We were dealing with all of these people and they were our partners. As a business, that industry changed. Five to six years ago, graphics was the main market we were exploiting. We're still working with these companies, but it's not in mobile. But going back to your first question, what is always my main concern is scaling up our business. For us, it was starting from 40 to 50 people and growing organically for many years, then embarking on the MIPS acquisition—which for us was a big one—as well as a few smaller ones, and scaling up without messing things up. That is a challenge. The MIPS acquisition has been fantastically successful. A number of people told us it was the wrong thing to do. They said that for MIPS it was too late, or that a British company buying an American company would never work. They were all wrong. The reason things worked out is that the technology was right, and fundamentally it was a bunch of engineers who like to do cool things. And it was the right thing for the industry. I thought it would take two years to work. That's not the case.
SE: So what is Imagination Technologies now? How you define the company?
Yassaie: Buying MIPS wasn't an accident or a random thing. I was looking at my slides from 1995. Other than the logos being a different color, our plan showed our road map with fundamental IP categories needed to build an SoC. We had one arrow that talked about our graphics and multimedia. Another said processors. And the top one said connectivity. We really believed that from a technology point of view, each of these has a similar kind of engine. Video and graphics are a bit different, and ultimately we did fixed function.
SE: But they're all processing, right?
Yassaie: Yes, exactly. We put a lot of focus on graphics and the first deal we did was Sega Dreamcast, which was fantastically important for a team of 60 people. The hardware was really amazing. And internally, we started an effort to develop processor cores. That was 1995 to 1996. We thought that anything we developed would have to be multithreaded. We ended up developing that. Connectivity didn't come until 2000. You have to make sure each leg is successful enough to support the others, so the CPU development for four or five years was very much stealth. We were doing a lot of work, but we were not putting marketing behind it. And then the connectivity came. MIPS was a blessing in disguise. I remember sitting in a strategy meeting saying we're doing great in graphics, the video is working, we have good connectivity because we have programmable technology, which will handle many standards. With processors, we were using the cores internally, but we needed a bigger engine. [Fomer MIPS president and CEO Sandeep Vij] came to see me at one of the shows and asked, 'What do you think about MIPS?' That led to how could we do a deal. We couldn't pay $500 million for MIPS. So they split the business from the patents. The patents got sold for $350 million. We bought the rest for $100 million. We got to strengthen our processor capabilities significantly. But our goal was to get the multimedia/graphics, connectivity, and the processor as a basis for a platform.
SE: So what's the strategy behind this? Is it to build an integrated platform?
Yassaie: We are about enabling platforms. Our job isn't to actually do platforms. We make a whole range of graphics. It's the same for connectivity. You can have Bluetooth or 802.11ac. You can mix and match. We do build some of these platforms as test chips internally to make sure the software works, but ultimately it's our customers that decide which piece they want. And if you look at our customer base, you'd be hard pressed to find a customer that uses only one IP. There tend to be at least two. And there are system companies that have some IP they care about, but they still need the processor or some other piece to get to market quickly.
SE: How does Imagination differentiate itself from ARM and maybe even Intel?
Yassaie: Intel is different because they're a chip company. They're not an IP company. When you look at ARM, we're in the same space. We were the graphics company, they were the CPU company.
SE: And now you have MIPS and they have Mali.
Yassaie: We've done a lot more work on connectivity, though. We're very much a solutions enabling company. In terms of differentiation, in CPUs they're the big guy. Our job is to create the balance that the industry needs. We're on target to achieve that.
SE: It's an ecosystem battle, right?
Yassaie: Yes, and on the ecosystem we've added companies like Oracle. We ship 700 million MIPS CPUs every year. That's not trivial. But it's a long-range partnership. And the prpl foundation also tells you people are interested to see MIPS succeed. A lot of these guys already use MIPS. When we made the announcement of the MIPS acquisition, we went to our customer base and found out what they cared about and what was working. We put a lot of that intelligence into MIPS. We put a new road map in place. We got Mentor, Synopsys, Cadence and Green Hills sorted out on tools. Those things have an impact. Since the acquisition, we've had 60 to 70 MIPS licenses. This is a different place than where MIPS was.
SE: You're pushing the very edge of Moore's Law with mobile. What kind of problems do you see there?
Yassaie: I've been in semiconductor engineering for a very long time. There were never enough silicon resources. I remember doing a video decoder. We had to use seven chips because they couldn't fit on one. Now, even at 28nm, a full-blown MPEG decoder is 2 square millimeters. We believe 28nm will be around for a long, long time. The IoT will be at 40 and 65nm. But if you're at the other end developing one of these applications processors that has to squeeze a huge amount of performance into a small footprint with as little power as possible, then the latest process is relevant to you. And the volume better be high. Still, Moore's Law as we know it is gone. It's no longer the price goes down as your area decreases 1.8 times. That's gone. Area is still going down and power is going down, but the cost is going down. You pay more per square millimeter. But it's okay because everything is integrated.
SE: Now that you do have a number of key pieces, are you planning on 2.5D and 3D packaging?
Yassaie: We don't do that ourselves, but we are working with multiple partners—foundries and companies—doing a lot in that space. We also will bring in ray tracing, as well. Even we could use bigger bandwidth, so Wide I/O structures is important. Packaging technology is going to become quite exciting. It was always interesting, but it's going to get much more interesting.
SE: What happens to your customer base? The fabless industry is under a lot of pressure these days.
Yassaie: Historically, the fabless semiconductor companies will pick up special areas and become really good at those. There were people who did MPEG-2 decoders. There were things that small-to-midsize fabless companies could do and make a business out of it. To do these mega-application processors for mobile you have to be really big. But if you partner to one side, such as the IoT or automation or what's going on in assisted driving or cameras—there's an awful lot of new emerging areas that need a lot of knowledge and specialization. The fabless semis of the future will be a combination of semiconductor guys and specialists in their markets. You can see some of these customers emerging now, and those opportunities are immense. From our point of view, a health care company needs to understand how the sensors work, what algorithms are needed to extract that data, the processing, and they need connectivity. There's no point for them to spend their time worrying about the processor or connectivity. Where they should spend their time is determining what that application requires and what SoCs tap that know-how. Semiconductor companies that combine applications, particularly with the cloud, have plenty of opportunities. But they do need domain knowledge about the area they care about.
Return to: 2014 Feature Stories