The Uncertain Future Of Fabless Semis

Some industry watchers believe there is a shift to system OEMs calling more of the shots while fabless semiconductor companies are taking a back seat to decision making. If true, what does this mean?

popularity

As with most things, perspective is everything, this is especially true when it comes to changes in the semiconductor ecosystem. Some industry watchers say indicators clearly point to a shift happening where system OEMs again make the decisions about what is in a chip, both software and hardware, pointing to Apple, Samsung, Microsoft and Intel as prime examples. As a result, the fabless semiconductor companies are taking a back seat to what goes into an SoC. If this is true, what does this mean from a technology and ecosystem perspective?

“The reason why you had fabless semiconductor companies as separate entities and specialists and then people buying chips from them was because the technology was changing so fast these fabless semi guys were the experts and it was much more efficient and cheap to use them,” observed Kurt Shuler, vice president of marketing at Arteris. “As an OEM you didn’t want to be trying to keep up with all the technology changes. Well, technology change in semis has really slowed down. Innovation has slowed down. Some of that is just physics and there are all kinds of reasons, but the reality is mainstream things aren’t changing as fast as before. There are some new technologies out there like EUV, 450mm and 3D, but those are really, really, really expensive. These are things that only one or two or maybe three companies in the world are ever going to invest in.

What that means for the traditional semiconductor vendors is consolidation and layoffs, and those chip designers are going to work for companies like Google or Facebook or Apple or Microsoft or you name it, he said. “Those companies have the option of either designing their own chips – hire some design services firm to do the back end stuff, send it to TSMC, or if they have enough power they can go to AMD. This is what Sony and Microsoft did. In the short term a lot of the fabless semi guys are really in trouble. You can look at the finances of some of these guys. They need to find their niche.”

But is that going to be enough?

Shuler believes that’s the only option. “The fabless semi guys are going to be more specialists than some general guy that feeds a fab, which is how it was in the old days. Like TI, they are really going analog and they are going into an area and will specialize in that. Freescale made some inroads into automotive, but the networking stuff looked better. For ST, it is sensors. Each vertical industry only has enough oxygen in there for one or two big players. Mobility on phones and modems—you can tell who the winners are going to be. If you don’t include the China market you’ve got Qualcomm, MediaTek, Apple, Samsung and Intel IMC – those are the really big winners.”

The tipping point for the shift is likely technology maturation, in addition to the fact that capital investment for the next manufacturing node is so tremendous that very few companies will invest on their own.

Not everyone agrees, of course. “On the one hand we have at least several customers who are not doing this or definitely not yet doing this, so we are actively doing architecture work,” said Vasan Karighattam, senior director of architecture at Open-Silicon. “It could be the types of markets, but we are we are definitely doing complete spec to parts in both multimedia and other applications. It may be the very big, like Apple in the case where they are making smart phones and [other products] — they are making their own calls. We don’t normally architect parts for those. But for all of the other areas that we are working on, we actively work on architecture with our customers and they seem to want to work with us right from the spec onward.”

Who can challenge Apple?
If it is true that innovation is stalled, how do products remain compelling? “In the case of smartphones, you tie features together really sweetly so complicated stuff is really simple for the end user,” said Karighattam. “How do you do that? You create your own software, you create your on-chip, you integrate it really, really well so even though you are using the same raw technology as somebody else using merchant silicon, yours is much better. That’s kind of a play from the Apple playbook. When the speeds and feeds fail you… how else are you going to innovate?”

He believes the fabless companies will run into some big challenges if they can’t figure out how to specialize quickly. “They are going to have to specialize and are not going to be responsible just for the chip. They are going to be responsible for the integration of the whole package. If you are creating analog stuff you still have to make sure that can be easily integrated for your customers. If you’re doing the network stuff you have to create the software that rides on top of it. The OEMs are not just going to buy a chip from you because the expertise is in how the complete system fits together, not the silicon widget. The silicon widget is worthless without everything else around it. It’s got to be integrated into the design and it’s got to have a complete software solution.”

In terms of fabless semiconductor players needing to specialize even more than they do today, some of it depends on where some of the companies do business, Karighattam said. “If they are front and center making a device for a smartphone, for example, then the question is apt. Such a company would have to probably start making more niche products in the spaces they play in but Open-Silicon is not doing that. We play in so many spaces – networking and multimedia ASICs—that companies are seeing us more as a consultancy in smart metering and other stuff for the home, as well as on the networking side, in data centers, and for multimedia applications. We are actually receiving more requests from customers as to whether we are interested in taking over the entire spec to parts.”

From an ecosystem perspective, there are three reasons systems houses will take over and actually build their own chips, said Kevin Kranen, director of strategic alliances at Synopsys:

  1. When they have huge volumes for a single chip;
  2. Where software has to drive the hardware;
  3. Where it is very purpose-built hardware, such as big iron.

“When you have a Cisco it’s natural for them to build their own chips and it is built to run their specific application needs,” said Kranen. “I don’t see the equation having change that much. It’s kind of been the equation over the years, but you do see guys like Apple getting into sufficient volume that it’s just a no-brainer to design the chips themselves.”

Kranen believes there are a few companies that develop and drive the software and decide what’s going to be on these SoCs. “The more functionality they put in with working software the better it is for the whole ecosystem. But it also stresses it because there is always additional functionality that is expected.”

Consider what happens, for example, when Apple decides to put a front- facing and back-facing camera on a cell phone and develops software to exploit it and use it. That puts pressure on the whole ecosystem to do the same thing, which equates to four MIPI ports for all of the cameras on the hardware. Kranen said that drives not only what Apple puts into its silicon, but what its followers have to do, as well. And it affects IP suppliers, which need to know in advance which standards are going to be popular and how the new stuff is going to be integrated. “It’s kind of like surfing the wave,” he said.



  • jasiojasio

    This an excellent article. Yet there is a lot of work to be done to lower costs of 28 nm (no 20 or 14 nm nodes): 1) Put all designs on ultra low power technologies like ARM. Go to Google images and see what is server farm=need of power plant. This why Goggle is teaming with fabless Qualcomm because a need of much less power hungry server farms , 2) There is IBM-Samsung-ST-Global foundries consortium of FD SOI which takes much less power and is poised to get better yields than Fin-Fets even on 14 nm process. No need really of 20 or 14 nm because on 28 nm FD SOI ST designed ultra low power chip which beats performance of Snapdragon 800.
    Those technologies do not demand EUV 450 nm or 3-D expenditures and they will help even smaller fabless to survive. Just my 2 cents I would add to this excellent article.