Semiconductor industry grapples with changes in markets, technologies and economics.
The semiconductor industry is undergoing sweeping changes in every direction, making it far more difficult to figure out which path to take next, when to take it, and how to get there.
The next few years will redefine which semiconductor companies emerge as leaders, which ones get pushed down or out or absorbed into other companies, and which markets will be the most lucrative. And that could change again as technology is used to interconnect markets that have so far never interacted with each other under the broad banner of the Internet of Things.
“Historically, the semiconductor industry replaces about half of the top 10 companies every 20 to 30 years,” said Wally Rhines, chairman and CEO of Mentor Graphics. “Most of those replacements will be new companies.”
Over the next few years markets, researchers and companies also will identify which technologies will be the next big drivers for chips, how they are made, and which applications they initially will focus on. Since semiconductors first emerged as a separate industry, this was a relatively linear and predictable progression. From mainframes and minicomputers to PCs and then to smart phones, the chip industry played a critical role in reducing the cost of computing and making it almost ubiquitous.
“The semiconductor industry lives on waves of new applications,” Rhines said. “We just got through wireless, where there were more than 50 chip companies and maybe 100 cell phone manufacturers. A bunch of them have been eliminated. Unit volumes have not declined, but prices fell. And now we’re at the tail end of the PC market. Everyone is waiting for the next phase to take off. In the short-term, it will be a tough market.”
That sentiment is being echoed across the industry—often with a large dose of consternation—as the smartphone begins to flatten and no other market of similar scale replaces it. IDC’s most recent forecast calls for 3.1% shipment growth in smartphones in 2016, down from 10.5% in 2015 and 27.8% in 2014.
What will replace it is unknown at this point. “We don’t know if it will be one application with billions of units, or a new class of products,” said Lucio Lanza, managing director of Lanza techVentures. “But the idea that technology drives the market is upside down.”
Lanza predicts that within five years, there will be one or more big new drivers for technology, such as a device that can handle multiple ways of perceiving and monitoring human health, or something that uses machine vision or some other electronic sensing technology to serve an end market. But rather than the semiconductor technology defining the market, as it did with the PC and at the beginning of the smart phone market, it will be the applications that define the technology. The challenge will be how to design these devices quickly and scale the manufacturing quickly enough to reach markets.
“The ability to move things quickly in many directions will be a big challenge to the industry,” said Lanza. “We will need to push processing in directions we never thought about. And it will be driven not by how quickly we can get to the next nodes, but by how fast we can get to market. That will require, in particular, adaptability for devices. The one who finds a new opportunity quickly is the winner.”
Waiting for the next big opportunity is hardly business as usual for the semiconductor industry, though. For the past 50 years there have been few lulls in end demand. Those that existed were driven by broad market downturns. There was never a question of how semiconductors would evolve.
“The only way to be successful right now is to win share from other people,” said Jack Harding, president and CEO of eSilicon. “There is a certain ruthlessness that has risen in the industry as a result of no growth coupled with consolidation. You find people who are merging, slashing R&D, and kicking the can down the road for the next CEO to explain why they’re not innovating fast enough. There’s also a set of behaviors that start to smack of desperation. The industry has always had tough people in it. But now there is an atmosphere that is a little bit less of a handshake and coopetition, and more of an ‘I win, you lose’ kind of sentiment.”
Physics only compounds the problem. The economic benefits of shrinking have stalled out at a time when other shifts are kicking in.
“After 14nm, the only thing that scales is density,” said Charlie Janac, chairman and CEO of Arteris. “Power, performance and cost per transistor do not scale. Improvements come from designing new architectures in the same node, and paying more attention to how to design chips to get a competitive advantage. You can see this with ARM’s big.LITTLE architecture. The smaller processor is doing most of the heavy lifting. Then you get the rest of the system to be cache-coherent. People are using this approach to make their own accelerators coherent for the CPU. You get more bandwidth and throughput with a single memory space.”
That doesn’t mean that shrinking features will stop. Tapeouts are happening at 10nm, test chips are being manufactured at 7nm, and work is underway to make 5nm possible. But it’s certainly getting harder and more expensive on all sides, and the number of players at these advanced nodes is shrinking.
“Foundries talk about version 0.1, 0.5 and 0.9 of their process, and in the past we engaged with customers starting at version 0.9,” said Antun Domic, executive vice president and general manager of the design group at Synopsys. “Now we engage at version 0.1. It puts more pressure on us, and it has raised questions among our customers about how much better output they can get from established technology nodes—28nm and 40nm are very important nodes.”
Domic said 2.5D could benefit, as well, once the industry agrees on the right approaches, particularly for mixing and matching of technologies developed using different processes. “If there is little more standardization, we certainly will move in,” he said.
Others are watching the changes, too, but there also is a level of disquiet over these kinds of shifts. “There is a nervousness in the industry when it comes to architectural changes,” said Sundari Mitra, CEO of NetSpeed Systems. “If you’re a startup you don’t care because you have nothing to lose. Or if you have really deep pockets, you can have two things going on at the same time. But people are getting more innovative to make up for not getting process improvements from generation to generation.”
What comes in the next phase is murky. Markets that will drive semiconductor demand are themselves in flux. Tesla, Apple, Google, and a number of Chinese startups are challenging the established players. The so-called “maker” market is chipping away at established players across a wide swath of end markets, often in lower volume using off-the-shelf components. And new markets that didn’t exist, such as robotics, or those which appeared to have bottomed out, such as image sensing, are showing explosive growth.
“A lot of guys still will need ASICs, and they will need to be low power, but they also may manufactured in quantities of 100,000 to 1 million,” said Dave Kelf, vice president of marketing at OneSpin Solutions. “Do they need to go to 7nm? No. But the question is whether they can get away with tools developed five to seven years ago, or whether they need some of the new features in tools.”
That isn’t clear, said Kelf, and it may vary greatly by market and by application. For level one baseband software, for example, the hardware and protocol stack need to be verified together. So even though the chip is not developed at the most advanced process, it may still require advanced tooling.
Frank Schirrmeister, senior group director for product management in the System and Verification Group at Cadence, points to a bifurcation in the tool market. For processors developed at the most advanced nodes, there will continue to be demand for the most advanced tooling. For smaller designs at IoT edge nodes, there will be many more of them.
“The key words are specification and application specificity,” said Schirrmeister. “So we will still deal with big customers and they will contribute to our business, but the tail is getting much longer. Those are typically designs that will not challenge capacity and complexity, but for any application domain, such as a heart rate tracker, it may not require the same tools but it will require a different balance of tools. So for analog/mixed signal that is interfacing with system-level design, you may encounter 1,000 things if you enter a new area of the network and the network needs to be configured. The system-level is not just the chip.”
Thinking well beyond the chip to the system-level is one of the big opportunities for EDA companies, and discontinuities and the need for more complex design beyond the chip could open significant opportunities for existing and new companies.
“The chip is just one component in the system,” said OneSpin’s Kelf. “This is where you could see some of the old HDL tools becoming more important because the permutations are greater than just hardware. It’s interactions and specified functionality. But the question is whether this will be done in layers, which is the current trend, or whether the hardware and software interact so tightly that it becomes more complex to test with the hardware.”
The influence of software on design already is on the rise, in part because markets are becoming more narrowly defined and in part because there are so many more software than hardware engineers. While that will not lessen the importance of semiconductors in years to come—software still needs to run on something—it could change the dynamics of the industry.
But there also is a middle ground alongside deep learning and artificial intelligence. “The idea is that you can train chips instead of programming them,” said Arteris’ Janac. “Google, Microsoft and others are heavily investing in that space. We’re just starting to see the top of the deep learning market. We have no idea of how big it will be or what shape it will be. The key will be to pick the winning segments for this.”
Also key will be understanding how to put together technologies more quickly to go after new markets. Advanced packaging technology is still in it infancy, but it will likely mature over the next several years to the point where building systems more quickly out of stock components will become much simpler.
“On the IP side, a couple companies already are starting to migrate their IP to deliver known good die,” said Bob Smith, executive director of the ESD Alliance. “If you look at the makers’ market, they’ve done interesting stuff buying small processors and doing their own software. There’s no reason they couldn’t spin into electronics. They pick up a board, canned libraries, and get a product to market. If it takes off, they build a chip. A similar argument could be made for FPGAs. There is enough interest in system scaling now to get a working group together on this topic.”
There are some large pieces missing from this analysis. For example, it is not clear is who will play and how far along they are with new technologies and approaches. Japan’s tech industry is rebounding, while Korea and China continue to challenge technology makers in the United States and Europe. There also is uncertainty about when interest rates will go back up, which will affect everything from continued consolidation to end market demand.
Moreover, rather than one or two factors that defined the winners and losers in the past—who could get to market faster with the most powerful processor or energy-efficient processor, for example—there is now a matrix of factors comprised of technologies that are at various stages of maturity, in markets that are increasingly interwoven and unsettled, and all with varying levels of financial visibility because much of the funding is private or small enough to go unreported.
On top of that, companies are trying out new ideas that so far have not reached the market in large enough volumes to identify trends. “There is definitely a shakeup in the industry right now,” said Anush Mohandass, vice president of business development at NetSpeed. “We are seeing new guys entering into the market and more vertical consolidation. We’re also seeing more platform solutions, not ASSPs, so if a platform works for 90% of your needs then you use that platform for differentiation.”
In that context—or perhaps because there is not enough context—making business and technology decisions becomes much harder. There are more factors, and not all of them are well-defined. But there is certainly a lot of interest in what’s coming next and how it will unfold.
Plotting The Next Semiconductor Road Map
Industry leaders examine enablers, implications, and perspectives for a changing technology ecosystem.
Rethinking Processor Architectures
General-purpose metrics no longer apply as semiconductor industry makes a fundamental shift toward application-specific solutions.
Will 5nm Happen?
Investments in finFET technology are hard to discard, but technical and financial challenges for getting there are huge.