中文 English

CEO Outlook: 2020 Vision

5G, China and AI are prominent, but big changes are coming everywhere.

popularity

The start of 2020 is looking very different than the start of 2019. Markets that looked hazy at the start of 2019, such as 5G, are suddenly very much in focus. The glut of memory chips that dragged down the overall chip industry in 2019 has subsided. And a finely tuned supply chain that took decades to develop is splintering.

A survey of CEOs from across the industry points to several common themes that will begin to take root in coming months. Among them:

  • An explosion of data generated by more devices everywhere is setting the stage for new compute approaches, different architectures, and better ways of utilizing that data.
  • That increase in data will widen the potential attack surface, and an increasing number of chip and system companies are taking security more seriously as a result of that increased exposure. Layered security is the new buzz phrase.
  • Faster, smaller and low power are still considered critical, but cheaper is no longer part of that equation. The new focal points are added value and performance per watt (because power budgets are fixed), but cost per transistor will be calculated differently as multiple process nodes are added into increasingly heterogeneous designs.
  • China will have broad impacts on the chip industry this year, driving a dual supply chain and forcing companies to reassess where they do business.

While all of these trends have been discussed for some time, what is surprising is the rate at which they are achieving acceptance. For example, the hurdles to getting 5G into the market are well documented, but the technology will roll out in multiple regions and in multiple versions in 2020, which is well ahead of recent projections.

Likewise, the trade war with China, which began in earnest in 2019 and started to unravel nearly three decades of intertwined business with the United States, is accelerating a breakup between the two superpowers. The deciding factor was a brief ban imposed on any company doing research in the United States that supplied IP to Chinese firms, particularly Huawei. China has since begun developing a separate supply chain, prompting the RISC-V Foundation to shift its primary business address to Switzerland.

Better performance everywhere
Until several years ago, the key metric for processors was lower power. That was driven largely by the huge chip volumes of the mobile phone industry, where performance was good enough but battery life was limited. But with the rollout of AI/ML/DL everywhere, and more sensors generating more data, the emphasis has shifted to higher performance within the same power budget. That will likely continue into the foreseeable future.

“Smaller, better transistors, smart everything and more communications will be dominant in 2020,” said Aart de Geus, chairman and co-CEO of Synopsys. “5G is coming faster than we thought, and that’s being driven by the amount of data being created and the amount of computing needed to get something intelligent out of it. So we are going to see more devices at advanced nodes with more capabilities. The opportunity for us is really positive, because you need a brand new generation of design tools.”

He’s not alone in that assessment. “We’re very bullish on 5G,” said Eric Starkloff, president and COO of National Instruments, who will take over as CEO later this month. “Sub-6GHz is happening faster than expected, but we’re also seeing significant investment in millimeter-wave technology. The most obvious application of millimeter wave is in the industrial IoT (IIoT), particularly on the factory floor. In automotive, most of the opportunity is around sub-6GHz, because you can model that on 4G and 3G use cases.”

There are two flavors of 5G. One is in the sub-6GHz range, which is basically a faster version of 4G LTE. Sub-6GHz signals bend around corners and behave roughly the same as signals at previous wireless standards, only faster. But at the millimeter wave frequencies, signals fall off quickly and they are easily disrupted by other objects and even by weather. To make millimeter wave technology work requires beamforming around those objects using multiple beams, phased antenna arrays rather than simple antennas that exist on phones today, and lots of repeaters to extend signals.

This is an area of intensive research and development on a global basis, and it is considered an essential element in moving large quantities of data generated by sensors for AI/ML, both in industry and in markets such as automotive.

“AI and machine learning will be big in 2020, but we’re also seeing a lot of activity around quantum computing and IIoT,” said Lip-Bu Tan, CEO of Cadence. “At the system level, we’re seeing a lot of interest in RF and millimeter wave for 5G. Design teams are so busy right now that everyone is feeling stretched. There’s 5G, IIoT, AI, machine learning, and cloud data centers. Automotive is moving a little slower, but it’s still growing.”

Market churn
What’s clear is that the old way of doing things no longer works, and that recognition has come on suddenly. While everyone has predicted the death of Moore’s Law at some point in time, it is more of a slowdown than a sudden drop-off.

Those who can afford it will likely continue shrinking so they can put as much as possible on a single die, but it won’t just be a processor and SRAM. In all likelihood, it will be a complex, heterogeneous system with many different processing elements and accelerators, and the most advanced-node development likely will be integrated into multi-die packages because not everything has to be developed at 5nm or 3nm.

“This is going to be a big packaging year,” said Wally Rhines, chairman emeritus at Mentor, a Siemens Business. “That includes chiplets, 2.5D, 3D and interposers. If you want to double performance per unit of area, the only way to do that is packaging.”

This approach is more expensive, and chips and packages now are amortized across a system rather than just priced as a chip designed for a socket. That has significantly limits the number of companies developing chips at the leading edge, because the cost of the design alone could be more than $100 million. Add manufacturing and packaging into that, and the cost swells even further. Long term, this may be addressed through a chiplet approach and more standardized packaging, where costs can be managed by choosing less expensive chiplets for non-critical operations. But in the short term, the cost of advanced packaging remains high.

“This is going to be the year of haves and have nots,” said Jack Harding, president and CEO of eSilicon. “For some, this will be the year of great differentiation. Anyone that is not generating real value will be pushed to the side. The cost of staying in the game is growing exponentially. If you don’t get to hit new records, you’re toast. You can apply that to the best EDA vision, the best OSAT vision, and the best foundry vision. The other big change is that there is no semiconductor industry anymore as we knew it. There is a micro-systems industry, and if you do not understand how to bring up products at the system level you don’t stand a chance. So you have to approach every engagement as top-down to transistors, versus through the lens of transistor providers.”

Harding said this is especially evident in machine learning, where the majority of expertise is on the data science side rather than semiconductors. “They don’t have a semiconductor vocabulary, so they have to talk about system problems. The real issue is that a lot of the systems are semiconductors, and for many companies there is no one to translate what’s going on. As an industry, we have to help them get there.”

China represents another important piece in this tapestry, and the trade war between the United States and China has set the stage for a dual supply chain after some companies with U.S.-based research were ordered to suspend all business with Huawei.

“China’s move to independence will create new opportunities for some companies,” said Naveed Sherwani, CEO of SiFive. “Now, for every tape-out you may have two tape-outs, one for each market. So there will be a chip that adheres to U.S. standards, and another one for Chinese standards. Two supply chains are more expensive, but business will continue to grow. And China has been pitching an agenda of ‘One Belt, One Road,” and that footprint isn’t just for China. That will extend into Africa, Asia, and help to broaden the semiconductor business. In the short-term, there will be some loss, particularly for U.S. companies, but as people use more devices, business will increase everywhere.”

Mentor’s Rhines agreed. “The trade with China has turned the United States into the vendor of last resort,” he said.

Another important change in 2020 involves data. The amount of data continues to balloon, and there is no end in sight for that trend. But most of that data is useless. The big challenge now is to clean that data more quickly, either closer to or at the source.

“We need better methods for cleaning data to make machine learning far more effective,” said John O’Donnell, CEO of yieldHUB. “Sophisticated, real-time and OEE (overall equipment effectiveness) analysis will become affordable to subcons. Systems that can integrate more easily with each other to improve yield and shorten lots on hold times will become very important. What also will change is that analysis will become much easier and faster in the coming year, both at the edge and in the cloud. Also, more and more larger companies will embrace the cloud for their manufacturing analysis.”

Security
Another key area poised for growth is security. Chipmakers and IP vendors have been warning about security risks for most of the past decade, but it wasn’t until the past year that the vast majority of companies began taking security seriously. Among the key proponents are automotive vendors, which are worried about liability, and industrial operations, where the big concern is IP theft. But there have been enough attacks that security concerns are spreading, and the problem is only getting worse.

“Security is becoming a must-have in all IoT designs,” said Anthony Ambrose, president and CEO of Data I/O. “In the past, people had a view that security was just a cost add-on. That’s changing. What we have to do now is simplify and scale security so it is easier to adopt for small and midsize companies. You’ll see a lot more of a recognition that security can save money for on-board devices and companies will be adding security in. In 2020 you’re going to see a lot more deployments of security. The semiconductor guys already are doing more deployments of more secure silicon and secure memories. What’s different is they can’t argue anymore that they can’t find the necessary tools to do the job.”

Autonomous vehicles have greatly accelerated the timetable for security, even if fully autonomous vehicles are a decade or more away from reality.

“In 2020, highway driving starts to become real for autonomous vehicles,” said K. Charles Janac, CEO of Arteris IP. “You’re also going to see more applications for machine learning and AI emerge. Right now, there is too much money being spent on this by big Internet companies that are doing a lot internally. Those investments will shift. You’ll also see 5G becoming very important. We will need that for the last mile. The other killer app is cyber security, and this is one that is somewhat worrisome because we’re starting to see 5G and machine learning being used to track entire populations.”

Conclusion
Change is accelerating, in part because there are so many new markets and technologies, such as automotive, industrial and 5G, and in part because the practicality and price of scaling are becoming barriers to many companies — and insufficient for those that can afford it. The result is massive innovation on multiple fronts, big increases in investments in startups, and accelerated business cycles driven by intense competition.

Each new inflection point opens the door for new startups, any of which may grow to the size of a Google or Facebook. But change is everywhere this year, and what the market will look like when the dust settles isn’t clear yet.

Related Stories
5 Major Shifts In Automotive
How new technology developments will change the trajectory of the automotive industry.
What Engineers Are Reading And Watching
Some new areas are emerging, while other tried and trusted topics continue to lead.
Where Technology Breakthroughs Are Needed
Gaps are emerging everywhere, from interconnects and memories to materials and manufacturing.
What Worked, What Didn’t In 2019
The year started out shaky, but EDA racked up solid growth and semiconductors show good promise for 2020.



Leave a Reply


(Note: This name will be displayed publicly)