Foundry Challenges in 2018

Growth will remain steady, but it’s getting harder and more expensive to move to the next nodes.

popularity

The silicon foundry business is expected to see steady growth in 2018, but that growth will come with several challenges.

On the leading edge, GlobalFoundries, Intel, Samsung and TSMC are migrating from the 16nm/14nm to the 10nm/7nm logic nodes. Intel already has encountered some difficulties, as the chip giant recently pushed out the volume ramp of its new 10nm process from the second half of 2017 to the first part of 2018, according to analysts. Time will tell if other chipmakers will have a smooth or rough transition at 10/7nm.

In addition, several foundry vendors are ramping up a new 22nm process, although the demand picture remains uncertain for this technology right now. On top of that, foundry vendors see enormous demand for 200mm capacity. Yet 200mm capacity will remain tight, if not sold out, throughout 2018.

It’s hardly doom and gloom in the industry, though. On the contrary, the foundry business is expected to see a solid year, growing 8% in 2018 over 2017, according to Joanne Itow, managing director for manufacturing at Semico Research. That’s exactly the same as 2017 growth projections. (Final tallies for 2017 won’t be available until the Q4 numbers available.)

“Growth drivers include AI, automotive and sensors,” she said. “Although the growth rate for smartphone unit sales has slowed, the smartphone BOM (bill of materials) still comprises the biggest portion of the foundry activity. That includes sensors, processors, image sensors, wireless and analog RF.”

Foundry vendors are also banking on growth in high-performance computing, power electronics and even cryptocurrency. But perhaps the most intriguing segment is China, where many foundry vendors are ramping up or building new fabs.

In the overall foundry market, TSMC continues to dominate the landscape with a 55.9% share in 2017, according to TrendForce, a research firm. GlobalFoundries is in second place, followed in order by UMC, Samsung, SMIC, TowerJazz, Powerchip, Vanguard, Hua Hong and Dongbu, according to the firm.


Fig. 1: Top 10 foundries by revenue. (Samsung, Powerchip numbers are estimates). Source: TrendForce

More numbers
Clearly, the signs are positive. In its most recent forecast, the World Semiconductor Trade Statistics (WSTS) group predicts that the semiconductor market will reach $409 billion in 2017, up 20.6% over 2016. The revenue growth has been fueled by an increase in average selling prices (ASPs) for DRAMs. Demand is robust for analog, flash memory and logic.

In 2018, the WSTS projects that the IC industry will reach $437 billion, up 7% over 2017. In total, IC units are projected to grow 7.1% in 2018, compared to 12.8% in 2017, according to VLSI Research.

The IC industry is changing, however. For years, the growth engine revolved around Moore’s Law, the famous axiom that states transistor density would double every 18 months. Adhering to Moore’s Law, chipmakers once introduced a new process every 18 months, ostensibly to lower the cost per transistor.

Moore’s Law is still viable, but it’s evolving. At each node, process cost and complexity are skyrocketing, so now the cadence for a fully scaled node has extended from 18 months to 2.5 years or longer. In addition, fewer foundry customers can afford to move to advanced nodes. In total, the average IC design cost for a 16nm/14nm chip is about $80 million, compared to $30 million for a 28nm planar device, according to Gartner. In comparison, it will cost $271 million to design a 7nm chip, according to the firm.

As before, though, the mobile phone represents the biggest market for chips. Cellphone IC sales, which account for about 25% of the IC market’s total revenues, are expected to reach $97.3 billion in 2018, up 8% from 2017, according to IC Insights. PC-related chips, the second largest market for ICs, are projected to grow 5% to $72.6 billion in 2018, the firm said.

Other markets are growing faster. For example, automotive IC sales are forecast to increase 16% in 2018 to $32.4 billion; meanwhile, IoT-related IC sales will rise 16% to about $16.8 billion in 2018, according to the firm.


Fig. 2: IC growth markets, in $billions. Source: IC Insights

“More and more customers are redefining their product portfolios to accommodate the IoT and/or automotive markets,” said Walter Ng, vice president of U.S. sales at UMC. “For the automotive segment, progress in infotainment, data security and advanced operational features are increasing the demand for MCUs integrated with embedded nonvolatile memory, RF components and MEMS sensors.

“In the IoT space, we see many different types of devices, but a predominant focus seems to be ICs that integrate MCUs with multi-protocol communications, including Wi-Fi, Bluetooth and sometimes even Zigbee. We are also seeing significant interest in home automation,” Ng said.

With those growth drivers in mind, foundry vendors must develop more and different processes to meet the growing requirements from customers. “The concept that one technology platform can service everything optimally from high-end IBM z System (mainframes) all the way to battery-powered IoT devices is unrealistic,” said Gary Patton, CTO at GlobalFoundries.

To meet these demands, foundries must increase their capital spending and R&D expenditures each year. But only a few foundries have the resources to develop a multitude of technologies. The smaller players are viable, but they must focus on a few select markets.

Besides the R&D dollar equation, here are some of the other challenges for foundries:

• Economic and political issues can impact the electronics sector.
• Lackluster demand and inventory issues tend to crop up in the first quarter, which could linger in subsequent quarters.
• There is an ongoing wave of merger and acquisition activity in the IC business. The consolidation translates into a dwindling customer base for foundry vendors.
• Silicon wafer availability is a concern. After years of oversupply, silicon wafer vendors see renewed demand. But vendors have not invested in new plants, and many have raised their prices.
• The packaging supply chain is another concern for foundries. Rising demand for chips is causing shortages of select manufacturing capacity, various package types and even some equipment.

10nm/7nm, 22nm migrations
In 2018, Intel is expected to ramp up 10nm. In addition, GlobalFoundries, Samsung and TSMC will begin to ship their respective 7nm finFET processes. Samsung also has announced various half-node offerings.

The node names are confusing. In simple terms, though, Intel’s 10nm technology is roughly equivalent to the foundry 7nm node.

Regardless, the node migrations are challenging. For example, several chipmakers took longer than expected to migrate from the planar nodes to 16nm/14nm. At 16nm/14nm, many vendors moved to a next-generation transistor type called finFETs. In finFETs, the control of the current is accomplished by implementing a gate on each of the three sides of a fin.


Fig. 3: FinFET vs. planar. Source: Lam Research

Generally, finFETs solve the short channel effects and other scaling issues, but the technology is more difficult and costly to fabricate.

Case in point: Intel was supposed to ramp its 10nm finFET process in the second half of 2017, but the schedule recently slipped. “It looks like that’s being pushed out until the first half of ’18,” said Abhinav Davuluri, an analyst with Morningstar, an investment banking firm, in a recent interview. “It could be a host of issues. But it’s more than likely that the issues they had at 14nm that took a while to solve have basically multiplied (at 10nm). Instead of double patterning, now you are doing self-aligned quad patterning. It’s just more steps as well as finer feature sizes. Those two issues are compounding upon one another.”

Time will tell if GlobalFoundries, Samsung and TSMC will face similar problems at 7nm. “It seems that all three foundries are making good progress with it,” said Samuel Wang, an analyst at Gartner.

Wang expects to see a sizeable ramp for 7nm in 2018, but it will pale in comparison to 10nm in the short term. In total, 10nm is expected to generate $5 billion worth of business in 2017. In comparison, 7nm is projected to generate sales from $2.5 billion to $3.0 billion in 2018, he said.

So how will 7nm play out over time? “It’s a more gradual ramp,” GlobalFoundries’ Patton said. “There are customers that are moving more aggressively (to the next nodes). Others will move more gradually.”

7nm, according to Patton, will be a long-lived node. “FinFETs will have a lot of legs. There is still a lot of room to extend finFETs,” he said.

In fact, TSMC says that the 7nm node could be as big as 28nm. “The initial applications for N7 are high-end application processors and high-performance computing. We expect more than 50 tape-outs by the end of 2018,” said C. C. Wei, co-CEO and president of TSMC, in a recent conference call.

Not all foundry customers are at the leading edge, however. Many are developing new chips and are exploring the idea of migrating to 16nm/14nm and beyond. Yet many companies are stuck at 28nm and above because they can’t afford the soaring IC design costs at advanced nodes.

Seeking to satisfy a potential gap in the market, GlobalFoundries, Intel, TSMC and UMC are developing a new 22nm process. 22nm enables faster chips than 28nm and is less expensive to develop than 16nm/14nm.

Not all 22nm technologies are alike, however. For example, GlobalFoundries is readying a 22nm FD-SOI technology. TSMC and UMC, meanwhile, are developing a 22nm bulk CMOS process. And Intel has a new, low-power 22nm finFET technology.

From there, foundry customers must weigh the various options. “It depends on what space you are in,” GlobalFoundries’ Patton said. “If you are focused on high performance and you are trying to make large chips, you would look at finFETs. If you are looking at something that’s more of a smaller chip, with a balance of cost and power, you go down the FD-SOI path.”

Bulk CMOS is another option. UMC is developing a 22nm process for use in RF, mmWave and other applications. “(This provides) the best combination of performance and cost,” said Raj Verma, associate vice president of the Specialty Technology division of UMC.

Still, many foundry customers will stick with 28nm for the foreseeable future. “The 28nm-class technologies offer an optimum combination of speed, power and cost,” Gartner’s Wang said. “There will be continued good demand for 28nm, generating $10 billion of annual revenue by the foundries.”

So, will 22nm take off? “Yes, 22nm is an extension of 28nm,” Wang said. “It offers better performance and density.”

200mm, specialty boom
Meanwhile, over the last two years, the IC industry has experienced an acute shortage of 200mm fab capacity amid a surge of demand for certain chips.

In 2018, 200mm capacity will remain tight, with 300mm expected to follow a similar path. “2018 will be the third year that 200mm technologies will be in very tight supply,” UMC’s Ng said. “And with 300mm technology capacities beginning to follow suit, we see customer sourcing strategies begin to take on a more strategic significance.”

A typical 200mm fab produces about 40,000 wafer starts per month. Generally, a 200mm fab produces a multitude of chips in processes ranging from 6-micron to 65nm. “Legacy MCUs, power discretes, PMICs, fingerprint sensors and display drivers still consume more wafers than existing capacity can meet,” Ng said.

All told, 200mm capacity will remain tight for some time, prompting chipmakers to look for new and creative ways to deal with the capacity crunch. For example, chipmakers have shifted some devices from 200mm to 300mm fabs. 300mm fabs already produce leading-edge chips.

“300mm technologies are expected to move toward much higher utilization, as well,” Ng said. “Legacy 300mm processes are being driven primarily by applications, such as power management, fingerprint sensors and display driver ICs, while more mainstream 300mm demand is driven by MCUs, wireless communications and storage applications.”

At mature nodes, foundries offer specialty processes, such as analog, bipolar-CMOS-DMOS (BCD), MEMS, mixed-signal, power management and RF.

Today, the specialty foundry business is undergoing a renaissance, thanks to the emergence of 5G and automotive. Power electronics and wireless remain growth markets, as well. “UMC is experiencing growth in BCD power management applications, driven by the global trend towards green power and energy reduction, which require increasingly more complete, secure and available power management solutions,” Ng said.

Meanwhile, the automotive market is still a small part of the overall foundry business, but the sector is growing at a fast pace. Consequently, foundries are scrambling to expand their efforts in the arena.

“We started to see a real inflection point over the last couple of years,” said Mark Granger, vice present of automotive at GlobalFoundries. “You can start to see the amount of semiconductor content in a vehicle starting to grow. With the addition of ADAS, there is certainly a surge over the last couple of years.”

Generally, the industry divides a car into five main domains—body, connectivity, fusion/safety, infotainment, and power train.

Foundries are seeing demand for chips in all domains. Some domains are growing faster than others. “In automotive, safety systems and electrification of the power train are driving specialty semiconductor content,” said Marco Racanelli, senior vice president and general manager of the RF/High Performance Analog unit at TowerJazz.

Foundry vendors are also gearing up for a push towards advanced driver-assistance systems (ADAS) and autonomous driving technology. ADAS involves various safety features in a car, such as automatic emergency braking and lane detection.

Luxury vehicles already are incorporating many ADAS features. The goal is to bring these safety features into mid-tier and entry-level models. That will depend on cost, which may take some time. Still, the advent of fully autonomous driving is years, if not decades, away.

Besides automotive, 5G is another potentially big market for OEMs, device makers and foundries. 5G is the follow-on to the current wireless standard known as 4G, or long-term evolution (LTE). It will enable data transmission rates of more than 10Gbps, or 100 times the throughput of LTE.

If 5G takes off, the technology will propel the development of new chips in both the infrastructure and the handset. “The need for faster and more mobile data is driving immediate demand for the expansion in hyperscale data centers and cloud computing hubs, and will create future opportunities in 5G systems for handsets,” TowerJazz’s Racanelli said.

“Data centers require specialty semiconductors from power management to high-speed optical fiber front-ends,” Racanelli said. “5G systems will drive more complexity in RF front-ends, requiring more RF-SOI switches, filters and amplifiers. Finally, 5G includes a 28GHz band for which our customers and partners have already created transceivers demonstrating 12Gb/s links in advanced SiGe.”

The big question is whether 5G will disrupt the landscape or fall short of its promises. “In the near term, there are questions on how the infrastructure will develop to support the currently defined 5G radio frequencies in the Wi-Fi and sub-6GHz bands for products like routers and smartphones,” UMC’s Ng said. “Longer-term concerns involve when the 5G infrastructure deployment will start and how long it will take to become viable for general public use. Associated with this will be the timing of when products that can utilize the 5G benefits will achieve mass market acceptance.”

China watch
Needless to say, China represents a big opportunity for foundry vendors. SMIC and other China-based foundries continue to expand. Then, on the multinational front, UMC has been ramping up a new 300mm fab in China. GlobalFoundries, TowerJazz and TSMC are building new fabs in China.

Foundries will not only make chips for multinational IC makers, but also for a growing collection of domestic fabless design houses. In total, China’s IC design industry revenue is expected to grow 22% in 2017, according to TrendForce. In 2018, the growth rate is expected to remain at around 20%, according to the research firm.

“This growth will be attributed to the IoT market, where AI and 5G solutions will be driving the expansionary momentum,” said Jeter Teo, research director of TrendForce. “Other revenue opportunities will come from emerging applications, such as biometric sensor hardware for fingerprint and facial recognition, dual-lens cameras and AMOLED.”


Fig. 4: Growth of China’s IC design industry. Source: TrendForce


Fig. 5 Ranking of China’s fab design companies by revenue. Source: TrendForce

While the industry is keeping a close eye on China’s semiconductor industry, it also should pay attention to the nation’s efforts in the application space. China is pursuing 5G, Industry 4.0, smart grids and other technologies, according to Lung Chu, president of SEMI China. “In some areas, (China) is moving faster than the rest of the world,” Chu said during a recent presentation.

Related Stories
The Race To 10/7nm
22nm Process War Begins
The 200mm Equipment Scramble
Foundries Accelerate Auto Efforts
China: Fab Boom or Bust?



5 comments

Mike Cowan says:

Hi Mark – thanks for the great overview of the semiconductor foundry business and its challenges – certainly daunting challenges.

Here are the details of the latest WSTS forecast numbers, published in November 2017.

Allen Rasafar says:

Great article, thank you for sharing this informative overview by Dr. Gary Patton.

EMVGUY says:

I dont think Intel has a real foundry business. Their revenue is probably ranked 20th (half of Dongbu Hitek) and it is dropping. Why are they mentioned here? Intel is a major customer of Foundries for wafers for Comms, FPGA, ASIC Controllers.

Mark LaPedus says:

Hi EMVGUY. FYI. Intel has a foundry business, called ”Intel Custom Foundry.” Go to this site: https://www.intel.com/content/www/us/en/foundry/overview.html So technically, Intel should be mentioned here. They don’t go after a broad base of foundry customers like TSMC. Intel’s foundry business focuses on a few customers. The real question is whether Intel has been successful in the foundry business. Based on market share figures alone, the answer is no. But market share is just one metric. However, I wouldn’t dismiss Intel–yet. They are still leading in chip manufacturing technology, but clearly, the competition is catching up….

EMVGUY says:

Agreed, That’s why I said “real” foundry business. 🙂

Intel still has trivial revenue and the ICF work was started over 5 years ago. Even Intel’s Altera group is still using other foundries for most of its volume.

What metric would show they are successful since the revenue is <$500M per year and they lose money?

Leave a Reply