Foundries See Mixed Future

Rising costs and a shrinking customer base make the leading-edge business difficult, but the overall chip market is growing.

popularity

Amid a tumultuous business environment, the silicon foundry industry is projected to see steady growth in a number of process segments in 2017.

As in past years, the foundry market is expected to grow faster than the overall IC industry in 2017. But at the same time, the IC industry—the foundry customer base—continues to witness a frenetic wave of merger and acquisition activity. Basically, the consolidation translates into a dwindling customer base for foundry vendors.

Adding to the uncertainty is a mixed picture for the overall IC market in 2017, plus some possible instability in the current economic and political climate.

For now, though, the foundry market remains a bright spot. 2016 was a relatively good year for foundries following a sluggish campaign in 2015. “The demand came back somewhere in the first half of 2016,” said Walter Ng, vice president of business management at UMC. “I wouldn’t say 2016 has been a gangbuster year for foundry. It’s been a good year.”

Going forward, foundry vendors are cautiously optimistic about 2017. “From what we see, customers are not expecting any significant correction per se,” Ng said. “Demand across the board looks relatively healthy. Demand for 8-inch is quite strong.”

In total, the worldwide foundry market is expected to grow by 6% in 2016 over 2015, according to Jim Feldhan, president of Semico Research. “2017 will see foundry growth improve to 10%,” Feldhan said.

Still, foundry vendors face some familiar challenges:

• The cost of cost to develop leading-edge processes continues to skyrocket at each node;
• Fewer foundry customers can afford to design leading-edge chips;
• The foundry customer base continues to consolidate, and
• There is constant pressure from customers worried about cost, cycle times and inventory control.

But foundry vendors also continue to innovate. They are expected to ramp up a range of new processes in 2017. At the high end, for example, Intel, Samsung and TSMC will ramp up 10nm finFETs in 2017. In addition, TSMC plans to roll out a new 12nm finFET derivative, with 7nm in the works.

In addition, the high-end planar process market remains robust. And other foundry markets, such as BCD, CMOS image sensor, embedded memory, mixed-signal, power semiconductors and RF, are strong.

To help the industry gain an insight into what’s ahead in 2017, Semiconductor Engineering examined foundry markets at the leading-edge nodes, planar processes and the specialty front.

More numbers
The IC market is a mixed bag. In one forecast, for example, Semico projects that the overall semiconductor market will decline by 2.5% in 2016, compared to 2015. IC units, however, are expected to grow by 4.2% in 2016, according to Semico.

Feldhan points to several macro trends that impacted the declining IC market in 2016. First, world economic growth was lower in 2016 than 2015; second, markets that previously sustained semiconductor growth are down or flattening, Feldhan said. The PC market continues to decline, while the smartphone market is saturated and slowing to single-digit growth rates, he said. On top of that, prices for chips related to PCs and smartphones experienced a significant degradation in 2016. And finally, the memory market, particularly DRAMs, experienced overcapacity and declining prices in the first half of 2016, he said.

But there is some good news here, as well. In 2017, Semico projects the semiconductor market will return to positive growth, increasing by 5.3% over 2016, with IC unit demand is projected to grow by 5.2% in 2017.

Semico’s Feldhan sees several macro trends in 2017. “World economic growth is expected to be slightly better. In addition, the smartphone market is expected to be slightly stronger, resulting in more stable pricing. And the memory market will achieve equilibrium, resulting in stable prices with increases in consumption for both DRAM and NAND,” he said.

All told, the industry remains cautiously optimistic heading into 2017. “Customers are continuing to proceed with business as usual, although there is some guarded cautiousness out there,” UMC’s Ng said.

For example, OEMs want to keep their inventories lean. But at the same time, they are under pressure to reduce their product costs. “What that means is (that a chipmaker’s) end customer are continuing to beat the drum on cost sensitivity,” Ng said. “So they are coming back to us and pushing us on continuous improvement in cycle time.”

It’s not all doom and gloom, though. For example, automotive, digital TV, IoT and medical electronics represent strong growth markets in 2017 and beyond, according to IC Insights.

screen-shot-2016-12-12-at-4-20-13-pm
Fig. 1. Where the growth is expected. Source: IC Insights.

The market for solid-state storage (SSD) drives is also strong, thereby fueling the growth of NAND. “Demand for server SSD will remain strong due to an increasing portion of high-density datacenter products and expanding adoption of all flash array in the enterprise,” said Chun Sewon, senior vice president of memory at Samsung.

Traditional planar NAND is no longer scaling, however. So, the industry is migrating from planar to 3D NAND. “We are still in the early days of 3D NAND,” said Yang Pan, chief technology officer for the Global Products Group at Lam Research. “(With 3D NAND), your density per unit area increases.”

What else is hot? “Looking further ahead, we are excited about emerging trends in virtual and augmented reality, big data, artificial intelligence and smart vehicles,” said Arthur Sherman, vice president of marketing and business development at Applied Materials.

At the high-end, more finFETs
Meanwhile, in the foundry business, Intel, Samsung and TSMC are expected to make the transition from 16nm/14nm finFETs to 10nm finFETs in 2017. Both Intel and Samsung believe that 10nm will be a robust node.

In contrast, TSMC says 10nm will be a short node and is putting more emphasis on 7nm in 2017. GlobalFoundries is skipping 10nm and moving directly to 7nm.

While that shows just how competitive leading-edge foundry market will be, it also points to some confusion. For one thing, the definition and specs of 10nm and 7nm are becoming murky. In addition, foundry vendors are chasing after a finite number of customers at those nodes at a time when design and process costs for these technologies remain too expensive for most companies.

“TSMC and Samsung have different strategies,” said Samuel Wang, an analyst at Gartner. “TSMC considers 10nm as a transitional node. TSMC encourages customers to use 7nm as well as 12nm, which is a derivative of 16nm. Only a few customers are using TSMC’s 10nm. Their 10nm wafer volumes will start to go up in 2017, but it will decline in 2019.”

TSMC is expected to move into 7nm risk production in early 2017. “Our 7nm has been adopted not only by high-end mobile customers, but also by high-performance computing customers for GPU, gaming, PC and tablet, virtual reality, server, FPGA, automotive and networking applications,” said Mark Liu, president and co-CEO of TSMC, in a recent conference call.

In fact, TSMC plans to beat its rivals to the punch at 7nm. To do so, it will utilize existing 193nm immersion and multiple patterning for 7nm, which is considered a complex scheme. The foundry would rather use extreme ultraviolet (EUV) lithography, which would simplify the patterning flow. But EUV won’t be ready in time for TSMC’s 7nm schedule, so it plans to insert EUV at 5nm.

In contrast, Samsung plans to wait until EUV is ready, and then will insert the technology for 7nm. “7nm, without EUV, will push the technology too hard,” said Hong Hao, senior vice president of the foundry business at Samsung Semiconductor. “It will result in a much higher mask layer count and a much higher complexity in fabrication.”

Dan Heyler, chief executive of eFusion, a global technology fund based in Hong Kong, said: “As for 7nm, the ramp schedule for TSMC and Samsung are highly aggressive. TSMC appears to be in the lead, but Samsung is committed to foundry.”

At 7nm, Samsung and others are banking on EUV. EUV is making progress, but it still isn’t ready for production. There are still issues with the power source, resists and masks.

“Multi-patterning works. Multiple foundries and technologies have already proven that,” said David Fried, chief technology officer at Coventor. “The single biggest excitement we’ll see in 2017 is some clarity on the commitment to EUV for 7nm. Some will commit to EUV for some levels in 7nm. Most people expect to see EUV at cut or via levels first. These are the places where EUV can be inserted into the flow as a substitution for complex multi-patterning flows.”

Don’t expect a broad adoption for EUV at 7nm, however. “The real density benefit will happen at 5nm with a more pervasive deployment of EUV throughout the process flow and an earlier commitment to that deployment,” Fried said.

Meanwhile, the 16nm/14nm finFET market remains vibrant. GlobalFoundries, Intel, Samsung and TSMC will continue to develop these processes. And in 2017, UMC is expected to enter the 14nm finFET fray.

“Many of today’s 28nm customers will move on to use 14nm/16nm finFETs,” Gartner’s Wang said. “The demand of 14nm/16nm finFET wafers won’t decline in the next few years.”

What else? “5nm research has to get into high gear (in 2017),” Coventor’s Fried said. “We’ve seen several device-level demonstrations of nanowires and other new devices. We’ve seen several patterning-level demonstrations of elements of 5nm technology. But, if any of the more revolutionary concepts are actually going to come to fruition for 5nm, 2017 is the year we need to see a more integrated demonstration, such as functional (perhaps small) SRAM arrays.”

Planar, embedded memory and RF
Not all foundry customers are moving to finFETs, however. Some will stay or move to the 28nm and 22nm planar nodes.

FinFETs are designed for high-end applications, while 28nm and 22nm address the price-sensitive and low-power markets.“Performance is important,” said Rajeev Rajan, vice president of marketing for IoT, automotive and new markets at GlobalFoundries, at a recent event. “But cost-effectiveness is also important.”

The leading-edge planar market is getting crowded, however. “28nm wafer demand will continue to be high,” Gartner’s Wang said. “But due to price competition, the revenue from 28nm may decline.”

GlobalFoundries, Samsung, SMIC, TSMC and UMC offer 28nm bulk processes. In addition, Samsung is also offering a 28nm fully-depleted silicon-on-insulator (FD-SOI) technology. And in 2017, GlobalFoundries is expected to ramp up 22nm FD-SOI.

There is room for both bulk and FD-SOI in the market. “One technology does not fit all,” GlobalFoundries’ Rajan said. “There are different needs in the market segments.”

And within those segments, the number of new applications continues to grow. “There is not just one driver of growth,” Rajan said. “There are a multitude of growth drivers.”

Embedded memory and RF, meanwhile, also represent growth markets for foundries. Today, the embedded market is dominated by traditional flash memory, based on NAND and NOR. Embedded flash is used in microcontrollers and other devices.

The mainstream market for embedded flash is 40nm and above, although some are migrating towards 28nm. On top of that, several foundries are working with various customers on the next-generation memory types, such as FRAM, MRAM, ReRAM and even carbon nanotube RAMs. Foundries want to offer next-generation memory IP to customers.

For example, GlobalFoundries is developing embedded MRAM for its 22nm FD-SOI process. Another vendor, SMIC, is developing embedded ReRAM based on the technology from memory startup Crossbar.

Still, the next-generation memory types are still waiting to take off after years of R&D. But when they finally fly, will these newfangled technologies replace existing embedded flash or traditional memory like DRAM and NAND?

“The economics and applications for embedded memory are different than standalone memory,” said Dave Eggleston, vice president of embedded memory at GlobalFoundries. “High-performance embedded NVM will co-exist with traditional DRAM and NAND standalone memory.”

Besides new memory IP, foundry customers also want integrated processes. In response, foundry vendors are integrating their processes with both nonvolatile memory and RF CMOS capabilities.

RF involves a number of standards, such as Bluetooth and the 802.11 variants. “There is a convergence of technology requirements for nonvolatile, wireless and low power,” UMC’s Ng said. “They are driving a controller with memory on-board and a wireless communication protocol. A lot of that is focused on 55nm and 40nm. Not so far down the road, some of the lead applications in this space will even look at 28nm.”

Specialty processes
The fragmented specialty process foundry market is also strong, especially bipolar/CMOS/DMOS (BCD), mixed-signal and high-end RF.

Some processes run on 200mm wafers, although some are moving to 300mm. “We see a lot of demand for 8-inch,” UMC’s Ng said. “Those applications are quite strong, whether they would be power management or RF SOI. Some of the older technologies like MOSFETs are a big demand driver.”

One technology, BCD, is heating up. Based on 0.18- and 0.13-micron technologies, BCD processes are used for power management ICs in smartphones and other products. More recently, BCD is moving into the automotive space.

GlobalFoundries, MagnaChip, STMicroelectronics and TSMC compete in the BCD market. More recently, UMC entered the business. “Everyone is chasing after automotive,” UMC’s Ng said. “We see a lot more customers focused on automotive. A lot of companies are focused on MCUs. Many are focused on infotainment. We are seeing a lot more interest in sensor technology.”

Indeed, the IC content continues to increase in cars. Infotainment and navigation are growth markets for ICs. Plus, several safety features, such as automatic emergency braking, lane departure/blind spot detection systems and backup cameras, are among the new market drivers. Autonomous driving technology also is hot.

In total, the automotive IC market is projected to reach $22.9 billion in 2016, up about 12% over 2015, according to IC Insights. Automotive still remains a small part of the total IC market. Its share of the IC market was only 7.9% in 2016, according to the firm.

Meanwhile, the high-end RF foundry market is also robust. In one big application, mobile carriers are moving towards a next-generation 4G technology called LTE Advanced Pro. The protocol requires smartphone vendors to support more frequency bands in their systems, meaning the RF content is increasing in the phone.

For that reason, demand remains robust for RF, especially the RF version of SOI technology or RF SOI. Used for switches and antennas in smartphones, RF SOI is produced by both IDMs and foundries.

In addition, several foundries, such as TSMC, UMC, Win Semiconductor and others, are chasing after the gallium nitride (GaN) foundry business, including both RF and power semis. In total, RF GaN was a $300 million business in 2015, according to Strategy Analytics. By 2020, the RF GaN market is expected to reach $688.5 million, according to Strategy Analytics.

At one time, base stations mainly used RF power transistors based on laterally diffused metal oxide semiconductor (LDMOS) technology. But today, LDMOS is gradually being displaced by RF GaN.

Foundries also are developing GaN for use in power semiconductor applications. Other foundries are pursuing another wide band-gap technology called silicon carbide (SiC).

Both GaN and SiC represent new growth opportunities in the maturing semiconductor industry. For example, the GaN-based power semiconductor market is projected to be a mere $10 million business in 2016, according to Yole, but that business is projected to grow at an annual rate of 86% from 2016 to 2021.

Related Stories
New Wave Of Consolidation
10nm Versus 7nm
Sorting Out Next-Gen Memory
RF GaN Gains Steam
GaN Power Semi Biz Heats Up



  • memister

    Will there be any fundamental differences between TSMC 7nm and Samsung 10nm?