Polysilicon forecast is cloudy for solar cell, IC industries
GaN CAN: HEMT devices represent one type of compound semiconductor that is built on gallium nitride wafers. (PHOTO COURTESY OF FBH)
After 17 years in the silicon side of the semiconductor business, Mike Chase knows a thing or two about Moore’s Law. But Chase now toils in the world of III-V materials, where the delighted Veeco Instruments executive learned that they do things a bit differently in an industry that has its own version of Gordon Moore’s famous dictum.
“The things that drive the LED business are a lot like the semiconductor business,” acknowledges Chase, the vice president of marketing for the Woodbury, NY–based equipment manufacturer. “Instead of cost per function, it’s cost per lumen. And instead of transistors per chip, it’s lumens per watt. That’s our efficiency: how much light do I put out for every unit of energy I put in?” Chase says. “That’s our Moore’s Law. It’s called LED efficiency.”
Light-emitting diodes, of course, are just one of the many optoelectronic and microelectronic RF, microwave, laser, and communications devices made by compound semiconductor manufacturers that use metal-organic chemical vapor deposition (MOCVD) equipment from Veeco and other suppliers.
Comparison with the well-established silicon side diverges, though, on everything from wafer size, to integration issues, and defect concerns. When dealing with the predominating sapphire and gallium-arsenide (GaAs) substrates and others made from materials such as gallium nitride (GaN) and indium phosphide (InP), Chase learned that the small differences are huge.
“For LEDs, feature sizes are actually getting bigger,” says the self-described “semiconductor veteran and ex-Varian guy…. If you’re going to light up a room with LEDs, the chips that make up the light array will tend to get bigger. At the same time, the cost per lumen has to go down.”
Compound device manufacturers have a different set of defect concerns than their silicon counterparts, Chase notes. One of the key problems in LED manufacture is lattice mismatch with the substrate. “One of the challenges in building an LED on GaN is dealing with the lattice mismatch between the sapphire substrate and the gallium nitride.” Manufacturers say there’s as much as a 16% gap between the lattice constant for GaN and sapphire.
“There’s a difference of opinion over substrate properties,” observes Marty Brophy, manager of foundry product engineering for TriQuint Semiconductor in Hillsboro, OR. “Some like higher resistivity, some like lower resistivity,” for example.
And yields? “Yields are actually pretty much under reasonable control. The old days of this being a 50% yield [business] are well past us. Anybody producing 100 million of something had better have 90%-plus or 95%-plus yield.” He points out that yields are not defect limited as they are in microprocessor and memory device processing.
Yields in LED processing are a function of wavelength uniformity, explains Chase. “MOCVD suppliers are always working to drive yields up,” noting that they are still not as high as mainstream semiconductor yields.
The looser defect restrictions are reflected in the cleanroom classifications used by compound semi manufacturers, according to Oded Tal, CEO of MAX International Engineering Group (IEG), a fab efficiency consultant based in Closter, NJ. Geometries are large enough that killer particles in the form of dust and other airborne contaminants hold less importance in compound chip manufacturing, he says.
There is no need for cleanrooms in classes typically used by silicon-based device manufacturers, notes Tal, whose company works with both silicon- and compound-based device manufacturers in order to improve their fab operations. He has yet to see any justification for rooms cleaner than Class 100.
GaAs wafer sizes had been increasing from 3 to 4 to 6 inches, says Dean Freeman, principal analyst for semiconductor market research at Gartner Dataquest in San Jose. The size has dipped back to 2 and 3 inches, to the benefit of what he calls “second-tier semiconductor [tool] companies,” such as Oxford Instruments and Ulvac, that “do a lot of III-V compound work.”
Business for GaAs and GaN device manufacturers and their suppliers is “pretty good,” Freeman believes. “The downside is that silicon keeps encroaching on gallium applications. It’ll even take over the indium phosphide applications and some of the more exotic compounds.”
Sematech recently launched a project to investigate the use of alternative materials to silicon in MOSFET channels. The consortium’s front-end processes division will explore the applicability of silicon-germanium and germanium as channel materials for enhanced mobility. The goal is to develop process technologies to replace silicon as CMOS scaling continues.
Sematech notes that germanium shares certain basic properties with silicon that once made it a rival material until integration issues became too much to overcome. Despite the mobility and scaling issues, industry experts believe silicon’s strength lies in its infrastructure and the industry’s ability to stretch the material’s capabilities with strained silicon and other technologies.
George Henry, the technical program chairman of this month’s International Conference on Compound Semiconductor Manufacturing Technology (CSMantech) in Vancouver, BC, says silicon “is the 500-pound gorilla in the room, if you will.” The equipment manufacturers, the fabs, and the huge financial investments give silicon an edge “because those guys have to look at the last tiny spec to get that 100% yield.” Henry, a senior advisory engineer at Northrup Grumman, points out that although many vendors are accommodating and helpful, the defense contractor has to adjust much of the equipment it uses because the tools are, essentially, silicon-centric.
Henry took note of Sematech’s announcement. “Silicon germanium with its siliconlike processes can be very competitive with the use of the available infrastructure.”
Freeman says that for every engineer who insists that switching to III-V compounds or other “exotics” is the road to improved performance, “you’ve got three people working with silicon to keep it going.” By way of illustration, the market analyst tells the story about his “communications guy, an old RF-CMOS [expert]. He says, ‘You know what, Dean? Don’t talk to me about that III-V stuff. I’m a silicon bigot, because we keep finding ways to make silicon work.’”
Is silicon encroaching on compound’s turf? “Certainly, we’re always watchful and concerned. Most of the market will be in silicon, but a percentage will use compound semiconductors,” explains Scott Davis of Sumitomo Electric Semiconductor Materials in Hillsboro, OR. “There’s rarely a knockout punch. Sometimes they’re encroaching on us, and, in other cases, we’re encroaching on them. Most of the switches in cell phones two years ago were silicon-based PIN diodes.”
Davis emphasizes, however, that as the architecture in cell phones becomes more sophisticated, the silicon-based diodes just can’t handle the more sophisticated needs of the switches. Using a silicon-PIN diode economically becomes “less and less possible. Silicon switches can be used in the low-end phones, but for the higher-end phones, the functionality of compound semiconductor–based products is better.”
Industry experts say that silicon continues to improve its frequency capabilities and performance, but the edge still goes to GaAs, particularly for high-frequency device requirements in the RF and microwave arenas. As more high-frequency communication applications crop up, compound semiconductors continue to remain highly relevant, since electrons travel up to 10 times faster in compound materials than they do in silicon.
Cell phone switches and power amplifiers—the primary drivers in the compound device markets—are being sold in modules, and the drive to lower costs is leading to some interesting approaches, says Davis. Most power amplifiers use a heterojunction bipolar transistor (HBT) structure, while most switches use a pseudomorphic high-electron-mobility transistor (pHEMT) structure.
Integration approaches differ, depending on the company. One approach is to have separate power amplifiers and switch die in one module. But some manufacturers have announced that they’re putting two different kinds of epitaxy—HBT and pHEMT—on the same wafer. TRW first published papers on this, he says, and more recently other companies have announced similar capabilities.
Earl Lum, a leading wireless industry analyst, explains that “in EDGE handsets, device manufacturers are finding that most power amplifiers have to be sold bundled with the transceiver because the designs are so closely integrated. Device manufacturers need to be able to sell a complete solution.”
Cell phones should continue to drive the market for compound semiconductors in microelectronics. In addition, Davis sees a bright future for the compound semi industry in optoelectronic and telecom applications. Compound devices emit light, which isn’t easy in silicon, he points out.
Blue lasers made from GaN are enabling the next generation of DVDs, he says. When asked about the two competing standards for next-generation video disks, he concedes that there will be a battle similar to the legendary one between VHS and Beta videotape formats. But, he notes, both new standards use GaN-based lasers. For materials manufacturerers, “it’s like selling bullets in a war.”
Another material to watch is indium phosphide, which has long been used in optical components that transmit light through fiber-optic cable, Davis notes. Developments with InP slowed down for quite some time after the Internet bubble burst, but that lull may end soon. “I will just say cryptically there is the potential for new, innovative applications which have not yet hit the press. Those volumes could be very good for compound semiconductors.” Davis expects the new devices “in the next year or two.”
The industry has not put a high enough priority on developing standards, an area that could help its growth and perhaps stave off silicon’s encroachment. Bettina Weiss, SEMI’s director of international standards, will address the upcoming CSMantech conference on the benefits of standards for the compound semi sector.
“I was asked to draw some parallels between the early success in standards in silicon and other areas and the opportunities we could draw for compound semiconductors,” Weiss explains. Cooperation and dialogue between customers and suppliers can, among other benefits, provide opportunities to reduce manufacturing costs “by streamlining processes and interfaces and making it easier for a device manufacturer to choose vendor A, B, or C.” Weiss says she will emphasize the advantages of standardized platforms for tools, tool performance, safety, facilities, chemicals, and the like.
The compound semi industry has had problems because most of the “big buyers with wafers use the same standards for equipment and automation as silicon people,” says Brophy of TriQuint Semiconductor. With his company’s focus on GaAs wafers, “it hasn’t exactly been very easy to get anything done. For a long time I tried to participate in that [effort] but it didn’t seem to go anywhere.
“Everybody expects different things. There’s a certain amount of black art in this.” The previously mentioned difference of opinion over substrate properties is one example. Confusion over wafer specs also leads to problems related to laser marks and notch requests. “All these things come together to prevent us from unifying on a good set of standards,” he adds.
The industry continues to grow, however. The GaAs industry takes in approximately $2 billion to $2.5 billion in annual worldwide revenues, according to Tal of MAX IEG. He says semiconductor industry analysts find it hard to validate the revenue figures because “a major part of the industry is selling to military applications” with captive fabs and restricted information. Analysts typically derive the estimates by examining semiconductor sales, but they don’t break them out by gallium arsenide versus silicon.
“They break it down to different niches, and it’s hard to trace back,” notes Tal, whose company advises silicon- and compound semiconductor–based fabs. “Some products are strictly compound semiconductors, so it’s easy to know they contribute [to the revenue total].”
Tal says there are “probably 15 significant players in the industry. Out of them in the commercial world there are maybe five that are really significant, and in the military [world] I would guess double that number.”
Chase of Veeco revels in the fact that in his new process world “something that’s the whole front-end-of-line for semiconductors is done in one machine for an LED.” The company’s two main MOCVD systems are basically “a fab in one tool” that in five hours can produce a device.
Chase relates that the compound semi business comprised about 25% of the equipment supplier’s revenues in 2004 and about 15% in 2005. The executive says there’s one big reason he moved to the compound business over a year ago: “This one has a lot of growth.” He says that, depending on the source, the sector will expand at a compound annual growth rate of 15–20% through 2012.
“It’s growing like semi used to.”—JC
CSMantech will be held April 24–27 in Vancouver, BC, Canada. Registration and hotel information is available at www.gaasmantech.org.
Search | Current Issue | MicroArchives
Buyers Guide | Media Kit
Questions/comments about MICRO Magazine? E-mail us at email@example.com.
© 2007 Tom Cheyney
All rights reserved.