At first glance, the Itami Works complex of Sumitomo Electric Industries Ltd. seems like any other generic Japanese manufacturing plant in this tidy little industrial suburb of Osaka. A cement wall, disguised by a row of shrubbery, encloses a sprawling complex of gray or tan buildings, some squat, some several stories tall. But take a closer look, and you might notice that the guards out front seem awfully vigilant, keeping close tabs on the comings and goings.

And for good reason. In this factory, hidden among the extruding machines, presses, and ovens churning out wires, brakes, and synthetic diamonds, Sumitomo has just quietly launched what will turn out to be one of the great semiconductor advances of the 21st century: the mass production of thin, crystal-clear disks of pure gallium nitride.

These disks, more properly known as substrates or wafers, are the foundation on which technicians fabricate chips or devices. And ones made of gallium nitride are key to long-lived and powerful laser diodes that emit blue or ultraviolet light. The lasers, in turn, will be the cornerstone of major new industrial and consumer electronics goods, notably next-generation DVD player-recorders and optical data-storage systems for computers. The reason: thanks to its short wavelength, blue light can write huge amounts of data on a DVD—up to 27 GB on a standard-sized disk, almost six times the storage capacity possible with ordinary red laser diodes. It's enough to store more than two hours of high-definition video or 13 hours of standard-definition video.

Blue semiconductor lasers have been available for several years, but until Sumitomo's breakthrough, they had to be built on substrates of sapphire or silicon carbide. Gallium nitride lasers deposited on those substrates are plagued by poor reliability, low production yield, and low output power. Semiconductor specialists have long known that the devices would work much better on gallium nitride substrates, but until now, no one could figure out how to make them. Kensaku Motoki, manager of the Advanced Material Group for Sumitomo Electric Industries, a diverse, multibillion-dollar company involved in everything from high-power electrical cables to a host of exotic semiconductors, has pulled it off. Here, during an interview at the Itami Works, he peers through the evidence: a transparent, colorless disk 50 mm in diameter [see photo, "Wafer Scale"]. To a semiconductor specialist, it's a little miracle, and with manufacturing just starting and R and D still ongoing, it has a price to match: US $10 000. For comparison, a silicon wafer with a diameter six times as large costs just $200. Roughly 1000 diodes can be produced on one 50-mm gallium nitride wafer, Motoki notes, but those devices are still so costly that you won't be seeing blue-laser DVD systems in your local Wal-Mart any time soon.

Nevertheless, the high prices aren't deterring consumer electronics companies from offering the first DVD recorders, with the blue laser diodes, albeit with price tags high enough to limit sales mainly to the well-to-do. Last spring, Tokyo-based Sony Corp. was the first consumer electronics company to begin offering next-generation DVD recorders, for roughly $3800. Sony leads a consortium called Blu-ray Disc, which is pushing one of two competing standards for the design of the discs, players, and recorders that use blue lasers. The group includes 10 major consumer electronics companies, among them Hitachi, Matsushita, Royal Philips Electronics, and Samsung. The other standard, Advanced Optical Disc (AOD), has been proposed by Toshiba and NEC, but at press time no AOD-based products were on the market.

Though the blue laser systems are pricey now, their costs will come down as their manufacturers gain experience, and it's expected that the market for blue laser diodes will climb, as will sales of blue light-emitting diodes (LEDs) used for solid-state lighting. Sales of the two are expected to reach more than $4.7 billion by 2007, according to Strategies Unlimited, in Mountain View, Calif.

Such projected growth is driving a whirlwind of research, much of it in Japan, into the development of blue laser diodes and gallium nitride substrates. And a hint of the fierceness of the competition surfaces during an interview at Sumitomo, when Motoki, pressed by this reporter to be allowed to tour the secret gallium nitride laboratory, offers a compromise. She can peer in a window, but must promise not to write about what she sees in there.

Infra-red and red laser diodes, such as those in today's CD and DVD systems, have been around for decades; blue ones have been around for less than 10 years and in commercial quantities for only about five years. Shuji Nakamura (M), now a professor of materials at the University of California, Santa Barbara, developed a blue gallium nitride LED in 1993 at Nichia Chemical Industries Ltd. (now Nichia Corp.) in Tokushima, Japan, and, two years later, the first gallium nitride laser diode.

In an LED's semiconductor layer, electrons recombine with holes—places in a semiconductor's valence band that are missing an electron—and give up energy as a photon. The color of the light is characteristic of the semiconductor material, and it is determined by the difference in energy between the material's valence and conduction bands, two bands that can be occupied by electrons in the substance. Gallium nitride is called a "wide bandgap" material because the difference between the valence and conduction bands is great—so great, in fact, that photons of visible light just slip right through the material without being absorbed. That's why substrates of gallium nitride are transparent.

A laser diode also works by recombining electrons and holes. But one difference between a laser diode and an LED is that in the laser, electrons and holes need a nudge from a passing photon to recombine. As the photons bounce back and forth between mirrored surfaces at each end of the laser, they nudge more and more electrons and holes to recombine, creating more and more photons, all with the same phase, direction, and polarity—the characteristic of a laser. One end of the laser is designed to let out a narrow beam of the photons, while keeping enough of them inside to sustain the lasing action.

With no substrates of pure gallium nitride available, Nakamura built his devices on a thin layer of gallium nitride deposited on a sapphire substrate. The thin layer was a single-crystal, meaning that the atoms occupy positions that form a regular pattern. Today, almost all gallium nitride LEDs and lasers are still made in much the same way, at places like Cree Semiconductor and Nichia.

Winner: Gallium Nitride

Goal: Build high quality gallium nitride wafers as substrates for blue and ultraviolet laser diodes

Why It's a Winner: The wafers will be the foundation for the next generation of DVD recorders

Organization: Sumitomo Electric Industries

Center of Activity: Itami Works, Itami, Japan

Number of People on the Project: 50

Budget: US $40 000 000

But devices built with sapphire substrates are plagued with a type of defect called a dislocation. Gallium nitride and sapphire have different crystal structures: the distance between the atoms in the sapphire crystal is 16 percent greater than that in gallium nitride. So when you form one crystal on top of the other, you get a mismatch where the two meet, a boundary area where the regular pattern of atoms suddenly shifts. It creates stress in the gallium nitride crystal that causes the atoms in the gallium nitride to misalign, producing the dislocations.

Like geological fault lines, these dislocations reach up into the gallium nitride, producing flaws that rob any devices built in the gallium nitride of performance and lifetime, according to Masami Tatsumi, Sumitomo's general manager of advanced materials R and D laboratories. Laser diodes built on a layer of gallium nitride that has been grown directly on a sapphire substrate can have dislocation densities of 10⁸/cm² to 10⁹/cm² and lifetimes of less than 100 hours. "That's not good enough for DVD players," he notes.

Dislocations are a grave concern for laser diodes because electrons collide with them, causing the electrons to recombine with holes without creating photons, thus destroying the lasing action.

Sapphire substrates have other problems as well. For one thing, they're difficult to split into pieces—a process called cleaving—having atomically smooth surfaces. That makes it hard to produce those mirrorlike surfaces on each end of the gallium nitride that reflect (or transmit) the photons. Lack of smooth surfaces reduces the amount of light coming out of the laser. Gallium nitride crystals, on the other hand, cleave smoothly, making it much easier to create smooth surfaces.

Another difficulty is that sapphire is an insulator. That's a problem because, after all, the laser devices made on it are diodes. They must have electrical connections on each side of the active region where photons are produced. But the fact that the substrate side of the device is an insulator forces device designers into some fairly unattractive configurations. To begin with, the designers have no choice but to put both electrical contacts next to each other on the side opposite the substrate. And with both of the laser's electrical contacts on the top of the device, the electrons are constricted to flow across a thin layer on the top surface, which has a high resistance, Tatsumi says [see illustration, "Making Light"].

The higher resistance means that more input power is needed to pump light out of the device, and higher power reduces its lifetime. Also, since both electrodes are side by side, the device is twice as large as it would be if one electrode could be placed on the back side of the substrate. Gallium nitride, in contrast, is a semiconductor, so one electrode can be placed on the bottom of the substrate below the laser, eliminating all those power problems.

With all the advantages of gallium nitride substrates over sapphire, why aren't they everywhere already? The reason is that the standard technique used to make single-crystal substrates of gallium arsenide or silicon simply doesn't work for gallium nitride.

For example, Motoki explains, single crystals of gallium arsenide are made by putting ordinary, polycrystalline pieces of the material in a glass vessel, called a boat, and melting a circular zone of the material by heating it in a resistance furnace. Moving the heater from one end of the boat to the other also moves the circular melt zone. As the material is left to cool, it forms into a single crystal. The cylinder that results, called a boule, is then cut into wafers, salami style, and polished. The technique can't be used for gallium nitride, Motoki explains, because the nitrogen evaporates out of the crystal as it grows, so the gallium and nitrogen atoms don't bond. To keep the nitrogen in, you'd need pressures of tens of thousands of atmospheres, which are difficult to realize in a commercial process.

In 1995, when Motoki first began to look into producing substrates for blue light-emitting devices, he realized that so many patents had been issued for manufacturing gallium nitride on sapphire that he had better look for a different angle. "I wondered what approach would be completely original and what market would be possible without encroaching on the patents of other companies," he told IEEE Spectrum. He decided to try to produce stand-alone gallium nitride substrates, an area where there were few, if any, patents, and where Sumitomo could be "on top of the world."

Basically, he puts gallium nitride on substrates of gallium arsenide, and then gets rid of the gallium arsenide to leave stand-alone wafers of gallium nitride. The trick, as always, is minimizing the dislocations between the two materials so that the final gallium nitride substrates are as defect-free as possible.

Among gallium arsenide's advantages is the fact that substrates of the material are readily available because they're used to fabricate high-speed circuits for cellphones. Sumitomo itself manufactures and sells them. Drawbacks include the fact that it melts at 1238 ºC; growing gallium nitride on top of gallium arsenide requires a temperature of more than 1000 ºC. At that temperature, so close to gallium arsenide's melting point, the material is very soft and reacts with the ammonia gas that supplies the nitrogen needed to form gallium nitride. "We had to prevent the reaction of ammonia and gallium arsenide," says Motoki. "I can't tell you how we did it, but we put in a lot of effort to establish the process."

To put the thick layer of gallium nitride on the gallium arsenide substrate, he uses a standard chip-making technique called hydride vapor-phase epitaxy. Then he dissolves and grinds away the gallium arsenide, and finally polishes both sides of the remaining gallium nitride layer to a clear shine. It's the same basic process he used in 2000 to produce the first sizable stand-alone single-crystal gallium nitride substrate.

But Motoki's work was far from over, because, just as with gallium nitride and sapphire, the atomic spacings in gallium arsenide and gallium nitride differ, and that, again, leads to dislocations. So Motoki developed a technique for producing regions with very low dislocation density by forcing the dislocations into a small area, leaving regions of low dislocation density elsewhere.

Here's how it works. Gallium nitride crystals grow in the shape of hexagons [see photo "Saying No to Disclocations"]. At first, each hexagon is separate. But as growth continues, the hexagons begin to merge, creating pits in the regions where the bases of the hexagons come together. The dislocations formed on the side surface of the pit propagate toward the center of the pit as the material grows. The upshot is that in the part of the crystal near the pits, the dislocation density is very high, while in the rest of the crystal, it's very low.

Using this technique, Sumitomo researchers reduced the number of dislocations in the low-density regions to fewer than 10⁶/cm². But the areas of high dislocation density were positioned randomly and it was difficult to make low-defect gallium nitride lasers, which are several hundred micrometers long. So over the next two years, they developed a method for positioning the location of the pits to leave larger areas relatively dislocation-free. By 2002, they were able to produce areas over 100 µm wide and far more than 500 µm long with dislocation densities of less than 2 x 10⁵/cm²—an area big enough for a laser diode. Defect densities of gallium nitride on sapphire substrates have come down to about 5 x 10⁶/cm².

At the 2002 International Conference on Solid State Devices and Materials, held in September in Nagoya, Japan, Sony researchers were able to report that, using one of Sumitomo's low-dislocation-density substrates, they had built and tested a gallium nitride blue-violet laser diode with an expected lifetime of over 100 000 hours. In the conference's extended abstracts, the researchers reported that the area of low dislocation density extended for more than 150 µm and the minimum density was 2.8 x 10⁵/cm². "This number," the Sony team wrote, "implies the presence of only 3 to 9 dislocations for a laser...with an area of 10^-5 cm²....Therefore, it is possible that a number of laser[s]...will contain no dislocations at all considering the 1.5 µm width of these [laser diodes]."

It's good, but not yet good enough. The researchers tested the laser at 30 mW in an ambient temperature of 60° C. That output power is required for recording on laser discs. But at that power level, it will take more than an hour and a half to fill up a laser disc with data. Increasing the recording speed requires higher output power levels, again, shortening the laser diode's lifetime. Therefore, to maintain acceptable lifetimes at faster writing speeds, the number of dislocations must come down still further.

Motoki and his team are confident that by improving the control of the manufacturing process, they will be able to reduce dislocations even more. Meanwhile, Sumitomo is shipping substrates to many customers developing blue-violet lasers. The company is now ramping up production with a goal of 500 substrates per month.

Masamichi Yokogawa, Sumitomo's general manager of the Epi-Solution Division, anticipates that demand for blue-violet lasers for next-generation DVDs will begin to take off as early as 2005. Eventually, demand will reach levels comparable to those for the red and infrared gallium arsenide laser diodes used in today's DVD and CD systems—that's about 30 million laser diodes per month for red and 100 million for infrared, and a combined market, in 2002, of $1.4 billion, according to Strategies Unlimited.

He expects that the market for industrial applications will begin to grow even sooner—as early as this year. Such applications could include medical instruments, printers, and maybe even light sources for semiconductor lithography tools that now use mercury lamps.

His company has taken a two-pronged approach to furthering the technology, building the manufacturing infrastructure at the same time as it pushes ahead with research. As a result, there's a good deal of mixing of research and development and manufacturing.

On the R and D side, engineers are developing the techniques for growing the crystals and then polishing the wafers. "This is a very special material, unlike gallium arsenide or indium phosphide. It requires a special polishing method," Yokogawa says.

On the manufacturing side, developing the special techniques and machines is difficult, because no company has ever done what Sumitomo is now doing. In manufacturing, the most important goals are to improve the reproducibility and the production yield of the substrates and to reduce the manufacturing cost. Yokogawa hopes engineers can cut the price of the substrates by a factor of 5 to 10 within two years.

That would bring them more or less in line with the price of gallium nitride-coated sapphire wafers, which are sold by Cree, Nichia, Toyoda-Gosei, and others, according to Deepa Doraiswamy, research analyst for Frost and Sullivan. At press time, prices for the wafers were not available.

With such a potentially lucrative market, it's certain that other companies are not standing still, either. Fumio Orito, deputy general manager of the Science and Technology Office of Mitsubishi Chemical Corp., in Tokyo, told Spectrum that Mitsubishi is working with universities and other companies to develop alternatives to sapphire as a substrate material for gallium nitride. It also wants to develop bulk growth techniques, which could be cheaper than the Sumitomo technique and result in fewer dislocations because there is no sapphire or gallium arsenide substrate.

Mitsubishi's approach to making single-crystal gallium nitride is a modification of the method used to make zinc oxide substrates. To make the gallium nitride crystals, gallium metal and gallium nitride powder are combined with ammonia and ammonium chloride at moderately high temperatures of from 300 to 500 ºC and pressures of 200 to 500 megaPascals (about 2000­5000 atmospheres). The work is still preliminary. But so far, the researchers have succeeded in producing only small shards of gallium nitride a couple of millimeters across and a centimeter or so long.

At the GE Global Research Center in Niskayuna, N.Y., researchers are pursuing a technique similar to Mitsubishi's, but with even higher temperatures and pressures: 600 to 1000 ºC and 500 to 2000 megaPascals (5000­20 000 atmospheres). They have been able to produce single-crystal gallium nitride substrates 1­2 mm thick with an area of 15 by 18 mm². Researchers have made considerable progress, says Mark P. D'Evelyn, manager of GE Global's Ceramic Processing Laboratory, and have measured defect densities less than 200/cm². But much more work remains before the product is ready to be commercialized.

Meanwhile, ATMI Inc., in Danbury, Conn., is using an approach similar to Sumitomo's, but with a starting substrate of sapphire instead of gallium arsenide. ATMI scientists grow the desired thickness of gallium nitride, then remove the sapphire to leave a free-standing layer of single-crystal gallium nitride. The company has begun to sell engineering quantities of 50-mm wafers with uniform dislocation densities of 10⁶/cm². An advantage of the uniformity, says Allan Salant, the company's manager of development engineering, is that laser diodes built on its wafers are not restricted in device size or location.

In total, the number of companies and research institutions developing gallium nitride substrates has climbed to more than 20. And the rate at which the technology is advancing is testimony to the importance of this material. At this time, there is little doubt that we will have commercial quantities of low-dislocation-density gallium nitride substrates in the near future. The questions are only by what methods and from whom.