TUCKED away in the outskirts of Kofu, a small city in the heart of Japan鈥檚 grape-growing region about 100 kilometres west of Tokyo, is a microchip factory with the oddest clean room in the world. In most chip plants, you would expect to find the clean room to be a model of uncluttered order, of anonymous cabinets connected by automated transporters. But not this one. At the Kofu factory, the unsuspecting visitor climbs into a bunny suit, steps through an air shower and enters a high-tech shambles that looks more like a dairy than a semiconductor manufacturing plant.
At one end of the room are three enormous bronze-coloured stainless steel udders. Each one sprouts a thicket of pipes and tubes, and under the udders, plastic trays catch drips from ice that sheathes the joints between teats and tubes. On top of one udder, a tiny triangular orange flag twirls incongruously. It means: this machine is working.
But the resemblance to a dairy is superficial. The milking going on here is for profit, not gold-top. The 鈥渢eats鈥� are actually receptacles containing aluminium, gallium and arsenic; the 鈥渦dders鈥� are high-vacuum reaction chambers in which ultrathin layers of crystal are grown on circular wafers of gallium arsenide, 75 or 100 millimetres across. To be more precise, the steel udders are the largest molecular beam epitaxy machines in the world. And if you own a satellite television dish, the chances are that the amplifier on your antenna was made in one of these machines.
Advertisement
The clean room and factory belong to Fujitsu Quantum Devices, a subsidiary of Japan鈥檚 largest computer maker. Every month, it produces between 2 and 3 million high electron mobility transistors (HEMTs), or modulation-doped field effect transistors (MODFETs) as this Fujitsu invention is more generically known. The low-noise, high-gain characteristics of the HEMT have shrunk satellite dishes to diameters of 30 centimetres or less. The factory also makes quantum-well lasers for fibre-optic trunk lines and submarine cables, gallium arsenide integrated circuits for supercomputers, and microwave devices for mobile phones.
Fujitsu Quantum Devices, established in 1991, is already responsible for between 10 and 12 per cent of the parent company鈥檚 semiconductor sales, which last year totalled around $4.5 billion. While the bulk of the semiconductor industry鈥檚 output is made from the single element, silicon, Fujitsu Quantum Devices specialises in compound semiconductors, which are produced by mixing various elements, most commonly gallium, arsenic, aluminium, indium and phosphorus, and the company is by far the world鈥檚 largest producer of them.
Promised land
Commercial success in selling products such as HEMTs and quantum-well lasers to fast-growing markets like satellite broadcasting and optical communications is perhaps the most obvious reason why Japanese firms in general, and Fujitsu in particular, are so enthusiastic about quantum devices. An ambitious, across-the-board research effort is under way at Japanese companies, universities and government institutes. For the microelectronics industry, quantum country could be the promised land. Nobody knows the way there; they are not even sure if such a place exists. But if it does, the Japanese are taking the big risks necessary to get there first.
Three levels
Japan鈥檚 leading expert in the field, Hiroyuki Sakaki of Tokyo University, has defined three levels of quantum device. Fujitsu鈥檚 HEMTs and quantum-well lasers belong to the first level: they are essentially conventional semiconductor devices whose performance is enhanced by making the paths along which electrons and photons travel so small that their motion is governed by the rules of quantum theory rather than classical mechanics. For instance, quantum effects help to explain how electrons tunnel their way through thin insulating layers that are regarded as impenetrable barriers in classical semiconductor devices.
Sakaki鈥檚 second level consists of devices that directly exploit quantum mechanical effects such as this electron tunnelling. The most advanced of these are resonant tunnelling diodes and transistors, in which electrons whose energy matches the so-called resonant energy level of the device can tunnel paths at extremely high speeds. For almost a decade, researchers at Fujitsu have been trying to harness this speed for applications such as supercomputer switches.
The third level consists of devices in which the separation between the energy levels of electrons is very small, and where the quantum effects are so subtle that they disappear at temperatures above 10 kelvin. This category is the least understood, says Sakaki, and therefore most in need of study. But it remains to be seen whether devices in Sakaki鈥檚 second and third categories will ever become more than what one Japanese commentator calls 鈥渜uantum toys鈥� for physicists to play with. As Sakaki is the first to concede, 鈥渦sefulness is a new addition鈥� to researchers鈥� objectives.
This exotic blend of fundamental physics and advanced technology is what makes up research into quantum devices. It is a fusion that 鈥渞eally suits Japanese characteristics鈥�, says Leo Esaki, the Nobel prizewinning physicist often referred to as the godfather of quantum devices. For Japan has a research culture in which scientists and engineers are happy to work together, and where both pay an attention to detail that is, by Western standards, incredible.
Also incredible to Westerners who have witnessed it is the sheer scale of the research going on in Japan. At the Basic Research Laboratories of Nippon Telegraph and Telephone in Atsugi, a suburb on Tokyo鈥檚 southwestern fringe, some 100 researchers are working on various themes related to quantum devices. 鈥淎 lot of groups here are doing very good work,鈥� says Bruce Joyce, director of semiconductor materials at Imperial College in London, who last year spent several weeks working at the NTT laboratories.
Across the road from NTT Atsugi is Fujitsu鈥檚 Quantum Electron Devices Laboratory, which houses another 100 researchers. The jaws of Western visitors drop when they see the dozens of systems for molecular beam epitaxy (MBE) and metal-organic chemical vapour deposition (MOCVD). Indeed, informed sources estimate that there are probably more advanced materials fabrication systems in just these two corporate laboratories than in the whole of Europe. Japan鈥檚 other giants 鈥� Hitachi, Matsushita, the Mitsubishis (Chemical as well as Electric), NEC, Sony, Sumitomo Electric and Toshiba 鈥� are also working on quantum devices, as are a horde of smaller players.
鈥淭he level of activity is very impressive,鈥� said Michael Pepper, managing director of Toshiba鈥檚 Cambridge Research Centre, who was in Tokyo recently. 鈥淚t鈥檚 very good to see that companies here have such a positive attitude to high technology and long-term research. The fact that they think that such research is going to pay off is very positive.鈥� Pepper says that quantum devices are also the focus of work at Japanese laboratories based in Britain, at both Toshiba鈥檚 and Hitachi鈥檚 research centres in Cambridge.
Thinking small
Meanwhile, Japan鈥檚 corporate counterparts in the US and Europe have either scaled back their activities (Bell Labs) or left the field altogether (IBM, Bellcore, Philips). In Japan, research conferences on semiconductor devices and materials feature big sessions on techniques for the fabrication of extremely small structures. Such sessions are rare elsewhere, according to Seigo Tarucha, a group leader at NTT鈥檚 Atsugi labs. 鈥淚t鈥檚 a great pity that compound semiconductor research is not supported by [US and European] industry,鈥� he says.
Why should Japanese companies support such hairy research? The simple answer is that conventional semiconductor technology based on silicon is sooner or later going to run out of steam. At the International Electron Devices Conference in San Francisco in December, NEC announced the world鈥檚 first 1-gigabit dynamic random access memory (DRAM) chip, which it expects to start delivering in 2001. The smallest features on this chip will be just 0.18 micrometres across. This compares with 0.5 micrometres on the 16-megabit chips that are currently the industry鈥檚 standard product.
A simple extrapolation to subsequent generations of DRAMs shows that the width of circuit lines on 4-gigabit chips will be 0.15 micrometres and on 16-gigabit chips, 0.1 micrometres. But quantum effects become an issue below 0.15 micrometres, says Pepper, as electrons start to tunnel at random between the input and output sections of a memory cell鈥檚 transistor, which makes it increasingly difficult to determine whether the device is off or on 鈥� in other words, whether the device stores a one or a zero.
Future attraction
So about ten years from now, chip makers are likely to be casting around for an alternative device structure. They reason that if quantum effects like tunnelling are the problem, then quantum devices that exploit these effects could well be the solution. Hence the attraction for researchers of devices like the resonant tunnelling diode and its derivatives. The first resonant tunnelling diodes were built by Esaki at IBM in the mid-1970s, but it is work done a decade later by Jerry Sollner at MIT鈥檚 Lincoln Laboratory that has sparked the latest round of interest. Sollner demonstrated that practical devices could be built out of resonant tunnelling diodes that operated at very high frequencies of up to 700 gigahertz. The downside was that most quantum devices work only at cryogenic temperatures where vibrations in the crystal lattice, which scatter electrons, can be frozen out. Then, in 1985, Sakaki鈥檚 group at Tokyo University discovered that resonant tunnelling would work at room temperature, and hence was of potential commercial significance.
Naoki Yokoyama has been one of the most active of the investigators pursuing this approach. He manages a group of 35 researchers at Fujitsu鈥檚 Atsugi laboratories, and in 1985 demonstrated a resonant tunnelling hot electron transistor (RHET). Its electrons are described as 鈥渉ot鈥� because they have sufficient energy to tunnel through the device. Since then, his group has demonstrated several ways in which RHETs could be applied to configure the basic building blocks of computers.
In 1991, Fujitsu announced that it had built an adder circuit from seven RHETs, a quarter the number of elements required to build a conventional, silicon adder. As well as being smaller, the circuit functioned six times as fast as the conventional device, and consumed half the power. Within the past 18 months, Yokoyama鈥檚 group has unveiled a stripped-down version of its device configured in the form of computer memory cells as well as logic gates. Fujitsu claims that memory chips made using this technology might require just one-tenth as many components as their conventional counterparts.
Thumbnail chips
鈥淚t鈥檚 becoming very important to decrease the size of the chips,鈥� Yokoyama says, because the bigger the chip the more expensive and difficult it is to make. The 1-gigabit memory chip announced by NEC in December covers 9 square centimetres, which is considerably larger than the thumbnail chips that have been the norm until now. The planned 4-gigabit memory chips will be a huge 25 square centimetres, about as big as an after dinner mint. 鈥淚f we can reduce the size of [memories] by one order of magnitude,鈥� Yokoyama says, 鈥渋t could have a great impact.鈥�
When Yokoyama began his work on quantum devices, his aim was to build a 1-gigabit memory chip 鈥� but now NEC has revealed that it can build such devices in silicon. 鈥淲e had to change the target to compete with silicon,鈥� Yokoyama explains. Now Fujitsu is shooting for 10 gigabits by the turn of the century. Yokoyama says his ultimate goal is to fabricate a 1-terabit RHET memory, storing 1000 times as much information as NEC鈥檚 record breaker. For such a device, the outputs would be made up of an array of quantum dots, confining electrons to individual spots just a few nanometres across.
So far, Fujitsu has not got any further than demonstrating its ideas about quantum devices and their potential usefulness. The big question is whether such devices are a practical proposition 鈥� whether so much information can be packed and tapped effectively in such a tiny space. There are significant obstacles to be overcome, most notably the need to keep the hard-working devices cool (though the device works at room temperature, its hot electrons need to be chilled). But Yokoyama, is not put off: 鈥淭he important thing is to explore, to look for new physics and some applications.鈥� And what do Fujitsu鈥檚 managers feel about the value of such basic research? 鈥淚 think they believe us, that something useful will be created from our lab,鈥� Yokoyama says.
Fujitsu鈥檚 research is one of six corporate projects funded by Japan鈥檚 Ministry of International Trade and Industry under its $50 million Quantum Functional Devices programme. The programme started in 1991, and is scheduled to run to the year 2000. Fujitsu is the company that has made most headway so far. The other five receiving support are the computer chip makers NEC and Hitachi; the consumer electronics companies Sony and Matsushita, which qualify on the basis of their work making HEMTs and semiconductor lasers; and an American firm, Motorola, which is investigating quantum devices at its corporate research laboratories in Phoenix, Arizona. MITI has also established a second, much larger project, the Angstrom Technology Partnership, which is concerned with the fabrication of nanoscale structures. Funding for it will be $250 million over the ten years from 1992 to 2001. Japan鈥檚 two other science funding agencies also provide generous support for several projects of their own.
Japan is investing so heavily in quantum related research that academics have been heard to complain that too much money is available. 鈥淧sychologically, people feel obliged to work just because the money is there,鈥� says Hiroyuki Sakaki of Tokyo University. 鈥淚t is not a bad idea for MITI to pick up a hot topic [like quantum devices] to see whether the stuff is real or not,鈥� says Sakaki, but he feels that universities ought to be looking further ahead. Other academics suggest that much of the money will be wasted. 鈥淵ou don鈥檛 have to think as hard as you do if you don鈥檛 have as much money,鈥� observes Bruce Joyce of Imperial College.
Whether the quantum device 鈥渟tuff鈥� will turn out to be real or not remains highly controversial. Critics point to the well-documented tendency of Japanese companies to flog dead horses. More than ten years after IBM decided that Josephson junctions were not going to be commercially viable, Hitachi and NEC are still trying to turn these cryogenic superconducting devices into practical switches for high-speed computers. And Fujitsu only recently pulled out of the field.
Short odds
All the same, fortune favours the well prepared, as the story of Takashi Mimura鈥檚 invention of the HEMT at Fujitsu in 1979 demonstrates (see 鈥淨uantum leaps鈥�). Had Mimura not been allowed to spend most of his career trying 鈥� and failing 鈥� to develop gallium arsenide transistors, and had Fujitsu not invested in Japan鈥檚 first MBE machine several years earlier and built up expertise in using it, the HEMT might still not have appeared.
No one can say for sure which, if any, of the current crop of quantum device proposals will prove to be a winner. Pepper reckons that no more than 15 to 20 per cent of research in semiconductor physics leads to any sort of product. But given Japan鈥檚 current level of effort, it鈥檚 a safe bet that sooner or later something useful will pop out. And when it does, Fujitsu will have a factory to exploit it.
Quantum leaps
MAX PLANCK, the formulator of quantum theory, died the year that the transistor was born at Bell Laboratories in the US. That same year, 1947, a young Japanese physicist called Leo Esaki graduated from the University of Tokyo and went on to do more than anyone else to connect Planck鈥檚 theory to the real world of semiconductors.
In 1957, Esaki invented the first quantum device, the tunnel diode, at Sony while trying to work out why the transistors the company was producing under licence were letting current flow in the 鈥渨rong鈥� direction. A dozen years later, he dreamt up the superlattice, or quantum heterostructure, again to take advantage of quantum effects as a way of controlling high-speed electrons. This proposal opened up a whole new field of what Esaki calls 鈥渄o-it-yourself鈥� quantum mechanics, and it has become the dominant theme in semi-conductor physics.
Esaki, who has been president of Tsukuba University since 1992, did his pioneering work on electron tunnelling at Sony. He had begun this line of research at another company, Kobe Kogyo, where he spent the first eight years of his career. It was Kobe Kogyo, not Sony, that was the first Japanese company to license the transistor. The company was later acquired by the computer maker Fujitsu.
Esaki鈥檚 invention of the tunnel diode, which became known as the Esaki diode, caused a sensation in 1958 when it was announced at a conference in Brussels. In his keynote speech, Bill Shockley, co-inventor of the transistor and winner of the 1956 Nobel Prize for Physics, singled out Esaki for praise. Quantum physicists fell in love with tunnelling because it was a fundamental effect 鈥� one that could not be explained in terms of classical mechanics and that validated the theoretical predictions physicists had made thirty years earlier. And because electrons tunnel about five times as fast as they normally travel in semiconductors, computer designers, too, got excited about the Esaki diode.
This speed was not much use to a consumer electronics manufacturer like Sony, though the company did build an Esaki diode radio. (At least one of them survives, according to Esaki鈥檚 wife, who says she listens to it in the kitchen.) But American computer makers, always on the lookout for ways to build faster machines, were eager to try to exploit the diode. In 1960, Esaki accepted an invitation to join the biggest of them all, IBM, at the corporation鈥檚 newly established Yorktown heights research centre, north of New York City.
As a practical device, the tunnel diode turned out to be a disappointment. It would work only over a very limited voltage range, which restricted its use, and was unsuitable for integrated circuits because it had to be so heavily doped that it was difficult to make. The device鈥檚 lasting importance has been the inspiration it gave. In Britain, at the University of Cambridge, Brian Josephson adapted tunnel diodes to superconducting materials, and went on to share the 1973 Nobel Prize for Physics with Esaki. In the US, at General Electric鈥檚 laboratory in Syracuse, New York, Nick Holonyak made tunnel diodes out of gallium arsenide, in place of the germanium Esaki had used. Holonyak鈥檚 work led to the light-emitting diode, and sparked off a trail of discoveries that ended in 1962 with his GE colleague Bob Hall producing the first semiconductor laser.
Hall鈥檚 laser was a simple homojunction 鈥� a diode made from a pair of oppositely doped (positive or negative) layers of the semiconductor gallium arsenide. It worked only at cryogenic temperatures, and even then could only generate brief spurts of light before it went phut. It was not until 1970 that two researchers at Bell Laboratories, the Japanese physicist Izuo Hayashi, and Morton Panish, an American chemist, came up with the first room-temperature, continuous-wave laser. Their trick was to make the device from an active, light-producing layer of gallium arsenide sandwiched between two layers of aluminium gallium arsenide. This arrangement, known as a double heterostructure, kept the optical wave confined to the active layer.
Meanwhile, Esaki had been wondering how to exploit the principles of quantum mechanics to design materials with desirable electronic properties. The effects that Esaki had in mind would emerge only when the electrons were restricted to travelling in one or two dimensions, which in turn would mean making devices out of very thin layers 鈥� less than 100 angstroms (10 nanometres), compared with 5000 angstroms that was the thinnest Hayashi and Panish could manage.
Then Esaki heard about a newly developed technique for growing crystals, a technique that looked capable of producing such ultra-thin layers. The technique, developed during the mid-1960s at Bell Labs by John Arthur and later Al Cho, was called molecular beam epitaxy (MBE). In 1969, Esaki and his colleague Raphael Tsu wrote a paper proposing 鈥渟uperlattices鈥� 鈥� crystalline 鈥渃lub sandwiches鈥� made up of different, and differently doped, semiconducting compounds. The paper was rejected by Physical Review on the grounds that it was 鈥渢oo speculative鈥� and involved 鈥渘o new physics鈥�. As Esaki now gleefully recounts, a shortened version subsequently appeared in the IBM Journal of Research & Development, and went on to become a citation classic. At the International Conference on the Physics of Semiconductors at Beijing in 1992, just under half the papers presented were on superlattices.
MBE provided the means to implement Esaki鈥檚 ideas, but the only institutions then able to afford to install such expensive, ultrahigh-vacuum equipment were IBM Yorktown Heights and AT&T Bell Laboratories at Murray Hill, New Jersey. At IBM in 1974, Esaki, Tsu and Leroy Chang used MBE to build a quantum device called a resonant tunneling diode. The key feature of this device is a quantum well, an energy barrier formed by sandwiching a layer of gallium arsenide between two very thin layers of aluminium gallium arsenide. Normally, electrons do not have enough energy to escape from the well and are trapped there. But applying a voltage at what is known as the resonant energy level causes the confined electrons to tunnel through the barrier in high-speed pulses. This effect can be exploited to create oscillators and switches that operate far faster than conventional devices.
At Bell Labs, Ray Dingle and Art Gossard used MBE to explore superlattice physics. They were joined by Russell Dupuis, who had used an alternative growing technique called metal-organic chemical vapour deposition (MOCVD) to grow the first quantum-well lasers. By 1979, Won-Tien Tsang at Bell Labs had demonstrated that MBE, too, could be used to grow quantum-well lasers, and announced his discovery to the world in a stack of papers. Meanwhile, a young postdoc in Dingle鈥檚 group called Horst Stormer had come up with modulation doping, a technique that persuaded electrons to drop from a doped layer into a thin, adjacent undoped layer. There they formed a two-dimensional electron gas that could whiz along the crystal lattice without being scattered by dopant atoms, and hence slowed down, as would normally happen.
Surprisingly, AT&T did not capitalise on much of the research being done at its laboratories. It was mostly Japanese firms that reaped the benefit, when they developed semiconductor lasers for compact disc players. The two dominant producers of CD lasers are Sony, which built on Dupuis鈥檚 work with MOCVD, and a small Kyoto-based company called Rohm, which was inspired by Tsang鈥檚 experiments with MBE. Millions of these devices are now produced every month.
The work on modulation doping was taken up in 1979 by Takashi Mimura, a young physicist in a group of creative researchers that Fujitsu had inherited when it acquired Kobe Kogyo. Mimura invented the high electron mobility transistor (HEMT) that Fujitsu now makes by the million every month for satellite TV and microwave communications systems. But in some respects, HEMTs have been a disappointment, too. They have not, as Fujitsu had hoped, become the technology for the next generation of supercomputers and mainframes. For this application, silicon has made unexpectedly large strides.
In Japan, original contributions to quantum device research have come most notably from Hiroyuki Sakaki, a physicist at the University of Tokyo. In the late 1970s, at Esaki鈥檚 invitation, Sakaki spent 18 months working at Yorktown Heights as a visiting scientist. While he was there he proposed the study of how electrons move in superlattices 鈥� not across the layers, which pre-occupied Esaki in his work on resonant tunnelling, but parallel to them. The proposal was applied by Dingle鈥檚 group at Bell Laboratories in the form of modulation doping. Back in Japan, Sakaki came up with the idea of one-dimensional structures called quantum wires. In an HEMT, electrons are allowed to flow in two dimensions; in a quantum wire, they are further constrained, so that they can only flow in one dimension.
Sakaki鈥檚 former student Yasuhiko Arakawa, of Tokyo University鈥檚 Institute of Industrial Science, continues to set the pace in the field. He has recently grown 鈥渮ero-dimensional鈥� devices, or quantum dots, which are just 7 nanometres across. They are, he claims, the world鈥檚 smallest.