Reinventing the Transistor
Hewlett-Packard is betting that it can build computers whose functionality rests on the workings of individual molecules. It’s blue-sky research, but if it works, it will push computing far beyond the limits of silicon.
Every Friday afternoon at Hewlett-Packard Labs in Palo Alto, CA, R. Stanley Williams, one of the most respected thinkers in the field of molecular electronics, gets his group of 25 research scientists together to talk shop. One by one, they make their way to the conference room. Williams walks in exactly on time, sits down in front, and leans back, frowning, his hands steepled. He was hired by HP in 1995 to rethink the basics of computing and has handpicked the team inside this room to do just that. Williams likes to wear jeans, and his hair reaches halfway down his back, so he gives a first, fleeting impression of quietude and informality. But he apparently never smiles, and his people work 19-hour days to meet his deadlines. Williams waits a few minutes for the habitual latecomers, then stands up. He speaks in an efficient monotone.
“We’re going to hear first from Gun-Young today,” he says. “What he has accomplished is magnificent. Everyone here owes him a lunch because his hard work has paid for our salaries for the last several months.”
Gun-Young Jung, a recent postdoc from South Korea, stands up and quietly describes his work on nano imprint lithography, a process that uses a physical mold to create features as small as six nanometers across on silicon wafers. That’s more than an order of magnitude smaller than the finest features achievable using today’s advanced photo-lithographic processes. Sometimes things stick to the mold, though. It’s like cake batter sticking to a pan, he says. His presentation lasts about ten minutes and is followed by two others.
Listening to these speakers, one after another, gradually conveys a sense of the group’s style. They enjoy self-deprecating humor and inject frequent expressions of bewilderment into their scientific explanations, like “I don’t know” and “it’s still a mystery” and “I still need to investigate,” and even “I am still quite a novice.” And despite their obvious expertise, this isn’t false modesty.
Williams’s group faces a monumental task: trying to make computers whose functionality rests on the workings of molecules. To do so will mean reinventing the transistor. While silicon and other inorganic semiconductors have always been the basic building blocks of microchips, it turns out that organic molecules can also have some potentially useful electrical properties. Indeed, over the last few years, researchers have learned to synthesize molecules that can function as electronic switches, holding binary 1s or 0s in memory or taking part in logical operations. And molecules have one significant advantage: they are really small.
Such work is critical to the future of computing, because conventional chip fabrication technology is on a collision course with economics. Today’s best computer chips have silicon features as small as 90 nanometers. But the smaller the features, the more expensive the optical equipment needed to manufacture them. A state-of-the-art fabrication plant for silicon microchips now costs some $3 billion to build. A chip in which silicon transistors are replaced with molecular devices, on the other hand, could in principle be fabricated through a simple chemical process as inexpensive as making photographic film. A circuit with 10 billion switches could eventually fit on a grain of salt; that’s a thousand times the density of the transistors in today’s best computers. A computer built from such circuits could search billions of documents or thousands of hours of video in seconds, conduct highly accurate simulations and predictions of weather and other physical phenomena, and do a much better job of imitating human intelligence, perhaps even communicating with us through natural conversation.
But no matter how tempting in theory, it’s speculative, blue-sky research, and investing in molecular electronics is a gamble few companies have been willing to make. HP’s confidence in Williams is a big reason it’s one of the exceptions, says Shane Robison, the company’s executive vice president and chief strategy and technology officer. “In addition to his ability to put together a first-class team of cross-disciplinary experts and an emphasis on how to turn science and technology into real products, Stan’s best quality is probably his eternal optimism,” says Robison. Of course, there’s also the lure of immense profits, should Williams’s technology ever displace conventional silicon chips. “Projects this ambitious are always a long shot, but we wouldn’t be doing it if we didn’t think there was a good chance of succeeding,” Robison says.
To be sure, the company has hedged its bet by being cautious with funding. Williams’s group has a four-year, $12.5 million grant from the U.S. Defense Advanced Research Projects Agency (DARPA), and HP provides matching funds, but about half of the DARPA funding goes to university research partners. Signs of economizing are everywhere in the lab, from a shortage of supplies in the coffee room to jury-rigged equipment. Nonetheless, the group has made one breakthrough after another-most notably, by proving that a “crossbar” design once common in conventional electronics can be resurrected on the molecular scale. In a demonstration last year, the group trapped molecules in the junctions between titanium and platinum nanowires arranged in an eight-by-eight, one-micrometer-square grid, and showed that the molecules can be switched “on” and “off” at specific junctions-a first step in building a working memory or logic device.
Spend some time in Williams’s lab and you start to understand why a lot about molecular electronics is still a mystery, beginning with the relatively simple question of what exactly the researchers are building. Yong Chen, a native of China and a member of Williams’s group since 1998, spends a lot of his time sitting in a stuffy, windowless, nine-square-meter room padded with thick foam. It’s the home of a delicate electron microscope, which uses electron beams to create a rough picture of the structures Chen creates in the laboratory down the hall.
Chen is the leader of the team that has given the group its biggest public success to date, the 64-bit crossbar memory. His team first imprinted eight parallel nanowires made of titanium and platinum on a silicon substrate, and covered these wires with a one-molecule-thick layer of a synthetic chemical called rotaxane. They then deposited a second set of titanium wires perpendicular to the first, creating the possibility of an electrical connection between the wires at any junction in the grid.
Each molecule of rotaxane-which was invented by chemist Fraser Stoddart at the University of California, Los Angeles-consists of a long axle with two lumps of atoms at each end, and a ring of atoms circling the axle. Stoddart and Williams’s groups theorize that when a voltage is applied through a specific, intersecting pair of nanowires, the rings on the rotaxane molecules between the wires “jump” from one end of the axle to the other and stay there until another voltage is applied. This could raise or lower the molecules’ resistance to electrical current, and these two states of conductivity would represent digital 1s or 0s. Now Chen, eager to see how small he can make such a device, is trying to print the individual wires even closer together. It’s painstaking labor, where you never know if you’re making progress until the moment it works.
Today Chen is open mouthed, rapt, focusing absolute attention on the monitor in front of him, while also trying to carry on a conversation. He is not entirely successful. Several minutes pass quietly as a question hangs in the air, unanswered. He increases the microscope’s magnification as he searches through a series of fuzzy, gray-on-gray images that look like satellite photos of a desert.
“After we finish the fabrication process, we come in here to check what kind of thing we have got,” he says. “I want to see if the wire is grounded to the substrate or suspended above it. There’s one. Oops, I lost it.”
Eventually he finds something that looks like a length of rebar on a pile of charcoal dust but is actually a wire, 35 nanometers in width, resting on the silicon base. He takes a picture, silent again, holding his breath since sound waves will affect the quality of the image.
“We can talk now,” he says. “Here, in fact, you can see this wire is broken. Too bad. This is a routine experiment, frankly.” Chen’s goal is to find a combination of materials-a “recipe,” if you will-that will impart a Teflon-like non-stickiness to the mold that deposits the wires on the substrate; otherwise they bulge and twist when the mold is removed. But sitting in this hushed, foam-covered room, watching one of the leading scientists in the field searching through grainy images, you realize just how difficult it is to work on this scale. Three weeks later, after five months of painstaking experiment and observations, Chen and Gun-Young Jung find the result they were looking for, bringing the possibility of molecular-sized circuits a small step closer.
“I miscalculated several things,” Chen says simply.
Now he can move on to the next problem.
Observing results, of course, is the last step in a train of events that traditionally begins with a theory about how things should behave. In the case of molecular electronics, though, very little has run a straight course from theory to experiment to result. Theories can languish for years waiting for tools precise enough to test them. In fact, chemists first proposed the idea of molecular electronics in the mid-1970s, but another 20 years would pass before anyone could begin to put it into practice. Lately, though, experimental results have begun to leapfrog the ability of theorists to explain them.
One puzzle is the lack of consistency in measuring experimental results, from lab to lab and even from experiment to experiment. Alex Bratkovsky, a theoretical physicist and native of Moscow who joined HP in 1996, says he was one of the first to realize that a molecule’s orientation between metal electrodes is critical to understanding its switching properties. “The current depends tremendously on how the molecule connects with the substrate,” Bratkovsky says. “The signal may go away, then come back, depending on the position of the molecule. We disregarded that fact for quite a while.” Since controlling the orientation of the molecule is still beyond current experimental tools, results vary widely from lab to lab, and scientists need to judge in many instances whether differences between their results have real meaning or can be explained by effects still outside of experimental control.
To understand the switching phenomenon, the HP researchers are studying a range of new molecules that might be controlled more easily than rotaxane, Bratkovsky says. Some of these are already being designed, but progress is slow. It can take more than two years to design, simulate, synthesize, and finally test a molecule for its electronic properties-after which researchers may find themselves beginning all over again.
Across the hallway from Bratkovsky, Duncan Stewart, an experimental physicist recently hired by Williams’s lab, spent more than six months on a contrarian experiment to help investigate why some molecules can act as molecular switches, changing their conductivity in response to an applied voltage. Instead of designer molecules like rotaxane, Stewart used a simple hydrocarbon molecule consisting of a chain of 18 carbons surrounded by hydrogen atoms. Stewart calls it the “Plain Jane of the molecular world.” It’s stable, inert, and theoretically should have no interesting electronic properties. But it switched anyway.
“I have heaps of data, and the story is that the data do not fit any model, or any existing theory. So even in the simplest case, we don’t understand how electrons are traveling through a molecule,” he says. “At times it’s extremely frustrating. You have to be very pigheaded, beat your head against a wall for six months, and eventually a single brick budges, and eventually the whole wall crumbles and you see another wall.”
The Dust in the Machine
If the materials studied by these researchers seem baffling and unpredictable, the machinery they use is even more so. Progress in molecular electronics is often at the mercy of unpredictable glitches in the experimental equipment. This is, after all, laboratory science and not engineering.
Tan Ha, a native of Vietnam, is in charge of the equipment used in the lab’s clean room. Two or three times a day he dons a clean-room suit and goes into the room to test, adjust, and modify equipment for what are in many cases first-of-a-kind experiments. We suit up. “Now we’re ready for chemical warfare,” he says. The mask over his face makes it difficult to judge whether he is joking.
Once inside we make a beeline for a machine called a chemical vapor deposition reactor. It looks like a big steel cylinder on its side, encased in glass. “I have a special relationship with this machine,” he says, and touches the glass with a gloved hand.
This type of reactor is standard fare in semiconductor fabrication facilities, but Ha has modified the machine to perform the ultraprecise experiments required by Ted Kamins, a member of Williams’s group since 1995. Kamins has worked for years on the ultimate dream of nano research: making devices “grow” in desired structures rather than building them piece by piece. His goal is to grow the nanowires required by molecular electronics, as an alternative to using nano imprint lithography. So far, Kamins has synthesized wires as small as 10 nanometers in diameter by exposing “nanoparticles” of various materials to a mixture of gases in the deposition reactor. In the ensuing reaction, long chains of silicon grow up around the particles, producing what looks under the electron microscope like a forest of needles.
Growing the wires required for molecular electronics is exciting stuff, but Kamins’s particular experiments almost didn’t happen. Ha tells me that he spent over a year of his life trying to make the machine work. “Every time we ran an experiment, contamination would destroy the process,” he says. It wasn’t that the machine was broken; it’s just that no one had ever needed to do the experiments that Kamins wanted to do. “It got to be a spiritual agenda for me,” says Ha. “Ted was frustrated. So was I. I’d be in here on my knees all day long, modifying things screw by screw. I’d go to bed at night and close my eyes and see the plumbing diagram on my eyelids. It turned out to be a problem in the exhaust system. I went home and told my wife, That’s it; I am a proven equipment engineer.’ That’s how happy I was.”
Picking a Winner
Much to Duncan Stewart’s disappointment, Williams asked him to publish his results with the hydrocarbon molecule after six months and concentrate on other work. Yet Williams encouraged Ha to keep working on his knees and dreaming about plumbing diagrams for a year, for experiments that Williams estimates are at least six years from fruition and may never yield a practical result. In a sea of competing theories and possibilities, and with the budget pressures he complains about with some regularity, how does he decide?
“It’s a matter of experience,” Williams says. “I’ve been down many blind alleys many times in my career. They’re so enticing. You can get into these things and think, okay, just one more step, just one more step. Other things feel like they are in the right direction, and I can see where we’re going.” In other words, he has learned to trust his intuition, because it’s all he has. “I’ve been through the cycle many times.”
Williams’s longest commitment to any idea in molecular electronics is to the crossbar architecture. But he admits that even this idea might be a blind alley. Will it ever be possible, for example, to cleanly trap molecules at the junction of two wires with complete confidence in their orientation? Then there’s the practical problem of gain, or turning a weak electrical input into a strong output; this is a critical capability needed both to carry out logic operations and to amplify the tiny currents crossing the molecular switches so that conventional silicon systems can detect them. And it’s a problem with no demonstrated solution.
“Stan is a smart guy, God bless him, and if anyone can solve these things, it’s going to be his team,” says James Tour, a Rice University chemist who is working on a competing approach to molecular computing. “But he’s got a tough problem. At every crosspoint the molecules need to be stable. Then they need to interface with all the wires coming out. There’s an enormous cost to that. They have a steep hill to climb.”
“It’s certainly possible that we are wrong,” admits Williams. Then he shakes his head and stops being humble for a brief moment.
“I don’t think so,” he says. “I think we’ve picked the winner, something that will allow this thing we call Moore’s Law to continue on for another 50 years. I used to think it was impossible. Now I think it’s inevitable.”
|Alternatives to Silicon|
|Technology ||computing elements||Leading Institutions|
|DNA computing||DNA and RNA strands in solution||University of Southern |
Institute of Science
|Molecular electronic devices||Molecules such as rotaxane||University Hewlett-Packard, Yale|
|Nanocells||Gold nanoparticles deposited in random arrays||Rice University|
|Nanotube electronic components||Carbon nanotubes acting as transistors, memory, and wires||IBM, Harvard University, NASA |
Ames Research Center
|Quantum computing||Quantum properties of electrons and molecules||MIT, IBM, Hewlett-Packard, National|
Institute of Standards