Scientists pursue miniaturization of chips by substituting chemical reactions for silicon.
A radical approach to semiconductor fabrication may soon lead to supercomputers the size of wristwatches. Scientists are developing logic gates based on molecular oxidation that could allow these building blocks of computers to be constructed of only a few molecules.
This advance may open the door to continued reduction of computer components when silicon optical lithography reaches its theoretical limits around 2010. Researchers conducting the work at the University of California-Los Angeles (UCLA) and Hewlett-Packard Laboratories, Palo Alto, California, believe the technology could replace silicon-based integrated circuits inside personal computers. The result could be entire memory chips that are 100 nanometers wide—smaller than a bacterium.
These logic gates are based on the oxidation and reduction of rotaxane molecules. Although still in their infancy in the laboratory, processors built with this technique could form the basis of supercomputers small enough to be woven into clothing. This development is critical as the demand for more powerful computers causes chip manufacturers to include more transistors and wires on processors, which drives up the cost of semiconductor fabrication facilities.
The new approach also offers a number of other benefits. Because these are chemical processors, there would be no danger of losing information when an electrical power source goes down. In addition to information technology applications, the minuscule processors would enrich the biomedical field by potentially offering the diagnostic community a means to get close enough to cells to identify specific diseases.
The rotaxane used consists of dumbbell-shaped molecules that have another ring-shaped molecule trapped around the narrow portion of the dumbbell. According to Phil Kuekes, computer architect, Hewlett-Packard Laboratories, the shape more closely resembles a folded dumbbell, “as if Arnold Schwartzenneger had bent it into a V.” The ring is a separate molecule that is not covalently bonded to the dumbbell and can therefore change its position. The total molecule has on the order of a hundred atoms.
“The potential virtue of such a molecule is that the ring can slip from one position to another when you reduce or oxidize it. One position represents a zero, and the other position represents a one,” Kuekes explains. Chemically, reduction is a gain, and oxidation is a loss of electrons.
The researchers have already produced functioning “and” and “or” gates using rotaxane. An “and” gate, a logic element with two or more input lines and one output line, outputs a one when all of the input lines carry a one. Otherwise, the output is zero. An “or” gate, which also has two or more inputs and one output, results in a one if any of its inputs is one, and a zero if none of the inputs is a one. Logic gates such as these are the basis of computers.
The underlying element of the molecular gates is a junction where two flat wires cross each other, forming a square region at the overlay point. In the intervening space is a monomolecular layer of rotaxane that is 20 angstroms thick.
“What we have built is about 10,000 molecules on a side. But if you make the wires smaller, the area of that little patch shrinks. If we can go to nanosized wires, which in fact do exist, carbon nanotubes being a perfect example, you would then have only a handful of molecules in that area underneath.
“In principle, a single molecule has all the equipment it needs to be both a bit and an open or closed switch, though practical devices would probably have three or four molecules,” Kuekes explains. The rotaxane monolayer remains stable for a considerable amount of time at room temperature. At present, the devices cannot be reconfigured. They can be set to be ones or zeros, but once set, they cannot be reset, he adds.
“These particular molecules are really ROMs [read-only memory], not RAMs [random-access memory]. You can write them once. We are doing research on related molecules to see whether we can get RAMs. We’re confident we will be able to. I want to emphasize that we have not yet done it.
“We can read and write the bit because it is essentially an electrochemical cell, just like a battery is an electrochemical cell. You can cause a chemical reaction to occur by putting a relatively high voltage across those two wires, ‘high’ meaning a volt and a half or 2 volts maybe, and that changes the configuration of the molecule. The ring moves. You oxidize or reduce it, actually add or subtract an electron. If you then remove the potential, the molecule stays in the new state,” Kuekes explains.
Because they operate as switches, the junctions potentially can be used in making a field programmable gate array (FPGA), which is a chip that contains a large number of “wires” and switches that can be programmed by downloading configuration bits to customize the FPGA to carry out desired functions.
“An FPGA is a special type of integrated circuit, a do-it-yourself circuit. They make one standard circuit, and then you essentially can put any wiring pattern into it. You burn them. But the point of reconfigurable gates is that you don’t so much burn them as you set a memory bit. Basically we’ve taken this idea and asked what if we could dramatically shrink this and in effect use a molecule as both a memory bit and a switch,” he offers.
Manufacturing FPGAs is a $1 billion business today. The arrays are sometimes used in classified military applications because the chip can be configured by the military itself; therefore the final circuit does not have to be divulged to chip manufacturers.
“The configuration bits are not in ordinary memory so no hacker can get into them,” Kuekes says. Because the conductivity of the device differs by a factor of almost a hundred between the two states, there is little danger of confusing one state for the other.
“The more important thing, which excites me as a computer designer, is that when the switch is closed, it has a very nonlinear-IV [current-voltage] curve, a very nonlinear resistance. For all practical purposes we can treat it as a diode. I’m not sure I’d go so far as to say it is a diode, but because of the nonlinearities it acts like a one-molecule diode. If you have diodes, even without a transistor you can build logic gates, which is what we have done.
“The performance theoretically is extremely high, simply because of the small size. We take advantage of all the performance improvements with small size that silicon gets. The capacitance and resistance just drop. Speed comes basically because you take very little energy to move a bit around. Because of the very small capacitance, it will switch very fast, certainly in nanoseconds, perhaps in picoseconds. We don’t have really solid theoretical numbers. Because of the very small currents through these switches, we don’t have any timing measurements at all. All our measurements are essentially taken at DC [direct current], showing that these are ‘and’ or ‘or’ gates,” Kuekes explains. There will be a lot of opportunities in engineering, he predicts.
“Computers keep getting better because it takes less energy to switch states. Designers use this in two ways. One way is to switch at the same speed and use less energy. This is desirable, for example, in laptops. But these chips don’t clock real fast. The other way to take advantage of the reduced energy needed to switch is to run Niagara Falls through them, pushing them to switch faster but dissipating more heat.
“We will be able to build very small machines. We’ve been talking about computer paint, the size of a grain of dust with the power of the machine on your desk. So if they’re cheap enough, you can imagine mixing them with paint and putting them on a surface—low powered. As you package them and build supercomputers, they’ll be tremendous. In principle there’s a factor of maybe a billion in efficiency, at the molecular scale,” Kuekes says.
The nature of molecular computing is such that tolerance of defects is crucial to success. Rotaxane computers will have the necessary tolerance, he adds.
“In the mid-1990s, I was involved in the design of basically a do-it-yourself supercomputer here at H-P Labs. We called it Teramac—the ‘tera’ because it did a trillion very small operations a second and ‘mac’ because it was a ‘multiple architecture computer.’ It was a reconfigurable machine. In effect we put together 864 FPGAs of our own design. These chips were about 16 millimeters on a side, so the total area of our own silicon, not memory and so on, but the chips we had designed, was almost a quarter of a square meter. The machine was massively parallel, with a huge number of configuration bits in it, and you could download all of these configuration bits and essentially rewire the computer to be any interesting, highly parallel machine you wanted. We built it for our own architectural research, but among other things it had a very nice property. It had so many wires and switches that we could find defects and reroute the machine to lock out the defects,” Kuekes explains. The chips were defective because they were cut from the wafer on which they had been manufactured without the usual routine testing.
When chips are built on silicon wafers, a portion is usually defective because of tiny imperfections in the silicon. The larger the physical area of the chip, the more likely it is that it will contain a defect. Normally, these are eliminated through testing. Because the idea of the research was to experiment with wiring around defects, no attempt was made to discard defective chips. In this case, nearly three-quarters were defective.
“Going around defects is a bit like locking out bad sectors on a hard drive. You have to have a lot of wires and particularly a lot of switches, a lot of choices to take another route,” Kuekes explains.
“You need to be able to lock out defects in molecular computers. Remember that these molecules are thermodynamically assembled, by things banging against each other. It’s very statistical, and we would tend to get yields of maybe 95 percent if we were lucky, unlike the semiconductor that gets yields of 99.999 percent per part. If the per part [yields] were 95 it would be disastrous without defect tolerance,” he adds.
Reconfigurability means that a rotaxane computer could be changed, for example, from a general-purpose machine into a signal processor in seconds by downloading a different set of configuration bits from a hard drive.
Kuekes believes practical devices will be available in less than 10 years. “The argument is that in roughly 10 years we’ll have something somewhat competitive with silicon. The problem is that silicon is a moving target, but that’s roughly when they are going to have a lot of problems. By then, we’ll have something with a lot of sophistication and complexity.”