Although it's a little like watching a chess match in slow motion, molecular electronics researchers are converging on viable circuit-fabrication methods.
Several approaches to building circuits with molecules reached the stage of at least rudimentary logic or simple devices, such as inverters or AND gates, last year. Ratcheting up the pace, Hewlett-Packard Laboratories has announced two patents that are said to solve some small but nettlesome problems with a molecular FPGA approach based on a switching molecule called rotaxane. The HP patents detail a practical method for connecting molecular-scale circuits to the outside world and a method for defining circuit sub-blocks in a massive crossbar array of nanowires. A third patent describes an approach to memory arrays using rotaxane.
The design rules of molecular-electronics schemes hint at the scale of the problems the researchers face. The erbium disilicide wires used in HP's process are 2 nanometers in diameter and are spaced 9 nanometers apart on a silicon substrate. The HP design is based on a Manhattan architecture in which a parallel series of north-south wires intersects a similar series of east-west wires. Each wire crossing becomes a location for a molecular switch. Billions or perhaps trillions of gates could be put on a silicon die.
HP's effort is only one of many approaches that enlist chemical processes to build circuits. In other recent work, researchers at IBM Corp. showed that carbon nanotubes can behave like transistors with gain; a group at Delft University, in the Netherlands, built basic logic circuits with carbon nanotubes; and researchers at Harvard University used indium phosphide nanowires to demonstrate logic circuits.
Notre Dame researchers are pursuing an essentially different approach with arrays of quantum-dot transistors that do not have to be wired up but instead influence their neighbors through electric fields. The arrays would perform computation by emulating cellular automata, which can compute a wide range of algorithms. Recently, the quantum cellular automata were shown to be capable of producing gain, said Craig Lent, one of the researchers on the project.
Lent doesn't classify quantum cellular automata as strictly molecular circuits, since the quantum dots are essentially metallic in nature. But metallic materials may make it easier to interface with classical electronic circuits, he observed.
Molecular methods are attractive because they solve the enormous problem of defining trillions of devices and their interconnects. "It is basically a 'shake and bake' approach to semiconductor processing," said HP researcher Phil Keukes. Chemical reactions, rather than computer-defined masks, determine the circuit and assemble it.
"The billions being spent on semiconductor fabs are due mainly to the problem of having to align structures to extremely small tolerances," he pointed out. The complex details of a circuit's structure are held in automated design databases. And the data-processing costs of storing and manipulating ever-larger databases is falling exponentially, even as semiconductor fab costs increase.
"When you look at the trends, the obvious solution is to find a simplified means of building the hardware while keeping the complex details of a circuit in software," Keukes observed. That has led the HP team, which is working in conjunction with a group at UCLA, to design a molecular equivalent of an FPGA. Once built on a molecular scale with an astronomical number of gates, complex circuit designs can be downloaded to the hardware.
The basic hardware design is simple: Organic molecules of rotaxane are introduced into a crossbar array of erbium disilicide nanowires. Wherever two wires cross, a rotaxane molecule will attach itself. The wires do not have to be precisely aligned, because the rotaxane molecules will seek out the crosspoints wherever they occur. The rotaxane is able to mimic all the functions of an FPGA cell.
In theory, the design represents a simple solution to many of the tricky details of building molecular circuits. It solves one universal problem: the inherent defects created by any chemical reaction.
"Current technology, such as a Pentium, demands about a 99.9999 percent success rate per transistor," Keukes said. "But with chemical processes, you might only get 95 percent yield not nearly enough."
The FPGA array can be used to implement a redundant wiring scheme in which defective cells are simply switched out of the network. In an earlier experiment, the HP researchers were able to demonstrate how that could work by deliberately building an FPGA processor that contained 3 percent defective cells. Called the Teramac, the resulting processor achieved near-supercomputer-level processing with an array of FPGAs containing 200,000 defective cells. Keukes and his colleagues developed an algorithm that tested the cells, switching out the defects and wiring up a desired computer architecture.
The recent patents add some important components to this circuit design approach.
One issue is how to build arrays of distinct circuit components. With the simple crossbar scheme, every wire is a global interconnect, a highly inefficient setup.
The answer was to include special "cutter" wires in the array. A positive potential is placed on a conducting wire that needs to be cut, and an equal negative potential is applied to the cutter wire that crosses it. All other wires are grounded so that only the specified junction sees the double potential, which is high enough to produce a chemical change in the conducting wire, making the junction a high resistance. That effectively cuts the wire at a predetermined point.
Keukes also believes that his group has solved the I/O problem, with a patent on a means for multiplexing between CMOS-level signals and molecular signals. The I/O problem could be acute because of the very large difference between macroscopic currents and molecular currents, as well as because of the extremely large number of devices that molecular systems could create.
Others have been working on the problem. James Tour, who heads the molecular-electronics effort at the Center for Nanoscale Science and Technology at Rice University (Houston), has applied a programmable, organic approach to the problem.
"We are developing a 'nano-cell,' wherein there is a plethora of organic/metallic material which can be set into differing switch states with external voltage pulses in a somewhat random yet static arrangement between multiple micron-size contacts," Tour said. "If that material can be programmed from the micron array that surrounds it, then the functionality could be enormous, although we do not know precisely where each element is within the nanocell. That's much like the brain, where we use it but we do not know the precise neuron interconnect pathway. The nanocell would be trained post-fabrication, much like an FPGA."
One advantage to the method is the ability, as in the HP effort, to use the programmable-gate function to put the complexity of the problem into software rather than hardware. "Others' approaches, in my opinion, are far more regular and suffer from the micro- to nano-I/O-size problems," Tour said. "Ours suffers from an enormous programming challenge, but the assembly is relatively trivial. So we shift the fab problem to a programming problem. Our simulations are encouraging nonetheless."
The advent of molecular electronics may thus be upon us, but in what form is not clear. Keukes of HP believes that molecular memories will appear in five years. One certainty, in his view, is that silicon will be up against insurmountable physical and economic barriers in 10 years. At that time, molecular methods may be mature enough to step in. Other experts in the field also believe the 10-year mark is probably realistic.
Notre Dame's Lent said he thinks "the memories from HP are the clear front-runners. They don't have truly single-molecule memories, though, and may not get them soon," he said. HP last year demonstrated a 16-bit molecular memory.
Give it up
Several parameters must be considered when determining how quickly various projects will get to practical results, Lent believes. "I would classify some of the ideas in terms of what they 'give up.' If you give up single-molecule devices, then many parallel schemes are possible using 104 or 106 molecules per device. They can then be current-driven without power dissipation being a showstopper."
On the other hand, such conglomeration will not provide the extremely high densities promised by molecular processes. "If you give up 'design,' you can have random single-molecule devices. The limitation is that if you just plop down a spaghetti-and-meatballs heap, then most of the area of the device is wasted," Lent said.
Again, density is the casualty with this approach since many of the elements will be in the wrong place or will be defective. "If you give up logic, you can go for just memory. This is certainly a good idea because it will succeed sooner," Lent said. "That's why I think the only practical demonstration so far is HP's memory, even though it's not yet single-molecule."
Rice's Tour said that his group's nanocell method would produce viable memory systems in one to two years.
Mark Reed, a molecular-electronics expert at Yale University, believes it is too soon to say which project will win the race, although he believes that memory systems will be the first out of the gate. His group also developed a molecular memory circuit last year.