Physics of the Future: How Science Will Shape Human Destiny and Our Daily Lives by the Year 2100 (34 page)

But perhaps the most important application of carbon will be in the computer business. Carbon is one of several candidates that may eventually succeed silicon as the basis of computer technology. The future of the world economy may eventually depend on this question: What will replace silicon?

As we mentioned earlier, Moore’s law, one of the foundations of the information revolution, cannot last forever. The future of the world economy and the destiny of nations may ultimately hinge on which nation develops a suitable replacement for silicon.

The question—When will Moore’s law collapse?—sends shudders throughout the world economy. Gordon Moore himself was asked in 2007 if he thought the celebrated law named after him could last forever. Of course not, he said, and predicted that it would end in ten to fifteen years.

This rough assessment agreed with a previous estimate made by Paolo Gargini, an Intel Fellow, who is responsible for all external research at Intel. Since the Intel Corporation sets the pace for the entire semiconductor industry, his words were carefully analyzed. At the annual Semicon West conference in 2004, he said, “We see that for at least the next fifteen to twenty years, we can continue staying on Moore’s law.”

The current revolution in silicon-based computers has been driven by one overriding fact: the ability of UV light to etch smaller and smaller transistors onto a wafer of silicon. Today, a Pentium chip may have several hundred million transistors on a wafer the size of your thumbnail. Because the wavelength of UV light can be as small as 10 nanometers, it is possible to use etching techniques to carve out components that are only thirty atoms across. But this process cannot continue forever. Sooner or later, it collapses, for several reasons.

First, the heat generated by powerful chips will eventually melt them. One naive solution is to stack the wafers on top of one another, creating a cubical chip. This would increase the processing power of the chip but at the expense of creating more heat. The heat from these cubical chips is so intense you could fry an egg on top of them. The problem is simple: there is not enough surface area on a cubical chip to cool it down. In general, if you pass cool water or air across a hot chip, the cooling effect is greater if you have more surface contact with the chip. But if you have a cubical chip, the surface area is not enough. For example, if you could double the size of a cubical chip, the heat it generates goes up by a factor of eight (since the cube contains eight times more electrical components), but its surface area increases only by a factor of four. This means that the heat generated in a cubical chip rises faster than the ability to cool it down. The larger the cubical chip, the more difficult it is to cool it. So cubical chips will provide only a partial, temporary solution to the problem.

Some have suggested that we simply use X-rays instead of UV light to etch the circuits. In principle, this might work, since X-rays can have a wavelength 100 times smaller than UV light. But there is a trade-off. As you move from UV light to X-rays, you also increase the energy of the beam by a factor of 100 or so. This means that etching with X-rays may destroy the wafer you are trying to etch. X-ray lithography can be compared to an artist trying to use a blowtorch to create a delicate sculpture. X-ray lithography has to be very carefully controlled, so X-ray lithography is only a short-term solution.

Second, there is a fundamental problem posed by the quantum theory: the uncertainty principle, which says that you cannot know for certain the location and velocity of any atom or particle. Today’s Pentium chip may have a layer about thirty atoms thick. By 2020, that layer could be five atoms across, so that the electron’s position is uncertain, and it begins to leak through the layer, causing a short circuit. Thus, there is a quantum limit to how small a silicon transistor can be.

As I mentioned earlier, I once keynoted a major conference of 3,000 of Microsoft’s top engineers in their headquarters in Seattle, where I highlighted the problem of the slowing down of Moore’s law. These top software engineers confided to me that they are now taking this problem very seriously, and parallel processing is one of their top answers to increase computer processing power. The easiest way to solve this problem is to string a series of chips in parallel, so that a computer problem is broken down into pieces and then reassembled at the end.

Parallel processing is one of the keys to how our own brain works. If you do an MRI scan of the brain as it thinks, you find that various regions of the brain light up simultaneously, meaning that the brain breaks up a task into small pieces and processes each piece simultaneously. This explains why neurons (which carry electrical messages at the excruciatingly slow pace of 200 miles per hour) can outperform a supercomputer, in which messages travel at nearly the speed of light. What our brain lacks in speed, it more than makes up for by doing billions of small calculations simultaneously and then adding them all up.

The difficulty with parallel processing is that every problem has to be broken into several pieces. Each piece is then processed by different chips, and the problem is reassembled at the end. The coordination of this breakup can be exceedingly complicated, and it depends specifically on each problem, making a general procedure very difficult to find. The human brain does this effortlessly, but Mother Nature has had millions of years to solve this problem. Software engineers have had only a decade or so.

One possible replacement for silicon chips is transistors made of individual atoms. If silicon transistors fail because wires and layers in a chip are going down in size to the atomic scale, then why not start all over again and compute on atoms?

One way of realizing this is with molecular transistors. A transistor is a switch that allows you to control the flow of electricity down a wire. It’s possible to replace a silicon transistor with a single molecule, made of chemicals like rotaxane and benzenethiol. When you see a molecule of benzenethiol, it looks like a long tube, with a “knob,” or valve, made of atoms in the middle. Normally, electricity is free to flow down the tube, making it conductive. But it is also possible to twist the “knob,” which shuts off the flow of electricity. In this way, the entire molecule acts like a switch that can control the flow of electricity. In one position, the knob allows electricity to flow, which can represent the number “1.” If the knob is turned, then the electric flow is stopped, which represents the number “0.” Thus, digital messages can be sent by using molecules.

Molecular transistors already exist. Several corporations have announced that they have created transistors made of individual molecules. But before they can be commercially viable, one must be able to wire them up correctly and mass-produce them.

One promising candidate for the molecular transistor comes from a substance called graphene, which was first isolated from graphite in 2004 by Andre Geim and Kostya Novoselov of the University of Manchester, who won a Nobel Prize for their work. It is like a single layer of graphite. Unlike carbon nanotubes, which are sheets of carbon atoms rolled up into long, narrow tubes, graphene is a single sheet of carbon, no more than one atom thick. Like carbon nanotubes, graphene represents a new state of matter, so scientists are teasing apart its remarkable properties, including conducting electricity. “From the point of view of physics, graphene is a goldmine. You can study it for ages,” remarks Novoselov. (Graphene is also the strongest material ever tested in science. If you placed an elephant on a pencil, and balanced the pencil on a sheet of graphene, the graphene would not tear.)

Novoselov’s group has employed standard techniques used in the computer industry to carve out some of the smallest transistors ever made. Narrow beams of electrons can carve out channels in graphene, making the world’s smallest transistor: one atom thick and ten atoms across. (At present, the smallest molecular transistors are about 30 nanometers in size. Novoselov’s smallest transistors are thirty times smaller than that.)

These transistors of graphene are so small, in fact, they may represent the ultimate limit for molecular transistors. Any smaller, and the uncertainty principle takes over and electrons leak out of the transistor, destroying its properties. “It’s about the smallest you can get,” says Novoselov.

Although there are several promising candidates for molecular transistors, the real problem is more mundane: how to wire them up and assemble them into a commercially viable product. Creating a single molecular transistor is not enough. Molecular transistors are notoriously hard to manipulate, since they can be thousands of times thinner than a human hair. It is a nightmare thinking of ways to mass-produce them. At present, the technology is not yet in place.

For example, graphene is such a new material that scientists do not know how to produce large quantities of it. Scientists can produce only about .1 millimeter of pure graphene, much too small for commercial use. One hope is that a process can be found that self-assembles the molecular transistor. In nature, we sometimes find arrays of molecules that condense into a precise pattern, as if by magic. So far, no one has been able to reliably re-create this magic.

The most ambitious proposal is to use quantum computers, which actually compute on individual atoms themselves. Some claim that quantum computers are the ultimate computer, since the atom is the smallest unit that one can calculate on.

An atom is like a spinning top. Normally, you can store digital information on spinning tops by assigning the number “0” if the top is spinning upward, or “1” if the top is spinning down. If you flip over a spinning top, then you have converted a 0 into a 1 and have done a calculation.

But in the bizarre world of the quantum, an atom is in some sense spinning up and down simultaneously. (In the quantum world, being several places at the same time is commonplace.) An atom can therefore contain much more information than a 0 or a 1. It can describe a mixture of 0 and 1. So quantum computers use “qubits” rather than bits. For example, it can be 25 percent spinning up and 75 percent spinning down. In this way, a spinning atom can store vastly more information than a single bit.

Quantum computers are so powerful that the CIA has looked into their code-breaking potentials. When the CIA tries to break the code of another nation, it searches for the key. Nations have devised ingenious ways of constructing the key that encodes their messages. For example, the key may be based on factorizing a large number. It’s easy to factorize the number 21 as the product of 3 and 7. Now let’s say that you have an integer of 100 digits, and you ask a digital computer to rewrite it as the product of two other integers. It might take a digital computer a century to be able to factorize this number. A quantum computer, however, is so powerful that in principle it can effortlessly crack any such code. A quantum computer quickly outperforms a standard computer on these huge tasks.

Quantum computers are not science fiction but actually exist today. In fact, I had a chance to see a quantum computer for myself when I visited the MIT laboratory of Seth Lloyd, one of the pioneers in the field. His laboratory is full of computers, vacuum pumps, and sensors, but the heart of his experiment is a machine that resembles a standard MRI machine, except much smaller. Like the MRI machine, his device has two large coils of wire that create a uniform magnetic field in the space between them. In this uniform magnetic field, he places his sample material. The atoms inside the sample align, like spinning tops. If the atom points up, it corresponds to a 0. If it points down, it corresponds to a 1. Then he sends an electromagnetic pulse into the sample, which changes the alignment of the atoms. Some of the atoms flip over, so a 1 becomes a 0. In this way, the machine has performed a calculation.

So why don’t we have quantum computers sitting on our desks, solving the mysteries of the universe? Lloyd admitted to me the real problem that has stymied research in quantum computers is the disturbances from the outside world that destroy the delicate properties of these atoms.

When atoms are “coherent” and vibrating in phase with one another, the tiniest disturbances from the outside world can ruin this delicate balance and make the atoms “decohere,” so they no longer vibrate in unison. Even the passing of a cosmic ray or the rumble of a truck outside the lab can destroy the delicate spinning alignment of these atoms and destroy the computation.

The decoherence problem is the single most difficult barrier to creating quantum computers. Anyone who can solve the problem of decoherence will not only win a Nobel Prize but also become the richest man on earth.

As you can imagine, creating quantum computers out of individual coherent atoms is an arduous process, because these atoms quickly decohere and fall out of phase. So far, the world’s most complex calculation done on a quantum computer is 3 × 5 = 15. Although this might not seem much, remember that this calculation was done on individual atoms.

In addition, there is another bizarre complication coming from the quantum theory, again based on the uncertainty principle. All calculations done on a quantum computer are uncertain, so you have to repeat the experiment many times. So 2 + 2 = 4, at least sometimes. If you repeat the calculation of 2 + 2 a number of times, the final answer averages out to 4. So even arithmetic becomes fuzzy on a quantum computer.

No one knows when one might solve this problem of decoherence. Vint Cerf, one of the original creators of the Internet, predicts, “By 2050, we will surely have found ways to achieve room-temperature quantum computation.”

We should also point out that the stakes are so high that a variety of computer designs have been explored by scientists. Some of these competing designs include:

•
optical computers:
These computers calculate on light beams rather than electrons. Since light beams can pass through each other, optical computers have the advantage that they can be cubical, without wires. Also, lasers can be fabricated using the same lithographic techniques as ordinary transistors, so you can in theory pack millions of lasers onto a chip.