Making smaller, faster, cheaper computers


Ralph C. Merkle
Xerox PARC
3333 Coyote Hill Road
Palo Alto, CA 94304

This is one of a series of Predictive Papers published in the Proceedings of the IEEE.

This paper is available on the web at

I Earlier forecasts

Previous predictive papers have treated readers to forecasts by some of the finest minds of the decade. They foresee major advances in computation, communications, biotechnology, nanotechnology, and a host of other areas -- advances which will give us (among other things): a networked world where communication, education, entertainment and commerce will all take place over a pervasive digital medium; computers approaching and even surpassing the power of the human brain; constellations of satellites peering down from the heavens tracking where we are and keeping us all in constant touch; better crops; vast improvements in disease prevention, diagnosis and treatment that will force us to cope with the societal issues of longer and healthier lives; cheaper and more abundant energy; etc.

Watkins' 1962 forecast saw two major developments: the ability to "grow" complex circuits and "construct highly complex three-dimensional circuits entirely within a single crystal of solid."

Rather than add to this plethora of predictions about what we'll be able to build, we'll instead hazard a few guesses about how we'll build them. The forecasts of Watkins and the others can only happen if we develop new and vastly superior ways of making things. What new principles, what new ideas will lay the foundations for this technological revolution?

II The very atoms

Perhaps the most remarkable advance of the past few decades has been our ability to see and move the very atoms from which we and all our manufactured products are made: scanning probe microscopes (SPMs) have not only opened up a whole new world for our inspection, but have also let us touch and move the parts of that world in a way that few in 1962 dared to dream would ever be possible. Our first steps have been halting but the trend is clear: ever greater precision in creating ever more intricate and varied patterns of atoms. We can already imagine high density memories where each bit is encoded by the position of only a few atoms -- or even by a single atom. Looking further, we see molecular switches made with the ultimate precision: every atom in precisely the place it must be for the switch to work.

Watkins' "single crystal of solid" must have the "defects, impurities, vacancies, interstitials, dislocations, precipitates, grain boundaries, etc." in just the right places: the places required for its function. While we want to build it with the same precision that a chemist brings to the synthesis of a molecule, yet it will require something very different from conventional chemistry. Today's SPMs have demonstrated the basic principle, that we can indeed (as Feynman put it) "...arrange the atoms the way we want; the very atoms, all the way down!"

III Small size, high speed, and massive parallelism

This idea of using some future descendant of the SPM to manufacture a complex computer immediately raises a problem: arranging a few atoms, building a small molecule, a single logic gate where every atom is in the right place, these things seem to fall within the realm of the feasible. But billions and billions of them? Mole quantities of miniscule switches? One of today's SPMs, moving (if we've scaled ourselves down to the molecular size range to watch) like some giant and stately inverted Eiffel tower over a small patch of atoms, could never hope to build more than a handful of switches let alone the vast numbers that we need to make.

Two things will save us: shrinking the Eiffel tower and making many of them. With smaller size comes faster operation and the possibility of massive parallelism. Very many very small SPM-like devices, rank upon rank of them, each orders of magnitude faster than any of today's SPMs, would give us a combined throughput equal to the task of making tomorrow's computers. Whether we use them to read and write molecular marks on a surface, thereby giving us the ability to make and read billions upon billions of marks per second, or whether we use them to make more complex devices and indeed whole arrays of computers, the principle remains the same. If one is too slow, use two. If two aren't enough, use four. If four aren't enough, use eight....

Have we, though, merely solved one problem by creating another? We wanted to make untold billions of switches, and solved our problem by asking for untold billions of miniature SPM-like devices. But how do we get untold billions of them? Lithography might let us make thousands or perhaps even millions, but untold billions? How can we possibly make so many, and make them cheaply and precisely to boot?

IV Self replication

Giant redwoods can grow to over 300 feet and have billions and billions of cells. Nature knows the secret of making cells in vast quantities: cells can divide and make more cells. One cell divides to make two, two cells divide to make four, four cells divide to make eight.... After some dozens of replications we have a towering giant. We know that self replication is possible, for nature has already done it.

This is not the first time we have been inspired by nature. Birds flying through the air demonstrated that heavier-than-air flight was possible. But when we actually made our own flying machines they were very different from birds. A 747 has no feathers, nor can it perch upon a branch nor flap its wings. Birds do not use jet engines, nor do they have a metal skin. To build airplanes we had to do more than copy the biological world, we had to develop new principles and new approaches more suited to our human capabilities and limitations.

In the same way, biological self replicating systems are both more and less than we want. We want to make, not squishy soft watery cells, but hard crystals with precise impurities and imperfections. We aren't seeking merely to copy nature's example, but to design something new: an artifical self replicating device able to make complex circuits within "...a single crystal of solid." Is this possible? Can we do it?

V Artificial self replication

The basic principles underlying artificial self replicating systems were studied by von Neumann in the 1940's, and are no more difficult than the basic principles underlying the computer. Von Neumann divided his artificial self replicating systems into two major subsystems: a computer and a constructor. The computer held within its memory a set of instructions which it interpreted and used to generate the control signals which directed the constructor. The constructor interacted directly with its environment to make another computer and another constructor. Finally, the computer in the original device copied the contents of its memory into the memory of the new device (thus avoiding an infinite regress, or a homunculus within a homunculus within a homunculus....)

The idea of a very small computer is one that we are already comfortable with. To this we must add a very small descendant of the SPM to act as the "constructor." Given a molecular constructor able to arrange and rearrange molecular structure in some programmable way, the difficulties of self replication become more pragmatic than conceptual. What does such a system look like? How will thermal noise influence our design? What sort of framework must exist around the constructor if it is to function correctly? People have skin, bacteria have bacterial walls, what protects the computer and constructor from a less than perfect environment? How will it bring raw materials through this "skin"? What sort of "molecular tools" will it use? These and other questions have already been the subject of many papers (see for further information). They pose no fundamental barriers but instead invite us to create, not just one answer, but whole classes of answers: they invite us into whole new vistas of research.

Watkins foresaw both engineers who would "grow" their computers and also computers made with amazing precision within a three-dimensional crystal of solid. He didn't foresee the merging of these two forecasts. The idea of a self replicating device able to arrange "the very atoms" was first advanced by Eric Drexler, who coined the term "assembler" to describe it. As we are forced to move beyond conventional lithography to some new and remarkably precise, remarkably flexible and remarkably low cost manufacturing technology, what could be more natural than to use self replication to make the many billions of assemblers with which to make the mole quantities of logic gates that future computers will require?

VI Safety

To mention "artificial self replication" is to conjure up images of Mickey Mouse from The Sorcerer's Apprentice frantically trying to stop ever more and ever smaller brooms from fetching ever greater floods of water. Are artificial self replicating systems an open invitation to worldwide disaster?

The only self replicating systems we are familiar with are living, and we unconsciously assume that artificial self replicating systems will be similar. But the machines people make bear little resemblance to living systems. The image of a 747 going feral, swooping out of the sky to clutch an unsuspecting horse in its landing gear, seems incongruous. Machines lack the wonderful adaptability of living systems. A 747 requires Jet A, a refined source of energy that is delivered to it by an elaborate system that includes oil fields, pumps, tankers, refineries, fuel lines, and trucks. It can convert this artificially refined fuel into energy using engines that can run on little else. Cut off from refined fuel, airstrips, maintenance crews, spare parts, navigational systems and all the other paraphenalia that keeps it flying and a 747 is just a large piece of scrap metal. A bird, in contrast, can live on berries, seeds, worms, insects, small rodents, fish and bits of bread tossed to it by amused tourists. Its living and remarkably adaptable digestive system can convert all these and more into energy and essential raw materials for power and self repair. It thrives in the complex and ever changing natural world.

It will be challenge enough to design an artificial self replicating system able to function in a controlled artificial environment using a single specific source of highly refined energy, let alone in the wild disarray of the natural world where it would have to adapt to whatever came its way. The fear that we will accidentally destroy the planet seems remote indeed. Yet there is cause for concern. We have a long and bloody history of settling our differences by war: new tools and new technologies will let us make new weapons, giving us new opportunities to express the less civilized aspects of our nature. Peace and global security might happen by accident, but prudence dictates that we take a more active role in preparing for our future.

VII Conclusions

Looking ahead and asking not what things we will make but how we will make them, we see two principles that will both change how we make computers and which will let us move beyond computers to change how we make everything else. The first principle is demonstrated (albeit in nascent form) by scanning probe microscopes, which even now let us move individual atoms and molecules. The second principle is self replication, which will let us inexpensively make things in enormous quantities. Taken together, these two principles will revolutionize how we make computers - and quite a few other things as well. Watkins' vision of growing complex three dimensional computers is moving closer to reality.

This page is part of the nanotechnology web site.