1989, Eigler made headlines by using his handmade STM to spell out the
letters "I-B-M" with individual xenon atoms. Before joining IBM as a
research staff member in 1986, he was a postdoctoral member of the technical
staff at AT&T Bell Laboratories. In 1993, he was named an IBM Fellow,
the company's highest technical honor. In addition, he is a Fellow of
the American Physical Society and the American Association for the Advancement
of Science. Among other prizes and distinctions, Eigler won the Dannie
Heineman Prize from Germany's Göttingen Academy of Sciences in
1998 and the Nanoscience Prize in 1999. He is a frequent lecturer in
the United States and abroad and even trains service dogs.
this December 2004 interview with MICRO,
he discussed his views on a range of subjects, from evolutionary versus
revolutionary nanoscience and the limitations facing CMOS technology
to molecular cascades, the quantum mirage, IBM's investment priorities,
and how nanoscience may lead to technological breakthroughs in the years
to come—including in the semiconductor industry. "You can do all kinds
of things with chemistry," Eigler remarked, "but you can't do what you
can do when you move atoms."—BM
future is in small structures," you said someplace. What do you mean
by that in a broad, theoretical sense? Not in the nanotech sense, but
in the nanoscience sense.
In the science sense: Science goes according to what you can do and
where the unknown lies. And you can sort of predict where science—except
for the serendipitous stuff—is going to make progress. It makes
progress where you can identify the unknown and you have the capability
of making progress—of changing the unknown into the known and
understood. In terms of the future of science of small structures, the
motivation for saying that is that starting about 20 years ago, we have
a whole new set of tools that allow us to both investigate and to create
small structures. That enables us to get answers to questions that we
could not have gotten before.
In addition, we recognize
that the science of structures that are small is interesting, different,
and unknown. Because for really small structures, at the atomic scale,
we have a pretty good idea. We just ask, "What is the physics or the
chemistry of a single atom." We've got that pretty well understood—not
fully, but pretty well understood.
If we ask about the
properties of larger structures, there are still some very fascinating
things to work on. But every once in a while, something really big comes
along. For example, the last really big thing there was high-temperature
superconductors. Every once in a while, somebody discovers something
really big—something like colossal magnetoresistance or beyond-colossal
magnetoresistance or super-beyond-colossal magnetoresistance.
If we look at structures
that are on the order of tens to hundreds of atoms in size—macromolecular
structures—or the structures that we build on that same length scale,
they have scientific properties that are not very well understood and
that we now have access to in a way that we never had before. That's
why over the last 10 years we've been living through what I would call
the golden age of nanoscience. Because it's a time when so much is being
discovered and understood. This is being done using all different kinds
of experimental techniques and laboratories all around the world. It's
also being done because there is the applications side—that exploration
process, which is at the very least tantalizing and at the very most
potentially revolutionizing in terms of how it could affect mankind.
That's sort of a long ways out.
is a really loaded word. Many things affect how that exploration process
The science part is really exciting. There's no doubt about it. At least
for the scientists working in the field, because there's a lot going
on, a lot that we just did not properly anticipate, or we look at something
and say, "Oh gee, that's new. I don't understand that. Where did that
spoke of nanotechnology as distinct from nanoscience and that nanotechnology
really amounts to, as you put it, "market applications," "the way the
world works." Those remarks [available on-line at www.nanotec.org.uk/evidence/oraleiglerDrdon.html]
were given at the Royal Society in 2003. I thought that was actually
pretty interesting. Obviously, coming from a trade magazine, our focus
would mainly be the nanoscience applications for semiconductor manufacturing.
What semiconductor-related applications are you working on?
There are some obvious connections to what I'd call evolutionary nanotechnology.
The way I look at it, the way IBM looks at what's happened to our semiconductor
business, is that we are manufacturing a lot of products where the properties
of things on the nanometer-length scale begin to really matter. And
the size of the structures is on that 100-nm-length scale and getting
smaller. Certainly, that's what we're manufacturing and delivering—or
about to. So in that sense, nanotech is already here. If you look at
what research has to do in order to keep that pipeline filled, in order
to try to extend those technologies to smaller and smaller length scales
in order to deliver—hopefully—improved performance or improved
cost, we have to be doing our research at the nanometer-length scale.
So that's what I would call an evolutionary kind of nanotechnology.
nanotechnology, and the way I think of that in terms of computation
is as an exploratory effort. It's largely an effort at trying to understand
if there are opportunities to create or bring into the marketplace new
technologies that are useful in doing computation, which are different
from just evolving our semiconductor technology. The challenges of evolving
semiconductor technology are evident; they're written about all over
the place. It's going to be extremely difficult to continue shrinking
CMOS in a way that we have become accustomed to shrinking CMOS. Not
only will we be making a very significant effort in improving our semiconductor
technology, bringing advantage to our clients that way, but we feel
it is incumbent upon us to search for alternative technologies that
would be useful in computation and would get around the problems of
semiconductor technology. That's the exploratory part.
What is taking place in that direction? Are you engaged in that?
I'm certainly engaged in that. Two aspects of our work go on simultaneously.
One aspect is to build a base of fundamental knowledge in areas that
we think are important to the future of our industry. That's the basic
science part of it. At the same time, we try to explore how we might
utilize that knowledge in order to do computation. That goes on sort
of in the same effort.
I can give you an
example of something that we did a couple of years ago. An observation
in the laboratory concerning what happens when you create certain arrangements
of carbon monoxide molecules on a surface led to the implementation
of the smallest logic circuits ever made—smaller by an awful lot compared
with anything that had ever been made. And not just a device, but circuits—circuits
that worked in a way that is very different from the way we do computation
using transistors. And at the same time, we did some very fundamental
work. These circuits were used in something we call molecule cascades,
which are arrangements of molecules—I should say metastable arrangements
of molecules—which are akin to setting up dominoes and then knocking
them over to create a cascade. You can think of that cascade of dominoes
as a way to transport a bit of information from place to place. Transporting
information is really important in terms of how we conventionally do
computation. We learned not only how to transport information using
these molecules, but how to do logic. In fact, we demonstrated everything
that you need in order to calculate an arbitrary logic function. We
built those circuits.
That was the applied
part of it. A fundamental part of it was studying the motion of molecules
and trying to understand what created the metastability, how long they
kept on moving. We've learned how to use that knowledge, how to leverage
that knowledge, to help us do computation. By understanding what's going
on, we get to the point where we can better engineer the next generation
of logic circuits.
Are you talking about molecular electronics?
The model I'm talking about is very different from molecular electronics.
Molecular electronics is doing electronics, but just using the devicelike
properties of molecules, the so-called devicelike properties of molecules.
Molecular electronics is also largely not understood and is overhyped.
There was a very interesting article in Science magazine by Robert
Service that came out I think a year or two years after Science
did its big issue on molecular electronics. It's interesting. That's
a very fruitful area of research because so much is not known and is
not understood. It's just not imminent by any means.
So that's something different from what you're talking about, where
you used carbon monoxide on a surface that actually succeeded in creating
Very different. We actually have real logic circuits. And not just something
that might have devicelike properties. We have real logic circuits;
we can really engineer them. The stuff that we were doing has its own
set of limitations, and it's certainly not at all ready for prime time,
nor did I think that it would readily be available for prime time as
we did it. But the important thing about that research was how it steered
our research effort, how it directed us and pointed out the importance
of going in the direction that we wanted to go in anyway, which was
understanding physics—the magnetic physics or the spin physics—of
very small structures. We did this cascade computation using molecules,
where the molecules would move from site to site in order to do the
computation. We're hoping that we'll be able to do the same kind of
cascade-based computation—which is very different from electronics—but
the thing that will be cascading will be the spin configuration on the
Has the motivation for the research you've done been the limitations
of traditional silicon technology? Thinking about something you said,
probably in your presentation before the Royal Society: Silicon technology
is slowing down as it matures, and, as I understood what you said, is
becoming more expensive. Is that a motivation for this research into
It certainly is. The thing that motivates our research is the needs
of our clients. When we can deliver a better solution to our clients,
we've done our job. That not only helps our clients to get their jobs
done better, but also makes IBM successful in the marketplace. So having
that focus on the needs of our clients is the number one answer to our
mantra. It's branded on our foreheads. It's one of the reasons why IBM
has been so successful over the course of the last decade or so.
Now, how do we do
that? One of the things we would love to be able to do is to deliver
things that compute better, faster, more reliably, and cheaper. That's
a standard answer, but it's one of the things that really benefits our
clients. But not only that: As we achieve that goal, the kinds of things
that we can think about doing with computational machines—with computers—expands.
There are so many things that you'd say can be done in principle, but
they can't be done in practice because they're too expensive or they
take too long to get done—where the chances that the machine would
survive long enough to get them done go to zero because the machine
has cooked itself. We would love to be able to do those things, but
we can't until we get better machines. So building the better machines
is clearly a driving motivation.
What kinds of machines do you have in mind? Does the scanning tunneling
microscope come in here? Or are you thinking about a research tool?
No. I'm talking about the things that compute, that do things that our
clients want to get done. I like to think very broadly. The first thing
for us is computers that do binary logic the way our conventional computers
do. But in the future, it may be that quantum computers will play an
increasingly important role. I do not in any way rule out the role that
analog computers can play or the roles that multilevel logic computers
can play—computers that work with more than just two levels of
This is going to
sound extremely odd: I look at this little black box over here, and
we think of it as a binary logic computer at one level. That's all you
need in order to understand it. But if you really want to understand
what goes on, you have to understand how analog devices work, because
when you really get down to the nuts and bolts of how a transistor works,
you have to think that a transistor is an electronic device, it has
a continuous response to a continuous set of inputs. It is not intrinsically
a binary device. We hook them up and we operate them in a way that from
the outside of the black box we can think of them—our logic devices,
our logic gates—as being binary devices. They have logical ones and
zeroes on the input and logical ones and zeroes on the output. That's
fine. But if you want to understand what goes on in the box, to figure
out how to be able to shrink that or make it cost less, you've got to
We think that in
the future the question—I think from our clients' point of view—is:
How do you get the job done? And I don't think our clients would really
care too much about the details of how the job gets done inside the
box as long as it gets done better, faster, cheaper, more reliably.
So our focus is broad when it comes to our research. Certainly there
is a very strong component of research that goes on at IBM that is devoted
to the evolutionary nanotechnology part that I talked about before—extending
CMOS down to smaller length scales—because that's certainly going to
be a real part, a deliverable part, of our strategy over the coming
years, certainly the immediate future. Because nothing will be ready,
that I know of, in that same period of time—certainly in the course
of the next five years, maybe a little longer.
But we also, in our
research portfolio, invest in longer-term, more-risky, less-known issues,
about how to extend—not so much how to extend semiconductor technology—but
how to extend the progress in computation that we've come to know and
love and almost economically rely upon. If our hardware technologies
level out, then the only extensions we'll be able to deliver are by
improvements in systems and in software. And there's lots of room for
improvement there. We have plenty of focus in those areas. I have this
focus both on the short term and the long term about what to do with
the hardware that's going to deliver improved performance or cost-benefit.
You have talked about possible niche areas for nanotechnology. You mentioned
sensors and embedded systems. How does that all fit into this?
I'm not sure that it's so important for computation. I think when I
mentioned that, I probably mentioned it within the context of being
asked to prognosticate about where and how nanotechnology—revolutionary
nanotechnology—might have its first really important impact on
things that are deliverable, things that will change how people do what
they do and make a difference in people's lives. And in small sensors:
I think there are some incredible things going on with the development
of new kinds of sensors to do bedside assays of what's going on with
your blood chemistry or something like that. So you get rapid, reliable
Another area that
I'm increasingly encouraged about—which is just really fascinating—is
the medical applications of things that people are doing with nanotechnology.
There's extremely encouraging work done by one of my colleagues at Rice
University [in Houston], where they have been utilizing what are called
gold nanoshells, which have this really cool property. It's a shell,
it's not a solid. On the inside there's a little sphere of silicon,
and in this sphere there is gold. And by choosing the diameter of the
silicon and choosing how thick the coating is, you can tune what's called
the plasmon resonance of that nanoshell—the shell of gold—over a broad
range of wavelengths. In particular, you can tune it to be at an infrared
wavelength, because infrared penetrates right through your body quite
are so small that they can readily move through your circulatory system.
And because the gold nanoshells seem to be biocompatible, don't seem
to be toxic—at least so far—and you can label these things with antibodies
that are specific to cancer cells, they've been able to inject mice
with them. These antibiotic nanoshells go throughout the mouse's system,
preferentially stick to the cancer cells, and six hours after they inject
the mice, the researchers come along with a little laser diode, shine
it on the mice, and the little mice say, "Hmm, it's nice and warm here."
What happens is the infrared moves through the mouse's body and basically
passes through except for where it hits the gold nanoshells, warms up
the gold nanoshells that are stuck to the cancer cells, cooks the cancer
cells, and out they go. It's the first test I've seen. The survivability
of the mice that were treated this way was 100%. They took these mice
that had these tumors and just knocked off the tumors.
I just gave a presentation
in Amsterdam using this an example. There may be some problems with
it: There may not be a viable cancer therapy because of problems with
biocompatibility, or with how the nanoshells get transported out of
your body, or whether they get transported out of your body, or something
else. We won't know until it's been through enough testing whether it's
going to be a viable therapy for cancer. But this is just one thing
that's being done.
The scope of things
that we can now try to do with nanoscale structures when it comes to
medicine, biology, and to nonmedical applications too—to computation
or to sensing—is just opening wide up because of our ability to understand,
to engineer, and to fabricate structures on the nanometer length scale.
That door is wide open right now. We're at the early stages. They used
to say that that was more of a statement of faith than anything else.
But as we increasingly see new stuff coming out and the applications
in this area growing, the laboratory results where people have really
done things—and these mice are still really alive—then you start saying,
"What is that—maybe it's not that far away." Nobody knows yet what
the—I'm going to make a pun—what the killer apps are going to be.
Maybe cancer-killing gold nanoshells are one of them. Pretty cool. It's
really fun to learn about this stuff.
Back to the evolutionary-revolutionary dichotomy that you talked about
a few minutes ago: The scanning tunneling microscope, the gizmo that
you guys used to write "I-B-M" in xenon atoms—what is the role
of that, for example, to extend CMOS?
Knowledge already has played a role.
The knowledge that you can see, as I think you put it, see and manipulate
at the atomic level?
The manipulation part hasn't played a significant role yet. But in order
to be able to extend CMOS technology, there are certain hunks of knowledge
that we want. We may need to understand and interface better between
a metal and a semiconductor. We may need to understand electronic properties
of that interface at a much more local scale. We may need to understand
the voltage distribution around an impurity in the semiconductor when
a flow of current passes there. And these tools—the scanning tunneling
microscope, the ballistic electron microscope, the atomic force microscope—are
tools that we can use in the laboratory to generate knowledge that may
allow us to test our fundamental knowledge. Or they can be test or inspection
instruments to help us know what structures we want to make, help us
develop the processes for building those structures, and help us inspect
and test those structures. So they really do play an important role.
But the key issue there is knowledge. Knowledge is hugely valuable.
Your laboratory, your colleagues have talked about the incredible shrinking
circuit—the shrinking circuit in accordance with Moore's Law ["IBM
Scientists Discover Nanotech Communication Method," available on-line
Frankly, that's not a law in a strictly scientific sense. It's really
a business construct.
That's correct. I don't know of any scientist who would disagree with
you, except perhaps a social scientist.
In any case, on such small scales, again you're talking about shrinking
device geometries: "Tiny wires don't conduct electrons as well as classical
theory predicts. So quantum analogs for many traditional functions must
be available if nanocircuits are to achieve the desired performance
advantages of their small size." That article goes into the quantum
mirage, which is described as a way to transport information on the
atomic scale that uses the wave nature of electrons instead of conventional
There's some meaning, some truth to that statement. When a structure
is small enough, its quantum properties begin to become important. It's
not only a question of how small, it's a question of how small and at
what temperature. At room temperature, the quantum properties of an
atom are everything. They are really important. As you get to too large
a structure, then the quantum properties begin to disappear. The macroscopic
properties are derived from the quantum properties, but there are issues
about the quantum properties that are very different.
Let me give you an
example of something that is between an atom and a larger structure
but still exhibits a very important quantum property: carbon nanotubes.
Carbon nanotubes transport electric current down them in a way that
is different from the way we normally think of a wire. The quantum properties
of electrical transport down a carbon nanotube at room temperature are
still important. They help to make the carbon nanotube a very good electrical
conductor. (At least metallic carbon nanotubes are very good electrical
conductors.) If you were to take a bunch of copper atoms and shrink
them down—not shrink them down, but build a wire out of them—which
have the same diameter as a carbon nanotube, I'm not sure, but at the
same diameter the wire might have important quantum properties, but
it would still be very different from a carbon nanotube. When you get
too-much-larger-diameter wire, then you can really start thinking of
it as being just like the wire that you use to wire your house—unless
you get it really cold. The quantum properties offer us some unique
capabilities that we don't yet know how to exploit, but if we could,
it might be really beneficial. So that's an example of what the quantum
mirage taught us.
We demonstrated something
using the quantum mirage that was just sort of a hint at what might
be coming if we can manufacture things on a small enough length scale.
For instance, if you look at a chip, your active device is a transistor.
It's all on one layer. And then you might have as many as 10 or 12 layers
of wiring on top of that. Why do you have so many layers of wiring?
Because wires carry information from place to place in the circuit.
They carry information in the form of a potential or a current. We can't
let wires touch one another, otherwise they stop carrying independent
pieces of information. They short out. We call it a short circuit, and
it just doesn't work. You have to have insulation between the wires.
So you have to get information from here to there and there to here.
You have to do a crossover. Crossover can't be done at the same level.
You can't have two wires—two conventional wires—be at the same level.
It sounds absolutely trivial, but believe it or not, it costs the industry
a hell of a lot of money to build all those layers of wires. It's not
trivial at all. It's expensive, and it's complicated.
So you can't have
wires—conventional wires—touch one another. What we use these wires
for is to send information. But if you think about it, we are quite
used to sending just tons of information right through other tons of
information with no difficulty whatsoever. For instance, right now there
are photons that leave your skin and come into my eyes so I can see
you. And I have photons leaving my skin going into your eyes so you
can see me. Those photons are passing right through one another, and
it's a lot of information. Take a room like this: If you were to look
at just a cubic centimeter of volume in a room lit at this level, there's
about the equivalent of 1018 or 1019 bits of information
per second going through that cubic centimeter. If you think of moving
information back and forth, how many gigahertz do you like—2 GHz? Let's
be generous: let's say 10 GHz, which is microwave frequencies. That's
just 1010 bits per second, maybe. It's not even that much;
it's actually down from that quite a bit, but you have to modulate it
some. I'll be generous: 1010 bits per second. The bits of
information per second going through a cubic centimeter in this room
is nine orders of magnitude greater.
We also do this whole
tunnel of sound. You can speak and I can speak and our sound waves go
right through one another, and we can hear each other. The reason is
that when we communicate using sound or light, we utilize the wave property
of sound or light waves, which pass through other waves. The way we
conventionally use wires, we do not design them to utilize the wave
properties of electrons in the wire. We just send electrons down the
wire the same way that we send water down a pipe. When you have water
moving down a pipe in one direction, guess what? You don't have water
moving up the pipe in the other direction. However, you could send sound
waves through the pipe in one direction and have sound waves go back
up the pipe in the other direction—sound going through the water.
What we demonstrated
with the quantum mirage was that we could do this with electrons in
a solid. When the solid becomes small enough, we can utilize the wave
properties of electrons in the solid to transport information from place
to place. And in a laboratory demonstration, we demonstrated the ability
to send two channels of information right through each other. On the
surface of the metal, we sent information from here to there and information
from there to here right through the same place. What this demonstrates
is an idea, or a principle, and it provides a motivation or a guide
or a spark.
Then we said, "OK,
how would we get that into a technology?" Actually, the first part was
basic physics and exploratory. It was a discovery of a phenomenon—the
quantum mirage. Then it was understanding the physics of that quantum
mirage that led us to think, "Well, this is really sending information
from one place to another place." We realized that because we were sending
it using waves, we could send multichannels of information through the
same volume of space and then demonstrate that. Then we're starting
to switch from the basic science to the exploratory technology part,
because we're trying to use that knowledge to do something. Then this
motivates us to think, "Well, what will be the next step? How do we
go from there toward something that we can deliver in the marketplace?"
Can the transport of information along the same plane or sending pieces
of information through each other potentially lead to the development
of a two-dimensional IC? ICs have lots and lots of layers, and that's
a big problem.
In principle, yes. I'm going to speak now like a physicist, not like
a guy who actually has to get something done and shipped for a client.
A physicist would say that there is no fundamental reason why computation
cannot be done in two dimensions—maybe even in one dimension,
but I want to pull my punch on that one. In fact, the molecule cascades
were done in two dimensions, and it was a fully general computational
circuit. Working in two dimensions using the molecule cascade, we know
how to do everything we need to do to calculate anything that this guy
[a standard computer] can calculate. So we have a proof of principle
that there is no fundamental need for three dimensions.
What the quantum
mirage stuff taught us is that we're going to use silicon and then get
information from transistor to transistor. I'm not suggesting that that's
what we would do, but if we were, we might have the silicon in one layer
and then we might move the information around, but just on one other
layer. We might need one other layer on top of that to distribute power
to the silicon. But for just moving information—at least in principle—it
can all be done. But that's sort of pie in the sky.
The kind of thing you're talking about is building circuits on a much
more simple level. It sounds like that would be a more simple, straightforward
thing. But is it something IBM is actually investing in or thinking
about for future technology development?
IBM's paying me to do this kind of work. I think we can call that an
investment. There's no doubt about that. And there are other people—I'm
not the only person doing this. To give you a more-balanced view of
our research effort: Our research effort, in many ways, is kind of like
an investment portfolio.
So you're talking about money—that's the important stuff.
Right. We have a very broad range of research investments that in some
ways make an investment portfolio. One word that characterizes it could
be a "wise" investment portfolio. And what we think is a wise research
portfolio is diversity in the nature of our investments. We make the
majority of our investment in things that are relatively near term and
where we think there's a pretty high likelihood of payoff. And that's
because we've got to drive the industry forward this year, next year,
and the year after that. But we also make investments in the things
that might drive the industry forward 5, 10 years or maybe even 20 years
out. Today you're talking to the guys doing the farthest-out research
among probably anybody in IBM research in terms of when it might impact
the marketplace. But there are other people with things that have to
be delivered this year, this month, or something like that, and they're
going to be in our products very shortly.
You talked about scientific research and money. I believe you said someplace
that it would be a good thing if it weren't all driven by the need to
make a profit: "Create a scientific culture less driven by money." Obviously,
IBM is in the business of making money, and ultimately what drives this
entire system is making profit.
It's not just making profit. If IBM were just interested in making profit,
I think we'd go out of business really quick. And so you have to think
about not only being financially successful as a corporation today and
this year and next year, but we intend on being around here 10, 20,
30, 40 years. And if the company is successful, we will morph into being
whatever we need to be in order to do that. That is something that we're
pretty good at. At least the last decade has shown us to be pretty good
at morphing what we are in order to be successful.
The long-term research that you were talking about, possibly 20 years
out: Is IBM at the top of that research effort, or who else is doing
that sort of thing?
In IT biz?
I don't know of anybody else in IT biz that makes the kind of investment
that we do. They're not even close. I could be really wrong about that,
but I can't think of any other company that comes even remotely close
to making the level of investment that IBM makes. Part of that, I think,
is our culture, part of it is our determination to drive the industry
forward. It's also part of a key component of our strategy. If you just
look at where we've positioned ourselves, at least in terms of the systems
that we build and sell, we sell at the high end, at the bleeding edge
of performance. That's why me make that huge effort to drive our technology
forward. If we don't, then we end up being in the commodity business,
and we just choose not to be in commodity hell.