A man and his microscope: IBM's quest to make atom-sized chips
The silver surfer speaks
El Reg: When you were doing this work, was it more to see just how far you could push things or where you thinking seriously about applications for semiconductor physics?
DE: When I first started working at IBM, the motivation was to lay a foundation of knowledge about surfaces because we knew that knowledge was going to play into our industry. We knew that we could identify areas where we really wanted to understand things better.
But, since we have learned to move atoms around, we have definitely changed the flavor of our research to being more applied. Now we ask more direct questions. Can we build things that compute that small?
And, we have learned to do that. We've built the smallest switch, which is just the motion of a single atom. Four or five years ago we came out with new kinds of logic circuits called molecule cascades, which are far and away the smallest logic circuits ever built, and they really do computation for you.
[Eigler discusses how a one atom switch works here.]
El Reg: What are you examining these days?
DE: My work has been more on the exploratory/revolutionary part. We have to get out there and look for the stuff to follow on. What will augment silicon or follow silicon. The work that has been going on in my laboratory has really been addressed at that level of questioning.
There is no guarantee that anything coming out of our laboratory will lead directly to something that is shipping in a product.
El Reg: What seems to be the most promising direction at the moment?
DE: One of the things that has me excited and that the group is working on is this. We really want to do computation with spins. Electrons have this property of having a magnetic moment of spin.
The way the transistor works is by controlling the flow of electrons. When electrons move from place to place, they bang into things, they lose their kinetic energy and that warms things up and creates all these problems such as resistance and leakage. That's the way the world does computation right now.
One of the buzzwords floating around is spintronics. This generally refers to using the flow of electrons to control the magnetic orientation of the spins - at some region - of a solid or using the spins to control the flow of electrons. It's kind of a way of switching things or doing computation using the combination of the motion of electrons and the spin of electrons.
Right now we are working towards being able to do computation only using the spin.
El Reg: Can you explain more how that would work?
DE: Imagine that you have a bunch of electrons. Suppose the spins of these electrons are all aligned and then I come to the end of the chain, and I grab one spin and knock it down. Can I make a type of reaction happen where they all go down?
One of the things I have just done is transport a bit of information from one place to another. Information transfer from place to place sounds very mundane, but, it's a huge issue. If you look at everyday semiconductor chips, the transistors are in an active layer of silicon and then on top of that in a CPU, there might be 12 layers of wiring. Why do we have all that wiring and what is it doing?
The vast majority of that wiring is doing nothing more than moving information from place to place. It is not doing logical operations on information. It is just moving information from a place where logical operations can be done to the next place where that information has to be processed.
So being able to transfer information just using the spin and not letting electrons move around is something we're interested in. But we are also interested in learning how to use the spins to do the logical operations. In fact, we are interested in learning how to do everything with the spins except for the input and the output.
El Reg: Does this type of technology really look that practical? Why spins?
DE: Well, for one, because it's there.
Two is that we think there is a possibility that it could be done in an extraordinarily dense way. We might be able to achieve three-dimensional integration as opposed to two-dimensional integration, which is what we conventionally do. And then we might be able to do it in a way that consumes very, very low power.
Do we know the answers? Can I tell you that any of those things are likely? No, we don't know the answers. But for somebody with a job like mine, when someone says we don't know the answer, we smile and say, "Job security." That's what we're paid to do is go out there and dig up those answers.
El Reg: How do you get one of those spins to go down, as you say?
DE: We're thinking about the techniques used with molecule cascades. They work the same way as dominoes in the sense that you can stack dominoes up, and they stay there assuming there is no big earthquake, and then you knock one over and have this cascade. You have to put in energy. With dominoes that's the process of standing them up. With magnetic spins you have to do the same thing globally with a magnetic field from the outside. We create a magnetic field that stands our spins up and once they are up, without moving them around, we should be able to knock over key dominoes at the input and dominoes/spins will flip.
Sponsored: Hyper-scale data management