Stan Williams, Hewlett-Packard

Stan Williams (Source: Hewlett-Packard)

Stan Williams is a Senior Hewlett-Packard Fellow and founding director of the Quantum Science Research (QSR) group (Palo Alto, Calif.), created in 1995 to meet the major challenges and opportunities lying ahead in electronic device technology as features progress toward nanometer scales. His primary research has been in the areas of solid-state chemistry and physics, then nanostructures and chemically assembled materials, with an emphasis on the thermodynamics of size and shape. Recently, he examined the fundamental limits of information and computing, which began his current research in molecular electronics. Williams has awards for scientific and academic achievement, including the 2000 Julius Springer Award for Applied Physics, the 2000 Feynman Prize in Nanotechnology, the Dreyfus Teacher-Scholar Award, and the Sloan Foundation Fellowship. He was a co-organizer and co-editor of the workshop and book Vision for Nanotechnology in the 21st Century, respectively, that led to the establishment of the U.S. National Nanotechnology Initiative in 2000. Williams is also adjunct professor of chemistry at UCLA, and of computer science at the University of North Carolina at Chapel Hill. He has a B.S. in chemical physics from Rice University, and master and Ph.D. in physical chemistry from UC Berkeley. The QSR group at HP Labs is focused on the fabrication of nanometer-scale structures and the measurement and understanding of their properties. Its research into molecular-scale electronics is driven by the realization that the fundamental limits to the power efficiency of computation lie as much as a factor of one billion beyond the presently known capabilities of silicon ICs. QSR researchers span a broad range of academic disciplines, including computer architecture, theoretical solid-state physics, electrical engineering, materials science, and physical chemistry.

SI: The QSR lab is a major player in nanotech development. Can you update us on the current projects?

Williams: We've grown rapidly over the last few years, and are expanding into new areas. We began with a primary emphasis in electronics, and have since gone into photonics and mechanics. We're doing work that allows us to control electrons, photons and atoms at the nanometer scale for the purpose of gathering, storing, processing and communicating information. At the finest scale that the laws of physics enable, we want to be able to perform ultra-precise measurements — measurements at the molecular scale — understand that environment, and then take that data and manipulate, process and store it, move it, display it — all of this using technology that is as close to the limits of fundamental science as is possible.

SI: Are you working on many projects?

Williams: About 30, but I'll mention a couple. We have been working on imprint lithography, and we continue to push the limits of this technology. We're trying to determine what the physical limits to building circuits and devices are; how small can you get. We have produced lab demonstration circuits with 15 nm half-pitch, and currently work at size scales that the International Technology Roadmap for Semiconductors projects would be useful around 2020.

SI: At that scale, quantum effects must get rather interesting.

Williams: Indeed! We are in a realm right now where the usual rules of the game no longer apply. We're working with a different set of physical laws. Our group was founded for this, with the understanding that technology would inevitably go into this very small size regime, and that once we got there, we'd find things like Ohm's Law no longer applying. We must work and understand this realm well before having to bring it into our products, and must know how to deal with it to take advantage of the new properties of matter that emerge at this nanometer scale.

SI: What sizes are you working on?

Williams: Currently, we are at 15 nm and intend to continue these shrinks. We want to explore the absolute limits of what can be done. We've some preliminary data, for instance, that shows that we should be able to make things as small as 4 nm.

SI: Not to sound corny, but you are going “where no man has gone before.”

Williams: Yes, we are. This was a strategic decision made 10 years ago by HP to ensure we have a strong presence in nanotech.

SI: Moving, as you do, in a quantum realm, have you encountered any surprises?

Williams: (Laughing) Practically every day! One of the specific structures that we built using these very small-scale features is a memory. We had ideas as to how and why the memory should work. Fortunately for us, when we've built them, they have worked. However, at this stage, we haven't completely figured out why. We're still doing some fundamental science to try to understand why these structures work in the way that they do. We're continually bumping up against mysteries, so we're doing a combination of very fundamental research and technology, and we iterate between the two.

SI: It must not be easy to find defects in your devices.

Williams: No, it isn't. An issue that we have alerted the semiconductor community to is an area we've worked on for some time, which we term “defect tolerance.” We understood early on that it isn't only quantum effects that create problems at these scales, but that the Second Law of Thermodynamics and entropy also become major issues. When you build wires where the entire width is maybe 20 atoms across, if just a few atoms are out of place, this can dramatically change the properties of that wire.

SI: And the Second Law of Thermodynamics guarantees that there will always be atoms out of place.

Williams: Exactly! You cannot make anything absolutely atomically perfect at that scale; thus, we have both quantum and entropy effects playing a part. So we decided that since it would be impossible to make perfect wires and perfect devices, we'd have to figure out how to make perfect machines using crummy — this is a technical term — components. ICs have been the closest things to perfection that man has ever made, and they've worked so well for so long that people forgot about this issue of having to try to build perfect machines out of imperfect components. We've had to revisit this at this extreme nanometer scale, and invent a range of new techniques that allow us to build a circuit (or machine) that will operate perfectly even though its components are imperfect.

SI: Mainstream developments at 65 and 45 nm must seem Brobdingnagian.

Williams: They do! One of our biggest problems lies in just looking at our devices. We must use special scanning probe microscopes. At that size scale, when we image our devices, we see individual atoms as lumps and bumps.

SI: For years, our industry has been dreading that the limits of CMOS are being reached. What do you see as the limits of nanotech?

Williams: There are definitely limits to what the laws of physics allow and those limits are set by, for instance, the speed of light. Information cannot travel any faster than that. Then there is also Heisenberg uncertainty issues. All these are unavoidable limitations to how fast one can process information.

SI: You make it sound a bit bleak.

Williams: Not at all! It turns out that those very fundamental physical limits are about a factor of 10⁸ beyond where silicon is now. So although we may talk about the end of silicon, we're very far from the end of physics. One of our primary jobs is to understand how to get as close to the limits of what physics allows. This may be by adapting silicon to do that, and if that's impossible, to see what we can add to silicon — or what we can replace it with — to allow us to get much closer to the fundamental limits of what is possible.

SI: Certainly, circuitry could be improved by some orders of magnitude.

Williams: That is correct. Consider that the effective thermodynamic efficiency of today's silicon circuits is about one part in 10⁸. Thus, only 0.000 000 01 of the electrical power that goes into a current-day computer is actually doing useful work in terms of computation. The rest is dissipated as waste heat. If automobiles were that inefficient, they would vaporize when you turn the ignition key. We're trying to understand what must be done to approach the thermodynamic limit of information processing. Just think of it, the human brain is about a million times more efficient than today's silicon chips!

SI: Is that why you think we can do better?

Williams: It is proof of that, and we know from fundamental physics that we still have quite a way to go. We may be running into silicon's end, but that is not such a serious matter since we are very far from reaching the fundamental end of physics.

SI: So how long do you think you think Moore's Law can be continued?

Williams: At least another 50 years. There are true limits to what can be done. Obviously, not everything is possible, but it is too early to worry about fundamental limits, and it's possible to improve the technology by at least a factor of a million from where we are today.

SI: Probably even more?

Williams: Probably. That's our daily job; we're using our ability to control electrons, atoms and light at this nanometer scale to make more efficient the way in which information is handled.

SI: How do you view the industry's effort to extend CMOS' life through nanotechnology?

Williams: There isn't gong to be a sudden revolution, where on Dec. 31st they shut down the fabs and on Jan. 1st they wake up to something else. The industry is very mature, has a lot of momentum, and a large installed base. Thus, any viable solution, to be economically feasible, must adapt to CMOS; it'll have to hybridize with and complement it. Much of what we're working is aimed at extending CMOS' life and improving it.

SI: You've said that a device you've created, the crossbar latch, could eliminate the need for transistors at the computational level. Is there any future for the transistor in computing?

Williams: Today there are still power supplies in computers. You take electricity from a wall socket and covert it to a usable form for the electronics. I think that tomorrow's transistors will perform somewhat more mundane tasks, such as distributing electrical current to other devices. They won't go away; they will still be an essential part of the system. After all, a computer today cannot work without a transformer; it may be a mundane component, but an essential one. Transistors will continue to be important, but not something that receives as much attention as they do today.

SI: You've announced a new way to design nanoelectronics using coding theory. How do things stand today?

Williams: This relates to defect tolerance. We've discovered that coding theory gives us a means for providing the mathematically correct amount of redundancy to make the circuit work. Think of it this way: I want my manufacturing process to be as inexpensive as possible. However, if it becomes too inexpensive, the devices I make will behave unreliably. We've found out that by adding a small amount of redundancy using coding theory we can take a circuit built out of many unreliable parts and make it very reliable, while keeping the manufacturing cost as inexpensive as possible. We discovered it some years back, almost by accident, and are still learning its implications for circuit design and optimization.

SI: You have written that your group is reinventing the computer at a molecular scale. Can you update us on this?

Williams: This is part of the need to understand where we are going to go beyond semiconductor physics. Silicon is facing some very real material limitations, but fundamental physics has no such limitations. Much of what we are doing is to understand the physics that will allow us to go beyond silicon. There has been much talk about “molecular electronics.” We prefer to call it molecular-scale electronics, because much of it may not involve an actual molecule in the usual sense of the word. However, the size of many of the components that we have been working on is at what I call the subviral range. We've made active devices that are smaller than the smallest known virus. Some of our devices use certain molecules as components. At this level, we're researching what Messrs. Schrödinger and Heisenberg will allow us to get away with, instead of trying to suspend the laws of quantum mechanics.

SI: That is a challenge getting close to the semiconductor industry.

Williams: Yes. As the industry scales down, it faces the hurdle of how to make its circuits behave as if they were classical systems. A standard FET device is a classical mechanical device — it treats electrons as if they were rocks. The problem is that we're getting into a region where electrons want to behave like waves, not rocks. We're investigating how to enlist the quantum behavior of matter at the nanometer scale and get our functionality out of that. In our devices, we use quantum mechanical tunneling.

SI: So, essentially, you're trying to turn quantum science into quantum technology.

Williams: I want to use that line! That is exactly what we are working on across a broad spectrum of applications.

SI: What can we expect from the QSR over the next three years or so?

Williams: In the various areas in which we are working, in electronics, we're investigating interesting new types of sensors and detectors, in particular chemical sensors, which are becoming increasingly important for many applications and reasons. We're defining manufacturing processes that will allow us to build at these scales; our imprint lithography is a good example. We think that our memory technology is turning the corner. In five years' time, there should be operational memories that will be commercialized for specific applications where regular memory needs to be outperformed. In the beginning, they will be premium components. There will be hybrid circuits for logic that incorporate mostly CMOS, but with a few interesting nano-switches and relays, making the circuit much more efficient. For the next 10 years at least we'll be building hybrid systems of these types. After that, CMOS will most definitely take a secondary role. There is no reason why the results of Moore's Law cannot continue for another half century.