Question 1: Tell us about yourself, your background, etc.

I got my PhD in 1968 from Moscow State University in Russia, and worked in that university until 1991, when I accepted a professorship at Stony Brook.

Question 2: What is your current best estimate for how long Moore’s law will continue?

It is mostly a question of economics. I am no expert on these issues, but I see economic difficulties mounting rapidly toward the end of this decade, in the 2008-2010 time frame. The continuing demand for high performance computing may defer the crisis through sustained massive investment in technology for a few years, but for how long exactly I am not the best one to ask.

Question 3: What is your assessment of quantum computing? Do you think that it will ever have broad applications?

This is an exciting research subject, but I don’t see any substantial market for that. Quantum computing can only do few things, such as number factoring, better than conventional computing, performance for other tasks being comparable. The market for factoring is not big enough to drive the technology, especially since the cost of implementation of quantum computing will probably grow exponentially with the size of the system. I think in order the practical quantum computing to happen, we need another revolution that would change the equation.

Question 4: Do you think that the semiconductor industry will be able to successfully transition to alternate technologies once conventional lithography techniques hit a wall?

It depends on how fast we will be able to come up with the alternative technology, chemically-directed assembly of molecular electron devices. We are working hard in this direction, as are many other people. But regarding the exact time of development of such revolutionary technology, there is no clear answer right now.

Question 5: Tell us about NOVORAM. Are terabit chips really feasible?

NOVORAM is a possible new memory technology that should allow scalable and very fast memories to replace DRAM (that is not scalable, and is running out of steam). If the NOVORAM concept is confirmed experimentally, then terabit NOVORAM chips should be feasible. The time frame for this is very short, I would say 5 to 7 years (or never).

Question 6: Tell us about single-electron transistor (SET) chips. How likely are they?

In terms of density, SETs would be the limit of computing. This technology is not especially fast, and there are problems with using them in logic. In particular, random background charge is one problem that is possibly insurmountable within the current computing paradigm. However, there is a new paradigm, that we call self-evolving neuromorphic networks. In this scheme SETs play a key role: our approach combines the chemically-directed assembly of SETs with nanoscale CMOS. If it proves successful, we will be able to put a hardware analog of the cerebral cortex on a 10x10 sq cm area of silicon.

Question 7: Are you familiar with the concept of a technological Singularity? What do you think of the concept?

I am not familiar with this term, but I think that the arrival of artificial intelligence (in a sense broader than implied by the common term AI) is inevitable, and I think that it is just a matter of time. Even if our approach turns out to be infeasible, somebody cleverer will come up with a different concept. Software problems are enormous, but so many people are working on that, that I would think within a century the success is almost certain. Perhaps within 50 years.

Question 8: Writers such as Ray Kurzweil argue that we can learn new computing techniques from reverse engineering the human brain. Do you agree?

Only partly. Of course what people are doing with neural networks is biologically inspired. However, when I look at the status of neurobiology (and I have been following this field for 30 years), I don’t see breakthroughs large enough to prove engineers with sufficient information. The problem with current high-resolution measurement technologies is that they are invasive, and you cannot measure activity of a thousand neurons simultaneously without perturbing it. Which means that I don’t think that literal reverse engineering would work. What we are going to do with our neuromorphic network architecture is to use the fact that these circuits should be able not only to operate a million times faster than the cerebral cortex, but also to self-evolve a million times faster. This is why we hope to reproduce the natural evolution of the brain starting from something very uniform and primitive. This is our idea, of course there could be other approaches.

Question 9: Tell us about your neuromorphic network architecture.

It is similar to neural networks, but with a twist. Initially we will try to use it for advanced image recognition, but in future this could hopefully lead to sentient machines. But at this point we haven’t proven that this system can work, even by modeling. One can say this is purely vaporware at this stage, but I am not aware of any reason why it shouldn’t work. Optimistically speaking, this could happen within 20 years.

Question 10: Some scientists argue that carbon nanotubes can serve as logic gates, interconnects, and molecular wires. These scientists argue that carbon nanotubes have the greatest potential to supplant CMOS. Do you agree?

Carbon nanotubes are mostly thought of as replacements for silicon transistor channels, because their surface can be smoother, and provide less electron scattering. But I am not very optimistic concerning this technology, for two reasons. First, right now people don’t know how to fabricate carbon nanotubes with reproducible properties. The real promise is with single-walled nanotubes, and when they come out from the fab with different helicities (i.e. the spiral of atoms on their surface has different arrangements). Helicity changes nanotube transport properties from metal to semiconductor to insulator, and nobody knows how to control helicity at fabrication yet. Second, for nanoscale devices, and we are talking here of devices of 10 nm, scattering is not the largest problem, and I think that silicon would still do very well. Unless we have a major revolution in carbon nanotube field, for example if somebody invents a process of fabrication of the nanotubes with fixed helicity, they are unlikely to seriously challenge silicon.

Question 11: What are your plans for the future?

We have the several projects that we are working on that I have already discussed, such as NOVORAM, SETs, and neuromorphic networks (including chemically-directed assembly of molecular devices). We are also working on modeling of nanoscale MOSFETs, to see seeing how far the conventional (and semi-conventional) CMOS can be scaled. From my point of view, these are all exciting research directions; my patience level is pretty low and I would not work on any project if it weren’t likely to bear fruit within the next decade or so.