Question 1: Tell us about yourself. What is your background, and what current projects are you working on?
I was educated at the University of Illinois, and my background is in computer science. I actually worked at Microsoft for about 7 years, from 1995-2001. I am currently the CEO of Apex nanotechnologies, a company that I founded.
Question 2: Tell us about Apex nanotechnologies.
We are a molecular design company. We make computer-aided molecular design software. We are in particular focusing on increasing the usage of computer-aided software design inside of corporate research organizations. Today, fewer than 5% of researchers do any of their research in software. Research in the chemical industry, the materials industry, and the biotech industry is done the same way that research has always been done – bench research, expensive lab instruments, and a lot of manual labor. We focus on moving more of that research into software, thereby allowing the dominant portion of corporate researchers who don’t know how to use molecular modeling to use these tools.
Question 3: Is this concept related to grid computing?
No. Grid computing is really about having shared computational resources that many people can utilize. So grid computing would be a way to get the CPU power to do some of this molecular modeling. But we can do this type of modeling on any thing from a desktop PC up to a small Linux cluster or a mainframe to a large grid. The two terms that people are most familiar with are molecular modeling or computational chemistry. We call it computer aided molecular design.
Certainly the ability to do grid computing, or “edge” computing, such as SETI@home has done, does give a huge computational speedup. However, in any scenario, you are also limited by the interconnect speed between the computers. So with most software problems, as you start to scale up, you need higher and higher bandwidth between the computers to keep scaling. People are currently very happy in this field if they can get a cluster of 200 machines and get close to a linear scale up. The biggest cluster that I’ve heard of using commodity PCs linked together, is on the order of a thousand machines. And even for that, they used very specialized networking equipment to connect them, to prevent the network from becoming a bottleneck, far in excess of a DSL line, let alone a modem. Even a DSL line becomes a bottleneck after you have a few dozen or so computers in the grid.
Question 4: It sounds like this concept is heavily hardware dependant, and greatly affected by Moore’s law.
That is absolutely correct. We have the ability to use nearly infinite computational power in this, in that we can always get higher accuracy answers by using up more CPU cycles. The models that exist all have some degree of approximations in them, but with increasing hardware power you need to make fewer and fewer shortcuts, and therefore you will have higher accuracy. But you’re going to need much more CPU power. The cheapest models scale on an order of n. That means that if you have n molecules, the CPU time you need is proportionate to that. If you double the number of atoms in the computation, you’ll double the CPU time. There is a whole range of algorithms that get increasingly accurate, and you can get up to some algorithms that scale to n7, where n is the number of electrons. So if you increase the number of electrons by a factor of two, you increase the required CPU time by 27 , so that is a very expensive but high accuracy method. The CPU time can go up astronomically. But there are intermediate methods that are fairly accurate.
Question 5: Are these intermediate models accurate enough to make working molecular models?
Yes, they are accurate enough to give you valuable answers at a fraction of the cost of doing actual physical experiments. They do not replace experiments, but they help to narrow down a range of possible experiments. They can make experiments more efficient and more productive by giving you a much better idea of what to look at.
Question 6: Will your molecular modeling techniques be able to completely replace experiments?
Maybe hundreds of years in the future, but not in the foreseeable horizon. It is possible that decades from now most of the work will be done via molecular modeling. But GM and Ford design cars using CAD software, and they have a really good idea of how a car will function before they build it, but they still build prototypes, and they still conduct physical tests. Nothing will ever be as accurate as actually doing the experiments. Molecular modeling might eventually be 95% of the work, but there will always be the crucial step of verifying things in the real world.
Question 7: What is your opinion of molecular nanotechnology? What is your opinion of Eric Drexler?
Eric Drexler is a visionary and deserves credit for popularizing the notion of building machines at the molecular level, with the suggestion Richard Feynman made. There is nothing that I see in principal that prevents the development of molecular nanotechnology. However, it does not appear to be anywhere close on the horizon, and I expect that for decades to come, there will be more prosaic accomplishments in simpler systems built at that scale.
Question 8: Would it be accurate to say that the tools that you are creating could be directly applicable in creating Drexler’s molecular assemblers?
Given many orders of magnitude more computing power than we have today, it would in principle be possible to use our software tools to model systems such as Drexler proposes. But the computational power required to do that will not be available for many decades, if you look at a Moore’s law curve.
Question 9: I noticed on your website a link to Ray Kurzweil’s The Age of Spiritual Machines. What is your opinion of Mr. Kurzweil, and what is your opinion of his book?
Ray is a very smart guy. He obviously has great credentials and background; he has done constructive work in the area of computer science and AI. The Age of Spiritual Machines is a very thoughtful book, and is one vision of a possible future. But predictions being what they are, it is very tough to endorse some of his prognostications. I think that whenever you start to look that far ahead into the future, there are many different paths that the world can take. His vision is an interesting and thought provoking one, but is it guaranteed to come out? Certainly not.
Question 10: Has Apex been able to perform any simulations of microscopic sensors or any other molecular devices?
No, we haven’t tried that at all. I think that it is quite viable to use our software to model relatively small systems like nanotube based sensors, given sufficient computer power. However, when you get into logic circuits, you get into other specialized types of modeling you need to do, in order to model types of circuits. There is a whole category of tools called SPICE tools that are used in the design of macroscale circuits. But we aren’t putting any effort into that.
Question 11: What are your plans for the future?
We are here trying to build a business, so we are working on a software product, and talking to prospective customers, and their needs. We are making sure that we can build a great business founded on real customers and real needs but also be centrally positioned to help facilitate a nanotechnology revolution.
