Singularity Institute

Justin Rattner discusses the future of Moore's law and Intel's methods of microchip design.

Justin Rattner is vice president and chief technology officer at Intel. He is also an Intel Senior Fellow and head of Intel Labs. In the latter role, he directs Intel's global research efforts in microprocessors, systems, and communications including the company's disruptive research activity.

Singularity Institute: How long can Moore's law continue?

Justin Rattner: Moore's Law is a belief structure, more than it's a physical law. Moore's Law is not directly governed by fundamental physics. It's really this observation that chip technology improves at this relatively constant rate.

Now, it happens because of competitive forces . If you think the guy down the street is going to be able to put twice as many transistors on his or her chip, then you are -- in the coming 12, 18, 24 months -- you have quite an incentive to make an investment to improve your technology and not get left behind.

So there's a competitive force at work here that drives people to make the investment and I'm not just talking about the financial investment, you know, the investment in time and intellectual horsepower to make it happen.

But we continue to be bullish on the rate of improvement in Moore's Law because we can look out about a decade and see the path that we'll take to continue those advancements. And that's no different that the kind of visibility we had a decade or three decades ago. We can always see about 10 years ahead. And so we knew how to get there and that decade horizon has remained constant for about 40 years. So, it's reasonable to expect that, if we can see out 10 years today, there's probably another 10 years beyond that.

SI: Where will computation be 10 years from now?

JR: Actually, Gordon didn't say anything about performance. Gordon was talking about how many transistors per chip. And more recently I think we've come to appreciate that there are factors that limit how fast we can make those transistors run, in particular, the thermal limits that have really kind of come up and beaten us in the butt. We have to be much more conscious about how to get the heat out. But we're continuing to build better transistors and it's really this balancing act between how fast you run those transistors and how much power, how much heat you can take out, take off the chip.

So we can run them a lot faster if we could cool them, but I think increasingly we want those systems to be portable, to be mobile and running off a battery and we're just limited by heat and energy in terms of speed.

SI: What happens to computation when we reach the limits for transistors?

The end is a long way off and, I think, we have evidence of this already that we will encounter physical limitations at various points. We just passed through one but we're still very creative, very inventive and we overcame those obstacles and I think we'll continue to do that.

The kinds of devices we'll be building, the kinds of transistors, the kinds of wires, may look very different 10 years from now or 20 years from now, but I think we'll still refer to the accelerating improvements as Moore's Law.

SI: What's terascale computing?

JR: We're more or less at the point where we can pack as much computing power in a single chip as would have filled a small-size house about 10 years ago. I actually worked on one of those machines. The first of those big machines that reached a trillion operations per second and now, about a dozen years later, Intel and other companies are getting ready to put chips on the market that pack that much power into a single chip, which I think just illustrates the power of Moore's Law as dramatically as you're likely to find.

When you pack that much power in a single chip you're really going to enable a new generation of applications. People will think about using computing power in ways that just seemed impossible and impractical in the past and we look at being able to build recognition systems that rival or perhaps even exceed human performance.

We talk about the ability to mine information, to search through terabytes, petabytes, exabytes of information at incredible speeds, to do it just - to mine not just text but to mine images and to mine video. And we talk about the ability to synthesize virtual worlds and virtual environments that, again, from a sensory point of view are likely to be every bit the equal of the physical environment in which we exist.

SI: Does the issue of the singularity ever come up in discussion at Intel board rooms?

JR: When I was working on my keynote address for the Intel Developer Forum last August, we, in fact, decided to look forward 40 years, because Intel was celebrating its 40th anniversary, you know, the singularity just seemed like a natural way to frame the conversation. And I think that most of the technologists, we don't have too many philosophers at Intel, but we do have a lot of technologists [laughs], I think most of the technologists in the room thought it was quite a reasonable proposition, that some time in the next 40 years we'll reach that point where human intelligence and machine intelligence equal one another and that machine intelligence quickly move beyond human intelligence.

Some of the other aspects of the singularity, like the ability to move human memory and thoughts and emotions into a mechanical substrate, I mean, we could only speculate on and we weren't sort of weighing the moral questions of that too heavily. But, I think we're quite willing to accept the premise that this point of equivalency is definitely within reach in the future or decades to come.

SI: What's the most amazing thing that's being at Intel today?

JR: You know there are a number of technologies that sometimes, you kind of have to pinch yourself and say, "Am I really looking at this?" You know one that's gotten people tremendously excited is this notion of wireless power transmission. In fact, I think we were all taken aback by how that idea would resonate with the average person.

Once we talked about our work publicly, we were just bombarded with emails and cards and letters and all of that with people interested in how we plan to do it and people concerned about the environmental hazard, the human health hazard that it might represent. Fortunately it uses magnetic resonance so it doesn't... if you could survive an MRI you could probably stand in a wireless power field without effect. [chuckles]

That's one technology and I think, again this notion of things being portable, mobile, go anywhere. The idea of never having to charge a battery, never having to replace a battery but being able to draw power from the ambient is one that really blows people away.

Something we're working on even further out, which we primarily work on in our Pittsburgh lab is this notion of programmable matter, which is the idea of packaging tiny bits of computer intelligence and memory in glass spheres, maybe just a few hundred microns in diameter and giving those individual spheres the ability to move relative to one another using electrostatic forces.

You can also give them the ability to change colors and, of course the ability to communicate with one another. So you can imagine vats of this stuff that you could literally program at an instant to take any shape and even to exhibit a variety of behaviors.

Your phone in the future might sit in your pocket as just sort of this slab of programmable matter and then when you take it out and bring it up to your ear it reconfigures or shifts shape, if you prefer the scientific analogy...or the science fiction analogy, it shifts shape and becomes a phone or maybe it turns into a display.

So programmable matter seems like an impossible idea that we as humans could create something matter-like. But in fact, we've done it at the centimeter scale. We've done it in two dimensions at the millimeter scale and now we're working on doing it in three dimensions at the millimeter scale and beyond that we'll try to go into the micron range with these tiny bits of programmable matter.

SI: Does Intel do work with quantum computing?

JR: We don't do much work in quantum computing. And the analogy I often give is; we don't do a lot work in optical computing either. I tend to group optical and quantum together to a certain extent in that, for problems that lend themselves to optical computers, there's nothing that will equal an optical computer.

The problem is not many problems neatly fold into that form. And I think the same is true, at least from what we know today, with quantum computers. They do some problems, particularly these cryptographic problems. They're unequal. But for much of what we think of as computing and in the future we think of as intelligence or machine learning we really don't know enough today to say whether quantum computers will really be appropriate for those.

SI: Why are you participating in the Singularity Summit?

JR: When I got the invitation, it just seemed like a great opportunity to share some of the thoughts we had at Intel and share our recent progress and some of the things we're thinking about for the long term with the community of people that have learned about the singularity or are curious about the singularity and perhaps want to test it a little bit to see if what they've read and what they've heard really holds up.

I think, as a working technologist and somebody who gets up every morning and tries to make sure that the laws of accelerating returns are working and working well, this seemed like a great opportunity to meet others in the community with the same interests.

SI: Why should someone support SIAI?

JR: We often find ourselves unprepared for events. We're frequently just overtaken and you know, "Gee, I didn't expect that." Or you know, "I hadn't heard about that." I think we sort of go through our daily lives often oblivious to these tectonic events that are taking place around us.

What's neat about the singularity is that it's something we can readily identify and we can begin to think about how it's going to impact us, how we want to prepare for it, what work we want to do, what the social and moral issues might be associated with that. In my way of thinking, that's really a unique opportunity that we're rarely afforded in our daily lives. The opportunity to sort of anticipate what we expect will be a truly remarkable event.

SI: What does the Singularity mean to Intel?

JR: Intel as a company I think, has been driven much of its life by the fundamental power of Moore's Law, which is a prime example, perhaps ÂŤtheÂť example of the law of accelerating returns so, so much of what we do, and we deal with it every day is the result of knowing that in a couple of years we're going to be able to put twice as many transistors down on a chip or maybe we're going to be able to move photons around on a chip as opposed to electrons.

So the underlying forces behind the singularity are really the same ones that drive Intel as a corporation. I think we understand that our success has been fundamentally rooted in that very important law and that our future still rests largely on the continued advance of that technology.