Moore's Law at 40: Part 4
- 2005-May-13
These captions and transcript were generated by a computer and may contain errors. If there are significant errors that should be corrected, please let us know by emailing digital@sciencehistory.org.
Transcript
00:00:01 ...Electronic Materials Business Group.
00:00:03 His efforts in this arena have resulted in the development of quite innovative chemicals and materials focused on the microelectronics industry.
00:00:12 Rohm and Haas has been and continues to be heavily involved in chemicals and technologies for preservatives, for adhesives, for sealants, for paints.
00:00:23 But over the last 10 years, through the directions of Mr. Gupta, very significant advances have been made in the development of pads and slurries for chemical mechanical polishing operations,
00:00:37 for photoresist materials for 193 nanometer lithography, and for polymeric materials for liquid crystal displays.
00:00:45 Mr. Gupta received a Bachelor's Degree in Mechanical Engineering from the Indian Institute of Technology, MS in Operations Research from Cornell University, and an MBA in Finance from Drexel University.
00:00:58 He joined Rohm and Haas in 1971, and over the next 28 years has had a succeeding array of responsible positions, including financial manager of the UK subsidiary, business director for plastics in Europe,
00:01:14 vice president of Rohm and Haas, and director of the Pacific region. He became chairman and CEO in 1999.
00:01:22 Mr. Gupta serves on the board of trustees for Drexel University. He's a member of the boards of Tyco, Vanguard Group, the American Chemistry Council, and the Chemical Heritage Foundation.
00:01:36 I'm pleased to present Mr. Raj Gupta, whose talk is entitled Electronics and the Evolution of the Chemical Industry. Raj.
00:01:46 Thank you, David, and good morning. I guess the start of the afternoon.
00:02:13 I just got the last 20 minutes of Elsa's presentation, and I can tell you I'm not a scientist, I'm not a chemist, but it was very impressive.
00:02:22 This is something I know about lithography and the challenges that we are facing.
00:02:26 So actually, my job this morning is to give you somewhat of a different perspective than some of the speakers you heard this morning or the ones you hear this afternoon.
00:02:36 As I said, I'm not a scientist, I'm not a chemist, and I certainly don't pretend to understand the details of semiconductor and nanotechnology.
00:02:44 However, I do lead a company that is based purely in science and chemistry, and electronic materials, as you just heard from David, is an important part of that.
00:02:54 And I'm a student of history and observer of trends.
00:02:58 Therefore, I'd like to take a few minutes to provide you my thoughts, somewhat offbeat perspective, about how electronics industry affects the evolution of the chemical industry.
00:03:10 In fact, the answer really is, it already has. Not small time, big time.
00:03:18 The more truthful thing is that both electronics and chemistry, what my company calls electronic materials, are today deeply intertwined.
00:03:28 And the pace of new technology, new products, new ideas to generate is getting faster and faster all the time.
00:03:40 How did this happen?
00:03:42 Chemistry evolved for sure, science evolved in general, and I think I also touched on the multidisciplinary approach that will be absolutely essential to go the next steps.
00:03:52 People remain curious, like Gordon and others, and both chemical and electronics industry led the way.
00:03:59 We have discovered new materials.
00:04:02 We drove deep into the very structures of materials, then learned to manipulate them in brand new dimensions.
00:04:10 And today the global chemical and electronics industry represents roughly $2 trillion in sales, bringing ever-changing technology to the marketplace.
00:04:22 Here's a very good example of old and new.
00:04:25 One of the first portable radios.
00:04:29 Being tuned by the father of frequency modulation or the FM radio, Edwin Armstrong, and today's newest portable music player, the iPod.
00:04:41 You probably don't recognize this guy on the beach.
00:04:44 Edwin, he is Edwin Howard Armstrong, one of the great engineers and innovators of the 20th century.
00:04:51 Most inventors of note are known for one, two inventions.
00:04:56 Mr. Armstrong had three.
00:04:58 He discovered how to make amplifiers, radios, transmitters.
00:05:02 He figured out how radios could tune in one station and tune out all others.
00:05:07 And he discovered frequency modulation or the FM radios.
00:05:11 That's pretty impressive.
00:05:14 And lithography.
00:05:15 It began with stones and wax and pigments and oils, evolved into sophisticated chemistry, and then with a change in light source and technology,
00:05:24 really became the enabling power for sharing thousands of images, ideas, and computations from your own laptop in the office or, for that matter, anywhere in the world.
00:05:37 So what's next?
00:05:40 How about intelligent, life-saving undergarments that can transmit information about heartbeats and breath sounds?
00:05:49 Or intelligent textiles that can change the color of handbags, shoes, even cubicle walls and point-of-purchase signs?
00:06:00 Just to give you an example.
00:06:02 More seriously about the textile industry and the impact of electronics industry and chemistry on it.
00:06:09 Most would say the textiles have been a fraying industry.
00:06:13 More than two-thirds of its workforce has disappeared in the United States in the past decade.
00:06:18 Plants have been shut down, more or less.
00:06:20 Yet, according to an article in the May 23 edition of Forbes magazine, there are 345 scientists at work in North Carolina reinventing this industry.
00:06:31 And they're doing it by using electronic material science and science and chemistry.
00:06:36 The results are startling.
00:06:38 This is the beginning of a resurgence of the U.S. textile industry.
00:06:42 This time concentrating on high-technology, flexible fabrics and smart clothing.
00:06:49 Good enough to attract investment from countries like Japan, Switzerland, and Israel.
00:06:54 They have been in a company called Nanotex, making carbon-based polymer molecule one-millionth the size of grain of sand to wick away moisture and resist stains.
00:07:05 A technology that can already be found in Brook Brothers and Eddie Bauer shirts.
00:07:12 Moving along, and just across the river from here in the New Jersey Institute of Technology, Dan Watts,
00:07:18 the executive director of the New York Center for Environmental Engineering and Science and Panasonic Chair in Sustainability,
00:07:25 and his team are putting together nanomachines into paints and coatings and making them smarter.
00:07:32 The technology in these paints can sense deterioration and breaks in the surface of coating
00:07:38 and then make repairs without any human intervention.
00:07:42 Self-curing coatings.
00:07:48 And there's every sign that the electronics market and the markets that support it will continue to grow.
00:07:55 Today my own company is focused on serving a segment of the market that represents 30 billion in sales.
00:08:02 That's as large as the entire global agrochemical industry as we know it today.
00:08:06 That's taken over 100 years to develop.
00:08:09 And it's sustained more than 10% growth, and we expect that growth to continue.
00:08:14 This actually does not even include all the plastics materials that go into electronics.
00:08:19 You can multiply it by that figure by many, many times over.
00:08:23 So the electronics industry has created a huge demand for chemicals, innovation, and science around the world.
00:08:32 And we have been fortunate at Romanhaas to have participated in the evolution of this science,
00:08:38 not only in terms of photoresist, lithography, and ancillary, but also CMP pads and slurries,
00:08:45 as well as the packaging technology and the circuit board technologies, and we continue to invest in it for the future.
00:08:52 In fact, electronic materials today represent about 20% of company sales and use,
00:08:58 or at least we invest about 40% of our global R&D dollars in electronics.
00:09:07 The point I want to make is a very simple one.
00:09:10 Chemical science may well have been the fountainhead of today's electronics industry,
00:09:14 but today that industry is giving back and making tomorrow's generation of chemistry-based products even more efficient,
00:09:22 useful, and, in my opinion, just downright amazing.
00:09:27 That gives me great hope for the future of both disciplines.
00:09:31 And as an aside, the future of Romanhaas Company relies on it, too.
00:09:37 I'm convinced that these sciences will remain successfully linked and evolve together for many, many years to come.
00:09:45 Gordon himself noted last year when he was awarded a parking medal right here in the city of Philadelphia
00:09:51 that chemistry was the fountainhead for the global semiconductor industry.
00:09:55 I would add in my humble perspective that the future of chemistry will rely heavily
00:10:00 and happily on the strength of innovations in the electronic materials industry.
00:10:05 I really appreciate your time.
00:10:06 There's always, before lunch, a hard time to giving a speech.
00:10:10 Probably the only worst time is before dinner or after dinner.
00:10:13 I didn't want to make it too long.
00:10:14 It was somewhat offbeat that you expected to hear from me,
00:10:18 but I'll be happy to really engage in any discussions or hear your comments about this.
00:10:22 This is a great undertaking, and the fact that Gordon himself is here to share the day with you
00:10:28 speaks volumes about the importance of this subject to our industry.
00:10:33 Thank you very much.
00:10:34 As I said, I'll be happy to engage in some discussions with you.
00:10:39 Applause.
00:10:48 Questions for Raj?
00:10:54 Question being asked.
00:11:24 Question being asked.
00:11:54 Question being asked.
00:12:24 Question being asked.
00:12:54 Question being asked.
00:13:24 Question being asked.
00:13:55 Question being asked.
00:14:09 I think this is a topic that every company and every industry is dealing with, and it's being globalized.
00:14:17 I think I'll give you my simple but pragmatic approach to this.
00:14:23 I think the idea that you can only take your order to the top of the list
00:14:29 is a flawed point of view, as we have seen,
00:14:37 because in China and India and all over the developing world,
00:14:41 they want today's or tomorrow's technology.
00:14:43 They don't want yesterday's technology anymore.
00:14:45 And it doesn't just apply to electronics.
00:14:47 It applies to every other field.
00:14:49 So we are all dealing with this.
00:14:50 I think the way we are trying to deal with it at Romanhaas is first and foremost priority that we have
00:14:57 is that the employees we have everywhere in the world are loyal, well-trained.
00:15:03 They understand the importance of our technology,
00:15:06 because the bulk of the technology leakage that takes place,
00:15:09 doesn't matter where it happens in the world, is through your own employees.
00:15:12 I think instilling in them that spirit and also having checks and balances in the system
00:15:20 and as well as fighting the cases.
00:15:22 We have been very fortunate and pursued it in China as well.
00:15:25 Successfully, I might say, that when there have been infractions, we go after it.
00:15:30 So I don't have a magic bullet answer to this.
00:15:33 In some cases in the past, and we continue to do that,
00:15:35 some of the parts of the processes are done outside those countries where you may have higher risk,
00:15:41 and we move the intermediates and process them.
00:15:43 But we have taken the view if we are going to move technology globally and move it fast,
00:15:49 we need to have other mechanisms of protecting our technology as well.
00:15:53 I think I'll give you an idea.
00:16:05 I joined, as David mentioned, 35 years ago at Roman House.
00:16:09 The reason I joined was my dad was alive.
00:16:11 He wanted me to get back to India.
00:16:13 That was 38 years ago.
00:16:14 But I'm glad I didn't.
00:16:16 But I have been back many times.
00:16:18 But I think what we have done, just to give you, when I started at Roman House,
00:16:22 we basically had one research facility located outside of Philadelphia,
00:16:28 and we had three or four technical centers around the world.
00:16:31 Today we have 35 centers of excellence in technology around the world,
00:16:36 about five or six significant R&D centers as well as significant technical service centers.
00:16:43 And the driver for that is relatively simple.
00:16:45 I think the pace of innovation has accelerated to a point,
00:16:49 and the adoption that customers want and the technical support they seek has to be on the spot,
00:16:55 closer to the marketplace rather than far away.
00:16:58 So I think that's been one driver for us.
00:17:00 The second driver for us has been just accessing the talent pool that's out there
00:17:04 in other parts of the world as well.
00:17:06 I think the combination of those two has driven us.
00:17:09 I think China Center is going to be a major investment for us,
00:17:14 about $30 million in first step and probably doubling that in the next step,
00:17:18 and we expect that in the next two, three years we'll have 300 scientists working there,
00:17:22 not only electronics but across the Roman House product range.
00:17:27 And we don't think of China Center just as providing products and innovation for Chinese market,
00:17:33 but really hopefully would be very well linked into a global technology center for Roman House.
00:17:39 And, in fact, John House is here.
00:17:40 He's not coming with us to China, but I'm traveling at the end of June, middle of June,
00:17:45 the House family, and we will be doing the groundbreaking there,
00:17:48 and then hopefully when our directors go out there next, middle of next year,
00:17:53 we'll open a technical center there.
00:17:58 Yes, it's not very far, actually, from there.
00:18:04 I think in a recent issue of the Business Week,
00:18:06 you talked about the disappearance of the chemical industry from the U.S.A.
00:18:12 and global solutions, and we would all be glad if that was produced everywhere around the world,
00:18:23 but looking from an American perspective, how do you view that?
00:18:30 Actually, you know, it was startling.
00:18:32 Actually, it came out on May 2nd, and I was having my board meeting in Philadelphia,
00:18:36 and I shared that with my board of directors,
00:18:38 which basically the tenet of this story was that in the next 10 years,
00:18:44 it'll be $50 billion-plus chemical plant investment in China and only one in the United States.
00:18:54 And I think my perspective on this is that it's really driven by a number of things.
00:19:00 First, the growth rate in the U.S. market is relatively small,
00:19:04 which can be met from the domestic production that's already out there,
00:19:09 plus the fact that more and more imported products, finished products, are coming from Asia.
00:19:13 That means it further depresses the rate,
00:19:17 so the installed base of the capacity of petrochemicals in the U.S. and Western Europe would still remain big.
00:19:23 But I think there is another message here,
00:19:25 and part of it is related to natural gas policy of the U.S. government making U.S. industry very uncompetitive,
00:19:32 but it's a bigger issue than chemical industry can solve, but I think it needs attention.
00:19:36 But aside from that, I think, to me, it's a natural evolution of things,
00:19:41 that things that are commoditized will have to be either close to the source of labor if labor is the driver,
00:19:50 or closer to raw materials where raw materials are the driver.
00:19:55 And clearly, in the petrochemical industry, drivers are raw materials,
00:19:59 and that's where you would see, you are already seeing the petrochemical industry moving to Middle East big time,
00:20:05 Malaysia, and eventually, in the next 20 years, I'm sure Russia would become a major,
00:20:10 once things settle down there, Russia would be a major source of petrochemical raw materials.
00:20:15 So I think that that is not necessarily bad or wrong, but I think it does mean something else.
00:20:20 It means for U.S. scientists, engineers to stay engaged, and our ability to attract talent will depend on innovation.
00:20:30 We don't have to. You talked about nano. I talked about smaller devices.
00:20:34 The amount of stuff we will need for making these things is going to go down every year, as far as we can see.
00:20:41 And that's not where the big innovation is. Innovation is going to be, how do you make these things work at nanoscales?
00:20:49 How do you find the science and innovation for the new generation of technology to come?
00:20:54 So I think in terms of knowledge workers and innovation, I think U.S. has tremendous advantage, and we need to continue to leverage it.
00:21:02 So I think that's kind of a long answer to your question, Arnold, that while things are changing,
00:21:09 I don't think it's a bleak outlook for innovation or science, particularly in electronics and other areas,
00:21:16 for the U.S. chemical industry and for the U.S. electronics industry.
00:21:24 Thank you very much, and enjoy your lunch.
00:21:36 Arnold, did you have anything that you wanted to add prior to break for lunch?
00:21:43 Third floor.
00:21:47 If Nick Mokoff is in the room, could he come up front, please?
00:22:03 I have the honor of being the chairman of the Eastern Technology Council,
00:22:07 and the unindicted co-conspirator of Arnold Thackeray and others here at the Chemical Heritage Foundation.
00:22:13 I will say I just learned chemically the results of stress on the Skook Hill yet again, racing down here.
00:22:21 We at the Eastern Technology Council pride ourselves on providing contacts, capital, information,
00:22:27 and celebration for leaders in the knowledge industry.
00:22:31 And we couldn't be more pleased than to be partnered up with the Chemical Heritage Foundation,
00:22:36 which is such a central player, and really an under-recognized player.
00:22:41 One of our missions for 2005 and 2006 is to try to bring more celebration
00:22:46 and more information about the Chemical Heritage Foundation.
00:22:49 We're delighted to be coupled up here.
00:22:51 Staff has very competently informed me that my primary goal is to keep things moving on schedule,
00:22:57 and that we have just about the brightest, best-educated audience we could possibly have here,
00:23:02 so there's no need for me to provide too much redundant information.
00:23:06 These super-famous and deservedly celebrated speakers have illustrious bios,
00:23:13 almost legendary bios, that are in your materials.
00:23:16 We hope you've had a chance to read them.
00:23:18 I will just remind you that Carver Mead is literally one of the seminal figures in microelectronics.
00:23:25 He's a leading commentator on the semiconductor industry
00:23:28 and something of both a living piece of history and a historian about semiconductor technologies.
00:23:35 He's the man who coined the term Moore's Law,
00:23:39 and he's been invited by the Chemical Heritage Foundation to do some personal reflections
00:23:46 and to look back on his 40 years of experience in microelectronics
00:23:51 to talk a little bit about how the life of Moore's Law has gone
00:23:56 and to reflect a bit on the importance of material sciences and chemical sciences to the microelectronics story.
00:24:04 Please join me in welcoming Carver Mead.
00:24:07 Thank you.
00:24:15 Thanks.
00:24:18 I met Gordon in 1960, and it's a fun story, but I'll tell it some other time.
00:24:31 At that time when we gave talks, we illustrated them with three-and-a-quarter-by-four lantern slides.
00:24:39 But Gordon's paper, of course, the story of Moore's Law, really starts in 1965,
00:24:45 and by then we had a new technology.
00:24:48 It was called 35-millimeter slides.
00:24:51 In the process of going back and looking at the material of how Moore's Law got started and how it evolved,
00:25:01 I was going through all these 35-millimeter slides,
00:25:04 and for some reason it just didn't seem right to come here and talk about that history with a PowerPoint presentation.
00:25:13 So I asked the Chemical Heritage people if they could come up with what was then the new technology,
00:25:21 and they were able to find a 35-millimeter projector.
00:25:24 So what we're going to have here this afternoon are the original slides that I dug out of the 40-year-old stuff,
00:25:33 and I thought that since it was about history, you'd like to see it that way.
00:25:39 This was my first introduction to Moore's Law.
00:25:44 It's a slide you already saw.
00:25:46 Harry Sello had it in his talk.
00:25:49 And Gordon had observed this evolution of the number of components on the chip,
00:25:58 and you can see just looking at this set of chips,
00:26:03 these were all at Fairchild before there was an Intel.
00:26:11 And Gordon at the time was in charge of the Fairchild R&D lab,
00:26:17 and he had asked me in 1960 if I would come up and consult with Fairchild,
00:26:22 and I had been going up there from Caltech every week,
00:26:26 and typically what I'd do is I'd fly in on Thursday night,
00:26:30 and then Gordon and I were both at the time early risers,
00:26:34 so I would get into the lab at 730 or 8 and have a half-hour chat with Gordon
00:26:41 about what was important and what I should look into and if I could be helpful in any way.
00:26:45 And then I'd go out and spend the day with the people in the lab,
00:26:48 and then about 5 o'clock I'd come back to Gordon's office,
00:26:52 and we'd have sort of a decompression session, and we'd talk about general things.
00:26:58 And every once in a while Gordon and Betty and I would go off to dinner.
00:27:04 And it was one of those sort of decompression sessions.
00:27:09 Gordon said, you've been doing a lot of work on electron tunneling, right?
00:27:13 I said, yeah, I've been doing a lot of work.
00:27:15 And now that happens when things get really thin, doesn't it?
00:27:20 I said, yeah, yeah, that's what it's about, all right.
00:27:23 And he said, doesn't electron tunneling limit how small we can make a transistor?
00:27:31 I said, well, yeah, it certainly would.
00:27:34 He said, well, how small is that?
00:27:37 Now this is a very typical Gordon Moore kind of thing.
00:27:42 Every single question was absolutely obvious, and I hadn't thought about it.
00:27:49 So that prompted me to do some thinking.
00:27:55 And it turned out I had coming up, they used to have these workshops
00:28:04 where you could get everyone who was working on a,
00:28:07 or at least the people who were doing leading-edge work in an area,
00:28:11 you could get them all in a room like this, and we'd have talks and rump sessions.
00:28:16 And there was one at Lake of the Ozarks where we had a bunch of people involved with device physics.
00:28:22 And I had been thinking about this question of Gordon's.
00:28:26 How small could you make a device?
00:28:28 What would happen?
00:28:29 And in those days, there were all kinds of papers saying if you made the devices smaller,
00:28:34 the cosmic rays would get you.
00:28:36 If you made the devices smaller, you'd have so many of them on the chip,
00:28:40 the whole chip would heat up and melt.
00:28:42 We actually heard a little bit about that from Pat.
00:28:46 And all of these sort of prophecies of doom.
00:28:50 And so I started thinking about it, and the more I thought about it,
00:28:55 the more it seemed like that wasn't really quite right.
00:29:01 So it turns out I have the slide here from my first talk that was given at this workshop
00:29:09 in, I think it was 67 or so.
00:29:14 And I'm sorry that this is a little bit busy, but let me just go through it because it's important.
00:29:21 The essence of this slide is, yes, I know it's not quite right,
00:29:26 but suppose you just did the simplest, stupidest thing you could think of.
00:29:32 And you took the MOS transistor, which is shown here,
00:29:36 and you scaled down all the dimensions by the same factor.
00:29:42 What would happen?
00:29:45 Well, you can't quite do that without thinking through some other things.
00:29:51 In the first place, if you just do that and don't change any of the doping concentrations in the silicon,
00:29:56 then the depletion layers around the source and the drain will overlap with each other
00:30:03 and the transistor will go into what's called punch-through.
00:30:06 It just won't be a transistor anymore, so you have to increase the doping in the substrate.
00:30:10 But you can do that, so that's okay.
00:30:13 The other thing is, of course, if you just do that and don't change the voltages,
00:30:18 then the electric fields start going way up and all kinds of bad things happen,
00:30:22 which I'll talk about in a little bit.
00:30:25 So you have to scale the voltages down too.
00:30:28 But you can do that.
00:30:30 And so this was the first sort of just dumb, simple, straightforward scaling.
00:30:38 Suppose we scale down the voltages and the dimensions and scale up the doping concentrations
00:30:46 such that the depletion layers are the same fraction of the size of the device.
00:30:51 What happens?
00:30:53 Well, the calculation is very simple.
00:30:56 The current is just the charge under the gate, the charge in the channel,
00:31:01 divided by the transit time.
00:31:03 And if you keep the electric fields constant, then the transit time,
00:31:13 the velocity doesn't change, but the transit time gets smaller.
00:31:18 But you're not asking for the physics to go to any place that it isn't already.
00:31:26 So the striking thing is, of course, that the power per device,
00:31:32 instead of melting like the IBM people were saying,
00:31:37 it actually scales in such a way that the power per unit area is constant
00:31:42 as you do this scaling.
00:31:44 Well, that's good.
00:31:46 It means the chip doesn't have to melt, which is very good.
00:31:50 But the other thing is, if you just work out the switching energy per operation,
00:32:00 which is this power per unit speed as the inverse of that,
00:32:06 it goes like the cube of the scaling factor.
00:32:10 So the performance, the real metric of performance,
00:32:14 how much computation do you get per unit energy,
00:32:18 is going like the cube of the scaling factor.
00:32:21 That's huge.
00:32:23 And I did this thing over and over and over
00:32:26 because it was obviously a violation of Murphy's Law, big time.
00:32:31 It was way too good to be true.
00:32:34 Well, I went ahead and gave this talk,
00:32:37 and I had people come down on me, all over me.
00:32:42 I mean, it was like, you have to be a stark raving mad.
00:32:45 Don't you understand that you can't do that because this, this, this, this, this.
00:32:51 And it was not a fun time.
00:32:57 Well, when we got to the same workshop the following year,
00:33:03 then there were three other talks.
00:33:05 There was one from Bell and one from IBM
00:33:07 and one from someplace else about scaling.
00:33:09 And they'd come to pretty much similar conclusions.
00:33:13 So that was good.
00:33:14 It means that now people were starting to think seriously about that.
00:33:19 So then we had to work out what Gordon had asked me to do originally,
00:33:24 not just how does it scale, but how far can you go.
00:33:26 And that took longer.
00:33:29 And eventually we published a paper, 71,
00:33:34 which we had worked out what we believed was a perfectly doable transistor.
00:33:42 And we gave two examples.
00:33:44 This is the larger of the two.
00:33:46 That's the only slide I could find,
00:33:48 which showed that you could do a quarter micron channel length
00:33:51 without any problem at all.
00:33:53 Now, remember, this was a time when the channel lengths that people were using
00:33:58 were, as Gordon says, in the range of mils, not microns.
00:34:07 And we also gave another example of a transistor which was perfectly workable,
00:34:12 which was 0.15 microns,
00:34:15 which is just now sort of workhorse dimensions in the industry.
00:34:22 So this was looking a long ways ahead.
00:34:26 We did not talk about the time scale with which this was going to happen,
00:34:33 except we were – my friend Dave Ferry at Arizona gave me this slide.
00:34:46 When he gives talks about it, he always makes fun of me.
00:34:50 And this is a plot of when doomsday is going to come as a function of year.
00:35:00 So different people have published the limits
00:35:05 of how small you can make transistors down through the years,
00:35:08 and Dave has gathered them all up.
00:35:11 And here are the various proposals.
00:35:14 Ours actually – he has ours a little – ours is actually down about here.
00:35:19 He took the larger of the two transistors we talked about instead of the smaller one.
00:35:24 So he said, no, you're no different than anybody else.
00:35:27 You're just way off the curve by, what, 20 years or something.
00:35:33 So that was always fun to have somebody poking fun at you
00:35:38 because it does help raise people's consciousness.
00:35:45 In that paper, we had this figure.
00:35:49 And this was as far – you notice there's a gap in here.
00:35:55 These are Fairchild points. Those are Intel points.
00:35:58 And I guess nothing had possibly happened when you guys were in the middle, right?
00:36:09 Anyway, that's the first of Gordon Moore's plots that I was able to actually go.
00:36:16 And I remember going to Gordon.
00:36:18 Gordon actually made these plots himself on this graph paper.
00:36:23 And he always apologized because he said, you know,
00:36:25 Intel's really a small company and you can't afford fancy graphics and stuff.
00:36:31 And so I would go and get the latest one.
00:36:34 So just as this paper was going to the press, I went and got the latest points.
00:36:37 As I recall, there were a few that you had just put on there when you gave me this particular plot.
00:36:44 And so that's the one that went into our paper.
00:36:47 And we said that at that time we had the idea in our head
00:36:54 that as you made the feature sizes smaller,
00:36:57 you're going to be more and more susceptible to dust and things
00:37:02 and the yield was probably going to be very hard to keep the same.
00:37:07 So the assumption was you were going to not make the chips bigger.
00:37:11 You were just going to make the devices smaller.
00:37:14 So this is, if you read components per unit area, it makes more sense today.
00:37:21 It's hard to see now how you could think that.
00:37:26 But at the time, that was pretty much the way we were all thinking,
00:37:30 that the yields were determined by things that go wrong,
00:37:36 dust and defects in one thing or another,
00:37:39 and there are obviously more of those because people had profiled
00:37:42 and there's more small ones than there are big ones.
00:37:45 So you probably had to be careful about making the chips too much bigger.
00:37:49 Now, of course, what's happened is the process people of the world
00:37:54 have not only got good at making the feature sizes smaller,
00:37:58 but they've gotten really, really good at cleaning things up
00:38:00 and making the yields better.
00:38:02 So it has been possible to make the chip bigger at the same time.
00:38:08 So that's why you should probably read this as devices per square centimeter,
00:38:12 roughly speaking.
00:38:14 And we said if you believe you can get to 0.15 microns,
00:38:21 and by then we said you're going to start to have leakage through the gate oxide
00:38:26 and you're going to start to have leakage through the junctions,
00:38:29 and I'll show you how that works.
00:38:32 But you can still make 10 to the 7 devices on a square centimeter size chip,
00:38:41 which was in those days just ridiculous.
00:38:45 Today it sounds like a pathetic little tiny chip,
00:38:49 but in those days that was an absurd thing.
00:38:52 So that was great.
00:38:54 And you notice this plot is due to Gordon Moore.
00:38:57 It doesn't say this is Moore's Law.
00:38:59 It says this plot was due to Gordon Moore.
00:39:02 So then what happened is there was a lot of resistance to this notion,
00:39:08 and so I started on sort of a personal crusade to go around the country
00:39:11 and try to convince people that it really was possible to scale devices
00:39:16 and get better performance and lower power
00:39:19 and all the good things that come with that.
00:39:22 And it was an uphill battle,
00:39:24 and every time I'd go out on the road I'd come to Gordon
00:39:27 and get a new version of his plot.
00:39:32 And that was so I managed to find a lot of these.
00:39:38 So here's the next one,
00:39:42 and notice he's still making that on his graph paper.
00:39:48 And that gets us up through 76.
00:39:52 So it's still going, and by the way,
00:39:54 the slope here has been commented before
00:39:56 as about a factor of two per year in number of components.
00:40:01 And then a little later,
00:40:04 Intel wasn't such a small company anymore,
00:40:07 and so there were some same points a little further along.
00:40:13 And at this point Gordon had already foreseen the fact
00:40:16 that what had happened is there had been sort of a pent-up capability
00:40:22 that hadn't really been fully exploited,
00:40:25 and people were sort of rushing in
00:40:28 as sort of a gold rush into this new integrated territory.
00:40:32 And so things were going faster than could be sustained forever.
00:40:39 And so probably this evolution curve
00:40:42 wasn't going to go a factor of two every year.
00:40:46 It was going to slow somewhat.
00:40:49 And he showed it slowing here.
00:40:51 Actually, you can see it already that for what in those days
00:40:55 Intel called logic devices,
00:41:00 they didn't recognize that a microprocessor
00:41:02 was something other than just a logic device.
00:41:06 I think they've realized that now, that it's a special thing.
00:41:11 At the time it was a logic device like something.
00:41:17 And already here you can see that the slope of those squares
00:41:22 has changed a little bit.
00:41:24 And so as far as I know,
00:41:26 Gordon is the only one who's pointed out the fact
00:41:30 that his curve had a break in it,
00:41:33 because what happened is then, of course,
00:41:37 the Intel marketing people got hold of this idea.
00:41:41 And it's not quite the same style.
00:41:46 And, of course, the world starts at 1971 now,
00:41:51 and the fact that the curve came down here
00:41:53 on a factor of two different slope
00:41:55 is carefully left off the slide,
00:41:58 because that's not good marketing.
00:42:01 And, of course, the text is bigger,
00:42:03 and it's much more...
00:42:06 But it's very nice, and this gets us out to the 80s.
00:42:11 And then this gets us a little closer to modern times.
00:42:16 And then this is very much the kind of thing
00:42:22 that Gordon had pointed out early on.
00:42:24 These were the so-called logic devices,
00:42:28 and these are the memories.
00:42:30 And, of course, a memory can be denser
00:42:32 because it doesn't have a lot of sort of special stuff in it.
00:42:36 It's very regular.
00:42:40 And you see a lot of the...
00:42:41 You saw the modern plot that's been put up
00:42:44 a couple times in this.
00:42:47 And it looks like it's going up at the end there,
00:42:49 and I think what's really happening there
00:42:51 is that the modern processors
00:42:54 are turning out to have a lot more memory on them.
00:42:57 And so they're getting...
00:42:59 They're going up from this curve to that curve.
00:43:02 I think that's what's really going on there.
00:43:04 But anyway, so this was a...
00:43:10 It was kind of neat to find all those plots
00:43:13 back in my catacombs.
00:43:16 But it's probably good if we talk a little
00:43:22 about what actually happens to a device
00:43:26 when you make it small.
00:43:29 And then I'll show you some numbers
00:43:31 and kind of how the predictions have gone
00:43:35 versus the actualities.
00:43:39 As I said before, the gate oxide you have to make thinner
00:43:43 as you make the device shorter
00:43:44 so that the gate still controls the potential at the surface
00:43:47 so it pulls electrons in or shuts them out.
00:43:51 And at some point you make a dielectric thin enough
00:43:56 and the electrons can tunnel through.
00:43:58 The wave function of the electron on one side
00:44:00 actually overlaps with the wave functions on the other side
00:44:03 and the electrons sort of leak off.
00:44:05 So it shows itself in a device as gate leakage.
00:44:08 It becomes more like a bipolar device
00:44:11 and less like an MOS device.
00:44:14 The other thing that happens, as I said before,
00:44:17 is to keep the device from punching through,
00:44:19 to keep the electrons in the source in the drain
00:44:24 and to allow the gate to shut off the transistor,
00:44:29 you need to keep the depletion layer
00:44:32 shorter than the channel length.
00:44:34 And to do that you have to make the number of doping ions
00:44:39 in the substrate here larger and larger
00:44:42 as you make the device smaller.
00:44:45 So those are the things that are going on.
00:44:48 So let's take a look at the actual numbers.
00:44:52 This is the actual numbers that have been...
00:44:56 I put together this talk about 10 years ago.
00:44:59 Some people asked me if I'd go back and do a retrospective.
00:45:02 At the time I wrote this it had been about 20 some odd years
00:45:06 since we had written the original 1971 paper
00:45:09 and it seemed like it would be good to look back
00:45:11 and see how well we did.
00:45:15 On these slides the big squares
00:45:19 are what we predicted 20 some odd years before
00:45:22 and the triangles and such
00:45:25 are the actual things that people did out in the factory.
00:45:30 And as you can see we're doing okay.
00:45:33 But the other thing I did at the time
00:45:35 is that over 25 years or so you actually have a trend.
00:45:40 So I just did a power law sort of fit to the trend
00:45:46 because that's what people have been doing
00:45:48 and typically when people get on an evolution curve
00:45:51 after they've gone for a while
00:45:53 they're going to keep doing that
00:45:54 until they run into some kind of a wall.
00:45:57 So it gives you some sort of modicum of basis
00:46:03 for extrapolating what's going to happen from there.
00:46:07 So these are all plots as a function of feature size
00:46:12 and this is a tenth of a micron here
00:46:16 and this is what the doping concentration is doing
00:46:21 and I'll show you what comes of this in a minute.
00:46:28 What happens as you increase the doping concentration
00:46:31 is it makes the depletion layer smaller
00:46:33 which is why you're doing it
00:46:35 but then the depletion layers get smaller
00:46:37 that means the electrons can tunnel through.
00:46:40 That shows itself as a junction
00:46:42 instead of being a nice rectifier like this one
00:46:44 where it goes out until it breaks down
00:46:47 it starts getting sort of soft
00:46:49 and then it becomes an ohmic contact.
00:46:52 In fact that's the way ohmic contacts work
00:46:54 and in fact in the process of doing all this work with Gordon
00:46:59 we worked out how ohmic contacts work.
00:47:01 It's funny there's more ohmic contacts in the world
00:47:04 than there are anything else
00:47:06 and yet it wasn't really understood how they worked
00:47:09 and so it was one of those things that was worth working out.
00:47:14 So well, a device doesn't work very well
00:47:17 when its source and drain look like that
00:47:19 so you can't go too far in that direction
00:47:22 or you get leakage to the substrate.
00:47:28 And the numbers are known.
00:47:30 We had been making junctions clear back in the 50s
00:47:37 that had large tunneling currents.
00:47:41 Leo Asaki had invented a device
00:47:44 which depended on electron tunneling
00:47:47 which is how I got into the whole thing
00:47:49 and so the physics of all that were well known and well understood
00:47:55 and a lot of us had worked on it.
00:47:58 So here's the analytical curve I used as an approximation to that.
00:48:05 And the other thing that happens, as I said before,
00:48:08 is as you make the oxides thinner
00:48:10 then the electrons can tunnel through the oxide
00:48:13 and so this is a function of feature size,
00:48:18 what the oxide's been doing.
00:48:20 Now in our very first paper in 71
00:48:24 I was very cautious about not making the oxide too thin
00:48:28 and I tend to still be cautious about making oxides thin
00:48:32 because I made a lot of thin oxides and did a lot of tunneling
00:48:35 and it bothers me when the control electrode of my device
00:48:40 has got a lot of current going into it.
00:48:42 Well, that hasn't bothered the industry so much
00:48:45 so they have gone actually on a curve which is steeper here
00:48:48 and actually it turns out that was a good move that they have done.
00:48:53 So, and once again, these are the points that we predicted
00:48:58 for what the oxides were going to be in the 0.25 and 0.15 micron devices
00:49:04 back when the device was up here.
00:49:08 So that was kind of fun.
00:49:11 And, well, how do we know when you're going to start
00:49:15 getting tunneling through thin oxides?
00:49:17 Well, it turns out that you heard Ed Snow's name mentioned earlier today.
00:49:27 He's one of the early Fairchild and then he never did Intel, did he?
00:49:33 I don't think so.
00:49:35 But he had done the original work with Bruce Steele, in fact,
00:49:39 on tunneling through silicon dioxide.
00:49:42 And so a bunch of us had worked on how that all worked.
00:49:49 So the physics, well understood, of oxides that thin.
00:49:53 So there's nothing new in that.
00:49:55 So you could predict what was happening there.
00:49:58 And then the industry has been very sporadic
00:50:01 about what to do with power supply voltages.
00:50:04 It never quite really clicked that you really do have to scale them down.
00:50:08 So they'd hang on to a power supply voltage until it was just overwhelmingly
00:50:13 the most limiting factor in going to the next generation.
00:50:16 And then all of a sudden, well, I guess we have to do something.
00:50:19 And so it's been sporadic, probably still will be.
00:50:23 It's problematic in a number of ways.
00:50:26 But that's an analytical approximation of what happens
00:50:29 if you scale a power supply voltage.
00:50:31 It's not linear in the scaling factor.
00:50:33 It's some power law.
00:50:35 I give the power law there for any of you who follow that sort of thing.
00:50:43 So if you now take those smooth curves
00:50:50 that were just approximations to what people are doing,
00:50:53 and you then extrapolate those,
00:50:58 suppose it just kept the technology smoothly evolved
00:51:01 along the lines it has been evolving.
00:51:04 What happens if you don't do anything else?
00:51:07 If you just keep making MOS devices the way we make them today
00:51:12 and the way we've made them all the way back
00:51:14 and don't change any fancy stuff, what happens?
00:51:19 Because it's important to know that
00:51:21 because that's what we've been doing all these years.
00:51:24 For 30 years, we basically made exactly the same device
00:51:28 and just made it smaller and smaller and smaller and smaller
00:51:31 without doing anything else.
00:51:33 Well, if that's what you do,
00:51:35 then these are the currents that you get,
00:51:38 and it's actually rather interesting.
00:51:42 This is the on current when the device is turned all the way on,
00:51:47 and this is the off current when the device is turned all the way off.
00:51:51 People think of the MOS transistor as a switch, but it's not.
00:51:54 I mean, the electrons don't know that they're being switched.
00:51:57 They just look at the potential barriers and do their thing.
00:52:01 So the ratio of the on current to the off current,
00:52:05 out here at a tenth of a micron, is very large,
00:52:08 so you don't worry about it too much.
00:52:10 But if you get down towards a hundredth of a micron,
00:52:13 I guess they would say 10 nanometers,
00:52:18 then the ratio gets to be closer and closer to 1.
00:52:23 In fact, it's a few, maybe an order of magnitude.
00:52:27 Still okay. The device still works all right.
00:52:30 It's not a problem. You can make logic with them.
00:52:33 You can do all the stuff.
00:52:36 There is a bunch of parasitic current.
00:52:42 The drain tunneling gets to be about the same as the off current,
00:52:50 and you see what I said about being conservative about gate thicknesses.
00:52:55 If I had allowed the gate to get a little thinner,
00:52:59 we would have had exponentially more gate current
00:53:02 and a bit less of the off current.
00:53:06 So that's the way that would trade off.
00:53:08 That is, in fact, the way the industry has traded it.
00:53:12 So this is interesting.
00:53:15 By the way, this is the current at threshold,
00:53:18 so it turns out that as the devices get smaller,
00:53:22 they're all going to run at some threshold,
00:53:24 which is a very interesting place to work.
00:53:29 So that's interesting.
00:53:33 The devices do get faster.
00:53:36 This is the delay time as a function of feature size.
00:53:40 They don't get faster super rapidly,
00:53:45 but they do get faster,
00:53:47 and they'll keep getting faster as time goes on.
00:53:51 So it's not like you're losing your shirt or anything.
00:53:55 That's all just fine.
00:53:58 Where does the power go in a device that's made?
00:54:03 Once again, this is just if you took what we do today and scale it.
00:54:07 No new materials.
00:54:10 Well, you see what happens.
00:54:12 This is the switching power,
00:54:14 the power that's actually doing real work.
00:54:19 This is the off current and the drain leakage current,
00:54:24 the power due to that,
00:54:26 and you see it's a factor of like 30 larger
00:54:29 than the power going into useful work.
00:54:33 So if this is all we did,
00:54:35 and we just kept scaling things down,
00:54:38 then the big hit we'd take
00:54:41 is that the power would get a lot more
00:54:44 for the same amount of useful work.
00:54:47 So that would not be improving.
00:54:50 Once again, I've been conservative on the gate.
00:54:55 So that's the big one.
00:54:58 So then the question is,
00:55:01 can you do anything about,
00:55:05 and I'll get to this one in a minute,
00:55:07 can you do anything about those parasitics?
00:55:12 And the answer is, of course you can.
00:55:16 And you heard a little bit of that today,
00:55:18 but maybe not in as much detail.
00:55:21 Let's take this one, the off current.
00:55:24 Well, you can, why is the off current as big as it is?
00:55:28 Well, the off current is as big as it is
00:55:30 because the gate doesn't have as good a control
00:55:33 of the potential in the channel as you'd like.
00:55:37 So it doesn't shut the transistor off.
00:55:40 So if you had a way of doing that,
00:55:43 it would cut down the off current.
00:55:47 Well, why can't you do that?
00:55:48 You can do that by making the oxide thinner,
00:55:50 but when you do that,
00:55:51 then the gate current goes up exponentially.
00:55:53 So it's this trade-off of gate current
00:55:56 versus this parasitic tunneling current.
00:56:02 Well, can you fix that?
00:56:03 Well, of course you can.
00:56:05 That's why people are working
00:56:06 on high dielectric constant dielectrics
00:56:09 because if you make something
00:56:10 with a high dielectric constant,
00:56:12 you can have a thicker layer
00:56:14 and still give the gate influence
00:56:17 over the potential down in the channel.
00:56:19 In other words, it capacitively gives you
00:56:22 more control of the potential in the channel,
00:56:25 but the electrons don't tunnel through it.
00:56:28 So there's a lot of interest now
00:56:30 in new kinds of gate dielectrics.
00:56:33 Now, it's no longer good old SiO2.
00:56:39 So that means that it's harder
00:56:42 because we were given a gift
00:56:44 when an acceptable semiconductor
00:56:47 had a fantastic insulator as its native oxide.
00:56:50 I remember in one of those discussions with Gordon,
00:56:52 he said, Carver, he said,
00:56:54 this isn't a silicon technology.
00:56:56 This is a silicon oxide technology,
00:56:59 and that's just certainly right.
00:57:02 There are many better semiconductors than silicon,
00:57:05 but none of them has an oxide like SiO2
00:57:08 that's just native.
00:57:10 I mean, that is just God's gift
00:57:12 to the semiconductor industry.
00:57:14 So, but yeah, it's probably not insurmountable
00:57:19 to use other kinds of gate things.
00:57:21 There will be a whole other set of,
00:57:23 you heard this morning about all the drifts,
00:57:26 the different drifts.
00:57:28 There'll be drifts associated
00:57:30 with other oxides, other dielectrics,
00:57:34 because they're not silicon oxide,
00:57:36 and we haven't worked through all those issues,
00:57:38 and we're going to have to learn it.
00:57:39 Can we do that? Sure, sure.
00:57:41 So that's the way out.
00:57:43 If you still want to build MOS devices,
00:57:46 that's the way out of the problem.
00:57:49 The other way out is the thing we heard this morning.
00:57:53 Why do people build gates on three sides
00:57:55 instead of only gates on one side?
00:57:57 It's because now you get the influence
00:58:00 over the potential is coming from all sides
00:58:04 instead of one side,
00:58:05 so we're sort of back to Shockley's
00:58:07 original analog transistor,
00:58:09 where he had to wrap the gate
00:58:10 all around the channel in the middle.
00:58:16 So the other parasitic here is the drain tunneling,
00:58:21 and that's an easy one.
00:58:22 That's tunneling into the substrate.
00:58:24 So you make an insulating substrate,
00:58:26 and it goes away,
00:58:28 and so that's the interest in insulating substrates.
00:58:31 Well, now there's a problem with that.
00:58:33 If you have an insulating substrate,
00:58:36 it's not only an insulator for electric current,
00:58:40 it's also more of an insulator for heat.
00:58:44 So now the device will have a larger thermal impedance
00:58:48 to the rest of the world,
00:58:49 and that means you can't cool it as well.
00:58:52 Well, okay, then you make the insulator out of diamond.
00:58:55 Diamond has a heat conductivity
00:58:57 four times that of copper,
00:58:59 vastly more than silicon,
00:59:01 so a device like this,
00:59:03 which is on a diamond substrate,
00:59:06 is a far more effective structure.
00:59:09 You can make CVD diamond now.
00:59:11 That's a well-known chemical process.
00:59:13 It's going to take some work.
00:59:15 Can we do it?
00:59:16 Sure, sure.
00:59:17 That's plain process development.
00:59:20 There's an enormous incentive for doing that.
00:59:23 So of course we can do that.
00:59:25 So the two big impediments
00:59:28 to going further down in dimension,
00:59:31 the tunneling into the substrate
00:59:33 and the tunneling through the gate oxide,
00:59:35 both have perfectly reasonable solutions.
00:59:38 They're going to involve a lot of physics
00:59:41 and a lot of chemistry in both of them,
00:59:43 mostly chemistry.
00:59:46 As Gordon said many times,
00:59:48 the semiconductor manufacturing industry
00:59:52 is a chemical industry.
00:59:54 It's a chemical process
00:59:56 that makes integrated circuits
00:59:58 through and through.
01:00:01 So it's kind of interesting,
01:00:05 if you think about it,
01:00:07 Moore's Law has been able to go
01:00:09 over all these orders of magnitude
01:00:13 without the physicists or the computer scientists
01:00:17 doing any work.
01:00:21 We got exactly the same device physics we had
01:00:24 when we started the thing in 1965.
01:00:26 We have exactly the same kind of
01:00:29 computer architectures that we had.
01:00:31 They're just on one chip now
01:00:32 instead of in a room this size.
01:00:36 We haven't had to think.
01:00:39 So we as physicists in computer science
01:00:42 have gotten really spoiled
01:00:44 because you processing people
01:00:46 bailed us out every time
01:00:49 or said another way,
01:00:51 whenever someone would come up with
01:00:52 a clever architecture
01:00:54 or clever programming techniques,
01:00:56 you could always beat it
01:00:58 by just making a faster Pentium.
01:01:01 Now, that's going to continue,
01:01:04 but as you heard this morning,
01:01:07 Pat was telling us,
01:01:09 finally people are starting to realize
01:01:13 that if you have more small computers,
01:01:18 it makes better use of the silicon
01:01:20 than one big computer.
01:01:22 It's sort of the lesson of the mainframe
01:01:25 all over again.
01:01:26 You know, we've been working and working
01:01:27 and working to make bigger and bigger mainframes
01:01:29 and still keep them on a chip.
01:01:31 And finally it's dawned
01:01:34 that parallelism is really the way
01:01:36 to get more computation.
01:01:38 You can see that on a slide like this.
01:01:41 This is a clock frequency.
01:01:42 It can't go up very fast on a scale like this.
01:01:46 And suppose you took a million transistors.
01:01:49 That's a tiny microprocessor in today's world.
01:01:52 And you just made a bunch of them.
01:01:54 Well, then you'd get this many operations.
01:02:00 So the number of operations is now going up
01:02:03 by the parallelism that you can get.
01:02:05 If you could use fewer transistors per operation,
01:02:09 it goes up even faster.
01:02:11 So the way microelectronics scales
01:02:15 and in fact a big lesson
01:02:18 for all of information processing
01:02:21 is as you distribute information processing,
01:02:24 you use more parallelism.
01:02:26 That's where you get the real power.
01:02:29 You use more and more computations
01:02:33 running in parallel.
01:02:34 We worked, what, 20 years ago now
01:02:37 to try to figure out how to do that.
01:02:39 And it was popular for some time
01:02:41 in computer science circles.
01:02:42 And then it turned out
01:02:43 that nobody would build machines like that.
01:02:46 And so it hasn't developed
01:02:50 as an intellectual art form in the way it could.
01:02:54 And in fact will have to
01:02:56 if we're going to really make full use of the technology.
01:02:59 And I was just delighted to hear Pat say this morning
01:03:02 that they're going to start putting more cores on a chip,
01:03:06 not just making bigger and bigger mainframes on a chip.
01:03:09 So that's been really nice.
01:03:12 So I always get asked,
01:03:13 you know there's all this stuff in the press
01:03:15 about Moore's Law's ending.
01:03:17 You know, it's a little like,
01:03:18 here's a picture of Piglet.
01:03:21 Here's Piglet and he's having this dream
01:03:24 about the heffalump, you know,
01:03:26 chasing the little Piglet here.
01:03:28 It's sort of the, you know,
01:03:31 the nightmares people have about the end of Moore's Law.
01:03:34 Is the Moore's Law coming?
01:03:35 Is the sky falling?
01:03:37 Is our industry coming to an end?
01:03:40 Is this tragedy about to befall us
01:03:42 because Gordon's going to fall off a cliff
01:03:44 and, you know, then what'll happen to any of us?
01:03:47 Well, A, there are many dimensions to innovation.
01:03:52 And we've been sort of
01:03:55 blindly following one particular one
01:03:58 because you people have bailed us out.
01:04:01 Every time you've been able to make these things
01:04:04 smaller and faster and lower power and all that.
01:04:06 And you've lulled us into basically
01:04:09 not thinking about the other dimensions of innovation.
01:04:12 But there are a lot of other dimensions of innovation.
01:04:15 And we have a lot of very smart people
01:04:18 who are responding to the need for innovation
01:04:24 and scaling the devices smaller is one of those dimensions.
01:04:28 And I'm sure your people are going to keep that up
01:04:30 just like Pat said this morning.
01:04:32 But as Pat puts more and more cores on each chip,
01:04:35 now we need programming techniques,
01:04:37 we need algorithmic work,
01:04:40 we need all those things that go with
01:04:43 a new style of computation.
01:04:46 And I just have to close by saying that
01:04:49 we have an existence proof
01:04:53 for enormous parallelism in computation.
01:04:56 And that is the brains of animals.
01:04:59 Brains of animals work with rather slow computing elements.
01:05:04 But they all work in parallel.
01:05:06 So there's an existence proof
01:05:09 for parallelism in computation.
01:05:12 And we know that as a vision system,
01:05:17 a housefly does a better job of vision
01:05:24 than all computing power we can throw at it today.
01:05:28 So there are dimensions of innovation ahead
01:05:31 that we simply haven't figured out yet.
01:05:34 But they're there to be figured out.
01:05:36 Our young people know that.
01:05:38 And Moore's Law is going to be healthy
01:05:42 and going in more dimensions than just device size
01:05:48 as long as we can see into the future.
01:05:51 Thank you.
01:06:01 Thank you so much.
01:06:03 That was stimulating and encouraging.