Digital Collections

Moore's Law at 40: Part 2

  • 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

00:00:00 Good morning. I'm Arnold Thackray, the President of the Chemical Heritage Foundation, and it's

00:00:19 my privilege and pleasure to welcome you all to today's sessions. Yesterday evening we

00:00:25 had a wonderful glimpse of the future of chemistry and electronics, and why it turns

00:00:32 out we all desperately need more computing power. Today we're focusing to the 50 years

00:00:42 essentially that changed the world. I think this technology, this area, has the claim

00:00:50 to be the most transformative technology of the last half century, and Gordon Moore,

00:00:59 who we're delighted is with us and will be speaking later in the day, is surely the central

00:01:04 architect, the vision, and the driver of what has taken place. When I've told people, it's

00:01:13 always interesting to do the cocktail party test. You know, you go into a cocktail party

00:01:17 and you say something to people and see what response you get, and go in and say,

00:01:22 well, Chemical Heritage Foundation is going to do Moore's Law 40, and the cocktail party

00:01:29 response is twofold. G, why the Chemical Heritage Foundation? And G, why in Philadelphia? And what

00:01:40 I hope to do as a strand that connects into the rest of the day is, in a certain way,

00:01:46 to answer those two questions, and to show the fateful role of a telephone call to a PhD

00:01:55 chemist that took place just 50 years ago, because I think that telephone call that we

00:02:02 celebrate the 50th anniversary of right now was really the transformative moment. Well,

00:02:09 here we are in Philadelphia. We're talking about chemistry. We're talking about electricity.

00:02:14 There's a certain inevitability that we're talking about Ben. And, of course, literally,

00:02:19 Ben had his home in our backyard here. And this is the portrayal of his famous lightning

00:02:26 experiment. The two cups move apart as the charge changes, as the one is connected with

00:02:36 the lightning. And Ben, of course, really made the whole territory of electricity quite shocking,

00:02:43 quite exciting, very much the subject of conversation and speculation. And Philadelphia

00:02:51 was very central to that speculation. Among Ben's discoveries, of course, was his discovery of

00:02:58 Joseph Priestley, a young, dissenting Protestant minister in England who he persuaded into the

00:03:05 territory. And Priestley's first scientific book was The History and Present State of Electricity,

00:03:11 with original experiments. I mean, Ben set him going. We are proud that among all the great

00:03:17 treasures that we have here in this building, we have this first edition of Joseph Priestley's

00:03:24 book. And incidentally, in the break, do go across the hall. We have a wonderful Priestley exhibit.

00:03:32 It's connected to two other exhibits, one of which is all about instrumentation ending up

00:03:37 with electronics. So you've got a sort of miniature cameo of everything just across the

00:03:42 hall. So Priestley is in this territory. And Priestley points out this seminal statement

00:03:52 that the great chemistry is the great field for the extension of electrical knowledge. For chemistry

00:03:57 and electricity are both conversant about the latent and less obvious properties of bodies.

00:04:01 And yet their relation to each other has been but little considered. And really, you might say,

00:04:07 as we approach the 250th anniversary of that statement, we're finally getting a grip upon

00:04:13 this territory of the relationship of chemistry and electricity. And as we tease out that strand,

00:04:21 of course, the next big event is Galvani and frog's legs. And Galvani, by taking a bimetallic

00:04:32 strip, two metals connecting it, makes the frog's legs twitch, even though the frog is obviously

00:04:38 not alive. And says, gee, you know, animals, electricity. As is usually the case in science,

00:04:49 there's skepticism around this. And Volta doesn't believe it for a minute. He says,

00:04:55 it's all just about your two metals. And so Volta seeks to do the experiment without the frog's legs

00:05:01 and connects metals and produces current electricity. And now we're into a different

00:05:11 world, of course. And current electricity is the great next sensation after Ben. And you can see

00:05:19 what a sensation it was, because here's Volta showing Napoleon how it works. I mean, this is

00:05:26 the great, great news. It's also the beginning of big science, because the guy who sees the

00:05:34 potential of what you can do for this is, I don't know if any of you, I don't know if this was a

00:05:41 Britishism, but we always used to know the rhyme that Humphrey Davy abominated gravy. He also

00:05:48 incurred the odium of having discovered sodium. And of course, he did this by putting, he did

00:05:57 big science. I mean, there are the batteries at the Royal Institution. And Davy had the insight

00:06:02 that if one cell is good, then hundreds must be better. And of course, he decomposed things in a

00:06:10 way that had never been done. And sodium and potassium were, of course, great sensations.

00:06:14 And perhaps other people in this room, like me, were partly attracted to chemistry by what happens

00:06:20 when you put sodium and potassium with water. And I don't think they do that in the, in the

00:06:26 elementary, middle school, high school lab today. So, so Davy, of course, is a great, great sensation,

00:06:34 that things are picking up. Davy, like, like Franklin, has a great discovery on the side,

00:06:40 of course. And he has the poor boy lab assistant. And the poor boy lab assistant is Michael Faraday,

00:06:46 who turns into his successor and makes the next great step of connecting electricity and magnetism

00:06:54 and producing the electric motor and the dynamo, which, of course, take a long, long time to become

00:07:01 practical realities. It's interesting that this period, the gap between the theory and the practice,

00:07:08 and here's fuel cells. I mean, fuel cells are, are obviously within a year of being practically

00:07:16 applied in 1842. And we, we still live in hope, I think. What, what was practically applied, of course,

00:07:27 was Charles Hall and the electrolysis of alumina to produce aluminum, this, this wonderful, wonderful

00:07:35 new metal. And here's Hall's electrolytic cells at the Pittsburgh Reduction Company. And here,

00:07:42 not very long after, now we've really connected chemistry and the electricity, here's plants at

00:07:51 Niagara Falls. And as we mentioned last night, by 1902, here in Philadelphia, you've got the,

00:07:57 the Electrochemical Society being founded. So things are moving right along. Another example,

00:08:05 Evans Mill, where brine was first electrolyzed. Evans Mill probably doesn't mean too much to you,

00:08:12 but, and you can't really quite see, but if I say the central figure in that photograph is Herbert

00:08:18 Dow, then you get the clue. Herbert Dow, fresh out of Case, Case Institute, is at the leading

00:08:29 edge of electrochemistry, and he's going to set up, he's going to, it's bromine, you know,

00:08:34 the point about bromine is the big commercial use was bromides. And I don't know if there's

00:08:41 anybody old enough in the audience to have taken a bromide, but bromides, of course,

00:08:46 were the tranquilizers, they were the popular medicine, and indeed Dow's first order came from

00:08:52 what today is a subsidiary of Merck here in Philadelphia. That was where he got his first

00:08:58 commercial start with his new Dow Chemical Company using electrolysis. Well, you can see we're moving

00:09:07 right along. Here's the obvious way to go, and it's the electric car. I mean, what could you

00:09:11 possibly not, why would you not do it this way? And Edison, you know, it's done a thousand mile

00:09:17 endurance run, so like the fuel cell, obviously, here it is. We're now in the future. I want to

00:09:26 put this one up because, of course, the thing that now further begins to complicate the territory is

00:09:32 electronics rather than electricity, and Arnold Beckman applies electronics and the thermionic

00:09:41 valve to the chemical world. Interestingly, Beckman, Dr. Beckman, some of you may have

00:09:48 heard tell the tale that he invented this thing, he took it to the ACS meeting, it happened to be

00:09:55 on the West Coast, nobody was particularly interested at all. He toured across the country

00:10:00 asking dealers, because all instruments went through dealers, could they sell any, and it

00:10:05 wasn't till he got to the end of his journey, three blocks from where we are now at the Patterson

00:10:12 Company, that he finally got somebody who said, yeah, you could sell 600 of those, and so he went

00:10:19 into business, and fateful consequences. Here's another, of course, another piece of going on the

00:10:31 electronic route. Here's the first real working electronic computer, just about a mile away from

00:10:37 where we are today, and as you see, there's an awful lot of people because, you know, valves burn out

00:10:43 at a high rate, there's a huge number of valves, and the main question is, can you make it work for

00:10:48 more than two minutes at a time, before some valve blows and you're in trouble? And of course, at

00:10:57 this very moment, the world is about to change, because a different way of doing the electronic

00:11:03 part is appearing in the first transistor, and again, we're right here on the East Coast in this

00:11:10 area, because of course, the first transistor at Bell Labs is here in easy commuting distance of

00:11:18 where we are today, and this is what takes us to the to the fateful telephone call, because of

00:11:26 course, Shockley is a Caltech graduate, and when he decides he wants to go out on his own, among the

00:11:35 people he calls to say, have you got any money, is his old professor, Arnold Beckman, and Arnold

00:11:42 Beckman is the person who says, sure, I will 100% bankroll what you want to do, it's not a big

00:11:49 expense, you're only going to have a small group of people, and because it's only a small group of

00:11:54 people, Shockley says, do you mind awfully if I don't come down to Pasadena, I'm very close to my

00:12:00 mother, my mother's in Palo Alto, and that's how the silicon got to Silicon Valley, because Shockley

00:12:07 set up shop in Palo Alto, but the new electronics was essentially in the East, Bell Labs, the big

00:12:15 companies, they were all here, and Philadelphia was the technical center, unsurprisingly, the

00:12:21 solid-state circuits conference met here beginning in 1954, and you can see when people don't realize

00:12:29 that forces are moving, if I were planning a meeting, an annual meeting in Philadelphia, I would not, I would

00:12:36 plan it for May, I would plan it for a day like today, but this group met in February, and I think Gordon can

00:12:42 testify that Philadelphia isn't at its most attractive in February, but the solid-state circuits conference

00:12:49 was here, Philco was here, and of course Philco was the largest producer of radios, and radios were the

00:12:54 largest consumer of thermionic valves, and so the territory connected to transistors, and so on and so

00:13:02 forth, and so it's unsurprising, every great insight in science, of course, always has precursors, and so some

00:13:10 of you may have read in the New York Times article recently, sort of saying, gee, Moore's Law, it had a

00:13:17 precursor when Engelbart in Philadelphia made some pronouncements at the solid-state circuits conference,

00:13:26 which is true, it was a sort of precursor, and so Philadelphia was where it was, but the real action, of course,

00:13:35 was in the thinking, the creative, different way of thinking that came to characterize Fairchild and Intel

00:13:46 coming out of Shockley. Shockley's firm blew up, and Fairchild, Intel, and so on are what came out.

00:13:56 Here's an early integrated circuit, and here, of course, Gordon, of course, is simultaneously the driver of the

00:14:06 transformation in what a chip is, and the prophet, and the creator, and the fulfiller of Moore's Law, and the story,

00:14:18 this story, that's surely not in Philadelphia, but we're delighted here in Philadelphia to put it into the longer

00:14:25 frame, and to look forward to our morning speakers now, who will really be talking about this time from the 1950s,

00:14:36 1960s, through to today. So let me, at this moment, it's my pleasure to turn the podium over to Miles Drake.

00:14:44 Miles is the Chief Technology Officer of Air Products and Chemicals. Air Products and Chemicals is the global

00:14:54 leader in the supply of electronic gases and materials. Miles and I have numbers of things in common.

00:15:05 Perhaps most obviously, it turns out, we went to the same two universities in England, and we both had the good sense

00:15:14 to understand the challenge, the adventure, and the opportunity of life in the United States. Miles.

00:15:30 Well, thanks, Arnold, and it's a pleasure here to introduce the first couple of speakers this morning, and just to continue

00:15:37 in Arnold's mode of sort of connections, I'd just like to introduce Pat Gelsinger with a few connections, perhaps.

00:15:45 He's the Vice President General Manager of the Digital Enterprise Group in Intel, and as Arnold indicated,

00:15:53 certainly Air Products and Intel have had a very long and fruitful relationship. In fact, Intel has really driven

00:16:00 Air Products into a whole global market that has been helped by Intel in terms of just the connections,

00:16:09 have been driven to do certain types of business. But the connections I want to illustrate is first that Pat is very

00:16:17 connected into this area, and I won't go through his details, but if you look, he actually started his education in the

00:16:25 Lehigh Valley. His parents and family are still in this area, so we did actually send someone from the Lehigh Valley to go

00:16:34 and make a tremendous contribution into the whole electronics industry. Second connection is just that, as he pointed out,

00:16:44 he is a sort of bridge between the more recent past that Arnold's talking about and the future, because he did have the

00:16:52 privilege and opportunity to work with people like Gordon Moore, Andy Grove, Robert Noyce. So he had a really good connection

00:17:02 into strategy, technology, innovation, and the whole drive at the early part of this whole semiconductor revolution.

00:17:09 So he really has a tremendous connection. We're looking forward to his insights. So he's both connected geographically here,

00:17:19 and connected through the technology. So I'd like to welcome him, but before I welcome him, I just want to mention that he did comment

00:17:26 that as an electrical engineer in training, and if you look at his contributions, one thing that we can all recognize easily is he was the

00:17:35 chief architect of the 486, and a number of people here might remember the joy of jumping up to a 486 in terms of the improved performance.

00:17:45 Of course, we've gone well beyond that, but as an electrical engineer, he's perhaps feeling, as you said, like a fish out of water with all these

00:17:52 chemists, but I think we can all show how warm and welcoming chemists can be. So I'd like to invite him up to give his talk. So thank you.

00:18:01 Thank you very much, Miles. It is a pleasure to be here. I do fear the chemists. There's a whole lot of domains that you know a whole lot more than me,

00:18:17 and we'll touch on a few of those in the conversation. So please, in advance, forgive my errors. Also, I sort of feel like the warm-up band, right?

00:18:26 At a concert you go to, the first guy doesn't have to be all that good, right? People are still getting their coffee and drinks, so please help yourself, and I'll try to

00:18:34 quickly move past to the real speakers of the morning, such as Harry, Elsa, and Raj, and Carver, Anna Lee, and of course the main actor here, Gordon, later today as well.

00:18:46 So we'll try to be as brief as possible so we get to the real discussions. I do want to start just quickly with a few Gordon stories, if I may. I have had the great privilege of working with Gordon,

00:19:00 as well as a number of other of the great people of the industry, and the first time that I had a very intimate interaction with Gordon, we invited him to be.

00:19:12 I was the architect, design manager, the person who drove the 486 into production, sort of the womb of the toon on that project, and we invited Gordon and his wife, Betty,

00:19:22 to be the guests of honor at our celebration dinner for the 486. You remember this, Gordon? Up in San Francisco, we invited them up there, and my wife is just beside herself about being with the founder,

00:19:35 being with this great man and his wife. What shall she wear? What shall she dress? How's this evening going to go? What should she talk about with Betty? Just very, very concerned about this.

00:19:46 So we get there, we have a wonderful evening, and by the end of the evening, Gordon's not a man of a whole lot of words. He's fairly quiet, unassuming, and so on.

00:19:55 Gordon and I are just talked out, and here's Betty and Linda just going on all night long, roses and daughters and kids and all of this kind of stuff, and Gordon and I are long ready to go home and get back to work or go to bed.

00:20:08 The women are having this wonderful evening, so I do think of Gordon and Betty as the nicest, most pleasant, most wonderful humans, and I've always said he's the nicest billionaire on earth, just incredible person.

00:20:22 Not only is he a wonderful person as an individual, but as a technologist, and I think of Robert Noyce as the greatest inventor, I think of Andy Grove as the greatest strategist, and I think of Gordon Moore as the greatest technologist of our age.

00:20:38 And as I was working on the 46 compaction, we were working, remember when the 46 at 33 megahertz was fast? Do you remember that? Yeah.

00:20:47 And we were working on taking it to a stunning 50 megahertz, okay?

00:20:51 And Yusef El-Mansi, the person who ran technology development for Intel at the time, and myself, were struggling.

00:20:59 We had a low-yield, do you remember this conversation, Gordon? No? No? Okay.

00:21:03 It's indelibly in my mind, we had a low-yield fallout issue on the 46 compaction, and we just, Gordon and I were just pounding our heads over what the problem with it was.

00:21:12 So finally, Gordon says, well, you know, come, I'd like to talk to you about it.

00:21:15 So we go in and we're laying out all of the experiments we had done, all the work that we had done with Gordon and so on, and he says, have you looked at this?

00:21:22 And Yusef and I spend the next 10 minutes explaining to him why that couldn't be the problem.

00:21:27 We had done this experiment, that experiment, and looked at it somewhat like this.

00:21:30 He says, look at this.

00:21:32 And of course, we're sort of humored. Okay, okay, boss, whatever you want.

00:21:36 So we're looking at it, knowing that that couldn't be the problem.

00:21:40 Well, we did a few short-loop experiments, and about two weeks later, here's Gordon.

00:21:43 He had no idea what the problem was.

00:21:45 Hadn't been working in the fab for years and years.

00:21:47 He's an incredible CEO, and guess what? That was the problem.

00:21:52 And I guess I just walked away that somehow, right, between brilliance, between innovation and insight, and just intuition, right, the greatness of the man that we're honoring today with Gordon Moore.

00:22:03 So, you know, I do consider it a great privilege to be here.

00:22:06 And so let's look forward rather quickly, if we could bring the slides up, a few comments about Moore's Law past, the challenges of Moore's Law present, and then looking forward to Moore's Law future.

00:22:20 Okay.

00:22:24 Yeah, that's a great slide, but mine are different than that.

00:22:32 Okay, very good.

00:22:37 Just a few examples of the impact of Moore's Law.

00:22:40 You know, exponentials that operate for a long time have incredible impact on the world, right?

00:22:47 I mean, I like this one, right, more transistors than grains of sand, and they're now a lot cheaper.

00:22:54 Second is, you know, this impact if you actually apply Moore's Law to other industries, right, the implications that this would have on the transportation industry.

00:23:02 And I sort of like the second one.

00:23:03 The only problem is each of us would have to be about the size of a pea, right, to fit in that craft to go that way.

00:23:09 And of course, right, the impact of, you know, just this production of computation, and we'll touch on that a little bit more later in the talk,

00:23:17 but just the incredible, you know, performance that we're now putting in the hands of developers, applications, and software industry.

00:23:25 And of course, this relentless march continues on.

00:23:28 I remember with Glee, you know, the 386, one of the chips I worked on, 175,000 transistors.

00:23:36 I think I personally touched almost every one of them.

00:23:38 Went to the 46, a stunning achievement at 1,120,486 transistors.

00:23:44 You know, we added a few at the end just to make the numbers work out just correctly.

00:23:47 And we thought crossing the million transistor mark was this stunning result, right?

00:23:53 Well, we've just produced, you know, we'll be producing this year the first Montecito chips,

00:23:57 our dual-core Itanium processors that have 1.7 billion transistors on a single chip.

00:24:04 And here's a wafer of the Montecito brought along.

00:24:08 And, you know, it's just stunning.

00:24:10 A billion, 1.7 billion transistors on a single chip.

00:24:14 I mean, a billion of anything is a big number.

00:24:16 And we're cramming it into this chip that we're going to produce in mass volume this year.

00:24:21 And we're marching relentlessly forward toward 10 billion and beyond.

00:24:27 The, of course, you know, as we look at that into the future, we realize that not all predictions came true.

00:24:33 I wrote a paper in 1988 called 2000, predicted what the microprocessor of 2000 would be.

00:24:40 I was a little bit off.

00:24:42 And we also noticed that some of Gordon's predictions were just a little bit off as well.

00:24:46 And we all quite enjoyed this one on wafer size.

00:24:50 Of course, it came out of an April Fool's edition of Intel Leads, our internal corporate magazine.

00:24:56 But we all quite enjoy applying Moore's Law to all sorts of places that it may or may not apply over time.

00:25:03 But what we have seen is that this march continues relentlessly forward.

00:25:08 And as we've seen challenges, barriers, other limitations to it, we've continued to move forward.

00:25:14 And one of the things that has moved forward as well, transistor counts has moved forward.

00:25:18 But we've also moved forward.

00:25:19 You remember when two-inch wafers were big, right?

00:25:22 So now, right, we're at the 300-millimeter wafer level.

00:25:25 And the wafer that I showed you for the Montecito chip, you know, that wafer by itself represents 120 billion transistors on a single wafer.

00:25:35 Right? 120 billion transistors on one silicon wafer.

00:25:40 Just incredible.

00:25:41 And today, so we're bringing 300-millimeter wafers into high-scale manufacturing today.

00:25:47 Intel has five factories up and running at this level.

00:25:49 And one of the important factors has been our continuing ability to scale the size of silicon wafers for mass production purposes.

00:25:57 And we're looking forward to, well, we don't have specific announcements today, about 450-millimeter wafers.

00:26:03 We already have research underway, what it will take to go up to the next wafer size.

00:26:07 You know what I mean?

00:26:08 You know, that's a small pizza.

00:26:09 Now we're going up to at least a good medium-sized pizza in the next generation of Moore's Law.

00:26:14 It's just quite stunning.

00:26:15 And the result of that has been that while there has been much said, you know, much said about the cost of fabs, you know, the cost of lithography, you know, the cost of many other aspects of Moore's Law, right,

00:26:26 the combination, right, of the scaling as well as the, you know, the scaling down and the sizes of the devices,

00:26:32 the mass production improvements that have occurred and the increase in the size and wafers, right,

00:26:38 there's been the inverse of Moore's Law is the cost per device, right, which has been, as you see by this graph, perfect against the long-term exponential.

00:26:47 And we're now crossing into the domain, right, of nanodollars per device, you know, an incredible result.

00:26:55 And while I was chatting with Elsa last night, I'm not sure where you are here, Elsa, but, oh, there, thank you.

00:27:01 She was complaining about the cost of lithography.

00:27:04 And, of course, I immediately said, who cares how much lithography costs?

00:27:08 This is the only thing that matters is are we producing cost reductions per unit device delivered?

00:27:14 And as you see by this, that continues unabated even though the cost of the manufacturing itself has been increasing by this volume production,

00:27:22 the increase in size of wafer size, the manufacturing improvements, it's continued unchanged going forward and sees no signs of chaining whatsoever into the future.

00:27:31 Now, the fact that it costs more to build a fab, Intel doesn't mind that, right?

00:27:36 It just means a lot of little players can't participate and only big guys like us get to, so we're not particularly bothered by that realization.

00:27:45 So Moore's Law passed.

00:27:50 We do see this incredible, incredible implication that the exponential has had on the industry and the world.

00:27:58 The semiconductor industry by itself sits somewhere around $200 billion industry.

00:28:03 That's not too bad, right?

00:28:05 Gordon, his innovations, his researchers produced the $200 billion industry on the planet.

00:28:10 The result of that has been the direct production of the entire IT industry, the information technology industry, today about a $1.2 trillion industry.

00:28:19 However, what is even more, to me, incredible is the global impact.

00:28:25 If you go look at the GDP, pick the United States or of the planet, there are different ways to partition the GDP,

00:28:32 but what we see is that we see just every element of the GDP has been touched by this exponential.

00:28:39 If we look inside, I like this one, while all of us hate taxes and all of the other implications of it, we're now doing it better, more efficiently, 50% worldwide.

00:28:49 We see the implications of new technologies like RFIDs, something absolutely impossible to occur without Moore's Law,

00:28:57 transforming retail and now moving into the manufacturing segment.

00:29:01 RFIDs are dumb passive devices.

00:29:04 We're very actively making, and here's a mote, which is essentially a little computer, complete storage,

00:29:09 complete interaction with the physical world, as well as a radio built into it, all into a little device like this,

00:29:16 which eventually will be cost reduced into just less than a dollar or down into the tens of cents,

00:29:22 so we'll be able to have active sensors as well.

00:29:25 And a very important, as we heard from Rodney last night, is working robotics.

00:29:28 Active sensors will absolutely transform and alter the domain between the electrical and electron world and the physical world.

00:29:37 Technologies like this will be absolutely transformational in that regard.

00:29:43 We look at forward.

00:29:44 Essentially everything that's happened in the automotive industry has been a direct result of Moore's Law.

00:29:49 All of the innovations that have occurred in terms of braking systems, navigation systems, et cetera, a direct result of Moore's Law.

00:29:56 We've seen transformations.

00:29:58 Some of it, of course, Napsterized, not good by the particular recording industry.

00:30:02 Others, but producing digital audio, digital video, have absolutely transformed all forms of the entertainment industry.

00:30:10 And while the semiconductor industry itself, this is one of my personal favorite quotes,

00:30:15 represents a fairly modest portion of the GDP of our nation, about 3%.

00:30:20 It has about an order of magnitude, as estimated by Harvard, some research by Dale Jorgensen at Harvard University,

00:30:29 an order of magnitude larger impact on the productivity improvements of the United States.

00:30:34 In other words, even though its direct impact has been small, it has been the transformational element of a huge amount,

00:30:42 some estimates are far higher than this, of the entire productivity gains of our nation and of our world.

00:30:49 And if we drill into just a specific example, if we look at the mobile phone industry,

00:30:54 we took about 100 years to get about a billion phone lines installed on the planet, connecting up our world.

00:31:02 And then something wonderful happened, the innovation of the mobile phone in 1973,

00:31:08 and forecasts say by the year 2010, we'll have somewhere between 2.5 and 3 billion mobile phones connect on the planet.

00:31:16 So we'll have tripled, in the course of about 30 years,

00:31:20 we'll have tripled the entire connectivity of the prior 100 years of the phone industry.

00:31:25 We'll have gone from less than one-sixth of the world's population being connected, being able to communicate,

00:31:30 and the results of the mobile phone, in the order of a half of the world's population.

00:31:35 Now, just a quick analysis would be, if Moore's Law hadn't progressed,

00:31:44 and we had one-and-a-half micron technology, what we had when we first did the 386 chip in 1985,

00:31:50 and if Moore's Law hadn't progressed since that, but we wanted to build today's modern telephone.

00:31:57 The modern telephone, the modern mobile phone that we would have today, would be about the size of a good-sized piece of luggage,

00:32:04 be hundreds of pounds heavy, and it would take about a car battery per hour to operate that mobile phone.

00:32:13 Now, do you think the mobile phone industry would have taken off without this continuing, assaulting progress of Moore's Law?

00:32:23 Of course not. No way would we have ever seen the innovations that have resulted.

00:32:29 So, Moore's Law impacted so much parts of our industry and our world.

00:32:35 Does it have the opportunity to continue to be valid as we look forward into the future?

00:32:41 The first is, and this is, speaking to chemists, this is actually the one slide that I'll say I won't try to elaborate on very far,

00:32:50 but in the 80s, we used a very small portion of the periodic table.

00:32:56 About a dozen elements from the periodic table were part of our manufacturing processes.

00:33:01 Throughout the 90s, largely, the manufacturing engineers took over, and we had geometric scaling.

00:33:07 We didn't need a whole lot of help from the chemists.

00:33:10 We had only gone up to using about 15 elements of the periodic table.

00:33:14 What we've seen is that as we've moved into an entirely new genre of Moore's Law,

00:33:19 we've seen an explosion in our use of the periodic table.

00:33:22 As we're in the 2000s now, and I think our specific count as we go to our 65 nanometer devices

00:33:28 that we'll be bringing into the industry at the end of this year,

00:33:31 we expect 51 elements of the periodic table are now being used in the manufacturing of silicon wafers.

00:33:37 We've seen this explosion, this resurgence of the criticality of understanding material science and chemistry

00:33:43 at the core of our processing technology.

00:33:47 As we think about this, sometimes you hear many comments on biological computing or optical computing

00:33:55 or all of these other types of computing emerging and replacing Moore's Law, replacing silicon.

00:34:02 Increasingly, when you look at this chart, what you see is that silicon is actually performing the scaffolding, the frame.

00:34:10 Just like we build a building, we put other materials inside it.

00:34:13 The picture I want you to walk away from is that we continue our assault on Moore's Law.

00:34:18 It continues to provide the frame or the scaffolding as we bring more and richer chemical properties,

00:34:24 materials understanding into Moore's Law of the future.

00:34:28 The result of that is we're able to continue scaling our devices.

00:34:32 Today, we're at 90 nanometer devices, active junctions in the order of 65 nanometers inside of those.

00:34:37 This year, we'll go to 65 nanometers, and we'll have active devices 45 to 50 nanometers in size.

00:34:44 That's really, really small.

00:34:46 Our assault continues as we go forward, 65, 45 nanometers.

00:34:51 We'll go into production about two years from now, about the end of 2007.

00:34:55 We've already demonstrated the 32 nanometer devices have already been publicly demonstrated at a variety of conferences.

00:35:03 We already have working models of the 24 nanometer devices underway.

00:35:08 As we've continued to look forward to Moore's Law, we've always had about a decade of visibility.

00:35:15 It's sort of like driving in the fog.

00:35:17 Our headlights only go so far, but as you proceed forward 100 meters, your headlights proceed 100 meters further into the future.

00:35:24 That's the picture of Moore's Law.

00:35:26 I don't know exactly how we're going to solve some of the problems we face 10, 12, 15 years from now.

00:35:31 But as we progress two or three years into the future, new breakthroughs, new ideas, new things emerge that allow us to continue this framework looking forward.

00:35:44 Also, it's very important to realize that Moore's Law isn't bound to a particular instantiation of the devices that we build using it.

00:35:52 Over this period of time, we've gone from bipolar devices to PMOS devices to NMOS devices to CMOS devices, and we're exploring what's next.

00:36:00 The basic nature of the technology built on that silicon scaffolding has allowed us to continue to move forward.

00:36:07 Thus, we believe very strongly that Moore's Law will be valid long past the period of the CMOS generation that has been so productive for us now.

00:36:16 We had some pre-conversations with Gordon as we were coming up on today.

00:36:22 One of Gordon's predictions several years ago, over a decade ago, would have said that the silicon oxide layer itself was the fundamental barrier to scaling.

00:36:31 As a result of that, we would have stopped about two generations ago on Moore's Law.

00:36:36 If you haven't noticed, the semiconductor industry is still going strong, no problem.

00:36:40 Gordon wasn't quite right in that prediction.

00:36:43 We're now looking at new structures, new materials like high-k dielectrics that fundamentally change the nature of the devices themselves.

00:36:49 We're looking to new forms of lithography, like EUV.

00:36:53 We're starting to experiment with other forms of devices, like the 3-5 device that's shown here, about three times the transistor speed and about one-tenth the power product of our current CMOS devices.

00:37:05 More advanced research in things like trigate devices, where you actually clad the entire active area of the transistor, three sides of it, with gate structure.

00:37:15 Things like carbon nanotubes, which should allow us to scale well into the single-digit nanometer levels, have been shown.

00:37:22 We're even exploring entirely other properties of Moore's Law, MEMS devices, things that we might be able to use, other properties.

00:37:29 Another one that I have here today is we've recently shown the first silicon photonics actually producing light out of silicon.

00:37:38 And this is our first silicon photonics lasing device that we've produced, applying yet different aspects of Moore's Law and being able to look at other properties that can be done as a result of our continuing exploration of both the chemical, physical properties.

00:37:57 The result of that is we believe very much that we can continue forward, decade, decade plus.

00:38:05 In fact, I'd say we feel more confident now about the next decade of Moore's Law than we've probably felt at any point in the last three decades of Moore's Law.

00:38:15 However, there are some very significant challenges that we are facing.

00:38:19 One of those is power.

00:38:21 And this is a graph that I showed in a speech I gave in 2001 at the International Solid-State Circuits Conference.

00:38:30 And I showed this incredible assault of power density.

00:38:33 And what we saw, what was happening was the devices were becoming leakier.

00:38:36 We were facing increasing challenges of RC delays on the chip.

00:38:40 So we had leakage.

00:38:41 We had on-chip connectivity issues and voltage scaling.

00:38:45 We had scaled from 5 volts to 1 volt very successfully.

00:38:48 We weren't going to go from 1 volt to a tenth of a volt.

00:38:51 And as a result, power density was just exploding on our chip.

00:38:54 And this was a huge wake-up call for the industry.

00:38:56 And a number of people said, oh, as a result of this and Intel and Pat's predictions here, Moore's Law was dead.

00:39:03 Well, this is actually what's occurred since then.

00:39:06 And the wake-up call to the industry has been quite successful, a fundamental shift in looking at the direction of Moore's Law, addressing the power issues associated with it.

00:39:17 And we at Intel call this the right-hand turn.

00:39:21 We made a fundamental shift in what we optimize for, what we build, and the kind of devices that we're operating on.

00:39:27 So today I'd like to, you know, in effigy see this slide burned.

00:39:32 It seems like an appropriate thing for a power slide to be burned because we have fundamentally as an industry seen a challenge and adjusted to it.

00:39:42 Just has been the case for the last 40 years of Moore's Law.

00:39:48 And some aspects of how we've approached this, one is this incredible challenge of leakage.

00:39:52 And what's happened is as, you know, we have gate leakage, bolt leakage, gate to drain, a whole bunch of effects that have created more leakage power.

00:40:00 Unfortunately, as the devices have gotten smaller, we've gone from having tens of layers of silicon dioxide to few layers, four, five, six layers of silicon dioxide.

00:40:10 There's more and more leakage current inside of it.

00:40:12 And what we've realized is if we, you know, move our center point of design, we can actually very much change the characteristics of the devices that we produce.

00:40:22 And in this case, what you've seen is that, and this is at a constant voltage, the graph is assumed.

00:40:27 So at one volt or, you know, some constant voltage, you can create devices that have characteristics of the amount of gain that the device produce.

00:40:37 That's the horizontal and the vertical axis is the amount of leakage that the device produces, okay?

00:40:44 And of course, right, you know, we want the horizontal, but you'd really not like to deal with the vertical axis.

00:40:51 And unfortunately, the vertical axis is exponential.

00:40:54 So minor movements there create huge shifts in the characteristics of the device.

00:41:00 But what we've seen is that you can use that curve and operate on it anywhere you like.

00:41:06 So in fact, if we're ready to have constant device performance, we can actually get dramatic reductions and incremental improvements in the devices themselves as we go to generation to generation.

00:41:19 And these are some of the techniques that we're now exploring for fundamentally addressing challenges such as leakage current.

00:41:28 We're also facing new challenges as we get these devices smaller and smaller is the random variations or the impact of variations on the resultant devices.

00:41:37 And, you know, when you used to dope a channel with hundreds of atoms, it was fairly uniformly distributed.

00:41:44 And now that you have tens of atoms, they're not very uniformly distributed, right?

00:41:48 And the result of that is you get very dramatic variations in the performance of devices.

00:41:53 Yet another challenge that we're currently dealing with as we look to the scaling of devices into the future.

00:42:00 So challenges, yes.

00:42:03 Optimism, confidence, maybe a self-fulfilling aspect of Moore's Law.

00:42:10 We see it as being valid for decades into the future because of the innovation, the capabilities of groups such as this.

00:42:19 Now, I'd like to explore finally the question of what do we use it for?

00:42:25 What, in fact, if it remains valid, you know, is there benefit and capability from the continuation of Moore's Law into the future?

00:42:34 And one of these, and, you know, it was touched on last evening.

00:42:37 Thank you, Dr. Brooks, for your comments in this regard.

00:42:40 He set me up perfectly for today.

00:42:42 But one of these is can we continue to produce performance?

00:42:45 And over the life of Moore's Law, you know, so we've had Moore's Law at an exponential rate.

00:42:50 And we've had the performance that was produced as a result of Moore's Law at an exponential rate but at a slower rate than Moore's Law.

00:42:59 So if Moore's Law would double over a two-year period of time, giving us twice the number of transistors, the performance wouldn't be 2x.

00:43:07 It might be 1.7x or 1.6x over that period of time.

00:43:12 And that has continued over a long period of time.

00:43:14 And as we face some of these challenges such as power and other things, that rate has been declining.

00:43:21 Moore's Law, doubling the number of transistors, the performance that has resulted from it has been declining.

00:43:26 So over the last four years, we've approximately tripled our performance.

00:43:31 And the question is what will happen as we look forward to our performance in the future?

00:43:37 How much realizable capability will you have?

00:43:40 And this is this dramatic shift to multi-core processors.

00:43:43 I believe it's the largest shift in architectural directions that we've ever faced.

00:43:50 And this is a picture of five of our dual-core processors.

00:43:53 We've said we have 15 of these under development.

00:43:55 I spoke to the Montecito development, our large dual-core Itanium, 24 megabytes of cache, big, huge die.

00:44:02 That's as big as wafers used to be, Gordon, right?

00:44:05 Just one die and some of our particular server and desktop and mobile products as well.

00:44:11 So the shift to dual-core occurring extremely rapidly in the industry.

00:44:16 First products are now coming available.

00:44:18 And nominally, as we go forward, we'll follow Moore's Law.

00:44:23 So at 65 nanometers, everything will be dual.

00:44:25 At 45 nanometers, everything will be quad.

00:44:28 As we go to 32 nanometers, everything will be oct.

00:44:32 We'll be able to essentially follow this.

00:44:34 Now we just have this little problem of programming, which, of course, Rodney will take care of us, take care for us, as we heard last evening.

00:44:41 But the results of that is going to be over the next four or five, maybe ten years,

00:44:46 we will see the fastest rate of delivered performance improvement as a result of Moore's Law and that transistor budget that we have had in the history of Moore's Law.

00:44:55 So if Moore's Law is out here at 2x every two years, our performance is actually going to be faster than that as a result of this shift to multi-core for the first time in the history of computing.

00:45:05 A stunning rate of performance delivered as a result of this shift.

00:45:09 And, of course, the question is, is this a field of dreams, right?

00:45:13 You know, if we build it, will they come?

00:45:15 You know, will there be some opportunity to take advantage of this capability in the future?

00:45:19 When I introduced the 386, I remember it as my chip.

00:45:22 You know, the first ads in the Wall Street Journal, I thought it was my baby, right, being described there on the pages of the Business Journal of the World.

00:45:30 But skeptics says, who will ever need 16 megahertz of performance?

00:45:36 And 32 bits, Carver, are you nuts?

00:45:39 Right, that's minis and mainframes.

00:45:42 And yet we've continued our assault forward.

00:45:44 And what we've seen is that as we go through the generations of capabilities of the products, we go through these steps where, right, the software capabilities and the demands of those are here, and Moore's Law keeps marching forward.

00:45:55 And then there's a big enough gap where a new domain of applications or algorithms is possible.

00:46:01 And then there's this leap in software capabilities.

00:46:03 We went from DOS, right, into the mega era where it became graphics, into the giga era where it became media.

00:46:10 And the question is, as we go into the tera era, we're able to deliver teraflops and terabytes of capabilities, what will be those new defining applications and the potential of those?

00:46:20 And we heard just a little bit about this.

00:46:22 And thank you, Rodney, are you here this morning?

00:46:24 Yeah, over there.

00:46:25 You know, we heard some, you know, saw a few examples of this.

00:46:27 I'll give just a few as well.

00:46:29 We've analyzed in Intel.

00:46:32 We've looked at literally hundreds of core algorithms and looked at the scalability across parallel architecture such as this, as well as the domains of those potential applications and trying to predict some possibilities for the future.

00:46:47 You saw a few specific examples last night.

00:46:50 We've characterized those into these three categories, recognition, mining, and synthesis.

00:46:55 As we go into the domain where we're delivering hundreds of gigaflops in high volume, inexpensive, what types of things can we possibly do?

00:47:02 And these are some of the areas that we're exploring and we think show great promise when we go into this tera era potential for applications.

00:47:11 And for recognition, for mining, you know, in mining I would say imagine that you can Google everything.

00:47:20 You know, literally as we saw Rodney describe last night, you'll be able to carry, you know, on your person huge percentages of the entire knowledge base of the world.

00:47:31 Imagine you can Google at any time and in any way possible.

00:47:34 And not just Google it in the text domain, but Google it in the speech domain, the video domain, or the model domain as well.

00:47:42 And then finally that we can synthesize that world, right, where literally you can't tell where the real world ends and the synthesized world begins.

00:47:51 And just one example of that, Rodney had his last night.

00:47:54 And what's going on here in this picture is, right, you know, we in this case, and this is what, you know, what you're seeing on here is literally, you know, a day's worth of computing.

00:48:06 But all of the physics, the actual physics are being modeled precisely with all of the equations associated with the physical devices that are going on here.

00:48:14 Right, so you see real time physics occurring.

00:48:16 Right, you see fluidics occurring.

00:48:18 You see the real photonic equations, ray tracing occurring.

00:48:23 So we're actually modeling the actual characteristics of light as well.

00:48:28 And, you know, as a result you see this and, you know, you can't tell where reality ends and the synthesized world begins.

00:48:36 And we believe things like this will now become possible as we move into this domain of terror computing.

00:48:43 And imagine some of the possibilities that will occur.

00:48:46 Some of them will be small.

00:48:47 Some of those will be big.

00:48:48 But some of the problems that we believe will be uniquely solved and possible will deliver new types of devices that will, you know, truly be able to cross the human computer interface boundary.

00:48:58 They'll be able to interoperate with the physical world in new ways.

00:49:02 They'll be able to deliver capabilities at the way that they're able to analyze data, access data.

00:49:09 The vision that I would have is that over the next 20 years we will be able to take advantage of Moore's Law such that we will touch every human on the planet.

00:49:19 Today all six and a half billion of them.

00:49:22 That we will be able to touch them seven by 24.

00:49:26 Some aspect of the technology we'd be delivering to them every minute of the day and in every modality of their life.

00:49:34 Right, in some way, shape, or form.

00:49:36 And if we take that definition, right, of our target market, we're maybe three or four percent of the way there.

00:49:43 And we think that is the great potential of the next decade or two of Moore's Law is to finish the task that we've started.

00:49:51 Now, I do, just in one, in closing, and, you know, every speech should have a call to action.

00:49:57 So what do I need you to do?

00:49:59 Well, there is one very important thing that we would challenge, encourage, or ask you to consider.

00:50:04 Are we laying the seeds for tomorrow?

00:50:07 And, you know, maybe in Arnold's words, as we just heard him kick off here, you know, are we putting enough sodium and potassium into our, you know, educational system for the future?

00:50:17 And, unfortunately, we think there's great fear in this regard.

00:50:21 And I, myself, Craig Barrett at Intel, and others have spoken quite profusely on this topic.

00:50:28 We believe, in fact, that our nation is headed to become a second world nation.

00:50:33 That the results of science, material science, device engineering, IT technologies, that, and the lack of core investments, the weakness of our educational system,

00:50:43 our national policies in this regard puts us on a very frightening trend, right, where we are not laying the seeds for tomorrow.

00:50:50 And I think people in this room can make a much larger difference to that picture for tomorrow,

00:50:54 because we're here as a result of those types of innovations and capabilities that occurred in the past.

00:51:00 And I think it's our job to lay those seeds for the generations that follow us and that we have yet to come.

00:51:06 And even more importantly, and more staggering to me, is every place else I travel in the world,

00:51:11 and I travel quite a bit to other places in the world, they are passionate about engineering, chemistry, science, technology,

00:51:18 and they are investing in it at an incredibly profusive rate.

00:51:22 They believe this is their meal ticket to become the United States of the future.

00:51:26 It's unfortunate that the United States doesn't believe that any longer.

00:51:30 In conclusion, who knows what tomorrow will hold.

00:51:34 We believe firmly, passionately, optimistically, that Moore's Law is alive and well,

00:51:42 and will continue to be for decades into the future.

00:51:46 We believe that, yes, in fact, if we build it, they will come.

00:51:49 There will be applications, capabilities that will uniquely take advantage of these types of devices that we're building.

00:51:55 And ultimately, I'd like us all now to take out our calendars and to book the date where we can be back here again

00:52:01 celebrating the 50th anniversary of Moore's Law.

00:52:04 Thank you very much.

00:52:17 Okay, we've got time for perhaps a couple of questions.

00:52:28 I'll wait here until somebody puts their hand up.

00:52:34 Do you want to come to the mic?

00:52:38 Do you think you can continue the less than 100 nanodollar per transistor trend with EUV?

00:52:44 Well, again, FAB costs have continued to increase generation by generation.

00:52:50 Now for us to do a modern 300 millimeter high volume 100k wafer start FAB is about a $3 billion investment today.

00:52:59 And the result of that is we've done some very careful modeling of this,

00:53:04 and assuming certain bounds of our ability to get what I'll call scaling benefit,

00:53:09 e.g. the device gets cost reduced as we go forward, and our ability to reuse that FAB equipment.

00:53:15 And we've modeled that very carefully across very wide parameters.

00:53:19 We see no change in that fundamental graph as we go forward for several generations into the future,

00:53:26 and we've modeled it in great detail to that as we've explored different boundaries of equipment reuse,

00:53:31 cost of equipment, and the cost of wafers, the number of layers of wafers, and so on.

00:53:38 And we see absolutely no change in that picture going forward.

00:53:41 Of course, there's lots of uncertainty in this business,

00:53:44 and our expectation is if you're not investing in capital at this 2 or 3 billion year per year rate,

00:53:52 you shouldn't be in the business.

00:53:53 It's simply you can't stay current on the technology,

00:53:56 and you need manufacturing scale to deliver nanometers or nanodollars per device.

00:54:03 But we see absolutely no change in our ability to do that.

00:54:06 Again, you know, decade, decades into the future.

00:54:16 I've got two questions.

00:54:17 First is how soon do you expect that we will run out of the optical lithography,

00:54:24 and we'll have to do something else?

00:54:26 Of course, x-rays will be cast away.

00:54:31 And the second is how soon will we have to consider cooling,

00:54:37 which is like liquid nitrogen or something else that will not be able to cool it with air or even water?

00:54:47 Well, I love your first question.

00:54:49 How many of us would have thought that we'd be able to stay on 198 nanometer light as long as we have?

00:54:56 Because today, with that generation of light, we're able to produce lines at 45 nanometers.

00:55:03 I still find that amazing.

00:55:05 I'm not an optics guy, but I still go and I look at that, and I'm still just amazed.

00:55:09 How do we do that?

00:55:10 And after you finish explaining to me, I say, well, tell me again.

00:55:13 How do we do that?

00:55:14 It just seems phenomenal to me that we're literally at the quarter wavelength of the light source that we're using,

00:55:22 and we're still producing patterning at that level.

00:55:25 When we go to EUV, we go to 13 nanometer wavelength.

00:55:29 So 198 down to 13 nanometers, a dramatic reduction.

00:55:34 And we haven't then, so now we're to a quarter of the size of the dimensions that we're now printing at that level.

00:55:42 And if we follow that, and we can take that to literally the same kind of half or quarter wavelength printing capabilities,

00:55:50 optimal proximity correction and all these other wonderful things that have been invented,

00:55:54 that should allow us to produce devices and be able to print them well below 10 nanometers,

00:56:01 probably very close down to 3, 4 nanometers.

00:56:05 So it's stunning, and that's a good 15, 20 years of lithography into the future.

00:56:10 In EUV, you can think of it sort of like a soft X-ray kind of domain.

00:56:15 So that does not seem to be a fundamental challenge right now, of course, getting EUV done.

00:56:21 And Elsa will talk about some of the material challenges, et cetera, associated with that.

00:56:25 Well, do you actually have resist and other things that work at that level?

00:56:29 But I assume she'll figure that out just fine.

00:56:32 And I'm sorry, what was the second question?

00:56:35 When will you have to go to cooling?

00:56:38 Cooling, never.

00:56:40 Far beyond air?

00:56:41 Never.

00:56:42 Never.

00:56:43 And that was sort of part of the point to the power density slide,

00:56:46 because fundamentally the thermal envelopes of these devices is well-defined.

00:56:52 We're not going to put chillers into our cell phones.

00:56:54 They're not going to happen.

00:56:55 And outside of domains and the cost domains of very high-end computing,

00:57:00 I think we're on air and other maybe closed-form liquid cooling systems

00:57:05 for things like laptops and so on might actually justify those kind of things.

00:57:10 Fundamentally, we've said instead of this continuing increase in power and power density,

00:57:16 we fundamentally altered that curve.

00:57:19 And now it's approximately, a notebook is a TDP, a thermal range of 30, 40 watts,

00:57:26 and we're going to stay in 30 or 40 watts as far as the eye can see into the future.

00:57:30 We're going to keep driving down average power because battery density

00:57:34 and delivered energy is only increasing 3%, 4% per year,

00:57:39 so not assuming any fundamental breakthroughs in battery technology,

00:57:43 but yet the customer demands are to continue to have a longer battery life.

00:57:46 Boy, that says we have to operate essentially at or below the power levels we are today

00:57:52 to deliver those kind of mobile computing experiences which are exploding into the future.

00:57:57 Now, of course, if you come up with some great new breakthroughs in being able to deliver power,

00:58:02 for example, at Moore's Law rate as opposed to at 3%, 4% per year, maybe things change,

00:58:06 but we don't see that on the horizon.

00:58:09 Thanks. I'll need to cut the questions off now, or Arnold will be after me.

00:58:12 Thank you so much.

00:58:13 I'd like to thank you again.

00:58:15 Applause