Semi technology, business models converging, IBM’s Meyerson says
By Ann Steffora Mutschler, Senior Editor -- 7/16/2008
SAN FRANCISCO--Pointing to IBM Corp’s announcement Tuesday afternoon that it will invest $1 billion in its East Fishkill, NY, semiconductor plant over the next three years, along with another $500 million on R&D and capital equipment for the University at Albany's College of Nanoscale Science and Engineering, Big Blue’s Bernie Meyerson began his keynote address Tuesday at the Semicon West tradeshow by noting that the news is “really about the challenge of keeping this industry going. The challenges are technical and the challenges are to the business model.”
By way of background, Meyerson, IBM's VP for strategic alliances and CTO of the company’s systems and technology group, referred to Moore’s Law to illustrate the economics of computing, reminding that Gordon Moore made a brilliant observation that has held true for 30 or 40 years which states that the number of transistors that can be placed on a chip increases exponentially, doubling approximately every two years, and “if you do this, you win. The technology becomes sustainable and more importantly, you continue this trend we all count on,” he said.
“We drive the economics of computing and we drive them hard. If you look through the years, what has happened is in the timespan let’s say from 1980 to 2008, our period of time, you’re looking at an industry who has seen the performance improve in IT by, roughly speaking, 7 or 8 orders of magnitude: tremendous, tremendous progress. And if you look at that, you’re looking at a 70/80/90% compound annual growth rate in IT performance. This is done based upon the fact that a thousand dollars of computation expense (what you spend) used to get you - give or take - about 103 (1,000) computations per second, now $1,000 is getting you over 109 or 1010 per second.”
In fact, IBM also has a machine called RoadRunner, which is the first machine that breaks the petaflop performance, which was made possible financially through Moore’s Law, Meyerson noted.
One of the critical regions on a chip is the gate oxide, which has gotten down to about five atoms. “What are you going to do when you scale it by 2x? Are you going to cut the one atom in half and do nuclear fusion? That’s not going to happen, so you’ve got to find another tactic. Furthermore, you can’t just ignore the recipe, or you’ll get in big trouble,” he pointed out. And if the gate oxide is not scaled, it is not possible to get the performance wanted.
Meyerson admitted that IBM has experience with overscaling, as it did just that with its bipolar technology and the power when through the roof. It moved to CMOS manufacturing technology, only to watch other companies make the same mistake with CMOS and overscale, causing the power to go nuts, he said.
“Beyond this, there is an opportunity. It’s called, ‘Oh my gosh, what do we do next?’ The bottom line is that this is an enormous challenge and it has had ramifications well outside [what we’ve known],” he went on to state.
Another issue Meyerson brought up is based on a set of economic data that shows the growth rate in spending on servers for the last six or seven years. “Server spending has sat at $50 billion – pretty much flat – for five or six years after the bubble. Meanwhile, the amount of power used to cool and run servers is growing 800% faster. In point of fact, if you go out another two or three years, you see approximately $55 billion a year on the servers, and about $55 billion spent to cool and run them. That’s a bit of a problem because you understand that you are now spending just as much to run them as you are to buy them, except running them is growing at 600 to 800% faster, and oh, by the way, this is before oil didn’t cost $140 a barrel. Life is going to get complicated. Power is an enormous issue, not just in semiconductors,” he said.
“So what do you do about all of this?” Meyerson questioned. “We used to worry when we innovated - we introduced one new thing in a generation, and we bit our fingernails and worried if it wouldn’t work. But now innovation is the entire game. To drive innovation of all the various materials that we’ve introduced and the ones we’ve yet to introduce, there must be a roadmap that gets those materials there,” he said, referring to high-k materials, and air-gap technology.
“We need to drive the folks who know to do this and are willing to spend that investment in the bottom so we can have those things on the top to move forward on,” Meyerson asserted.
“The impact of all these physics on your financials becomes horrendous. If you look at the trend in the industry going back about three years, you basically had a situation where R&D costs were growing at 12.5% a year and revenues about 6.5%, [according to data from VLSI Research]. The interesting thing is, to put a number on it, in the 1950s for every dollar you invested, you got about $60 of revenue in the year; nowadays, you put a dollar in, you get $6 of revenue. That means if you are running at 16% NIBT, you have to put 100% of your margin back in just to do the development. That’s not going to fly as an industry and we know that,” he offered.
What has also happened is that the cost of developing a node are going through the roof because of the complexities. “Remember, it used to be you invented one new thing per generation, and now you are doing 10 or 15. It’s amazing what that does to your costs: it skyrockets. And the estimates are maybe $2 to $2.5 billion to develop the 22-nm node alone, the key word being alone. Not a lot of people are working alone,” Meyerson went on.
To be sure, the industry landscape is now filled with collaborations and the number of players continues to shrink due to consolidation with the business model now one of "coopetition" or competitive cooperation, he reminded. The size of engagements is huge because the costs of R&D are enormous.
These moves reflect two fundamental paradigm shifts. On the business side, it is coopetition – the radical collaboration with the pooling of physical and intellectual capital. “Why is it radical? You’ve got IBM, Chartered, and Samsung, which are head-to-head competitors in the foundry business and they are all part of this alliance. You have IBM, STMicro, and Freescale all working together, even though we are head-to-head competitors in the PowerPC area. TEL, AMAT, Hitachi Hi-Tec – all of these folks compete in tooling but they are all there leveraging the base knowledge, the fundamental physics that we get out of this, and the ability to test their equipment at its native dimensions. Similarly, with the materials folks, there is a community developing,” he observed.
“This is becoming a semiconductor ecosystem which is where we had aimed it and where we seemed to have wound up. You do have a challenge. The outcome is very good, you are getting tremendous progress, the challenge is that engineers tend to hate the idea. They’re not always the best of friends and they are naturally competitive. The way they get over it is that we treat them as one team. When possible, when working on the basics, you are one team. That has worked very well for us,” Meyerson pointed out.
The road forward, however, will look very different than the road travelled. “Technology shrinks are going to abate as a lever within the decade. It’s not that people won’t try to make things smaller, denser, faster, cheaper – of course they will, that’s called running a business – but if you were counting on that to drive the performance, you are wasting your time. There will be many, many other things that will gate the IT performance,” he said.
The pervasive semicnductor technology of the future will come more from how things are integrated, combined with the base technology, all of which will be driven by architectural advancements, Meyerson concluded.
© 2009, Reed Business Information, a division of Reed Elsevier Inc. All Rights Reserved.