Posted by Tom Foremski - October 31, 2011
Chipmakers have to be expert in chemistry because every time chip designs move to new, smaller geometries, they often need to discover new materials because the old ones don't work the same way. There are very few elements in the periodic table not used in chip manufacture.
Here is an interview with Intel's Mike Mayberry, who leads its process technology research and keeps Moore's Law alive.
By Intel Free Press
How do you respond to those who say Moore's Law is going to end soon?
My usual comment is, "Well, then I'll be fired first. And since I haven't been fired, then there's at least some period of time before it will end." We typically say we have about 10 years of visibility. We don't know what will happen beyond 10 because you have to invent things along the way.
We know that things will not end in the next 5. And we're pretty sure they won't end in the next 10. That was true when I started 27 years ago. There were things that people said in the '80s that would cause life to end in the '90s. And in the '90s, there were things that would cause life to end in the '00s. We keep extending that 10-year visibility.
Any kind of technology has some natural lifetime. And eventually, you have to replace it with something else. That's what we're working on today. What are the things that come after what we know and love today?
How does Intel Labs intersect with the work of your group's researchers and technicians?
Intel Labs looks at what kinds of products would be interesting to people in 5 to 10 years. We build things. We work with the Labs in some cases because they say, "We can do something really cool if we had a certain building block or certain fabrication capability." And we may say, "Yeah, we know how do to that. Let's work together." Or, we may say, "No, that's impossible. What are the alternatives? What can you do differently?"
We explore technology. Sometimes we find things that work. Sometimes we find things that don't work. We like to do sufficient research to find the fatal flaws and kill our own projects before we've spent millions of dollars.
Intel develops process technologies on its own, while others share the burden as part of an alliance. What benefits do you see to the Intel approach?
No one really works in isolation. Our development relies on equipment that does certain things. If the equipment was not capable, it doesn't matter how great the idea is. So, our research and development engineers work with suppliers. We work as well on basic materials that hopefully can be retrofitted to their tools. We work with universities on very novel ideas and hire the best students.
When people say we work on it ourselves, what they're really referring to is the process of integrating all the pieces together, not that individual pieces that were all created internally. Nobody can do that. In fact, we're intentionally reusing a lot of things for cost reasons from one generation to another.
But that said, the final integration that our technology development group does is unique, pulling in these building blocks that may have come from somewhere else. And they work on a lot of different things to get to high performance and high yield and low cost.
The consortium that works on these same things has both advantages and disadvantages. There's cost sharing. You don't necessarily have to buy one piece of equipment for every company. You share it. Now, by sharing it, you also get fewer hours on it. Your cycles of learning are typically much slower. Our development is optimized for very fast learning cycles and that's a key competitive advantage for us versus these alliances.
What's your secret sauce? How did you get to the right solution on Tri-Gate and high-k metal gate?
We're allowed to start early. We're allowed to take risks. We're allowed to fail. And when we find something with promise we can harness all the great talent at Intel to turn it into a production reality, not just issue a press release.
We tried a thousand combinations of materials to make high-k/metal gate and only a few worked. But then we made those few work in very high volume and we've shipped hundreds of millions of units ahead of our competitors.
What do these advances mean for the customer?
We keep things going, right? The original Moore's Law observation was about lowering the cost of a function. We also enable more complex products which can do new things people want to do. An audio recorder or a music player or a video player, for example, would be delivered at lower cost for a given amount of capability.
Most of these building blocks are about making things either more capable -- say a faster or smaller operating transistor -- and therefore potentially cheaper to manufacture, like a smaller wire or a different kind of function that we're integrating together that hasn't been possible in the past.
It wouldn't do any good to say, "Hey, we're going to deliver something that's not quite as good as what you had before. And it's going to cost more, isn't that great?"
In the research pipeline, what are some of the long-range options that are being explored?
We're looking at discovering the dimensional limits of materials. What happens when you can count the atoms? How do you measure, where do things break? We're optimizing novel devices for efficient operation at lower voltages, which let you pack still more functions within a given power budget.
We're also exploring non-traditional computing elements, which someday may allow very fast pattern recognition or other problems that are tough to solve today.
Are you nervous about Intel's future, the need to continually evolve and push the limits of technology?
If I were nervous, I wouldn't have this job. This is not a job for a nervous person or a timid person because there are lots of things that have to be invented. I'm glossing over all of the challenges that we have along the way. If you try in your head to think of all the challenges simultaneously, your head would explode. This is really hard stuff, but we need to make it look easy when it's done.
Clearly, it's a very different world. We have to recognize that technology is changing how we do things. And then we also have to realize that technology is changing the world and adapt our products in a similar way.
From a management point of view, we have to question why we've been doing something this way for 10 years. Do we still have to? Why did we start doing it this way? Does it still apply?
I'm big on the why behind it, the "I know you don't know how to solve the problem, but why are we working on it?"
What do you think Intel has to do to remain successful over the next 40 years?
The trends are that we have very complex technology that people want to ignore. They want it to just happen by itself. Arthur C. Clarke says, "Any sufficiently advanced technology is indistinguishable from magic." Some of the things we do on an everyday basis and take for granted would have been considered magic 25 years ago, right? Twenty-five years from now, the things that people will do will be magical to us.
We want to continue to make things ubiquitous, embed them everywhere. We want to continue to make things very easy to use, taken for granted, make things more autonomous.
In your 27 years at Intel, what has surprised you most about technology?
In my career, we've gone from no Internet to, when you ask your child to write a paper without using the Internet, they ask, "Uh, how do I start?" If you count all the characters in the books in the Library of Congress, we're building more memory in an hour than that. The point is we're using memory for more than characters in books. We're using it for sound. We're using it for video. We're using it to store experiences.
I think people are not exponential creatures. It's very hard to grasp things that change cumulatively over so radical a rate. We don't really appreciate that.
If urgent: send text or call 415 336 7547
Bacon's names Silicon Valley Watcher one of the
most influential blogs in the US.
SF Publicity Club's ninth annual awards
celebrating excellence in media.