Tuesday, August 26, 2003

Delaying forever

Intel Fellow and Director of Circuit Research Shekhar Borkar speaks....

The transistors are becoming so small that the atoms and molecules that used to look like a continuum now look like discretes. We will shift from the deterministic designs of today to probabilistic and statistical designs of the future.

if you take a thousand samples of IntelĀ® PentiumĀ® 4 processors, there is some speed variation from chip to chip. Some high-speed ones, some low-speed ones, some average ones, all in a nice Gaussian distribution.

In the future, a chip will have millions, if not billions of transistors. Each transistor has a probability of meeting the performance target. In the past, the spread of this probability was very small. But in the future, due to variability in individual transistors, the whole chip could yield different results

. I want to be connected anytime, anywhere, wherever I go in the world. If I connect with a wire, I can get 10 Gbit over the wire. But if I remove the wire while I'm sitting in the airplane, at least I can get 100 megabit/second using a radio. The radio is free in the sense that I don't even know there is an integrated radio in my device. I can either hook up my cell phone or put up an antenna. But it's seamless. I remove my wire and I don't even know it. It might be a little slower, but that's OK.

Radio Free Intel is the vision of adding wireless capabilities to every device by integrating the radio circuits and systems directly into every component. By using the cost and high-volume advantages of standard CMOS (complementary metal-oxide semiconductor) processes, Radio Free Intel will make wireless connectivity ubiquitous and ultimately, easier for people to use.

Today you are limited by power. The practical limit is around 75 watts, because if you look at the cost of cooling, that too will start increasing exponentially. No one is going to buy a thousand-dollar refrigerator to cool a thousand-dollar PC!

So you have to go back and ask the question, "How can I improve my computing efficiency for the power?"

That's where the threading comes in. Right now, there is a wide disparity between processor speed at 3GHz and bus speed at 400 or 800Mhz. This gap results in the processor waiting for a long time when it makes a request to the external memory. With threading, when the processor's execution unit is waiting for data from external memory, a new thread kicks in and uses the execution unit.

During the research, our mentor provides any help and guidance that the student needs. The student comes here for internship. Our mentor is usually on the student's PhD committee, so the mentor is very engaged . Roughly half of our staff in circuit research has come from the mentoring that we have done. We know the quality of employees we're getting, because we helped build their expertise.

Our vision is that in the future, you will have a CPU with a general-purpose processor, and a bunch of power-efficient fixed-point hardware such as the TCP/IP Offload Engine, MPEG encode/decode, and a graphics engine to handle dedicated functions that don't change at all. The general and special-purpose processors will be on the same die, doing the things that they are good at doing, and doing them power efficiently. This is like the System-on-a-Chip (SOC) initiative that the industry is talking about.

Gordon Moore showed a slide saying that for every ant in the world today, there are 100 transistors. Today, transistors already outnumber ants. But my job is to grow the transistor number to 10,000 transistors for every ant. Think of each transistor as a mini-calculator. We'd like to see 100 times more transistors in the world for more computing power

today Intel products are being developed on 65nm technology. In the labs, our research is focused on 45nm and beyond. That's four to six, or eight years out. Now we're working with the universities to do research on 30nm and beyond for products that are at least eight to ten years out in the future.

And they'll succeed. If you never tell a PhD student "this is impossible," they'll just quietly go and do it.

0 Comments:

Post a Comment

<< Home