[tw]

Originally Posted by Bitman

Intel's a bad example to use in this comparison; the original mandate for the Pentium 4 was "more megahurtzes at any cost," but the engineers failed. The 130nm chip contained circuits that just barely worked; when they moved to 90nm, they had to overhaul quite alot, doubling the number of transistors required. That alone burned up all the advantage that 90nm brought.

First off, Intel completely bypasssed 130 nm technology - shocking the semiconductor tool manufacturers who were counting on 130 nm sales. Intel went right to 90 nm technology for their future products while other manufacturer's (IBM and AMD) are still trying to develop 130 nm products.

Intel remains ahead of the competition in manufacturing smaller transistors. That was Intel's crown jewel - their semiconductor magic resulted in smaller transistors with less power and higher yields. Now that gates are routinely on the order of as little as three atoms thick, power reduction was obtained by shortening the transistor - requiring lower voltages. Some parts are now running on as low as 0.8 volts. Even that voltage reduction is approaching a limit - another part of the brick wall.

Two technologies that hoped to continue a quest for Moore's Law were strained silicon and high K dielectrics to replace glass - Silicon Dioxide (where the transistor is three atoms thick). (BTW glass is the primary reason why semiconductors are based in Silicon.)

Strained silicon is marching forward slowly and running into manufacturing problems due to mechanical changes in the substrate. The many high K replacement materials have failed. IBM's high K solution (from Dupont) was pealing off - causing a default on an IBM contract. Little has been spoken of Intel's high K material based in a secret halfnium material, even though Intel have successfully built a 40 nm static Ram. Intel microprocessor design always begins with a prototype static RAM.

But these new technologies are not appearing fast enough. Intel has hit a brick wall - probably in part due to their (fortunately outgoing) president who, BTW, is an MBA. Many of Intel's design groups have been doing the safe method rather than what Intel normally does. All symptoms of top management that did not come from where the work gets done. Intel's product line for the next four years may not maintain their previous technology advances. The question why is the question. Without a breakthrough, other semiconductor manufacturers will suffer the same fate.

Why Intel has hit a brick wall is interesting speculation. For example, even with warnings that the next Pentium would clearly have heat problems, Barnett did not encourage (or permit?) a fundamental shift in design objectives. Therefore years of development work were trashed only because Barnett and his top management did not listen to warnings from the little people (a factor common in MBA type managers). Suddenly a multi-chip processor must be rushed into design maybe without the proper architectural evaluation. Itanium group was making promises they could not keep. Top management stifled the Intel 64 bit solution much too long in a blind worship of what they were told. Suspect the current technically less saavy Intel management could not see that technical details did not jibe with summary declarations. Another symptom of MBA management. With management that naive, then what inside the semiconductor process development group was also stifled?

It is quite possible that Intel's stumbles are more due to Barnett - who does not come from where the work gets done and who is being removed rather early. But again the question - has CMOS technology hit a brick wall?

Bottom line is that others (ie IBM and AMD) have yet to use the concentration of technologies that caused so much heat in Pentiums. Competitors also have not succeeded in developing technologies that hopefully avoid these heat and speed problems. All manufacturers will see a sudden increase in heat as they too begin using the same smaller technologies.

Maybe the term brick wall is not appropriate. Maybe mountain is a better term. The mountain has suddenly become very steep and is only getting steeper. We know the CMOS transistor is approaching its limits. Limits imposed by the size of atoms, the amount of quantum tunneling that occurs even through glass, and a lowest limit voltage that a silicon transistor can use. Without some major breakthrough, faster processors based upon speed increase will not be possible. Currently the most promising breakthroughs just are not happening.

BTW you don't have to know about semiconductor technology to appreciate what is being stated. You don't care what quantum tunneling is. Just that quantum tunneling is a problem AND that solutions for that problem are not appearing. You don't care why processors use lower voltages. Just that lower voltages were another solution to the speed / power consumption problem - and voltages are approaching minimum values. Bottom line is that no one in the industry really has succeed in the labs with a solution to these speed and power consumption problems. As we know from previous discussions, everything takes four to 10 years to develop. This suggests a serious problem will be confronted in the next four years if something breakthrough does not happen.

Currently the only promising solution is something called strained silicon. That means less power consumption only when transitor is not on. Less power at the same speed. What then? What comes next?

How does one make money in the stock market? You are seeing it demonstrated here. Place your bets. Select your numbers. The wheel will spin again in January.