We heard about "TeraHertz transistor" as early as 11/2001:

We heard about "TeraHertz transistor" as early as 11/2001:


http://www.intel.com/pressroom/archive/releases/20011126tech.htm

"Our goal is to overcome these barriers and produce chips that have 25 times the number of transistors of today's microprocessors at ten times the speed with no increase in power consumption."

"Power consumption as a limiting factor"

"Intel is expected to begin incorporating elements of this new structure into its product line as early as 2005."



Things looked even better in 11/2003:


http://www.intel.com/pressroom/archive/releases/20031105tech.htm

"Intel Corporation today announced it has identified new materials to replace those that have been used to manufacture chips for more than 30 years."

"Intel said the new high-k material reduces leakage by more than 100 times over the silicon dioxide used for the past three decades."

"Transistors with these new materials are an option targeted to be integrated into future Intel processors as early as 2007, as part of the company's 45-nm manufacturing process."



At first, I think Prescott was engineered with frequency scaling in mind. Prescott was very well engineered -- same IPC even with 50% longer pipeline. The first iteration even at 103W is tolerable. But by 2005, some elements of "TeraHertz transistor" for frequency scaling need to be there especially for Tejas per the 11/2001 announcement. So is there really going to be a frequency scaling problem?

Let's say there is. Then the problem could from engineering failing to replicate research results, bean counters failing to buy new equipment, or marketing killing something because they cannot think of a way to market it. Intel has always overcome engineering problems. But Intel has also always been run by Ph.D.'s. So the later two would be more worrisome. In fact, I seem to remember Otellini saying that a feature wouldn't be put in unless it can be marketed. (And to this day, I still cannot figure out how "net" or "burst" has anything to do with Hyper-Pipeline.)

Or there actually isn't going to be a frequency scaling problem. This means Jonah really is going to be a better solution. Or, at least it will be easier for marketing to tell people "Buy Jonah, you are getting two processors for the price of one". This is great news to shareholders. In the old days Intel just had two design teams. Now the best ideas can be shared among and the best result be picked from at least four design teams (California, Oregon, Israel, India, Texas?, Arizona?).

(OK, I am bad mouthing marketing too much. My guess is that it's a combination of several things, including Tejas is late. But I really don't think frequency scaling is the main cause.)

What's more curious is the IA-64 and IA-32 merge plan. My impression has been that Prescott has 64-bit X86 support that can be turned on if necessary. But the real stuff is that Tejas will have IA-64 support. Now Tejas is no more, how/when is a X86 processor going to be able to start executing IA-64 code?