WR :
Its NOT about TSMC, which is the FASTEST changing in process nodes, and came from 130 when Intel was on 65nm , to 40nm, which Intel hasnt and cant currently do. Its not about Theo, I dont care where info comes from, what he wrote aligns with what EETimes had written as well.
JD, Theo's article is full of errors which then trickle down to what you write. There are way better sources to express the basic plan that TSMC isn't using HKMG till <32nm, and in selecting good sources, you may come to a better understanding of why that has to be.
TSMC has a cheap process at 40nm. It doesn't perform well enough for any mainstream DT/server CPU. Intel also has its own cheap processes - its flash is at 34nm at the moment, probably used in its X25-M flash device. Nothing supports the notion that TSMC jumped ahead of Intel process-wise, as Theo's title indicates, or that Intel is ahead of TSMC, as raw sizes indicate, because Intel does not build the same products as TSMC, so its flash process does not need to be qualified for a broad range of products.
The earlier TSMC uses HKMG, the better its products. See, TSMC could delay HKMG indefinitely, lowering product frequencies as needed where conventional silicon fails to scale. Or it could throw money at HKMG right away. However, HKMG is an investment that does nothing if the work orders are not there. The timing of this investment is only a general indication of how much value they place on a better performing process. TSMC is a bulk foundry, and the bulk of revenue these days is in cheap parts, not high-performing, low-volume ones.
The latter part of Theo's article goes into a GPU discussion where he appears completely blind to the fact that without a breakthrough in chip packaging technology or off-die memory controller design, a small GPU die is also a lower bandwidth die.
Im going on the understanding that HKMG would have a profound effect on gpus, as theyre THE most power hungry and hottest components in your PC. I was reading about Intel using its HKMG in small processes for low powered devices as well, and as we know TSMC moves in this market as well. What Im saying is, the crossovers between Intel and TSMC exists in a few areas, and it appears Intel already begun the use os HKMG is these areas, where itd benefit TSMC as well, and then like Ive said, the gpu arena would benefit hugely by it,
I agree with you on one point tho. It comes down to necessity. Not because HKMG is so great tho, but because the need isnt as great as the lowering of the process is in TSMCs eyes, and their customers, thus HKMG is somewhat overrated, and certainly is as was portrayed previouly here as to its needs at smaller nodes, and even posssibly its overall benefits vs costs, which all has to be taken into consideration in overall evaluation of its use, and usefulness
Taken from here
http://www.tsmc.com/download/english/a05_literature/1_Corporate_Overview_Brochure_2007.pdf
Technology Leadership
TSMC has the broadest range of technologies and services in the industry. The proven path to success embodies a platform
strategy that bundles together process technology options and services. TSMC collaborates with partners to ensure that
all services supporting those technologies represent the best practices in the industry. To that end, TSMC and its partners
deliver the largest portfolio of process-proven IP and libraries, and the industry's most advanced design ecosystem.
TSMC's platform strategy stretches across both advanced and mainstream technologies.
Advanced technology platforms include a wireless SoC platform, a consumer technology platform, and a PC and network
technology platform.
Mainstream technology platforms cover specific applications including power IC, display driver IC, CMOS image sensor,
microcontroller, and RFID.
Ive seen Intels use of HKMG for its SoC packages in process, and plans for TSMCs usage of it as well, but currently, using a smaller process than Intel does, and going into production with it, TSMC isnt using it ...yet
What youve failed to mention about Theo was, he was 100% right on when he said AMD was collaborating with oceers on Deneb, before this all happened, and again, he was ridiculed for it, but was right on the mark. Now, if Id posted his reverse HT crap, and stood by that, yes, include me in with his foolishness, but if you weed out the truth and balogna, and sometimes downright crap, you can use it as a gage for a larger picture, which IS what I did here
I guess if you include only failings and never include success, you get a distorted view of whats going on, and either way, its those that do this that suffer for it, becoming more single purposed, and more closed minded. As many here were wrong in their assessment about HKMG, and its effects on the 45nm node, or lack of it, and the predictions of what would occur without it, those same people have helped others, have brought alot of new info, and taught alot of us many things, so I commend those people for that, but at the same time, just like me or anyone else, they can be wrong, and paint an overall impression, one thats way off the mark, if no one is independent in their thoughts, and just pile on, damaging the overall image of anyone or anything that lacks in their claims, without the higher understanding it requires to independently learn on their own, and see a bigger picture