Very interesting set of posts from Chevyman1963 on Yahoo
Any comments from any of the process guys?
===========================================================
Hey dummy, Dresden can run flash as well as any other product AMD makes. It is all run on the same type of machines and on the same type of wafers. I should know I worked in Fab25 and I ran both there as well as a number of the communication parts. The process of doping etching difussion etc is controlled by the recipes of the machines. A difussion furnace, etcher, implanter do not care what mask is on the wafer it will run via the recipe time and controls that the recipe gives it.
==========================================================
The real problem with copper is contamination not is placement or useage. Copper is so soft as to become a problem with everything it touches causing a transfer that leads to contamination in tools and transfer equiptment. Copper could be used in anything that needs a circuit or contact but, you have to wiegh the cost vs results. If the copper does not enhance the chip enough why use it considering the extra trouble to keep the contamination issues out. There is enough money in a high end CPU chip to make the cost vs rewards significant enough for the effort required.
===========================================================
I am not sure we will ever reach .13 with the current litho technology. We are probably going to have to change to xray litho tech to get that low. I have however seen post in tech mags that say it is possible with current tech. used at U of T. the problem at .13 is in the contacts that connect the layers of circuits etched on a chip. At .13 you run into the problem that the contacts are so narrow that getting the contact material (whatever it is) to flow down and fill the contact without a void is nearly impossible. The reason I say that is because of the size of the atoms of current materials used make it a real problem to fill the voids. The voisd of course causse a yield problem. That is one of the many reasons for the change to copper at this time. It folws better that th Au tha tis currently being used and the Au currently being used will not fill those gaps without a problem. Someone earlier posted .09mu possibilities, I say not witha todays tools and technology. Hell look at he great Intel..It's having considerable problems with .18mu. Look for new materials to be used in the future and look for taller chips. The k5 was a 5 layer metal chip..k6 a 6 layer metal chip..k7 now has 7 layers of circuits etched on it. What will the k8 be????
Just my WAG EOM
======================================================
I wonder if any of you remember that this fab is less than 4 years old, is state of the art and is only 1 of 3 that AMD still has in Austin. Yes it will remain open, yes it probably will continue to run other products just as it does now. Finally look for AMD to shutdown Fab 14 and Fab15 in Austin since they are over 25 years old already and use outdated equiptment. AMD shutdown Fab10 in Austin 2 years after Fab25 went online. Now that Dresden is online I really don't think they need Fabs 14 or 15 since Fab25 can run all of the product without alterations.
Just my OPINION of course. i worked in Fab25 for 3 years.
========================================================
Just thoughthat I would remind you that the tools in Fab 30 have been there and qualling for a year now. Do you think in a year that they could figure the tools out and get them up and running?
Yep they can!! Dresden is up and running production and not a measly 800WSW. Hell 800 wafers a week leaves the tools idle and the employees asleep since most tools have multiple chambers allowing them to run an average of 62 wafers per tool per runcycle. The numbers are etchers 50wafers per run per tool, difussion furnace 500 wafers per run of 10 hours,( and they have at least 6 furnaces), Sinks to remove mask after etch run 200 wafers per hour each, implant tools 50 wafers per cycle,(a recipe cycle is typically 1-2 hours on everything except furnace) litho for mask Applied Materials newest 200wafers per run. Anyway as you see my numbers are probably a little low as to wafers per tool per hour. Now remember all the metrics are completed and the tweaking is over so ther is no reason thet FAb30 can't run full out. Yield is not a problem as evidenced by the speed grades already out therefore all the tweaking is over and the Fab just has to run the wafers. In case you don't remember I worked in Fab25 during the ramp up there and I also helped train the German techs. The German techs left Austin 13 months ago to run FAb30 and qual all the newly installed tools. The second half of Fab30 will not be completed all at once. Just as with Austin, tools are only brought in as production increases are needed. Since both FAb25 and Fab30 are identical as far as the production floor is involved then all AMD has to do is follow is previous plans for installing tools. The difference this time is that they don't have to develop the process as we did in Fab25. Most of the delays in the k5 and k6 were strickly tool problems. I know, one problem tool that cost AMD months and months of delays was my tool set. It was an experimental tool and was rushed to the fab too early. It cost AMD 4-5 months delay itself. In fab25 everything was new, the process, workers,tools, engineers,etc. It's different this time they know the tools abilities now, they know the process and design now and they are not having to deal with experimental tool sets on production runs like we did in Fab25 for the first 2 years. Just my opinion of course..but I am long ..have been for quite awhile. i expect to see AMD print a lot of money in the future.
===========================================================
First off ACE i have been in Fab25 (was there 3 years) and never have they had 4k people in the fab. Austin has 3 fabs and about 18 other buildings that those 4k people work at. The fab is a state of the art automated factory that can be and is run by approx. 300/400 people per shift. The entire etch dept only had 23 people at full cap per shift. That was for the largest process part of the chips: 65% of production time is spent in etch because of the 7 layers of metal, the contacts/Local interconnect etc. My point is the other post was that AMD does not have to reinvent every process and tool theis time. Also Fab30 has been open for 12 (count them 12 months). This is certainly adequate time to tweak tool sets that they had already been trained on. Linearity of the line is controlled by having multiple chamber tools that can allow you to run multile items in your area at the same time. The bottlenecks you are looking for were addressed in Asutin.
They are not at full capacity but they are not at 800WSW either.
They are most likely in the 3k range and you will be suprised very soon to see how close I am on this. I said 3 months ago that I knew about a 750mhz k7 test and yesterday I said look for mid OCT. This is all just my opinion of course but at least I have been to and in the fab.
=====================================================
I have been trying to tell people that AMD has been running this fab for a year and there is no reason that it not producing other than the fact that they run production on a as sold basis. The pass thru time for a processor from start to finish is ~50+ days and AMD's policy for manufacturing si to not start production until there is orders to fill. Otherwise you just waste wafer, time chemicals elec. etc. the fab should be in some sort of production run and not at just a measly 800 WSW like AMDACE would have you believe. Hell they have had a year to get it going since the tools were installed and the tecs came back from training in Austin. Just my opinion however. I am long and looking for a great 2000
======================================================
You say it will take a few years to reach 5000WSW. Well amd did 5000WSW in Fab25 in its first year and a half. Dresden has been funning a year and they were already familiar with the tools since they are the same as the ones in Fab25. That should have taken some of the lead time out of the mix. I will concede some time to the shrink for .18 however and agree that full capacity probably not for a few more months.
Gooo Chevy
Gooo AMD |
