Gate last better? Not so fast say some

Even if they go gate first instead of gate last, they are still following suit. Intel claimed that HK/MG was the best way to go in order to continue the process advancements. AMD and most other semiconductor related companies said it wasn't needed. Now AMD and the others are all trying to get their own HK/MG going.

Of course they claim no problems but the push backs and delayed technology says otherwise. There has been little to no word on BD being out in Q1/Q2 2011 and not even a small amount of news. I would imagine that IF GloFlo, IBM and the others had no problems with HK/MG that it would be on track for a release near the same time a SB hits.

Some people think that Intel just magically came upon this but its not quite like that. They did years of reasearch using many different materials and it took them a lot of time to get it just right.

Until we see it in action, Gate first/Last will be fought out. I doubt there is anything significant to gate first. Just like SOI, there is a reason why Intel went the other way. SOI has great potential in some areas but sucks at smalled nodes when it comes to yield. So there is a reason why Intel went to gate last instead.
 
All I know is, theyre all copying IBM using silicon too.

Also, no one said it was never needed, it when when they thought theyd need it applied.

Also, since its being mainly driven by GF, and its turmoil of ownership, some delays are expected there, as well as the other partners non need to have it as well, in the DT cpu environ, yes its different, but AMDs voice is one of many here.

Ive left this open as a question, because til its here, we simply dont know

Claiming to know BD is just waiting in the wings for this, I cant make such a claim, can anyone here?
Either AMD is kickin butt in dev, and is waiting, or theyve timed it out just right, and still, we dont know how good itll be today
 
BD has to go to at least 32nm. I doubt it will work as well on 45nm. In fact all their info from AMD has shown its said to be 32nm HK/MG. Unless they found a way to also do 32nm SOI non HK/MG at the same time as they are pushing 32nm HK/MG, then it will be HK/MG.

But I tend to watch Intel as a idea of whats going to be next in the process technology. While their arch may not always be the best (Netburst) their process tech is pretty much top of the line no matter who you look at or ask. Intel stayed away from SOI for a reason. AMD went with it because they rely on IBM for newer stuff due to not having the deep pockets Intel does. Its also why they have a consortium for HK/MG.

If Intel didn't go gate first there is a reason. As I siad before, Intel started researching this quite a bit back because they saw a problem using just Silicon. Hafnium was the answer after many trials and errors.

SOI was skipped by Intel because they saw that the yields were not worth the investment. I would imagine its something much like that for gate first or something else.
 
Hmm, the only actual fact we know is that GF has delayed their SOI 32nm node due to "yield ramp" problems. This does tend to agree with what TSMC (and Intel) pointed out as problems with the gate-first approach - having to cook the hafnium formula to survive the annealing temperatures, which according to Intel anyway means that the Hi-K gate is tuned to survival rather than optimum leakage.

But, I agree with JDJ that we need to wait until shipping products can be compared to each other, as all the article quotes & promises are from paid mouthpieces :D.
 
True, but also, that was then, and that was Intel, as each does things slightly different.
Who knows, it could become a very nice process, one that may be better or not
Tho, since both sides have alot of monies tied into their own approaches, its not surprising each promotes its own, and it was Intel that said AMD would need it at 45nm also, and they also use bulk, and shunned soi years back, but people learn
 
^ From what I've read about the diminishing returns with SOI at lower nodes, I tend to agree with Jimmy about SOI not being worth it, esp. given the 1/3rd extra wafer cost.

There was the IEEE article I linked to over a year ago, about the 2 giga-pascal strain benefits that gate-last gave Intel.
 


yeah, it's those genius engineers!

Oh wait - that's AMD!!

Sorry jennyh :D...

 
Correct me if I'm wrong, but didn't/isn't IBM been using HK/MG since ~2005ish? Would the not have perfected HK/MG by now after all these years?

And yeah, I agree with jimmy. Intel has been doing damn well with their HK/MG (since what? 65nm?Which would mean 3 gens?). With the research budget that Intel has (albit a bit smaller compared to IBM's full R&D budget IIRC) I'm pretty sure Intel found out about something that AMD/IBM,et al are missing about HK/MG.
 

eyefinity

Distinguished
Mar 6, 2010
1,106
0
19,310


SOI and HKMG aren't mutually exclusive, that's why Bulldozer has both.

Intel stayed away from SOI for a reason.

Yes, cost. Cost per wafer is higher with SOI, and intels business revolves around selling many wafers worth of chips. Much more than AMD can make.


AMD went with it because they rely on IBM for newer stuff due to not having the deep pockets Intel does. Its also why they have a consortium for HK/MG.

IBM are in the Fishkill Alliance and they have deeper pockets than Intel. The Fishkill Alliance exists to share the future cost burden of chip manufacture.

SOI was skipped by Intel because they saw that the yields were not worth the investment. I would imagine its something much like that for gate first or something else.

No, intel has never used SOI because it would be too expensive for them. They had never used immersion lithography until 32nm where they were forced to use it. This is the same as AMD being forced to use HKMG at 32nm - both are required at 32nm or it just doesn't work. SOI is different, it's a choice made by AMD which allows them to pay a small extra cost per wafer, but gain more in other ways like lower power draw. This is what lets AMD stay close even with a year or so lag on process.
 

eyefinity

Distinguished
Mar 6, 2010
1,106
0
19,310


GF hasn't delayed their 32nm SOI node, AMD delayed Llano because of yields learnings. Llano has a powerful gpu on-die, and no gpu has been made using SOI or HKMG yet.

It's got nothing to do with gate first.
 
Hmm, unless you actually work at AMD or GF, you're limited to the same sources we are, which is Dirk Meyer's statements at the Q2 earnings report:

"Llano - our Fusion APU offering aimed at the higher end of the client market - is generating positive customer response. However, in reaction to Ontario’s market opportunities and a slower than anticipated progress of 32 nm yield curve, we are switching the timing of the Ontario and Llano production ramps. Llano production shipments are still expected to occur in the first half of next year," said Dirk Meyer, chief executive officer of AMD, during a conversation with financial analysts.

...

"We have seen the rate of yield leaning below our plans on 32nm. [...] We take a bit more time to work on the 32nm yields up the curve. So, the effective change [...] to our internal plans on Llano amounts to a couple of months," said Mr. Meyer.

I don't see anything in his statement that denies problems with GF's 32nm node. And XBitlabs hasn't ruled out problems with GF: "Earlier it was expected that AMD will start to ship its Llano accelerated processing units (APUs) for revenue already in Q4 2010 with official product launch taking place sometime very early in 2011. It is not clear whether AMD has problems wedding the Llano design to the 32nm SOI fabrication process, there are issues with the process itself or the design of the Llano has certain flaws."

So, do you work for AMD or GF??
 
OTOH there's also the issue of AMD changing the Bulldozer date from Q1 2011 to 1H 2011 to just "2011" on the roadmaps. Actually a google search turns up Meyer statements about "ramping up" the first 32nm chips in "mid-2010", but those were dated in early 2009:

http://www.xbitlabs.com/news/cpu/display/20090305165448_AMD_Delays_Bulldozer_Processors_to_2011.html

AMD Re-Schedules Bulldozer Processors to 2011.

AMD’s First “Bulldozer” Named “Orochi” Aimed at 2011

[03/05/2009 04:54 PM]
by Anton Shilov
Advanced Micro Devices on Thursday said that the company had reset plans for the launch if its new micro-architecture code-named Bulldozer and the first truly next-generation processors will only emerge in 2011. This slightly contradicts the optimism caused by comments of the company’s chief executive officer, who promised to “ramp up” the first chips produced at 32nm node in mid-2010.

“Our guidance for the Bulldozer CPU core was reset to expect test silicon in late 2010 and product in 2011,” said Damon Muzny, a spokesperson for AMD, in a brief conversation with X-bit labs.

Dirk Meyer, chief executive officer and president of AMD, said in a recent interview that the company expected to ramp up the manufacturing of central processing units (CPUs) using 32nm process technology sometime in mid-2010. Since Mr. Meyer said back in early 2008 that the first samples of processors featuring code-named Bulldozer micro-architecture will be made using 45nm fabrication process would be out in 2009, the claim regarding the ramp up of chips at 32nm node caused expectations that Bulldozer processors will be out earlier than expected. However, this is not correct.

Also note the revised 2011 roadmap immediately below the quoted portion.

Since there isn't any GPU on the initial (server only) version of BD, but it *will* be on 32nm SOI gate-first HKMG, then that sorta leaves the "first-time fabbing the GPU on HKMG" scenario out in the dark..

Anyway, we're having fun here speculating so don't rain on our parade :D.
 

eyefinity

Distinguished
Mar 6, 2010
1,106
0
19,310


No I don't work for AMD or GF. But, I can read the source Jaydeejohn linked in the OP.

Gregg Bartlett, our SVP of technology and R&D, addressed a number of these misconceptions in his presentation at (recent) GTC 2010 (conference). You can see that there are no Vt stability issues and gate-first offers comparable or superior performance when compared to gate-last approaches. Our 32-nm high-k/metal-gate ramp is in early production at Fab 1 and we are confident in our ability to deliver for our customer and maintain our time-to-volume leadership position in the foundry industry. We are currently accepting designs for all of our 28-nm technologies. Multiple customer designs have already been silicon-validated, and many more test chips are in prototyping at Fab 1 on the way to early risk production late this year.''


Also,

Samsung Electronics issued this response:

''As you may recall, Samsung announced in early June the full qualification of 32-nm low power high-k metal gate in our S Line. That qual included a full 1000 hr high temp operating life (HTOL) and have experienced no such problems.''

Apparently it all started with some Barclays financial analyst repeating year old "news" about the gate first problems. There are no gate-first problems, it's the superior technology.

Llano is the first GPU ever made by GF, the first GPU ever made with SOI and the first GPU made with HKMG. It's also the first true fusion processor so really its not a surprise to see yield learning issues with it. However, it is sampling and expected in Summer 2011.
 

eyefinity

Distinguished
Mar 6, 2010
1,106
0
19,310

I'm not sure what you mean by this but Bulldozer is on 32nm, not 45nm. That's old news, Bulldozer was delayed - more like changed to 32nm instead of 45nm -, but not because of GF.

Also note the revised 2011 roadmap immediately below the quoted portion.

Since there isn't any GPU on the initial (server only) version of BD, but it *will* be on 32nm SOI gate-first HKMG, then that sorta leaves the "first-time fabbing the GPU on HKMG" scenario out in the dark..

Anyway, we're having fun here speculating so don't rain on our parade :D.


But Bulldozer isn't having any problems, in fact it is sampling soon according to John Fruehe.
 
Thats sorta the way I read it too
The old guard is regurgitating old news.

What Ive read is, it makes for a leaner outcome, as in, smaller, and easier to transition as well, for properly scaled chips.
Now, is this info exclusive to older non HKMG chips only? and not including Intel, and its current HKMG?
This is but one reason to wait and see, and not to put all chips in on the old guard only
 


I don't remember the current Bulldozer being slated for 45nm at all. Its possible that the original idea for Bulldozer, which was dropped at one time, was but current Bulldozer is all set and has been for 32nm.

From what I can gather, BD has been delayed more than likley due to the process technology. From all the news we have gotten, BD seems to be fine on the arch side. In fact most of the news we have gotten has been about the arch and not the process which should also be important since its the next step that will lower thermals, power usage and increase performance as well.

Of course you wont hear anything from anyone working on the 32nm HK/MG process claiming it has a problem or is bad. Remember Prescott? Intel knew it wasn't better than Northwood but they would never admit it. AMD knew Phenom I was pretty much crap but they never would admit it.

In the business world you never slander or talk bad about your own product.

I guess we will see in Q1 of 2011 if BD is coming or not.
 
But its different when old claims come about on spurious infos at that, by others/non committed.
Im sure we will hear the story, maybe not in full, but enough to gleen the truth, as to why any delays, and even better, whether itll be just another path,inferior or superior

 
I just go based on what I have seen and from that info from the period when BD was re-announced to now, info has changed. And the arch info has been trikling down much like it always does. But the process info has not.

Thats why I come to the conclusion that IF BD is delayed past Q1 2011 then it is due to the process having either problems or yield issues (which AMD has been known to have yield issues on a new process).
 
Thats what I alluded to earlier as well
But beyond this, which isnt my OP, or its thrust, as therell be many things non AMD using this, Samsung as example, are close to ramp.
So, its not waiting for BD thatll tell the story here, or at least fully tell it all.

This could be a smear, truthful, or even set down to laziness, and following the leader
If this approach is good, itll change quite a few things