28nm Chips Remains in Short Supply at TSMC

[leeashton]

[citation][nom]phatboe[/nom]I thought TSMC scrapped the 32nm node to focus on and accelerate the 28nm node. Now they claim to want to be more conservative? Yeah this may save TSMC a few bucks here and there but it is destroying it's partner's profits. Mean while Intel is gaining market share in almost every sector that they are involved in. It's so sad that the TSMC is holding back progress for so many companies all to save a few bucks.[/citation]
I dont think they are ding this to save money, I think its a physical limitation of thier ablitites, this is not the first time TMSC has had yield issues.

 

[billcat479]

When I was unlucky enough to work for Intel back in the mid 80's they had a huge R&D plant next to their production fab plant. Intel has a lot of money and have fabs dedicated for new designs which I would guess TMSC doesn't have anything to match it.
And by just looking at numbers people miss the point of how close they are to working with atom's because as they keep shrinking they have to use more and more advanced tools to be able to make transistors and related items on chips while trying to get all the atom's placed onto their chips. The cost in upgrading from 32 to 28 is huge. Then they have to run test after test, try this equipment or another that will work at those size levels.
It's crazy hard and they can only go so far until they have to take radial new idea's if they keep this shrinking going. One day soon they will hit that atomic roadblock and then if they can't deal with quantum issues that size shrinks stop there. It's so easy to see why it's taking longer and longer to bring out smaller IC's and at a huge tech/learning/machines to get the size farther down.
AMD would have needed a lot more money if they tried to keep doing this on their own. I can see why they got out of that side. It would cost them way to much to keep doing it. We are already seeing the new quantum computing being developed and soon we will have computers where even the makers of the chips won't really know how they work even when they do. It will be interesting to say the least. But for the rest it's just way to easy to see why they are having problems and low die yields because all of the steps have to work so much harder to keep the process from any screwups in their construction. It's like walking on a tight rope with 100mph winds, one very small misstep and off it goes.

 

[hannibal]

This is something that has been going on for a while. It is harder to make things cheaper, because development costs are going so high. It is not impossible situation where new chips are getting more expensive (relative to the speed) than old. The would allow smaller footprint, less heat and low powerusage, but maybe not more speed for the money, because they will be so much more expensive to produce.
Who knows, maybe we will see some new innovations, but atom level size is very hard to bypass...

Maybe we will stick in same production node much longer time and just fine tune it and get more speed via chips architectural design changes. The progress would not be so fast, but same machines would be used longer time in production, and so allowing more profit and cheaper costs...

 

[-Jackson]

Sigh, the realities of business.

 

[eddieroolz]

This certainly won't help the availability of Kepler nor Ivy Bridge. iSad.

 

[husker]

[citation][nom]dreadlokz[/nom]short supply? sure, just as diamonds and every other expensive thing in the world... ppl who believe that is just dumb! shortage is how they do money[/citation]
I disagree. I can walk into any jewelry store and have my pick of several expensive diamonds. That is a lot different from an "Out Of Stock" sign in every store window. Like Jewelers, nVidia makes money by actually selling an item, not by driving customers to their competitors because they have no supply.

 

[hector2]

So does "can't meet demand" actually mean that TSMC's die yield is so bad that they need a lot more wafers to produce the same amount of die ? It would make sense that they won't want to ramp wafer starts if the process isn't healthy and need to improve die yields first

 

[hector2]

[citation][nom]eddieroolz[/nom]This certainly won't help the availability of Kepler nor Ivy Bridge. iSad.[/citation]

TSMC's problems have nothing to do with Ivy Bridge availability. That's made by Intel and it matters little to them how healthy TSMC's process is. It's already in Intel's interests to ramp IB quickly because 22nm lowers die cost and improves performance/power

 

[hector2]

[citation][nom]SM4RT3R_TH4N_U[/nom] Fool: I guess you're going to tout that Intel's process nodes are not only ahead of everybody else by 6 to 12 months, but you are going to suggest that they are actually better at the same geometry?That's completely laughable, Intel's 45nm didn't outperform GloFo's or TSMC's 40/45nm in any metric. Even the 45nm SOI process had less leakage than 45nm HKMG, which was what HKMG was supposed to address.Then there was Intel's 32nm shenanigans, they spent an eternity selling budget dual-cores and ultra-low-volume 6-cores before they could finally get yields to the point where they could sell high-volume mainstream quad cores. Meanwhile, GloFo ramped their 32nm much more quickly, with moderate availabilty of the entire line available from day 1.But, where you lose this argument is Ivy's specs: It's barely more than a die shrunk Sandy Bridge, and they only gained little to no clock-speed, at less than 20% power savings? That doesn't sound like a very good 22nm process to me, are you sure it will be any better than competing 28nm processes? Based on evidence from current 28nm GPUs and known specs of Ivy Bridge, 28nm appears to tie or beat 22nm in every metric, even transistor density.[/citation]

You're really in serious denial. You're so far off in all of your statements it's laughable. What bizarro world do you live in ? LOL