Increasing capital expenditure
Chipmaker TSMC said that it will increase its capital expenditure in a bid to get enough gear to make 28-nanometer chips.
A number of its customers are furious that the outfit did not have enough machines to handle the 28 nm process. TSMC said that it will be spending more than US$7 billion in an effort to satisfy customers’ strong demand for its advanced process. That is an increase of about 17 percent on an estimate of US$6 billion that TSMC made three months ago.
It might be too late already Nvidia and Qualcomm have said that they are looking for a second source, such as south Korea’s Samsung, UMC and even GlobalFoundries. TSMC chairman and chief executive Morris Chang told a technology forum in San Jose, California that the company had some difficulties with 28-nanometer chips, but those difficulties had to do with not having enough capacity, not the yields.
The company expects to complete 132 tape-outs of 28-nanometer chips for customers this year after finishing 36 at the end of last year, Chang said in January.
>So, Dneni, you are saying all this stuff is total nonsense about TSMC being in trouble because, in fact, they will be at 20 nm while Intel is still at 22 nm?
Intel is at 22nm with Ivy bridge, great piece of silicon, AMD is toast.
Intel is pimping 32nm capacity for Atom chips. Where are the 22nm Atom chips? Where is the 22nm foundry customer silicon?
So right now it is Intel 32nm vs TSMC 28nm foundry business. Wake me up when it is Intel 22nm Atom vs TSMC 28nm ARM. After that it will be Intel 22nm Atoms versus TSMC 20nm ARMs.
14nm is gonna be much more interesting. Intel will probably rule 14nm unless TSMC or Samsung pulls a rabbit out of the hat.
We really really believe you Morris Chang. Darn "capacity"
"TSMC chairman and chief executive Morris Chang told a technology forum in San Jose, California that the company had some difficulties with 28-nanometer chips, but those difficulties had to do with not having enough capacity, not the yields."
If there are no yield issues shouldn't the capacity be as planned? Then where is difficulty coming from? There is still something fishy. Their is certainly scrambling to catch up with Intel on process, and by rushing things they are going to mess up further. Risk to $7B additional capex going to waste is high. And who will take that risk?
If there are no yield issues shouldn't the capacity be as planned?
I think that's logic but not everybody is logic:
you can NOT improve yields by adding capacity;
"Regarding concerns that Qualcomm won’t be able to get adequate volume of chips produced at feature sizes of 28 nanometers, Agarwal writes, “Our checks indicate that TSMC is aggressively working to improve yields thru higher capex.”
I think Wally was right about this one. TSMC was thinking that the industry would stay at 40NM while it took it's time to get the yields right at 28nm. Only premium products that could afford to pay premium prices for the newer 28nm products would demand them. Thus low yields and low volume shouldn't have been such an issue.
Instead we have some of their largest customers who want 28nm for mainstream low value high yield production such as smartphones. The chairman is right, their largest problem isn't that they can't get the yields right, it's that they don't have enough equipment to satisfy demand.
So the new formula will be to multiply their yield issues by the addition of more machines. This is like having a restaurant that has a slow kitchen fix it's issues by opening more restaurants!
I'm sure the 28nm ARM processor is wonderful. Right now it's a collector's item.
They are screwed. Buying a bunch of new machines to throw at the problem will multiply the problem as then we'll start seeing chip quality drop. It's not like you can just buy new machines and the problem is gone. More machines equals more manufacturing headaches equals more shop floor arguments equals CHASO.......I love it!
They are in trouble.