28nm @ TSMC: Very expensive or just wafer-limited?

[UniversalTruth]

China tablet chipset vendors shifting 28nm process orders away from TSMC, say sources

Following the steps of Qualcomm, some China-based tablet chipset vendors reportedly have also slashed their 28nm orders at Taiwan Semiconductor Manufacturing Company (TSMC) as they are switching part of their orders to other foundry houses which can offer better foundry pricing, according to industry sources.

Rival companies including Globalfoundries and Samsung Electronics both have managed to improve the yield rates of their 28nm process nodes, enabling them to offer better prices to lure clients away from TSMC, notably such as Qualcomm, said the sources.

Affected by decreasing orders, TSMC's 28nm lines are currently operating at 80% of their capacity

 

[UniversalTruth]

TSMC seeing utilization rate of 28nm processes fall

IC foundry Taiwan Semiconductor Manufacturing Company's (TSMC) capacity utilization of its 28nm processes has fallen to 70% due to a slowdown in orders for high-end mobile chips, according to industry sources

So, wth is going on. At least 2 hardware retailers (newegg and hardwareversand) have no 290Xs in stock

Anyone any ideas? Any?

 

[AlNom]

290X doesn't seem like a high end mobile chip, unless I'm mistaken.

 

[UniversalTruth]

I would guess so too. But it doesn't mean that the process for mobile chips is utilised at 40% while the one for high-end desktop chips at 100% either. They speak about the overall processes utilisation... which would give an idea what is going on...

 

[jimbo75]

What's going on is that TSMC is getting less orders because less smartphones are selling and also Globalfoundries is taking some customers off them.

 

[Grall]

Well, we can hope that should mean no shortages of next-gen consoles then...unless Sony becomes GDDR-limited perhaps.

 

[tunafish]

So, wth is going on. At least 2 hardware retailers (newegg and hardwareversand) have no 290Xs in stock

Anyone any ideas? Any?

The time from making a chip order to store shelves is around 3 months iirc. "X is not on store shelves" is usually caused by failed prediction of demand rather than being strictly supply constrained. If AMD sees more than expected demand now, they can't really do all that much about it as if they did a huge order now, by the time the chips would actually be available they might no longer be in demand.

 

[UniversalTruth]

3 different 16nm processes coming at TSMC:

16nm FinFET (entering trial production in the end of 2014);
16nm FinFET+ (entering trial production in the end of 2014);
16nm FinFET Turbo (expected sometime in 2015, 2016).

http://www.digitimes.com/news/a20140328PD213.html

 

[Deleted member 13524]

So TSMC's 20nm will only last for a little over a year as highest-end process?

 

[McHuj]

So TSMC's 20nm will only last for a little over a year as highest-end process?

yes, it's a stop gap until 16nm is ready. Call the TSMC 20nm, 16nm-Beta, as it's is really 20nm with FINFET transistors, it's not a completely new node from the ground up. I think most will switch as fast as possible toit thanks to the improved power consumption of the FINFETs.

 

[Alexko]

3 different 16nm processes coming at TSMC:

16nm FinFET (entering trial production in the end of 2014);
16nm FinFET+ (entering trial production in the end of 2014);
16nm FinFET Turbo (expected sometime in 2015, 2016).

http://www.digitimes.com/news/a20140328PD213.html

I wonder if that has anything to do with NVIDIA's decision to delay Parker and Volta.

 

[UniversalTruth]

They can use FinFET and FinFET+ with later Maxwell GPUs, and in 2016 Pascal on FinFET Turbo.

 

[Erinyes]

3 different 16nm processes coming at TSMC:

16nm FinFET (entering trial production in the end of 2014);
16nm FinFET+ (entering trial production in the end of 2014);
16nm FinFET Turbo (expected sometime in 2015, 2016).

http://www.digitimes.com/news/a20140328PD213.html

16FF should be entering mass production in Q4'14.

16FF+ has already been announced and it said to bring a 15% improvement in speed with no power or area penalty - http://community.arm.com/groups/soc...-finfet-and-arm-s-64-bit-biglittle-processors

Regarding 16nm FF Turbo..this is the first time I have heard of this third supposed 16nm process and I'm not entirely sure this news is accurate.

yes, it's a stop gap until 16nm is ready. Call the TSMC 20nm, 16nm-Beta, as it's is really 20nm with FINFET transistors, it's not a completely new node from the ground up. I think most will switch as fast as possible toit thanks to the improved power consumption of the FINFETs.

Yep..16nm FINFET is based on 20nm and does not have any major density improvements..only around 5%. Note that cost per wafer and hence cost per transistor is significantly higher for FINFET so we may see some vendors stay at 20nm for a while.

Apart from lower power consumption, FINFET will also allow higher frequencies compared to 20nm, which has limited gains.