@double-o-nothing, yes, you're right...this could be another one of those great examples for the parallel universe. I would have never really give much thoughts to how important 40nm/45nm market is untiil I talked to T.Y.
Thanks for your thorough, detailed answers. As I go back to my notes, I do see RockChip are moving onto 28-nm for a couple of their quad-core APs and one dual-core AP this year; so is Amlogic. Certainly, MTK is moving to 28 with their quad core. But you are absolutely right. Most of them -- and their flagship chips -- remain at 40nm.
Minimum Order Quantity (MOQ) is not something vendors often talk about, but you shed good light on the issue. Thanks!
>@baybal, you are also right about the strong demand from >China's AP processor guys. Some of them are still using >40-nm process, butthey are definitely ;moving toward 28nm. >The question is, as you pointed out, SMIC's capacity.
>T.Y. says, "With Beijing and Shanghai combined, our 28 nm >capacity is 15,000wafers per month."
>Do you think that's enough?
The question is not so much about SMIC's 28nm capacity, but about their total capacity that they can market to AP makers, which includes service on few generations old nodes. When bigger AP makers will move towards 28nm, they will free up a lot of capacity on 45nm which will become available to lower tier fabless companies (think Actions/Infotmics/AMLogic/Nufront).
If these second-tierers will no longer need to go across the straight for capacity, they will get a very very significant cost advantage. Most of mainland fabless have no problems lagging few nodes behind the market. For them, going below 65nm does not change a lot as they are constrained by MOQs more than by per chip costs.
>@baybal, it's interesting you say that it is wise for >SMIC to stop chasing TSMC...I've heard others say >that too. So far, that strategy is definitely working. >The issue is what might happen three years from now...
It is unlikely that mainland fabless will need <28nm more than 45nm service in 3 years.
As you probably know, foundries give MOQs for wafers, and not for individual chips. You can make significantly more chips per wafer with 28nm than with 45nm, but you still need to reach the MOQ for wafers which will always be higher for a service on a node with higher market demand.
@baybal, you are also right about the strong demand from China's AP processor guys. Some of them are still using 40-nm process, but they are definitely moving toward 28nm. The question is, as you pointed out, SMIC's capacity.
T.Y. says, "With Beijing and Shanghai combined, our 28 nm capacity is 15,000 wafers per month."
@baybal, it's interesting you say that it is wise for SMIC to stop chasing TSMC...I've heard others say that too. So far, that strategy is definitely working. The issue is what might happen three years from now...
SMIC market position has definitely improved through post-crisis years. SMIC managed to weather the economic downturn and following waves of bankruptcies by entrenching themselves in stable niche markets (sensors) and generic 180nm+ service, something that TSMC were not interested in.
TSMC, on the other hand, made a possible error by putting all resources into 450mm/<40nm service that, after 5 year of development, has not reached the market. TSMC lost a sizeable portion of mainstream service market exactly because they were going towards <28nm too fast and trying to skip a node.
It was a wise decision on SMIC's part to stop chasing Taiwanese and to seek growth opportunities beyond TSMC dominated niches. While SMIC has no real way to compete with TSMC heads on in the high end VLSI/SoC sector, their current tactic is valid. SMIC is not out of risk zone yet, and they can be in red again the next year.
The competition in between SMIC and TSMC for mid-end sector seem to be already pushing prices and MOQs down. It will be interesting for me to see how much fab capacity SMIC's expansion will open to PRC's application processor market as it is crucial to PRC's domestic fabless brands (and to my income =).