The following comment is speculative and (hopefully?) provocative.
This SMIC -Qualcomm announcement could be the beginning of pivotal change for the semiconductor industry. In my recent article, "28nm – The Last Node of Moore's Law" we have pointed out that dimensional scaling beyond 28nm would not provide reduction of SoC cost and, accordingly, 28 nm could the preferred node for many years. And recently we reported that the industry logic-SOC leader -- Qualcomm "Calls for Monolithic 3D IC ... to make up for the loss of traditional 2D process scaling as wafer costs for advanced nodes continue to increase". Now we have SMIC announcing that it will bring to the market a monolithic 3D for Qualcomm ("Going forward, SMIC will also extend its technology offerings on 3DIC and RF front-end wafer manufacturing in support of Qualcomm Technologies.") So yes, at this time "only a small portion of semiconductors consumed in China are actually produced in China" and yes, "SMIC isn't exactly known for cutting-edge process technologies, compared to other foundry giants." But as the value of the more advanced nodes is diminishing, this announcement suggest that SMIC is positioning itself to lead in the next generation technology driver - monolithic 3D, using the most effective node for years to come. If the rest of the foundries will ignore it, they may find themselves trailing behind SMIC in few years, in what by then could become THE technology driver.
Really great point you make here, @Or_Bach. Thank you. Yes, I did notice in the release, SMIC is talking about 3D IC, which I found interesting. But now that you add more context to it, yes, this could be interpreted as the beginning a big change for the semiconductor industry. Makes sense!
I would say Monolithic 3D will indeed become pivotal - for Global Foundries. They've got specialists from CEA-LETI over in the Albany area. Qualcomm already expressed their interest in this development and also joined the LETI M3D R&D program. I'm sure SMIC will catch up eventually, but they still have a long march to go.
My friend Zvi Or-Bach makes persistent comments about cost no longer declining in advanced nodes. This may or may not be correct, however. Many experts disagree and consider it a "mythology" based on a long discredited graph that is reproduced over and over.
More importantly, the argument may be irrelevant. It is indeed correct that gate density remains same in 20nm planar and 14/16nm FinFET nodal processes. However, power and leakage will profoundly decrease in FinFET process. And power and leakage reduction may be far more imortant than density increase -- for both smartphones and mega data centers.
3D ICs are certainly intriguing and are clearly approaching in few years in memories (HMC, etc.). In mobile processors bandwidth requirements are certainly increasing and drive new type of packaging solutions. Still - even here there are alternative approaches. Leading smartphone OEMs use non-standard memories with ball pitch with 0.4mm and even 0.35 mm pitch. This allows use of traditional package-on-package approaches.
Hi Sranje, my blogs about cost no longer declining are only reporting the accross the idustry public information from vendors like Globalfoundries, ST, and from Fabless like NVidia, Broadcom, AMD, and just few weeks ego Qualcomm - quoting: "One of the biggest problems is cost. We are very cost sensitive. Moore's Law has been great. Now, although we are still scaling down it's not cost-economic anymore. It's creating a big problem for us,". And we are pleased to see that qualcomm actually taking action as we recently reported in our blog:"Qualcomm Calls for Monolithic 3D IC", firs developing EDA with help of Georgia Tec, than support the proccess development at CEA Leti, and now setting up volume production with SMIC.
Yet, this is the first announcement of moving to monolithic 3D for SOC and logic devices. Clearly this could be the start of a pivotal change not only in technology but might be even to the composition of vendors that will lead the industry to the next decade.
Dear Zvi, my apology to have to disagree once again - there is NO universal agreement whether cost are or not declining in advanced nodes.
-- Intel certainly confirmed that costs are declining.
-- At GSA Silicon Summit on April 10, 2014 at Computer Museum in Mountain View in a panel discussion titled "Advancements in Nanoscale Manufacturing" all five expert panelists agreed that cost are and will likely continue to decline
- Bob Aitken, ARM
- Adam Brand, Applied Matls
- Peter Huang, TSMC
- Nick Kepler, now at VLSI Research
- John Kibarian, PDF Solutions
As for HMC and similar -- high volume production is about 3-5 years away
Most importantly-- the drive into fine node FinFET is likely driven by MAJOR power and leakage reduction rather than by increase in gate/transistor density. This would be more likely "the pivotal change in IC industry" that you are talking about.
As for TSMC's concerns -- Samsung (with GF) foundry services may be a more pressing concern than either Intel or SMIC
We will all see -- to you and EET readers --- Happy 4th
Hi Sranje, I agree, there are some who still believe otherwise, and Intel is the one most notable. As Intel is the last IDM in the logic space it might be that they can do something the others can't. Our blog "Intel vs. TSMC: An Update" provides an elaborate discussion that I hope you will enjoy reading. The year delay of Intel meeting it 14 nm production release does add doubts. And you can draw your own conclusion from the fact that non of the major fabless companies announced shifting to Intel for their 14 nm products.
As for your reference to some verbal discussion at the GSA Summit panel, I like to point out the non of the panelist represent companies who actually buy silicon devices. Yet all the major fabless companies has made public statements, and presentations with charts clearly showing that 28 nm would be the lowest cost node for many years. So you are welcome to hold to your views, I believe the facts are clear, and most importantly, the foundries - who are now way ahead of SMIC, do know these facts, and accordingly should pay close attention to this new collaboration: Qualcomm - SMIC - Monolithic 3D IC
As for HMC yes it is 3-5 years away, but it is not monolithic 3D. I was refereeing to 3D NAND, which is a monolithic 3D memory, also known as V-NAND. It is now in mass production by Samsung for SSD application.
>Does this imply SMIC needs to dedicate (100%) 28 nm >capacity to Qualcomm?
May be, there are simply no big enought consumer for <=28nm in mainland china. Consumers in mainland China are not as picky to such things as power consumption, and they don't require extreme densities for their last gen designs. In additions to that Qualcomm will get a significant logistical advantage (no need to deal with PRC customs office) and associated cost reduction.
And you can draw your own conclusion from the fact that non of the major fabless companies announced shifting to Intel for their 14 nm products.
Intel is just dipping their toes into the water as a foundry, taking on small carefully selected customers to learn the ropes of the foundry business. The fact they haven't taken on any big fabless customers for 14nm or any other node says more about the pace at which they intend to pursue the foundry business than it does about Intel's 14nm process.
If Intel had signed up several big fabless customers at 22nm and no one signed up for 14nm then I could see your point. But there are much better explanations for the "lack" of large fabless customers for Intel's 14nm than the one you're proposing.
Unlike other foundries, Intel is in competition with many of the largest fabless customers. Is Intel likely to agree to fab CPUs for AMD or SoCs for Qualcomm? I think not. Whether they'd even consider taking on someone like Apple, who doesn't compete with Intel since they don't sell their SoCs, is unclear. Intel might harbor crazy dreams of winning them over as a customer for Intel's own mobile SoCs, and refuse to act as a foundry for Apple's SoCs on that basis.
I could see three reason why Qualcomm is diverting some of their 28nm production to SMIC;
1. Appeasing the Chinese government, getting some 28nm production to SMIC is very lucrative business and might clear the debacle between the parties.
2. There is a shortage of 28nm production capacity all over so Qualcomm has no choice but to resort to second tier production capability.
3. Qualcomm is not happy with TSMC pricing and trying to see if they can get better pricing at SMIC.
I'm not sure which one out of the three I mentioned above is the right one, but I'm sure one of them is right.
@beinglass, alternatively, the answer could be "all of the above" you mentioned in your post.
But had Qualcomm not been singled out by the Chinese antitrust authority, I don't think Qualcomm would have found SMIC to be their best choice (provided, Qualcomm were indeed looking for another foundry -- other than TSMC.)
As I mentioned in my previous message I'd like to yet make another comment. An additional motivation might have been the need of Qualcomm to find production partner for monolithic 3D. In the recent blog "Qualcomm calls for Monolithic 3D IC" posted by EE Time we can find the very recent Qualcomm massage in DAC 2014: "One of the biggest problems is cost. We are very cost sensitive. Moore 's Law has been great. Now, although we are still scaling down, it's not cost-economic anymore. It's creating a big problem for us." Looks like TSMC is not taking the lead in 3DIC, it might be that SMIC with Qualcomm after establishing good relationship in 28nm will continue to true monolithic 3D IC.