引用:
|
作者teszd@2014-02-08, 11:46 PM #184
就算20nm沒辦法降低成本
對很多廠商來說還是必須的
尤其是有競爭對手的廠商
比如qcom/mtk
NV/AMD(顯卡晶片) |
目前除了Altera與Xilinx，TSMC 20nm製程最受矚目的客戶當屬Apple了。
引用:
|
作者weiter5494@2014-03-19, 02:16 PM #222
|
蘋果下半年推出的新款iPhone及iPad所採用的20奈米A8應用處理器，4月開始拉高在台積電的投片量，且台積電第2季及第3季將獨吃A8代工大單。三星電子，因20奈米製程良率仍低且產能不足，暫時並無接獲代工訂單。
- chinatimes@Apr 14'14
With the ramping of
20-nanometer, which just started in a second quarter, we will have
a very, very small volume shipment. We'll have
much more volume in the third and fourth quarter at any new nodes, starting
with low margin. So, we expect there will be some dilution to corporate level margin starting from second quarter. The magnitude of that, it will impact by 1% in the second quarter. It will be slightly bigger than 1%, will be a very low single-digit impact on our second half. For the whole year, we still expect to see a slightly higher SGM compared to last year.
A8處理器TSMC本季小量投入，從第三季開始放量且利潤不高。
As for System LSI, revenue declined from the previous quarter as mobile AP demand decreased under weak seasonality. For System LSI, Samsung expects high-end mobile AP demand will remain weak through the second quarter, but earnings are expected to improve during the second half of the year as Samsung plans to seek growth momentum with solid sales of high-density CIS, aka ISOCELL, as well as the launch of new 20-nanometer AP products.
Samsung下半年這個20nm產品，有可能是Galaxy S5 Prime搭載的雙四核Exynos 5430跟iPhone 6的Apple A8，符合TSMC放量時程。
再看Qualcomm與Mediatek！前者64-bit旗艦晶片S808與S810雖採20nm製程，上市點卻是在今年底與明年初間，而MTK則是僅表態會有20nm產品在年底上，具體細節不明。
Nvidia has jettisoned their plans for a 20nm Maxwell by the end of the year and are instead taping out 28nm Maxwell chips by the end of this month to launch at the end of this year or early next year. It was believed that Nvidia wanted to hold out for 20nm with their mid and high end GPUs. Unfortunately, that plan doesn’t seem to be working out for Nvidia as TSMC is still having issues with their 20nm process, forcing Nvidia to launch their high end GM204 and mid-range GM206 at the 28nm process node as well. Similarly, AMD recently confirmed it is not moving to 20nm across all product lines (including GPUs) this year, although AMD hasn’t confirmed whether this is due to to lack of 20nm supply from TSMC or if AMD doesn’t yet have a competitive new architecture to put on the 20nm process. Either way, despite TSMC’s claims of having a solid 20nm manufacturing process along with 16nm FinFET soon to come, we simply haven’t seen a single, mass produced 20nm TSMC chip on the market.
AMD跟nVidia皆確認今年產品仍繼續使用28nm打造。換言之，就靠Apple撐起20nm一片天了。
Our 28 nanometer high-k metal gate has three options, 28HP, 28HPM and 28HPC. And this year these 28 high-k metal gate technology will be about 85% of the overall 28 nanometer in terms of the wafer.
引用:
|
作者weiter5494@2014-03-19, 02:16 PM #222
...法巴則認為技術問題已獲解決，良率來到五成。
台積20奈米已在1月下旬、於南科的Fab 14正式投片量產，而即使在上個月台積於20奈米量產上遭遇了些許問題，不過 經由選擇另一種不同的研磨液(slurry)，台積已順利解決了在CMP(化學機械平坦化)製程的挑戰，台積的20奈米製程良率，也一舉提升至50%。
- BNP Paribas@Mar 18'2014 |
4月份TSMC法說會前夕，法人圈傳出消息，TSMC的20nm良率已由2月的50％逐步提升，平均良率約55~60％，高點突破60％，不但超越蘋果要求的45％，更遙遙領先三星20奈米的35％~40％。雖在良率有所提升，在化學機械平坦化（CMP）製程中，因選擇了不同的研磨液（Slurry）材料，導致驗收卡關，這材料問題雖然不直接影響良率，但卻會影響產品的電信穩定度及使用壽命，在電信測試部分恐難過關，影響出貨。此說法遭TSMC官方駁斥為顛倒黑白，無稽之談，因為良率突破關鍵固然在slurry，但不良原料早被驗退而未投產。
引用:
|
作者weiter5494@2014-02-10, 03:47 PM #193
...16FF產品今年看得到？
When the first 16nm FinFET ARM implementations will hit the market？ We’re squinting at this year.
- Pete Hutton, EVP and president of ARM’s product group |
We have achieved
the risk production milestone of 16-FinFET in November 2013, November last year. And this month, we should pass the
1,000 hours so-called the technology qualification. So the technology is
ready for customer product tape-out. Our 16-FinFET yield improvement has been ahead of our plan. This is because we have been leveraging the yield learning of 20SoC. Currently 16-FinFET SRAM yield is already close to 20SoC. And with this status we are developing
an enhanced transistor version of 16-FinFET plus, with 15% performance improvement. It will be the highest performance technology among all available 16 and 14 nanometer technology in 2014. The above progress status is well ahead of Samsung.
- Mark Liu, President & Co-CEO of TSMC@4Q13 Earnings Call
在年初法說會（Jan 16）釋出16FF升級版的16FF+製程的風向球後，TSMC上月法說會（Apr 17）進一步揭開來年先進製程戰略。
We have a first 16FF product tapeout this month(Apr), it will ride on 16 FinFET process. 16 FinFET plus will be qualified in September. And for those customers taped out in the second half, I would say mostly, will be riding on the 16 FinFET plus. Majority of our process customers will run on 16 FinFET plus. And looking into the volume for the next year, I would say that most of the product will be run on 16 FinFET plus. The volume of 16 FinFET plus will happen in 2015. Even for some of the customer, initially their products sits on 16 FinFET, they also would like to migrate their second if market opened up indeed to upgrade their product. So I will see the vast majority will be 16 FinFET plus.
去年11月完成製程認證的16FF，上月已完成首件設季定案；而16FF+預定在9月份完成製程認證，明年投入量產。
綜觀製程沿革，20SoC、16FF、16FF+共用BEOL，是一脈相承的系列製程，主要是電晶體差異。從20SoC到16FF是planar到FinFET外加微縮；16FF到16FF+是電晶體特性的改善。相形之下，20SoC是過渡製程，而16FF更多屬於雛型製程。
而各擅研發及行銷的兩大聯席CEO對16nm製程的解讀，則將是TSMC未來經營績效的關鍵。
The
tool commonality between 20 and 16 are 95%. So when the customer move from 20 SoC to 16 FinFET, we only need
to increase a very marginal amount of the capital to suffice that demand. The ASP and the product will be more competitive for our customer. So I'll
put it together as one node. But it does provide our customer their product grade or product spec to continually improve year-after-year.
- Mark Liu, President & Co-CEO of TSMC
20 SoC and 16 FinFET are
totally different device structure. Even the back-end design rule are similar, you cannot port one to the other easily. No, it won't be easy.
Porting from one foundry to the other foundry you are getting harder and harder. The real reason is all the device characteristic right now is related to the strain, all those kind of things, which you cannot just copy. You cannot de-layer or you cannot do the reverse engineering to look at it what is the device structure and get the same kind of IC characteristic. So, it's quite very hard -- actually very hard.
- C.C. Wei, President & Co-CEO of TSMC
The porting it happens and I just want to make two points.
We just have to live with it, deal with it. Let me point out two points. One is porting is cost more and more nowadays for this generation. I think it takes a lot of R&D resources to do that. Secondly, we just have to do better 16 technology. There are reasons to do that, because, first of all, we ramp 20 SoC massively this year and lot of learnings, a lot of process window control, a lot of design and collaboration with our customer, we will build ahead of our competitor. So we believe our 16 FinFET right on top of our 20 SoC, given they are
same design rule, will be more mature at the time compared with our competitor.
- Mark Liu, President & Co-CEO of TSMC
這廂，Samsung的進度呢？上季已完成14 FinFET製程認證，年底投產，明年上半開始貢獻營收。
As the company
completed qualification of 14-nanometer process during the first quarter, we plan to ramp up production of the
14-nanometer process which is expected to begin by the end of 2014 and
earnings contribution is expected from the first half of 2015.
PS：Samsung NAND記憶體製程已進入低於10~20nm節點。
引用:
|
作者weiter5494@2014-02-08, 10:41 PM #183
檯面上的消息顯示， ...多年以來，IBM一直扮演的是技術開發的角色（涉及商業化的自身產品頂多是Power處理器），默默支撐著Common Platform聯盟夥伴的商業運作，是Samsung跟Glofo的精神導師（其實TSMC最痛恨的人應該是IBM），foundry領域的訂單肉搏戰反而與其無關。
長期來看，出售或合資的結果或有不同程度的影響。出售的話，IBM淡出半導體領域，換人操盤，最壞的下場，Common Platform聯盟可能瓦解，TSMC獲益最大；縱使合資，IBM的影響力也會削弱，影響聯盟運作效率。當然，若Samsung與Glofo若已能獨當一面，如此衝擊相對小，譬如說全球能參與ASML EUV光刻機投資案的，Samsung就是其中一個。半導三國鼎立，可視為雙I一T（Intel、IBM、TSMC）陣營之爭。 |
龍頭禪位，IBM聯盟新布局浮現。
同源自美國本土的IBM與Globalfoundries向來友好，過從甚密。兩者與紐約州立大學的College of Nanoscale Science and Engineering組成的研發鐵三角不僅撐起紐約州半導體產業一片天，其位於境內East Fishkill的晶圓廠，也牽線了戰友Globalfoundries在境內的Malta設立Fab 8晶圓廠與造價2億鎂的14nm研發中心，以後8個月內還將帶來100億鎂投資與將高科技就業機會自2200個增加至3000個，而IBM實際上也為此晶圓廠的良率提升付出重大貢獻。在IBM決定退出晶圓製造領域之際，Globalfoundries亦順理成章被視為最有希望的買家，執掌當地半導體產業的大旗。
IBM also held talks with chip makers Intel and TSMC, but TSMC has dropped out of the talks after having studied the IBM chip plant and doesn't plan to add a new plant now in an overseas market. While Intel is still involved, Globalfoundries appears to have a stronger interest, they said.
- the Times Union
而Samsung接棒IBM成為技術領頭羊，授權予Globalfoundries的Fab 8，使南韓的兩座（華城、器興）與美國的兩座（Austin/Texas、Malta/New York）晶圓廠共享基於20nm的14FinFET製程，同步待產。
You
design once and then you can
go to either Globalfoundries or Samsung or both.
鑒於Made in USA替紐約州所帶來的商機與就業（Samsung代工Apple處理器的德州晶圓廠S2，在2013年貢獻30億鎂營收），當地媒體顯然樂觀此一策略聯盟。此一家國情結對將來Apple處理器訂單動向的所形成的變數當不容忽視。
The potential to share manufacturing output could give the two companies a leg up because businesses such as Apple place a high priority on having the capacity to react to market demand. Samsung currently makes Apple's iPhone and iPad chips in Austin, but Apple, a major competitor with Samsung in the smartphone and mobile device market, has been looking to move some or all of its chip production to another company ouside the United States.
IBM在NanoCollege的研發團隊將持續加入10nm製程開發計畫。
Both Samsung and Globalfoundries continue to work with IBM through a (research and development) presence in Albany as we explore technology options for 10 nanometers and beyond.
- Jason Gorss, GlobalFoundries spokesman.
這廂TSMC又是如何規劃10nm呢？
And for the 10 nanometer, we haven't announced it, but we did communicate with many of our customers that that will be the aggressive scaling of technology we're doing. And so, in the summary, our 10 FinFET technology will be qualified by the end of 2015. 10 FinFET transistor will be our third generation FinFET transistor. This technology will come with industry's leading performance and density.
- Mark Liu, President & Co-CEO of TSMC@4Q13 Earnings Call
10 FinFET will offer 2.2X of density improvement over its previous generation, 16 FinFET plus. So, currently, 10 FinFET development progress is well on track, but risk production will be in 4Q 2015.
- Mark Liu, President & Co-CEO of TSMC@1Q14 Earnings Call
14/16nm的關鍵，3D電晶體人人喊做，問題是，除了Intel外，誰辦到了？
引用:
|
作者weiter5494@2014-04-28, 10:29 PM #256
...半導體設備商科磊上季$8.32億營收與$1.23鎂EPS皆優於市場預期，但肇因晶圓廠在3D電晶體製程良率問題的本季展望卻令市場十分失望。
Now in logic and foundry, with the introduction of the new 3-D gate architectures, the yield issues our customers are grappling with today are proving to be the most challenging that the industry have ever faced, and even the smallest variation and process margin can cause significant yield losses for these devices. Some of the issues of these processes people are dealing with are unique defect issues … new etch steps that are all part of the FinFET process. That’s just a small part of the complex technical challenge associated with bringing new 3-D device architectures to market... |
安啦！TSMC出品，必屬佳作，掛保證的。
About the message that ASML（此應為提問之分析師對另一設備商KLA Tencor的誤植）saw
mobile FinFET delayed, I really don't know what it means. Talking about the FinFET technology difficulties, I must say our
16 FinFET technology development is well on track and our
new improvement is well on track and we are working with customers closely and we expect
to ramp up 2015. But I think one unique feature of our 16 FinFET is, our 16 FinFET has the
same design rule, back-end design rule like 20. So, we can leverage all the new learning, all the massive work in this year into next year 16 FinFET.
- Mark Liu, President & Co-CEO of TSMC
引用:
|
作者weiter5494@2014-04-28, 10:29 PM #256
...而據瑞信分析，對10nm製程攸關至要的EUV光源，原訂商用時程目標（2017年）會再度跳票。製造商ASML上季營收��14億，EPS達��0.57；展望本季��16億，低於市場預期的��16.9億，同時未出訂單僅��10.7億，也大幅低於市場預期的��16.4億。
The company is “pushing out longer term targets” for its “extreme ultra violet” lithography systems, and “without the success of EUV, ASML has limited earnings growth potential.”
- John Pitzer with Credit Suisse
|
關於EUV進展遲緩的事實，TSMC也不予否認。這意味著在16nm製程後，接下來仍無可避免降低成本的難度。
Today,
EUV is not up to the production spec. Actually the most recent breakthrough was a 30 watt and now they have a higher 80 watt machine that we're still working toward that goal. So
EUV will not be inserted in 10 nanometer at the start, because it will slower the pass-through window. So our EUV team is still continuously working on EUV, hopefully to insert a few layers
after the 10 nanometer process start to qualify as a follow up process simplification. And that will meant to be
second half next year.
- Mark Liu, President & Co-CEO of TSMC
黑卒吃過河，TSMC終於越線，一腳踩入封裝業地盤，ASE、SPIL如臨大敵，SiP、CoWoS（與TSMC撞衫？）齊出。不過片語隻字，TSMC仍讓觀眾對InFO與CoWoS仍墮五里霧。
The difference between the
InFO（Integrated Fan-Out）and the CoWoS is actually
the geometry to connect multi-dies together. In the CoWoS, actually we are using very small geometry, actually 65 nanometers of geometry to connect the multi-dies together. In InFO, we're using the larger geometry, which are still technical confidential information. But the cost is much, much lower. We will see inFO product out, but that will be our customer's schedule. Technology is close to be
in production ready probably early next year.
- C.C. Wei, President & Co-CEO of TSMC
如此神器，外資居然唱衰？
While the InFO or vertical stacking fan-out may be able to offer a lower cost and a simplified process flow to the chipset makers to replace the current PoP package, the technology is still at a very early stage and the risk of InFO remains way too high. The molding compound process may warp the wafer and induce an extremely high yield loss, which is likely to undermine the cost structure.
- Roland Shu of Pritchard
非專業觀點，TSMC藉由向下整合的2.5D封裝加值模式，來彌補客戶對28nm以後的尖端製程成本降幅不佳的疑慮，其差異化服務以區隔競敵的固樁搶單動機十分明確。然觀察CoWoS發布逾年的進展，加以此刻再推InFO，顯然客戶買單狀況欠佳，努力空間依然。只是，這般師法面板業者整合觸控功能的策略，究竟是加持，或是樹敵？同時，對手已實現以TSV在DRAM作同質裸晶堆疊，未來邁入異質堆疊，挑戰InFO的潛力恐難以低估。Wait and See！
PS：
TSMC今年28nm出貨（in terms of the wafer）中，28HKMG（28HP、28HPM、28HPC）將占比85%。
28HPm - 提速30% vs 28LP
20SoC - 提速20% vs 28HPm
16FF - 提速40% vs 20SoC
16FF+ - 提速15%、節能30% vs 16FF
16FF+ - 提速40% vs 20SoC