According to latest earning call TSMC is on track to start volume production of 7nm early 2018. This gives it a two year lead over Samsung which is due to have 7nm in 2020. Does that mean TSMC is going to effectively have the market cornered for two years, or am I missing something?
TSMC to gain marketshare for 7nm?
- Thread starter lefty
- Start date
hist78
Well-known member
Yes, it is what they're saying at the earning conference call yesterday. The 7nm risk production will start at 1Q2017. There are 15 semi companies utilizing the 7nm node and they are expected to reach tapeout stage starting in early 2Q2017. The full conference call video and presentation materials can be viewed at:
Taiwan Semiconductor Manufacturing Company Limited
Additional highlights:
1. 5nm will use EUV extensivelly . TSMC made some significant progress on EUV and expected to start 5nm risk production in first half of 2019.
2. For 3Q2016 revenue, it achieve 17% quarter to quarter and 22% year to year growth.
3. 4Q2016 Guidance: Revenue will be flat. Gross profit margin 50.5% ~ 52.5%, operating profit margin 40% ~ 42%
4. For 2016 worldwide semiconductor industry (exclude memory) will have 1% growth, foundry market will have 7% growth, and TSMC revenue will have 10% growth in US dollar or 11~12% growth in NT dollar
5. TSMC's revenue in each smartphone sold in the world will increase to around $10 next year, from about $9 this year.
6. TSMC wants to encourage its customer to enter high performance computing market by carefully managing the profit margin on the 7nm node process.
Last edited:
Exactly. It really is going to be 28nm deja vu all over again. Only this time TSMC sees it coming and will build the appropriate capacity so there will be no shortages.
Judging by the 7nm tape outs that I am aware of, ALL of the top fabless companies will use TSMC for 7nm in 2018-2020 including QCOM who has been at Samsung for 14nm and 10nm. AMD is working with GF on 7nm but I no longer consider them a top fabless semiconductor company.
My feeling is that the Morris Chang less investor calls have been very conservative in forecasting so TSMC will continue to beat for the next few quarters, unless the world economy crashes or there is another natural or unnatural disaster.
hist78
Well-known member
The official conference call transcript has been posted:
http://www.tsmc.com/uploadfile/ir/quarterly/2016/3tBfm/E/TSMC 3Q16 transcript.pdf
http://www.tsmc.com/uploadfile/ir/quarterly/2016/3tBfm/E/TSMC 3Q16 transcript.pdf
At a forum last week with the semiconductor equipment people it was openly discussed that Intel is now in production at 10nm and have moved the process from OR to Israel. The yield is at 80% which is low for Intel to start production but I'm guessing Intel feels pressure from TSMC.
You also have to wonder how AMD will compete with their 14nm Zen CPU if Intel is shipping 10nm CPUs to the PC market. I'm not convinced that AMD will compete with the Intel 14nm CPUs much less the 10nm ones.
You also have to wonder how AMD will compete with their 14nm Zen CPU if Intel is shipping 10nm CPUs to the PC market. I'm not convinced that AMD will compete with the Intel 14nm CPUs much less the 10nm ones.
Cannonlake is only for the 2-core Y-series and U-series initially, because of the low yield. Zen CPUs on the other hand are 4/8 core CPUs for highend desktop and servers. So, even if Intel release Cannonlake unexpectedly ahead of schedule, they it will not be competing against Zen.
Also the latest leaked road maps show a new family called Coffee lake which is the cannonlake architecture on 14nm. That means it won't be until 2018 that we see the high end 10nm parts.
Also the latest leaked road maps show a new family called Coffee lake which is the cannonlake architecture on 14nm. That means it won't be until 2018 that we see the high end 10nm parts.
It depends. For the big chips would be a great yield, but unfortunately that should not be the case. Most likely those are the mobile Y and U CPUs (smaller ones). On top of that, it could even be the shippable yield and not the prime one. Israel did a great job with the 22nm production ramp up, let´s see if they will repeat themselves.
It will take much longer to see Intel desktop CPUs at 10nm in the market, for the exact same reason as above, yield. I´m instead pretty sure that 14nm AMD CPUs are competitive against Intel 14nm ones. Yet one quarter and then we'll know.
Strange to say but you are right, the Israel Intel fab gets much better results so I have much more confidence in Intel 10nm than 14nm. I also know that Intel is relying more on their partners and suppliers than ever before which is good. So Intel 10nm truly is a collaboration.
hist78
Well-known member
It's been many times people mentioned that Intel's Israel division can take on difficult challenges and can make things happen.
This is the thing I don't quite understand. For big semiconductor companies, such as Intel, TSMC, or Samsung, they do move people, instruments, and talents around and collaborate globally. Producing semiconductors is an engineering project based on disciplines and precision. It's something can be replicated and transferred from one site to another and the quality should be the same across multiple sites globally. Why this is different in Intel's case?
Last edited:
Not exactly. Process engineering of semiconductors is quite delicate that a tiny variation in the environment will make the outcome totally different. The same process parameters will not yield the same results even using the same sets of equipment at different geographical locations. Some tuning is always required and the final yield will not necessarily be the same. I guess that's why TSMC is improving its yield in a much faster pace as all its fabs are located relatively close geographically, while Intel's are not. That also means Intel's Israel engineers do a better job to tune the processes to achieve a better yield.
Exactly. Intel Haifa started in the 1970's and was the first Intel development site outside of the United States. I spent some time there in the 1990's and was impressed with the military like precision of their operations. I have also spent time with other Israeli companies such as CEVA and Mellanox and can say the same thing about them.
hist78
Well-known member
Understand your explanation. But I'm wondering is there any any other situation that is unique to Intel?
TSMC's Nanjing 16nm fab will start production in the second half of 2018. Most of the equipment in the new fab will be coming from TSMC's fabs in Taiwan. It will be interesting to see how fast TSMC can ramp up at that most advanced fab in mainland China.
Speedfriend
Member
TSMC was supposed to be way ahead for 14nm and 10nm but Samsung was there too. Why should 7nm be any different?
Samsung is on a two year process node cadence so they should be at 7nm in 2019. Samsung has chosen to hold 7nm for EUV to keep the costs down. TSMC will use EUV for 5nm in 2019. YOu can check this chart to see the density comparisons for the different nodes since the names no longer are the length of the drawn transistor.
https://www.semiwiki.com/forum/content/6160-2016-leading-edge-semiconductor-landscape.html
It's funny that date keeps changing. Before it was 2020, now it's 2019 and some other reports say it's 2018
I just spoke to Samsung about it yesterday so this is the most recent update. It is highly dependent on EUV but since TSMC announced EUV for 5nm in 2019 I have much more confidence in the Samsung date. My guess is that TSMC 5nm and Samsung 7nm will be comparable in density.
I think it could be close.
Last edited:
Or these nodes could be skipped. The buyer of the most EUV tools delayed its use to 2022.
Last edited: