TSMC Investor Call Discussion: What is Apple doing next?

[Daniel Nenni]

The TSM transcript is up and definitely worth reading especially since it pertains to Apple and EUV.

Mark Lui:
Now my second key message is about 7-nanometer that include our N7 and N7+ and its outlook. TSMC's N7 has passed technology qualification in April, right on our schedule. N7 is widely adopted by our customers. We expect to have 13 new N7 product tape-outs this year. So far, N7 yield is ahead of our plan. We expect to have a very fast and smooth N7 ramp-up in 2018, with its yield learning even better than our 16-nanometer.

Our N7+ technology is under development based on this robust N7 technology. We inserted several EUV layers in N7+ to take the most benefit from EUV technology with minimum risk. Our N7+ technology will be again the most advanced technology in the foundry industry in 2018 in terms of density, performance and power. This N7+ will ride on N7 learning to have a very steep yield learning curve. We also have enabled the design porting from N7 to N7+ to be very easy by carefully design rule arrangement and our strong EDA utility and IT support. We expect our 7-nanometer node, N7 and N7+, to be a major and long-life technology node. It will be used in mobile, high-performance computing and automotive markets. It will cover premium, mainstream as well as low-cost products in all those markets.

Then I will briefly talk about our advanced technology progress. I would like to update our EUV progress. Both our EUV tool development and the lithography process development have progressed very well in the past six months. We have worked closely with our tool partner, ASML, who has demonstrated 250-watt EUV source power in their labs. In TSMC, EUV lithography process development also went on very smoothly. We run several EUV layers with our N7 SRAM vehicle with the same yield level as non-EUV process.

Lastly, our 5-nanometer. Our 5-nanometer technology development is well on track, already with SRAM functional yield. We will offer 5-nanometer in 1Q 2019.

http://www.tsmc.com/uploadfile/ir/quarterly/2017/2wc5q/E/TSMC 2Q17 transcript.pdf

 

[simoncc2]

Interesting to notice :

- Oct 2011 : " TSMC has also said the 28-nm production is taping out [...], with over 80 customer product tape-outs already". 28 nm in volume production says TSMC | EE Times
- Apr 2015 : " in 16FF+, The foundry will have more than 50 tape-outs by year’s end" TSMC Outlines 16nm, 10nm Plans | EE Times
- Jul 2017 : " We expect to have 13 new N7 product tape-outs this year. " this article

how many in 5nm in 2019, 5 tape-outs ? this competition between GF, Samsung and TSMC to the denser node will eventually make it non profitable for one of them, if not all due to lack of customer... who will be first to exit ?

 

[yulizi]

Gokul Hariharan - JP Morgan Chase & Co, Research Division - Head of Taiwan Equity Research and Senior Tech Analyst
First question on 7-nanometer. Just wanted to get the numbers right. I think last time, Dr. Liu, you mentioned 15 tape-outs. Right now, I think it's30, 3-0, in terms of 7-nanometer tape-outs that you're seeing. Could you talk about what proportion of that is HPC? Because last time you mentionedalmost half of those are HPC-related products.

Looks like there will be 30 N7 tape outs by the end of this year.

 

[name99]

Interesting to notice :

- Oct 2011 : " TSMC has also said the 28-nm production is taping out [...], with over 80 customer product tape-outs already". 28 nm in volume production says TSMC | EE Times
- Apr 2015 : " in 16FF+, The foundry will have more than 50 tape-outs by year’s end" TSMC Outlines 16nm, 10nm Plans | EE Times
- Jul 2017 : " We expect to have 13 new N7 product tape-outs this year. " this article

how many in 5nm in 2019, 5 tape-outs ? this competition between GF, Samsung and TSMC to the denser node will eventually make it non profitable for one of them, if not all due to lack of customer... who will be first to exit ?

Your argument makes no sense. The issue is not how many different customers use a process, nor even how many different designs use a process. The only metric that matters to TSMC is how much revenue is generated by a process. Even if Apple is the ONLY customer for a particular process, as long as Apple is able to cover TSMC's costs for the process, then the process can happen. (Likewise for, say, QC+internal keeping Samsung going, or IBM+AMD keeping GF going.)

There may be other consequences to the ever higher costs of designs, but your particular metric and concerns strike me as the wrong lens through which to view the problem.

.........................................................................

More interesting to me is how aggressive the 5nm timing is. To put it in the context of Apple the timeline one might have imagined would be something like
- 1H17 A10X 2H17 A11 10nm
- 1H18 A11X 2H18 A12 7nm
- 1H19 A12X 2H19 A13 7nmEUV
- 1H20 A13X 2H20 A14 5nm
but TSMC seems to indicating moving 5nm around half a year to a year earlier than fits this cadence. Which makes one wonder just HOW fast Apple and TSMC will ramp up. Will we instead see something like:

- 1H17 A10X 2H17 A11 10nm
- 1H18 A11X 7nm 2H18 A12 7nmEUV
- 1H19 A12X 7nmEUV 2H19 A13 5nm
or some variant like this? For all the talk of 10nm being a short lived node, will it in fact be 7nm@DUV that is the short lived node, with Apple moving to EUV immediately and hardly bothering with 7nmDUV except perhaps for the lower volume A11X?

 

[lefty]

Looks like there will be 30 N7 tape outs by the end of this year.

That's a good point. TSMC said this at the 1Q earnings call:

And we expect about 15 tape-outs in this year with volume production in 2018.

Now it's down to 13, so 2 customers backed out?

 

[Fred Chen]

Some reports Samsung plans 2019 6nm to steal from TSMC. The node marketing war continues.

台積電7nm將批量生產 Samsung加快6nm製程！ - 香港新浪

 

[name99]

To follow up on my point about Apple cadence, we have this:
Samsung to produce chips for Apple’s new iPhone next year
which claims Samsung is making a play for Apple's business in 2018 with 7nm EUV.

IF this claim is true, then it seems like 7nm really will be a very short-lived node. Is there any reason at all (technical or economic) to prefer 7nm over 7nmEUV?
Presumably the only gating factor will be how fast ASML can supply machines?

 

[Fred Chen]

To follow up on my point about Apple cadence, we have this:
Samsung to produce chips for Apple’s new iPhone next year
which claims Samsung is making a play for Apple's business in 2018 with 7nm EUV.

IF this claim is true, then it seems like 7nm really will be a very short-lived node. Is there any reason at all (technical or economic) to prefer 7nm over 7nmEUV?
Presumably the only gating factor will be how fast ASML can supply machines?

If there is no EUV or any further EUV delay, 7nm will be long lived.

 

[count]

Interesting to notice :

- Oct 2011 : " TSMC has also said the 28-nm production is taping out [...], with over 80 customer product tape-outs already". 28 nm in volume production says TSMC | EE Times
- Apr 2015 : " in 16FF+, The foundry will have more than 50 tape-outs by year’s end" TSMC Outlines 16nm, 10nm Plans | EE Times
- Jul 2017 : " We expect to have 13 new N7 product tape-outs this year. " this article

how many in 5nm in 2019, 5 tape-outs ? this competition between GF, Samsung and TSMC to the denser node will eventually make it non profitable for one of them, if not all due to lack of customer... who will be first to exit ?

My guess is neither of those three.

At some point, not in the next 2 years, maybe not in the next 5, but probably within the next 10 years market forces will drive Intel to either go fabless or fail. They are already losing their process lead and if you compare their cadence to the foundries, they are going to be 2 nodes behind 10 years from now. They simply won't have the volumes to fill their own fabs and invest in leading edge processes. Their chip volumes will be probably be split between 2 of those three names, and that will prop up volumes at the leading edge. On the trailing edge, there are lots of smaller companies that won't be able to compete and their volumes will be captured by the big three, extending the life of older processes.

 

[lefty]

To follow up on my point about Apple cadence, we have this:
Samsung to produce chips for Apple’s new iPhone next year
which claims Samsung is making a play for Apple's business in 2018 with 7nm EUV.

IF this claim is true, then it seems like 7nm really will be a very short-lived node. Is there any reason at all (technical or economic) to prefer 7nm over 7nmEUV?
Presumably the only gating factor will be how fast ASML can supply machines?

I wouldn't worry. Any rumour from the Korean press concerning Samsung and TSMC is nearly always made up.

 

[raghu78]

My guess is neither of those three.

At some point, not in the next 2 years, maybe not in the next 5, but probably within the next 10 years market forces will drive Intel to either go fabless or fail. They are already losing their process lead and if you compare their cadence to the foundries, they are going to be 2 nodes behind 10 years from now. They simply won't have the volumes to fill their own fabs and invest in leading edge processes. Their chip volumes will be probably be split between 2 of those three names, and that will prop up volumes at the leading edge. On the trailing edge, there are lots of smaller companies that won't be able to compete and their volumes will be captured by the big three, extending the life of older processes.

Well said. Intel's process lead is over. Intel's monopoly in the PC and server industry allowed them to own their fabs. I think that monopoly is going to end. It will not happen overnight but it will happen over the next few years. AMD is back with a competitive x86 architecture and more importantly with TSMC and GF they will have access to process nodes on par with Intel at roughly the same time. Moreover Intel process node cadence is slowing down compared to the foundries like TSMC who are picking up speed. I think its quite obvious now that TSMC will have the process lead by 2019 with N7+, followed by N5 with 2020. TSMC is unlikely to relinquish their process lead given the furious pace at which they are marching.

 

[simoncc2]

Your argument makes no sense. The issue is not how many different customers use a process, nor even how many different designs use a process. The only metric that matters to TSMC is how much revenue is generated by a process. Even if Apple is the ONLY customer for a particular process, as long as Apple is able to cover TSMC's costs for the process, then the process can happen. (Likewise for, say, QC+internal keeping Samsung going, or IBM+AMD keeping GF going.)

I was just pointing out that with less and less customers at the leading edge nodes, the risk of having no customer for one of the big three is getting bigger,as well as getting only a too poor / small one only to justify the race seems possible. This is what happened at 28/40nm for many companies like SONY or TI I suppose. AMD spending on silicon can justify the building of a 10B (15 in 5nm?) fab ? is TSMC really making money out of Apple or just expect to amortize the cost over the years with other customers? what if SEC loose market share to chinese mobile companies ?

 

[Arthur Hanson]

Follow the money, it tells the truth, all the rest is just talk. Apple and TSM are the best at keeping secrets and as a result can surprise, usually to the up side. TSM and Samsung are going to rule the nanotech manufacturing world. Right now it's a side business, but in the not to distant future that will change dramatically.

 

[benb]

TSMC are killing it, plain and simple. It is unclear exactly what they do differently, and better, than competitors, but the uniqueness of their culture will hopefully be one day dissected and understood, the way Toyota' unique culture has been studied by process improvement experts. They are undoubtedly employing some lean tools, at a scale and to an effect greater than Intel or others can match.

Lean is the science of eliminating waste and shortening cycle times. This is of key importance in semiconductors, yet, fabs have not really leveraged it, until now, seemingly, at TSMC.

Toyota uses lean (their unique version of it) to develop a new model in 18 months. By comparison, the fastest others can manage is 36. Toyota is more profitable and develops technology more easily because of lean. The comparison between Intel and TSMC is similar: It takes TSMC a year to develop a node, whereas, Intel needs 3.

It will be interesting to see if TSMC can translate their lean tools developed in Taiwan into their new fab in China. I think that will be a major headache for them over the coming years. But, here again, Toyota's example provides guidance. Toyota was able to translate the lean culture from Japan to Kentucky.

One more thought about lean and TSMC: Lean is profoundly anti-automation. It makes use of human ingenuity and flexibility. Robotics will never be consistent with lean. Cultures that attempt to turn humans into robots, which is what is practiced at Samsung, and by second-hand accounts, Intel, are on the wrong track.

 

[count]

One more thought about lean and TSMC: Lean is profoundly anti-automation. It makes use of human ingenuity and flexibility. Robotics will never be consistent with lean. Cultures that attempt to turn humans into robots, which is what is practiced at Samsung, and by second-hand accounts, Intel, are on the wrong track.

Lean isn't anti-automation. Automation in lean is often refered to as autonomation and is meant to empower people the organization to do more rather than replace them, but it's very pro automation - with a "human touch".

 

[Daniel Nenni]

Remember, TSMC already has a fab in China just putside of Shanghai so there should be no surprises there.

I was in Hsinchu last week and I must say that the mood is good and there is much excitement around 7nm, especially 7nm EUV. From what I understand there is some secret sauce involved so just because one foundry does well with EUV does not mean another will.

The parking garages at the fabs were also full 24/7 which is a good sign. My guess is that Apple will be 20% of TSMC revenue this year. It will also be interesting to see how the other iPhone semi suppliers do. Here is a quick list from the iPhone7:
Cirrus Logic (CRUS): AAPL represents ~75% of sales
Skyworks Solutions (SWKS): AAPL represents ~40% of sales
Qorvo, Inc. (QRVO) AAPL represents ~35% of sales
Broadcom Ltd. (AVGO): AAPL represents 20% of sales
Qualcomm (QCOM): AAPL is 15-20% of sales
Intel (INTC): Took 50%+ of modem business (from QCOM)
WDC/SanDisk (WDC): AAPL represented 17% of SanDisk sales
Texas Instruments (TXN): Apple is ~10% of sales
Analog Devices (ADI): Apple accounts for ~5% of ADI sales
Maxim Integrated Products (MXIM): AAPL is <5% of sales
NXP Semiconductors NV (NXPI): AAPL is <5% of sales
My guess is that it will be a very good Christmas for all... ho ho ho...

TSMC are killing it, plain and simple. It is unclear exactly what they do differently, and better, than competitors, but the uniqueness of their culture will hopefully be one day dissected and understood, the way Toyota' unique culture has been studied by process improvement experts. They are undoubtedly employing some lean tools, at a scale and to an effect greater than Intel or others can match.

Lean is the science of eliminating waste and shortening cycle times. This is of key importance in semiconductors, yet, fabs have not really leveraged it, until now, seemingly, at TSMC.

Toyota uses lean (their unique version of it) to develop a new model in 18 months. By comparison, the fastest others can manage is 36. Toyota is more profitable and develops technology more easily because of lean. The comparison between Intel and TSMC is similar: It takes TSMC a year to develop a node, whereas, Intel needs 3.

It will be interesting to see if TSMC can translate their lean tools developed in Taiwan into their new fab in China. I think that will be a major headache for them over the coming years. But, here again, Toyota's example provides guidance. Toyota was able to translate the lean culture from Japan to Kentucky.

One more thought about lean and TSMC: Lean is profoundly anti-automation. It makes use of human ingenuity and flexibility. Robotics will never be consistent with lean. Cultures that attempt to turn humans into robots, which is what is practiced at Samsung, and by second-hand accounts, Intel, are on the wrong track.

 

[user nl]

....................

I was in Hsinchu last week and I must say that the mood is good and there is much excitement around 7nm, especially 7nm EUV. From what I understand there is some secret sauce involved so just because one foundry does well with EUV does not mean another will.

...................................

Is this 'sauce' what you are pointing at, the EUV resist ?

TSMC used an unnamed “novel resist” chemical to replace five immersion masks with one EUV mask at pitches ranging from 26-30nm. Liu said the company currently expects EUV could compress as many as 16 immersion masks to four or five.

EUV rolls into TSMC'''s 7+nm node in 2018 - EE Times Asia


If TSMC is already doing 1500 Wafers/day with 125 Watt using the NXE3350, they should do indeed quite well in throughput with 250 Watt EUV source power in the NXE3400 being installed in 2017 and the following years in their Fab. And the EUV source has been shown already to work at close to 400 Watt at ASML's site so throughput should be getting much better than the 1500 W/day the coming years...

Some June 2017 info on new developments on EUV resist were presented by JSR:
https://www.euvlitho.com/2017/P43.pdf

and by Inpria:
https://www.euvlitho.com/2017/P50.pdf

I am no expert in this area but going through the slides it seems there are quite some `chemical` knobs to play with regarding the novel EUV resists in both litho and etch steps....

User nl

 

[benb]

TSMC used an unnamed “novel resist” chemical to replace five immersion masks with one EUV mask at pitches ranging from 26-30nm. Liu said the company currently expects EUV could compress as many as 16 immersion masks to four or five.


A few comments about leaning a process from 16 to 5 steps: First its a huge change from the direction the industry has been heading for at least 5 years. It's a turn-around, perhaps a resumption of Moore's law type economics. Secondly, automation doesn't give you results like this; you have to have something else; chemistry or physics insights, driven by a scientific approach to manufacturing progress. TSMC is different in this regard from others; always trying new things with an openness to new, novel approaches.

 

[Daniel Nenni]

TSMC July is down Y-o-Y Bad News for Apple?

Hsinchu, Taiwan, R.O.C. – August 10, 2017 - TSMC (TWSE: 2330, NYSE: TSM) today announced its net revenues for July 2017: On a consolidated basis, revenues for July 2017 were approximately NT$71.61 billion, a decrease of 14.9 percent from June 2017 and a decrease of 6.3 percent from July 2016. Revenues for January through July 2017 totaled NT$519.38 billion, an increase of 3.5 percent compared to the same period in 2016.

[table] style="width: 100%"
|-
| align="left" valign="top" style="font-family: Arial, Helvetica, sans-serif; font-size: 11px; line-height: 15px; color: rgb(153, 153, 153); padding: 12px 0px" | Issued by: TSMC
Issued on: 2017/08/10
|-
| align="left" valign="top" style="font-family: Arial, Helvetica, sans-serif; font-size: 12px; line-height: 18px; color: rgb(68, 68, 68)" |

|-
[/table]

 

[raghu78]

Hsinchu, Taiwan, R.O.C. – August 10, 2017 - TSMC (TWSE: 2330, NYSE: TSM) today announced its net revenues for July 2017: On a consolidated basis, revenues for July 2017 were approximately NT$71.61 billion, a decrease of 14.9 percent from June 2017 and a decrease of 6.3 percent from July 2016. Revenues for January through July 2017 totaled NT$519.38 billion, an increase of 3.5 percent compared to the same period in 2016.

[table] style="width: 100%"
|-
| align="left" valign="top" style="font-family: Arial, Helvetica, sans-serif; font-size: 11px; line-height: 15px; color: rgb(153, 153, 153); padding: 12px 0px" | Issued by: TSMC
Issued on: 2017/08/10
|-
| align="left" valign="top" style="font-family: Arial, Helvetica, sans-serif; font-size: 12px; line-height: 18px; color: rgb(68, 68, 68)" |

|-
[/table]

Daniel I think its nothing to worry. If we look at 2016 revenues by month July 2016 fell Y-o-Y versus July 2015 . Overall for year to date 2017 is up on revenues over 2016.

Taiwan Semiconductor Manufacturing Company Limited
Taiwan Semiconductor Manufacturing Company Limited

The revenues from Aug 2017 onwards are what I would be looking forward to. The revenues from Aug 2017 to H1 2018 will determine if Apple is on to another super-cycle with the launch of OLED iPhone.