Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/tsmc-q3-2021-discussion.14761/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

TSMC Q3 2021 Discussion

Daniel Nenni

Admin
Staff member
Listen to the call, read the transcript and lets talk. Exciting times in the semiconductor industry, absolutely!


CC Wie opening statement:

"
Based on the midpoint of our fourth-quarter revenue guidance, our full-year 2021 revenue is expected to grow about 24% year-over-year in the U.S. dollar term. On the inventory front, we continue to expect our customers and the supply chain to gradually prepare a higher level of inventory in the second half of this year as compared to the historical seasonal level. Given the industry continually needs to ensure supply security, we expect the supply chain to maintain a higher level of inventory for a longer period of time."

Companies are hoarding, absolutely. But for how long? When will the inventory correction happen? Mid 2022?

"In the near term, we continue to observe short-term imbalances due to interruptions in the supply chain brought over by COVID-19. We also continue to observe the structure increase in long-term demand, underpinned by the industry megatrends of 5G and HPC-related applications, and the higher silicon content in many end devices, including automotive, PCs, servers, networking, and smartphones."

Yes, the semiconductor industry is alive and well. Double digit growth numbers ahead, absolutely. I would also add that AI is hitting most every chip and that will increase die size and performance demands from new process technologies.

"While the short-term imbalances may or may not persist, we believe our technology leadership, will enable TSMC to capture the strong demand for our advanced and specialty technologies. And we expect our capacity to remain tight in 2021 and throughout 2022."

Agreed. TSMC is dominating 5nm HVM and 3nm design starts. This will be another 28nm market burst all over again. TSMC has clearly dominated the Foundry FinFET era, next is GAA and that could be anybody's game. Exciting times!

"Next, let me talk about TSMC as a long-term growth driver and return. We are entering a period of higher structural growth. The multiyear megatrend of 5G and HPC-related applications are expected to fuel massive requirement for computation power and propel the greater need for energy-efficient computing, which demand the use of leading-edge technologies.

This megatrend will not only spur unit growth, but also drive increasing semiconductor content in HPC, smartphone, automotive, and IoT applications. COVID-19 has also fundamentally accelerated the digital transformation making semiconductors more pervasive and essential in people's lives. With our technology leadership, manufacturing, excellence, and customer’s trust. TSMC is better positioned to capture the course from its favorable industry megatrend. We sell differentiated technologies
."

Translation: TSMC is the one stop wafer shop with a massive design enablement ecosystem...

"Towards raise the structural increase in the long-term market demand profile. TSMC is working closely with our customers to plan our capacity and investing in leading-edge and specialty technologies to support their demand. Our capital investment decisions are based on four disciplines; technology leadership, flexible and responsive manufacturing, retaining customers' trust and earning the proper return."

This is a critical point and the key to TSMC's success, close collaboration, the massive ecosystem, enabled by TSMC's refusal to compete with customers or be frenemies. IDM foundries continue to struggle with this.

"At the same time, we faced manufacturing cost challenges due to an increase in process complexity at DD Note, new investment in mature notes, expansion of our global manufacturing footprint, and rising material in basic commodity cost. As we continue to work closely with our customers to support their goals, our pricing strategy will remain strategic, not opportunistic, to refer to our [Indiscernible] creation. We will also continue to work diligently with our supplier to deliver, and cost improvement.

Even us, we showed a greater burden of investment for the industry, by taking such actions we believe we can achieve a proper return that enables us to invest to support our customers, of course, and deliver long term profitable goals with 50% and higher gross margin for our shareholders. Now let me talk about our Japan fab print. We are expanding our manufacturing footprint to extend and enhance our competitive advantage in providing industry-leading technologies.

The world's largest [Indiscernible] capacity, efficient and cost-effective manufacturing, and to better serve our customer. Our global manufacturing expansion strategy is based on customers' needs, business opportunities, operating efficiency, and cost-economic considerations. After conducting due diligence, we announced our intention to build a specialty technology fab in Japan, subject to our Board of Directors' approval. We have received a strong commitment to supporting these projects from both our customers and the Japanese government. This fab will utilize 22, 28-nanometer technology for semiconductor wafer fabrication.

Fab construction is scheduled to begin in 2022 and production is targeted to begin in May 2024. Further details will be provided subject to the board's approval. We believe the expansion of our global manufacturing footprint will enable us to better serve our customer's needs in the rich, global talent, while early in the proper return from our investments and deliver long-term profitable quarters for our shareholders."


TSMC building fabs in the US, Japan, India, and elsewhere will bring us closer to world peace than ever before. It was said some time ago that as long as there is common business interests countries will stay at peace. McDonalds, Burger King, Pizza Hut, KFC, Starbucks, etc.... Let's hope semiconductors is better for the world than fast food franchises.

"Finally, I will talk about the N-3 and N-3E status. Our N-3 technology will use FinFET transistor’s structure to deliver the best technology maturity, performance, and cost for our customers. Our N3 technology development is on track. We had developed complete platform support for both HPC and smartphone applications. N-3 risk production is scheduled in 2021, and production was starting in the 2nd half of 2022. We continue to see a high level of customer engagement at N-3, and expect a newer tap-out for N-3 for the first year as compared with N-5. We also introduced N3E as an extension of our N3 family."

"
N3E while feature improved manufacturing process window, with better performance, power, and yield. Volume production of N3E is scheduled for one year after N3. Our 3-nanometer technology will be the most advanced foundry technology in both PPA and transistor technology when it is introduced. With our technology leadership and strong customer demand, we are confident that the N3 family will be another long and vast -- will be a large and long-lasting node of TSMC. This concludes our key message. Thank you for your attention."


As I said before, the Apple version of N3 is on track for iProducts in 2022 with the commercial version N3E the following year. N3E will have less EUV steps for higher throughput and greater efficiencies from what I have been told.

Now let's talk about the Q&A:
 
Just curious, Is AMD's use of 5nm wafers basically predicated on Apple getting 3nm capacity?
 
I notice TSMC only mentioned the upside of N3E, not the downside. Since there's no such thing as a free lunch, people may wonder what they could have done which improves process window, power and yield while having fewer EUV steps...
 
Just curious, Is AMD's use of 5nm wafers basically predicated on Apple getting 3nm capacity?
Not really. Big customers like Apple, AMD and Intel basically have their own TSMC capacity and commitment.
Apple, AMD and Intel already committed their 3nm qty to TSMC possibly in 2020. Not all customer want to have 3nm as early as possible. They have their own roadmap and schedule.

Apple will get 3nm first not because of AMD.
 
Not really. Big customers like Apple, AMD and Intel basically have their own TSMC capacity and commitment.
Apple, AMD and Intel already committed their 3nm qty to TSMC possibly in 2020. Not all customer want to have 3nm as early as possible. They have their own roadmap and schedule.

Apple will get 3nm first not because of AMD.
I think you misread my question - I know Apple isn't getting 3nm because of AMD, I was just curious if basically TSMC's 5nm capacity frees up for AMD once Apple is on 3nm. I assume getting 5nm a year or two after release provides cost benefits to AMD too.

Are you saying that TSMC builds new fab capacity for each customer along the way?

Thanks!
 
Isn't Samsung have the edge since it is working with 3nm GAA right now?
The first question in the Q&A was actually regarding TSMC's future roadmap and GAA. CC actually DID NOT verbally commit to using GAA for their future 2nm process node...only that "the GAA structure is being considered". CC actually stated that unlike his competitors (Intel & Samsung), he was not ready to release information regarding TSMC's future roadmap yet. However, CC did say that:
I can share with you that in our 2 nm technology, the density and performance will be the most competitive in 2025.
...
Let me conclude in one sentence, we'll become very competitive and we are confident that our technology leadership will be maintained.
 
I think you misread my question - I know Apple isn't getting 3nm because of AMD, I was just curious if basically TSMC's 5nm capacity frees up for AMD once Apple is on 3nm. I assume getting 5nm a year or two after release provides cost benefits to AMD too.

Are you saying that TSMC builds new fab capacity for each customer along the way?

Thanks!
Again, AMD will get 5nm in 2022 mostly because AMD already committed and provide this schedule to TSMC 2-3 years ago. AMD and Apple all on schedule moving to next node.
I will say AMD moving to 5nm free up more capacity to AMD 7nm products to PS5/Xbox.

" I assume getting 5nm a year or two after release provides cost benefits to AMD too. " ..........yes
But for 3nm, more companies are willing to pay more and take the first 3nm capacity.

"Are you saying that TSMC builds new fab capacity for each customer along the way?" ........in some way yes. We all know that some of the new big factories are for Intel.
 
As I said before, the Apple version of N3 is on track for iProducts in 2022 with the commercial version N3E the following year. N3E will have less EUV steps for higher throughput and greater efficiencies from what I have been told.

The following comments from C.C. today contradict this:
----------------------------------------
C. C. Wei

Okay. Great. Hi. This is C. C. Wei. Let me answer the second part of the question first. [Indiscernible] costs definitely it is higher than N5. That is because of technical complexity and we have to use many new pieces of equipment which is -- cost higher.

But then the ramp-up is very similar to the previous node. We saw many customers engagement actually is higher than what we observed in the previous node. The second half of 2022 will be our mass production but you can expect the revenue will be seen in the first quarter of 2023 because it takes a long tech cycle time to have all those wafers out.

Brett Simpson


Okay. So basically on three-nanometer, this won't be -- typically, you see your first revenue Q2 or Q3 it's going to be later next year. Is that right?

C. C. Wei

That's right.

----------------------------------------

So I think your prediction is wrong here. It looks like next year's M2 and A16 will be on 4nm, and then Spring 2023 is when the 3nm M2X(?) would arrive, followed by the A17 iPhone.
 
Last edited:
So I think your prediction is wrong here. It looks like next year's M2 and A16 will be on 4nm, and then Spring 2023 is when the 3nm M2X(?) would arrive, followed by the A17 iPhone.

Could be. I need to dig more on this. There are no secrets in the ecosystem, absolutely.
 
The following comments from C.C. today contradict this:
----------------------------------------
C. C. Wei

Okay. Great. Hi. This is C. C. Wei. Let me answer the second part of the question first. [Indiscernible] costs definitely it is higher than N5. That is because of technical complexity and we have to use many new pieces of equipment which is -- cost higher.

But then the ramp-up is very similar to the previous node. We saw many customers engagement actually is higher than what we observed in the previous node. The second half of 2022 will be our mass production but you can expect the revenue will be seen in the first quarter of 2023 because it takes a long tech cycle time to have all those wafers out.

Brett Simpson


Okay. So basically on three-nanometer, this won't be -- typically, you see your first revenue Q2 or Q3 it's going to be later next year. Is that right?

C. C. Wei

That's right.

----------------------------------------

So I think your prediction is wrong here. It looks like next year's M2 and A16 will be on 4nm, and then Spring 2023 is when the 3nm M2X(?) would arrive, followed by the A17 iPhone.
The N3 delays that customers have been notified of could certainly make it too late for Apple to meet the requirement to have mass production in time for Xmas 2022 so it's been said that they will use N4 (shrunk N5 for Apple?) but of course neither Apple nor TSMC will confirm this.

It certainly seems that TSMC want to move customers from N3 to N3E to "improve manufacturing process window, power, performance and yield" -- presumably meaning that they've pushed the design rules too hard and want to relax some of them a bit (bigger pitch = lower capacitance and higher yield), like Intel did for 10nm but hopefully nothing like as disastrously. The tradeoff must be lower gate density...
 
The following comments from C.C. today contradict this:
----------------------------------------
C. C. Wei

Okay. Great. Hi. This is C. C. Wei. Let me answer the second part of the question first. [Indiscernible] costs definitely it is higher than N5. That is because of technical complexity and we have to use many new pieces of equipment which is -- cost higher.

But then the ramp-up is very similar to the previous node. We saw many customers engagement actually is higher than what we observed in the previous node. The second half of 2022 will be our mass production but you can expect the revenue will be seen in the first quarter of 2023 because it takes a long tech cycle time to have all those wafers out.

Brett Simpson


Okay. So basically on three-nanometer, this won't be -- typically, you see your first revenue Q2 or Q3 it's going to be later next year. Is that right?

C. C. Wei

That's right.

----------------------------------------

So I think your prediction is wrong here. It looks like next year's M2 and A16 will be on 4nm, and then Spring 2023 is when the 3nm M2X(?) would arrive, followed by the A17 iPhone.
Seems to imply a 2.5 year cycle to go from 5nm to 3nm then (A14 on 5nm started selling Sept 2020)
 
Seems to imply a 2.5 year cycle to go from 5nm to 3nm then (A14 on 5nm started selling Sept 2020)

Remember, Apple is the process pipe cleaner for TSMC. The process version Apple uses is not the same as what is used by the mainstream TSMC customers thus the delay from the release of the iProducts to the releases of AMD, QCOM, and NVDA products.

If Apple were to reuse the same process then there is less activity within the ecosystem in regards to IP requalification etc... I have yet to see that happen so Apple does in fact use new processes every year. They are not publicly named processes so it does not necessarily track with the TSMC nodes names N7, N7+, N6, N5, N4, N3 etc...

From what I understand Apple is using an enhanced version of N5 this year, which will be called N4 next year. I'm not positive what Apple will use in 2022 but it will not be the same process as this year. Apple always get's a new process no matter what the name ends up being.

TSMC does a careful balancing act with the fabs. They build fabs specific for Apple but Apple moves on quickly to new processes so TSMC backfills the Apple fabs with new products from AMD, QCOM, NVIDIA, and other big runners. This is significant capacity so it not being available if Apple stayed on a process for two to three years would be noticeable in the supply chain and TSMC revenue stream.

It will be interesting to see what happens next year but with the transparency of TSMC we will know one way or another.
 
Remember, Apple is the process pipe cleaner for TSMC. The process version Apple uses is not the same as what is used by the mainstream TSMC customers thus the delay from the release of the iProducts to the releases of AMD, QCOM, and NVDA products.

If Apple were to reuse the same process then there is less activity within the ecosystem in regards to IP requalification etc... I have yet to see that happen so Apple does in fact use new processes every year. They are not publicly named processes so it does not necessarily track with the TSMC nodes names N7, N7+, N6, N5, N4, N3 etc...

From what I understand Apple is using an enhanced version of N5 this year, which will be called N4 next year. I'm not positive what Apple will use in 2022 but it will not be the same process as this year. Apple always get's a new process no matter what the name ends up being.

TSMC does a careful balancing act with the fabs. They build fabs specific for Apple but Apple moves on quickly to new processes so TSMC backfills the Apple fabs with new products from AMD, QCOM, NVIDIA, and other big runners. This is significant capacity so it not being available if Apple stayed on a process for two to three years would be noticeable in the supply chain and TSMC revenue stream.

It will be interesting to see what happens next year but with the transparency of TSMC we will know one way or another.
Thanks Daniel! That explains a lot (and very clearly). Cheers!
 
Thanks Daniel! That explains a lot (and very clearly). Cheers!
On another note, performance per watt is the critical Apple criteria so that is what we should be looking at when comparing the Ax SoC series. I have the new iPhone 13 Pro and I can tell you that it's a fine piece of technology. I upgraded from an iPhone 10. First and foremost the connectivity is tremendous. I use it for navigation when I sail, drive and fly. It is noticeably faster with better connectivity. I also use my iPhone as a personal hotspot when sailing for my laptop and other devices and it does a terrific job. Battery life is also excellent thus the performance per watt comment. It really is all about the battery. I can easily go a full day of navigation without a charge. I had to keep my iPhone 10 plugged in quite a bit.
 
On another note, performance per watt is the critical Apple criteria so that is what we should be looking at when comparing the Ax SoC series. I have the new iPhone 13 Pro and I can tell you that it's a fine piece of technology. I upgraded from an iPhone 10. First and foremost the connectivity is tremendous. I use it for navigation when I sail, drive and fly. It is noticeably faster with better connectivity. I also use my iPhone as a personal hotspot when sailing for my laptop and other devices and it does a terrific job. Battery life is also excellent thus the performance per watt comment. It really is all about the battery. I can easily go a full day of navigation without a charge. I had to keep my iPhone 10 plugged in quite a bit.
The A15 looks like a pretty decent jump in efficiency and performance over the A14, despite given the "same 5nm node" advertised: https://images.anandtech.com/doci/16983/SPECint-energy.png

As there were some IPC gains in addition to frequency (while also using less power), I assume some of that is reworking the silicon a bit (IPC), but getting higher clocks at lower power sounds like that "preview" node you are talking about.

It's refreshing to see the vendors go for better battery life over "only performance gains" in recent years..

Another topic for another time but with the rumblings of key engineers leaving Apples Semi team (some of the original PA-Semi guys IIRC) I'm curious what effect that's going to have on the next few generations of Apple processors.
 
From what I understand Apple is using an enhanced version of N5 this year, which will be called N4 next year. I'm not positive what Apple will use in 2022 but it will not be the same process as this year. Apple always get's a new process no matter what the name ends up being.
I believe your first statement is correct (and the general comment that Apple's PDK is different) but I don't agree with the assumption this is "N4". There are clearly differences between A14 and A15 in terms of implementation, but TSMC explicitly stated that N4 would offer cost reduction (assumed through reduced EUV layers) and scaling via optical shrink. These are not present on A15. I would rephrase this as TSMC continues to release BKM improvements within each "node", and those improvements can show up in basically any device. AMD's version of N7 in 2019 was different than Apple / HiSi's in 2018, and Apple's 2019 version was yet different still. None of those were "N7+" because they didn't absorb EUV or the 3:2 M1 pitch ratio. Another case would be recent MTK D1200 which is supposedly N6 but also not picking up any of the expected density improvement (CPODE), so ??? maybe they absorb some FEOL BKMs that make it N6-esque.
 
It certainly seems that TSMC want to move customers from N3 to N3E to "improve manufacturing process window, power, performance and yield" -- presumably meaning that they've pushed the design rules too hard and want to relax some of them a bit [...] The tradeoff must be lower gate density...
My speculation on this (as well as N4) is that TSMC took a very prudent approach of using multi-patterned EUV for metal pitches that should be possible to achieve with single-shot once the other pieces are in place (higher source power, better resists, etc). However on the DRs, this may mean some difference in line-end spacing, and certainly is going to affect electrical if there are no more A/B metal patterning. So they may not be pushed to relax pitches explicitly, but the resulting P&R impact may still cause a loosening of the routed utilization. I guess we'll see what's under the hood next year.
 
  • Like
Reactions: VCT
"...................... I think we have been consistently saying that N3 will begin production in the second half of 2022. That has been a consistent message since we first introduced N3 in 2019."

So TSMC doesn't think it's a delay.
 
"...................... I think we have been consistently saying that N3 will begin production in the second half of 2022. That has been a consistent message since we first introduced N3 in 2019."

So TSMC doesn't think it's a delay.
So they are probably talking about N3 will have 3-6 months longer (not delay) of ramp up than N5 as planned.

It's kinda funny some commentators are piling up on this. And there's a little bit of sadness that people gradually lost interest on how many months or years Intel is delaying a new product rollout. TSMC and Intel each is really working in their own universe with two very different mindsets, performance targets, and ecosystems.
 
Last edited:
Back
Top