TSMC's New N4C

[Daniel Nenni]

The TSMC Symposium was full of surprises. I do not remember a process from TSMC branded as cost effective but here it is:

And here is where it fits in the updated TSMC roadmap:

TSMC N4P is a big node so I understand the reasoning behind N4C and I hope it will be the N4 version that Arizona settles on next year. We shall see.

 

[IanD]

The TSMC Symposium was full of surprises. I do not remember a process from TSMC branded as cost effective but here it is:

View attachment 1877

And here is where it fits in the updated TSMC roadmap:

View attachment 1878

TSMC N4P is a big node so I understand the reasoning behind N4C and I hope it will be the N4 version that Arizona settles on next year. We shall see.

It all depends where the biggest markets for the Arizona fab turn out to be -- since the driver for N4C is cost and the Arizona fab is likely to cost more to run than the Taiwan fabs, it might be that N4P/N4X gets more custom. Of course the same fab can make any of them or a mix, so it's not like it's an either/or choice, it'll all be driven by customer demand...

 

[Daniel Nenni]

It all depends where the biggest markets for the Arizona fab turn out to be -- since the driver for N4C is cost and the Arizona fab is likely to cost more to run than the Taiwan fabs, it might be that N4P/N4X gets more custom. Of course the same fab can make any of them or a mix, so it's not like it's an either/or choice, it'll all be driven by customer demand...

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

 

[IanD]

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

It's going to be a difficult choice, and could depend on all sorts of things -- do (smaller?) US customers get a higher priority (e.g. faster TAT?) in a US fab, as opposed to playing second fiddle to the likes of Apple in a Taiwanese one? How much priority do they place on security of supply (earthquakes, China invading Taiwan)? Will pressure to fab in the US come not from the chip builders but their customers?

The bleeding-edge TSMC customers like Apple (who are already on N3 and developing in N2, not N4) are always going to fab in Taiwan because that's where the processes get brought up first, because that's where the process development is done -- that's not what the US fab is competing against, it's the "fast followers" or those customers for who the absolutely latest process is too difficult or expensive. Still *lots* of potential business though, assuming they don't all still go to Taiwan because it's a bit cheaper... ;-)

 

[hist78]

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

The US pricing strategy will have a high possibility to succeed because TSMC Arizona fab capacity is relatively smaller than TSMC's Gigafabs. The first two fabs will have a combined capacity of 50,000 wafers per month. Under the political and strategic consideration, Apple, Qualcomm, Nvidia, AMD, and Broadcom can easily allocate a small percentage of their orders to TSMC Arizona and help it to reach a high utilization rate.

Yes, the cost will increase but it can average out easily with the gigantic order quantity they placed in those non-US TSMC fabs.

 

[Paul2]

Would your company pay 10-20% more for local manufacturing?

That company is well known for walking away from contracts for 1%-2% difference.

Why would Cook negotiate for double the price, when he personally supervises Apples' procurement and goes on doing truly Chinese level bargaining.

 

[TBiggs]

Imagine you are taking delivery of a car or other dual sourced product and it can come from place A or place B. Would you accept that A is x% higher because it was higher cost center?

You as a buyer would say that is the supplier’s problem. This is especially true of said supplier had high GM and got huge multibillion subsidies and loans and tax breaks to offset and said costs. Really why do I care about your structure that is the suppliers problem.

It’s separately if supplier raises prices across the board and didn’t matter site A or B, that is different.

 

[nghanayem]

The TSMC Symposium was full of surprises. I do not remember a process from TSMC branded as cost effective but here it is:

12FFC was to sweeten the lack of scaling between 16FF and 20nm in spite of the added cost. I would assume N6 is in a similar camp given the lower mask count from EUV adoption and that it came out after N5 was spinning up.

View attachment 1877

I am curious to see what this looks like once we see teardowns. Based on the wording of better yield when using the optical shrink option of N4, that sounds like improved OPC/illumination/masks to me. But Fred would have a better idea. Maybe they added some lower cost metal layers, or some relaxed pitches on some of the layers for lower cost? I also wonder if it would be possible to do a non SiGe PMOS for folks who value cost over PMOS mobility without requiring designers to do too much work?

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

If the gap is that big, it is on TSMC to eat the margins since they would have to be doing something wrong to get 20% higher cost. If memory serves Scotten thought 7% would be a conservative estimate on the cost adder.

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

While A and M series SOCs would definitely be on bigger and better things by then smart watches, camera co-processor, security chipsets, and the fabled modems that always seem to just be a couple of years out will likely be on older processes for cost, yield, and technology platform features. There was also that talk of M4 using SOIC. Maybe N5/N3 base dies are on the menu for disagrigating SOC features (like Lakefield). Or maybe they are disagrigating their large system cache (in a step beyond what AMD does with V-cache)? It would also be neat if they had their on package DRAM 3D stacked, but maybe the cost benefit analysis prefers using COWS or POP/wire bonding for that specific use case.

 

[Barnsley]

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

Isnt it all mainly cycetime and price that people are competing on?

 

[akumar]

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

In Pharm industry Western Pharma companies pay higher price for CDMO (Contract Development & Mfg Org) services when delivered from Western World. Pharma CDMO is similar to foundry in semis. Customers pay one price when CDMO is executed out of China by Wuxi & another when done out of Ireland (& soon Singapore). A reason is Western customers are under pressure from their Boards to reduce China supply chain risk. Could similar argument work for TSMC Arizona?

 

[tooLongInEDA]

In Pharm industry Western Pharma companies pay higher price for CDMO (Contract Development & Mfg Org) services when delivered from Western World. Pharma CDMO is similar to foundry in semis. Customers pay one price when CDMO is executed out of China by Wuxi & another when done out of Ireland (& soon Singapore). A reason is Western customers are under pressure from their Boards to reduce China supply chain risk. Could similar argument work for TSMC Arizona?

Not convinced. Customers are far more aware of and sensitive to consumer pricing than for pharma products (where they usually don't have many alternatives or much choice).

 

[hist78]

The US pricing strategy will have a high possibility to succeed because TSMC Arizona fab capacity is relatively smaller than TSMC's Gigafabs. The first two fabs will have a combined capacity of 50,000 wafers per month. Under the political and strategic consideration, Apple, Qualcomm, Nvidia, AMD, and Broadcom can easily allocate a small percentage of their orders to TSMC Arizona and help it to reach a high utilization rate.

Yes, the cost will increase but it can average out easily with the gigantic order quantity they placed in those non-US TSMC fabs.

I'm not convinced the US pricing plan is going to work. So much negative feedback at the Symposium on that. In my experience cost is everything when it comes to wafers. Maybe some of the big systems companies will pay extra but most chip companies who have margin pressures will not.

Would your company pay 10-20% more for local manufacturing?

Even though Tim Cook of Apple said they would use TSMC AZ I have a hard time believing that. Apple can certainly afford the upcharge but Apple uses a customer process versions and is the first one on a node. By the time that technology hits AZ Apple would be on the next one.

We can do some calculations to see if this US pricing is feasible.

Here are some facts and assumptions:

1. Apple sold 427.1 million units of devices globally in 2023 on iPhone (231.8 million), iPad (135.3 million), Mac PCs (24 million), and Apple Watches (36 million, estimated).

2. TSMC Arizona Fab 1 has a monthly 20,000 of 300mm wafer capacity on N5/N4.

3, Apple 2023 revenue is $383 billion and net profit is $97 billion.

4. Assume Apple will commit to use 30% of TSMC Arizona Fab 1 capacity.

5. Assume each 300mm wafer will produce 400 Apple designed chips.

6. Assume each Apple chip will cost $20 more than the same chip made in Taiwan. The assumption of $20 cost increase for each chip produced in TSMC Arizona is significant. I have seen various cost estimates about the Apple designed/TSMC made chips. One study put the A17 Pro chip (used in the iPhone A15 Pro Max) at $130 a piece. I estimated the average Apple designed/TSMC made chips is less than $40 a piece across all Apple product lines.

Use above numbers and assumptions, Apple may need to pay $576 million more a year for chips produced in the TSMC Arizona fab. It seems to be a big number but:

1. It adds only $1.35 additional cost per unit to all Apple devices sold globally or just $2.48 more to each of the 231.8 million iPhones sold, using 2023 numbers.
2. It's 0.15% of Apple 2023 revenue.
3. It's 0.59% of Apple 2023 net profit.

This is a small price that Apple can manage easily in order to:

1. Proudly Made in USA.
2. Enhance supply chain resilience.
3. Create more high-paying jobs on US soil.
4. Deflate criticism about its economic and social impacts.
5. Better relationship with politicians and regulators among local and federal government, senators, and congressmen.

We can use the same approach to calculate the potential impacts for those who already committed to TSMC Arizona production such as AMD, Nvidia, Qualcomm, and Broadcom. Some of them may gain additional benefits, like national security related applications, by allocating a certain percentage of their orders to TSMC Arizona.

One of the key factors of this approach is that TSMC strategically positioned the Arizona Fab 1 with only 20,000 wafers per month while a typical TSMC Gigafab has a 100,000 wafers monthly output capacity. With such smaller capacity, the Arizona fab is easier to get customer orders without jeopardizing customers' cost structure.

My conclusion: TSMC US Pricing policy will work and the politicians, government, unions, suppliers, TSMC US customers , and TSMC will all happily claim a victory.

 

[TBiggs]

We can do some calculations to see if this US pricing is feasible.

Here are some facts and assumptions:

1. Apple sold 427.1 million units of devices globally in 2023 on iPhone (231.8 million), iPad (135.3 million), Mac PCs (24 million), and Apple Watches (36 million, estimated).

2. TSMC Arizona Fab 1 has a monthly 20,000 of 300mm wafer capacity on N5/N4.

3, Apple 2023 revenue is $383 billion and net profit is $97 billion.

4. Assume Apple will commit to use 30% of TSMC Arizona Fab 1 capacity.

5. Assume each 300mm wafer will produce 400 Apple designed chips.

6. Assume each Apple chip will cost $20 more than the same chip made in Taiwan. The assumption of $20 cost increase for each chip produced in TSMC Arizona is significant. I have seen various cost estimates about the Apple designed/TSMC made chips. One study put the A17 Pro chip (used in the iPhone A15 Pro Max) at $130 a piece. I estimated the average Apple designed/TSMC made chips is less than $40 a piece across all Apple product lines.

Use above numbers and assumptions, Apple may need to pay $576 million more a year for chips produced in the TSMC Arizona fab. It seems to be a big number but:

1. It adds only $1.35 additional cost per unit to all Apple devices sold globally, using 2023 numbers.
2. It's 0.15% of Apple 2023 revenue.
3. It's 0.59% of Apple 2023 net profit.

This is a small price that Apple can manage easily in order to:

1. Proudly Made in USA.
2. Supply chain resilience.
3. Create high-paying jobs on US soil.
4. Deflate criticism about its economic and social impacts.
5. Better relationship with politicians and regulators among local and federal government, senators, and congressmen.

We can use the same approach to calculate the potential impacts for those who already committed to TSMC Arizona production such as AMD, Nvidia, Qualcomm, and Broadcom. Some of them may gain additional benefits, like national security related applications, by allocating a certain percentage of their orders to TSMC Arizona.

One of the key factors of this approach is that TSMC strategically positioned the Arizona Fab 1 with only 20,000 wafers per month while a typical TSMC Gigafab has a 100,000 wafers monthly output capacity. With such smaller capacity, the Arizona fab is easier to get customer orders without jeopardizing customers' cost structure.

My conclusion: TSMC US Pricing policy will work and the politicians, government, unions, suppliers, TSMC US customers , and TSMC will all happily claim a victory.

Brilliance, every single source Apple supplier is sharpening their pencils and licking their chops

 

[Daniel Nenni]

We can do some calculations to see if this US pricing is feasible.

Here are some facts and assumptions:

1. Apple sold 427.1 million units of devices globally in 2023 on iPhone (231.8 million), iPad (135.3 million), Mac PCs (24 million), and Apple Watches (36 million, estimated).

You have to consider timing as well. Apple is on N3 in 2023 and 2024 and 1H 2025 so 2 years of big volumes per node. Apple is now designing to N2 which they will probably be on in 2H 2025 and 2026 and 1H 2027. I'm sure Apple is still selling N16 products in some parts of the world but not big volumes.

When will N3 and N2 be in AZ? Not until after the big Apple runs.

Apple can certainly claim US made but the majority of the wafers will still be made in Taiwan?

 

[TBiggs]

You have to consider timing as well. Apple is on N3 in 2023 and 2024 and 1H 2025 so 2 years of big volumes per node. Apple is now designing to N2 which they will probably be on in 2H 2025 and 2026 and 1H 2027. I'm sure Apple is still selling N16 products in some parts of the world but not big volumes.

When will N3 and N2 be in AZ? Not until after the big Apple runs.

Apple can certainly claim US made but the majority of the wafers will still be made in Taiwan?

What products would you say are on N16?

 

[hist78]

You have to consider timing as well. Apple is on N3 in 2023 and 2024 and 1H 2025 so 2 years of big volumes per node. Apple is now designing to N2 which they will probably be on in 2H 2025 and 2026 and 1H 2027. I'm sure Apple is still selling N16 products in some parts of the world but not big volumes.

When will N3 and N2 be in AZ? Not until after the big Apple runs.

Apple can certainly claim US made but the majority of the wafers will still be made in Taiwan?

I was talking about N5/N4 nodes used at TSMC Arizona Fab 1. There are plenty of products across AMD, Nvidia, Qualcomm, Broadcom, and Apple using N5/N4 technology. Apple alone has several A, M, H, R, and S series processors manufactured with N5/N4. N5/N4 will be a long-lived foundry node at TSMC Taiwan and Arizona.

I believe TSMC N3/N2 at Arizona Fab 2 will be another long-lived foundry node too. It was announced the TSMC Arizona Fab 2 will go HVM around 2027-2028. There is enough time to do implementation and adjustments.


"Apple can certainly claim US made but the majority of the wafers will still be made in Taiwan?"

Yes, you got the point. Every person and organization involved in the TSMC Arizona Fab projects can claim a victory of their own. Isn't that beautiful?!

Except Intel may not be too happy about it. The supply chain diversity and resilience advocated by Pat Gelsinger will be greatly improved by the arrival of TSMC Arizona Fab 1.

 

[jorgequinonez]

We can do some calculations to see if this US pricing is feasible.

Here are some facts and assumptions:

1. Apple sold 427.1 million units of devices globally in 2023 on iPhone (231.8 million), iPad (135.3 million), Mac PCs (24 million), and Apple Watches (36 million, estimated).

2. TSMC Arizona Fab 1 has a monthly 20,000 of 300mm wafer capacity on N5/N4.

3, Apple 2023 revenue is $383 billion and net profit is $97 billion.

4. Assume Apple will commit to use 30% of TSMC Arizona Fab 1 capacity.

5. Assume each 300mm wafer will produce 400 Apple designed chips.

6. Assume each Apple chip will cost $20 more than the same chip made in Taiwan. The assumption of $20 cost increase for each chip produced in TSMC Arizona is significant. I have seen various cost estimates about the Apple designed/TSMC made chips. One study put the A17 Pro chip (used in the iPhone A15 Pro Max) at $130 a piece. I estimated the average Apple designed/TSMC made chips is less than $40 a piece across all Apple product lines.

Use above numbers and assumptions, Apple may need to pay $576 million more a year for chips produced in the TSMC Arizona fab. It seems to be a big number but:

1. It adds only $1.35 additional cost per unit to all Apple devices sold globally or just $2.48 more to each of the 231.8 million iPhones sold, using 2023 numbers.
2. It's 0.15% of Apple 2023 revenue.
3. It's 0.59% of Apple 2023 net profit.

This is a small price that Apple can manage easily in order to:

1. Proudly Made in USA.
2. Enhance supply chain resilience.
3. Create more high-paying jobs on US soil.
4. Deflate criticism about its economic and social impacts.
5. Better relationship with politicians and regulators among local and federal government, senators, and congressmen.

We can use the same approach to calculate the potential impacts for those who already committed to TSMC Arizona production such as AMD, Nvidia, Qualcomm, and Broadcom. Some of them may gain additional benefits, like national security related applications, by allocating a certain percentage of their orders to TSMC Arizona.

One of the key factors of this approach is that TSMC strategically positioned the Arizona Fab 1 with only 20,000 wafers per month while a typical TSMC Gigafab has a 100,000 wafers monthly output capacity. With such smaller capacity, the Arizona fab is easier to get customer orders without jeopardizing customers' cost structure.

My conclusion: TSMC US Pricing policy will work and the politicians, government, unions, suppliers, TSMC US customers , and TSMC will all happily claim a victory.

They could make chips for the Apple Watch, Airpods and other chips on N4C. Since you get more of those chips per wafer, the cost would be considerably be less. It doesn't have to be an A series SoC.

 

[Daniel Nenni]

What products would you say are on N16?

iPhone 6 and 6+ and iPads might still be sold in 3rd world countries.

 

[blueone]

iPhone 6 and 6+ and iPads might still be sold in 3rd world countries.

Not a bad idea, but seems very unlikely. Apple stopped supporting the iPhone 6 about eight years ago. Even the iPhone 8 became unsupported four years ago. Several industry analysts are saying Apple needs to offer low-cost iPhones in some markets. Unfortunately, I can easily think of a lot of incremental costs involved, especially for chips redone in a cheaper fab process, additional IOS development for less capable chipsets, testing, and support. I find it difficult to see Apple doing this.

 

[IanD]

If cost for a given functionality is most important not absolute performance/density/power -- which is the case for many *many* products -- then you absolutely don't want to be in a bleeding-edge node, the cost per gate is higher to start off with and they haven't started to slide down the cost curve. This is especially true for N3/N2/A16 today. N4C is a cost-optimised N4 which is a shrunk cost-optimised N5, and is likely to be significantly cheaper per gate than anything 3nm or below.

The next super-cost-optimised process up is probably 12FFC, if you don't need the higher density and lower power of N4C -- these are two process nodes targeted at cost-sensitive products, one using immersion and one using EUV. Which one is cheaper will very much depend on the product, specifically how much higher density N4C is for the total chip -- logic scales pretty well with node, RAM less well, SERDES less well still, analog and I/O less well still...