Samsung Foundry Faces Yield Struggles and Client Losses, External Push for Spinoff and U.S. Listing

[kevin01]

Can you elaborate on this? BSPD is part of 18A. Customers that are running shuttles at Intel now have access to BSPD. Intel 14A will be 2nd generation BSPD for Intel. In areas like fin-fet and strained silicon Intel have been able to show significant improvement in their 2nd generation implementation. Why would you thing that this is an advantage for TSMC?

We are all just speculating at this point. However, if IFS can successfully deliver the 18A node in 2025, I would say IFS and TSMC have equal chances to lead in 2026/2027.

This endeavor is extremely challenging; one stumble (by either party) would yield the lead to the other.

 

[Daniel Nenni]

We are all just speculating at this point. However, if IFS can successfully deliver the 18A node in 2025, I would say IFS and TSMC have equal chances to lead in 2026/2027.
This endeavor is extremely challenging; one stumble (by either party) would yield the lead to the other.

The N2 vs 18A battle has already been won. TSMC won by a landslide. 100% of the semiconductor ecosystem is behind TSMC, that is the landslide.

The difference between Intel BSPD and TSMC SPR is customer involvement. I can assure you TSMC's N3 and N2 customers are behind SPR and that is big. You also have to factor in 16A NanoFlex. Those additional double digit gains in power and density could be the frosting on the TSMC victory cake.

 

[kevin01]

The N2 vs 18A battle has already been won. TSMC won by a landslide. 100% of the semiconductor ecosystem is behind TSMC, that is the landslide.

The difference between Intel BSPD and TSMC SPR is customer involvement. I can assure you TSMC's N3 and N2 customers are behind SPR and that is big. You also have to factor in 16A NanoFlex. Those additional double digit gains in power and density could be the frosting on the TSMC victory cake.

We may have a different definition of "lead". Your view seems to be more about commercial success, and I think TSMC has a big advantage in this regard, even if TSMC cannot deliver its new node on time, simply due to customer inertia.

My view of "lead" is more about whether each company can deliver its technical goals (PPA, yield, etc.) within the planned timeframe.

I am not discounting commercial success, which is very important and the ultimate goal. However, if a company cannot deliver its roadmap on time, fortunes can turn quickly. Samsung's foundry business is an example here.

If IFS can successfully deliver 18A in 2025, there is no reason to believe that TSMC can deliver its next node while IFS cannot, or vice verse. It is a fair game.

 

[Daniel Nenni]

We may have a different definition of "lead". Your view seems to be more about commercial success, and I think TSMC has a big advantage in this regard, even if TSMC cannot deliver its new node on time, simply due to customer inertia.

My view of "lead" is more about whether each company can deliver its technical goals (PPA, yield, etc.) within the planned timeframe.

I am not discounting commercial success, which is very important and the ultimate goal. However, if a company cannot deliver its roadmap on time, fortunes can turn quickly. Samsung's foundry business is an example here.

If IFS can successfully deliver 18A in 2025, there is no reason to believe that TSMC can deliver its next node while IFS cannot, or vice verse. It is a fair game.

I think we have a different definition of lead. I think the proof is in the pudding. If 18A is only doing Internal Intel designs then there is no pudding. The Intel 18A PDK worked for internal design (PPAY) but not external? Hopefully that gets fixed for 14A but until it does there is no pudding. TSMC takes a low risk approach for a reason, they have customers collaborating who cannot be late. Intel did not take a low risk approach to 18A BSPD. Do you blame me for waiting for the pudding before I determine who is in the lead?

Oh great, now I want some chocolate pudding, the kind that used to start from a box

 

[Artificer60]

I think we have a different definition of lead. I think the proof is in the pudding. If 18A is only doing Internal Intel designs then there is no pudding. The Intel 18A PDK worked for internal design (PPAY) but not external? Hopefully that gets fixed for 14A but until it does there is no pudding. TSMC takes a low risk approach for a reason, they have customers collaborating who cannot be late. Intel did not take a low risk approach to 18A BSPD. Do you blame me for waiting for the pudding before I determine who is in the lead?

Oh great, now I want some chocolate pudding, the kind that used to start from a box

Thanks for the clarification. When you say lead, I think technology. It seems you are looking at it from a more commercial sense. I would agree that from a commercial perspective that TSMC clearly has the lead, and will continue to hold it easily for the foreseeable future.

I suspect Gelsinger's goal is just to capture enough of the market to support leading edge development, which is a much more achievable goal in my mind. Though that is just my attempt to read the tea leaves.

You mention the 18A PDK. Does it not include BSPD for external customers? My impression was that it does, but I haven't seen it, and honestly wouldn't know how to interpret it if I did. I'm not a designer.

 

[dkr1986]

I think we have a different definition of lead. I think the proof is in the pudding. If 18A is only doing Internal Intel designs then there is no pudding. The Intel 18A PDK worked for internal design (PPAY) but not external? Hopefully that gets fixed for 14A but until it does there is no pudding. TSMC takes a low risk approach for a reason, they have customers collaborating who cannot be late. Intel did not take a low risk approach to 18A BSPD. Do you blame me for waiting for the pudding before I determine who is in the lead?

Oh great, now I want some chocolate pudding, the kind that used to start from a box

But there are external customers on 18A. They seem to have made the 18A PDK work for them.

Yes, only Intel internal designs are on 18A for HVM now (PTL being the lead product and CWF later). But if those internal products turn out to be better performant (PPA) than competitors that are now behind TSMC for N2 (which is also coming on a little later iirc), wouldn't that convince some of them to port their designs to IF? By that time, the 18A PDK and 18A process is probably a bit more refined too (risk off with internal designs).

Isn't that how the process leadership of 18A (as claimed by Intel) will drive more growth for IF? Or do we currently know 100% that 18A will definitely trail N2 in performance to call this race is over?

I don't think it is realistic to expect IF to topple TSMC right away. Intel themselves never claimed that, they are only looking to be no 2 foundry iirc. Customers are not going to sign up with Intel even if the node offerings is good, right away. They will need to see Intel execute before they entrust IF with their own designs. In 18A's case, Intel's own products will serve to measure that execution.

So I think the most important thing for Intel right now is to execute on its own designs on 18A and bring all the outsourced products back to IF.

 

[hskuo]

I think we have a different definition of lead. I think the proof is in the pudding. If 18A is only doing Internal Intel designs then there is no pudding. The Intel 18A PDK worked for internal design (PPAY) but not external? Hopefully that gets fixed for 14A but until it does there is no pudding. TSMC takes a low risk approach for a reason, they have customers collaborating who cannot be late. Intel did not take a low risk approach to 18A BSPD. Do you blame me for waiting for the pudding before I determine who is in the lead?

Oh great, now I want some chocolate pudding, the kind that used to start from a box

There are technology lead and commercial lead typically. Sometimes it can be winning the face or the reality. For example, Samsung used to lead in technology like 1st EUV, 1st GAA... It wins face but seems not to win commercial reality. As we know, when the company is in catch-up mode or as second supplier, there will be switch-cost incurred. The incentives like lower die cost, faster delivery/volume or better performance/cost usually are paining for the follower, which means deeper pocket is needed. It is rare to see the follower will catch up immediately in N+1 node and in 3-5 years. When we see people talk about DTCO, STCO and more, it implied more challenges and integration to be conquered. Good luck for all advanced semi manufacturers.

 

[kevin01]

There are technology lead and commercial lead typically. Sometimes it can be winning the face or the reality. For example, Samsung used to lead in technology like 1st EUV, 1st GAA... It wins face but seems not to win commercial reality. As we know, when the company is in catch-up mode or as second supplier, there will be switch-cost incurred. The incentives like lower die cost, faster delivery/volume or better performance/cost usually are paining for the follower, which means deeper pocket is needed. It is rare to see the follower will catch up immediately in N+1 node and in 3-5 years. When we see people talk about DTCO, STCO and more, it implied more challenges and integration to be conquered. Good luck for all advanced semi manufacturers.

Samsung is never in the technology lead, which is judged by superior PPA with a good yield rate. Claiming to be using the first EUV or the first GAA are not the tech leadership.

 

[Daniel Nenni]

Thanks for the clarification. When you say lead, I think technology. It seems you are looking at it from a more commercial sense. I would agree that from a commercial perspective that TSMC clearly has the lead, and will continue to hold it easily for the foreseeable future.

I suspect Gelsinger's goal is just to capture enough of the market to support leading edge development, which is a much more achievable goal in my mind. Though that is just my attempt to read the tea leaves.

You mention the 18A PDK. Does it not include BSPD for external customers? My impression was that it does, but I haven't seen it, and honestly wouldn't know how to interpret it if I did. I'm not a designer.

I guess I should say proven foundry technology. Comparing an IDM to a foundry is not apples to apples. I have not been able to find customers that are using Intel 18A BSPD. I have not found IP companies that are working with BSPD either. Several IP vendors have done the 18A port but they have no customers. I think 18A BPSD is an Intel only thing for now.