Samsung Reportedly Lands a 4nm Mega Order – Why is AMD Switching to “Dual Foundry Mode” for Its Next-Gen Chips?

[Daniel Nenni]

News outlets have been spreading this "rumor" as fact. Personally I do not believe it, not even close. Does anyone know different? I will be at an industry event next week and will find out.

According to TechNews’ report, there are recent rumors indicating that AMD’s next-generation chip, with the Zen5C architecture codenamed “Prometheus,” will adopt a “Dual Foundry Mode.” This means it will simultaneously utilize TSMC’s 3nm and Samsung’s 4nm processes. This move suggests that AMD aims to diversify chip manufacturing, avoiding reliance solely on TSMC for its upcoming products.

[News] Samsung Reportedly Lands a 4nm Mega Order – Why is AMD Switching to “Dual Foundry Mode” for Its Next-Gen Chips? | TrendForce Insights

According to TechNews’ report, there are recent rumors indicating that AMD's next-generation chip, with the Zen5C architecture codenamed "Prometheus,"...

www.trendforce.com

Rumors swirl around AMD’s upcoming chip architecture, codenamed “Prometheus,” featuring the Zen 5C core. As reported by TechNews, the chip is poised to leverage both TSMC’s 3nm and Samsung’s 4nm processes simultaneously, marking a shift in the competitive landscape from process nodes, yield, and cost to factors like capacity, ecosystem, and geopolitics, are all depends on customer considerations.

[News] A battle on 4nm: AMD Teams Up with Samsung, while Google Weighs Supplier Split | TrendForce Insights

Rumors swirl around AMD's upcoming chip architecture, codenamed "Prometheus," featuring the Zen 5C core. As reported by TechNews, the chip is poised t...

www.trendforce.com

According to a report from Wccftech, senior executives at Samsung have indicated that major players such as super-scale data centers, automotive original equipment manufacturers (OEMs), and other clients have been in contact with Samsung, considering the adoption of Samsung’s foundry services to manufacture their designed chips.

[News] Samsung Reportedly Secures AMD and Tesla Orders for 4/5 nm Chips | TrendForce Insights

According to TechNews' report, during a recent financial conference, Samsung revealed its plans to diversify its sales structure by expanding its clie...

www.trendforce.com

 

[blueone]

I've been reading rumors about this for a while. While I always read The Register with suitable skepticism (meaning a lot of it), I remember this article from July:

AMD looks beyond TSMC as it worries about supply chain risks

TSMC has too much capacity when China has made no secret of its desire for Taiwan

www.theregister.com

I can think of a few realistic possibilities why AMD might do this:

1. Diversify fab use to avoid a future re-do of past TSMC supply constraints.
2. Respond to geopolitical concerns regarding Taiwan.
3. Samsung made them a deal they couldn't refuse.

 

[Daniel Nenni]

I can think of a few realistic possibilities why AMD might do this:
1. Diversify fab use to avoid a future re-do of past TSMC supply constraints.
2. Respond to geopolitical concerns regarding Taiwan.
3. Samsung made them a deal they couldn't refuse.

Reasons not to:
1. AMD would pay more for TSMC wafers due to reduced quantities.
2. AMD would no longer be a trusted partner for process development.
3. AMD would be required to have duplicate design teams since they cannot work on both TSMC and Samsung processes.
4. AMD would be required to have duplicate IP teams since they cannot work on both TSMC and Samsung processes.
5. AMD would have to use Samsung for packaging since TSMC will not package other foundry wafers.
6. It would upset TSMC.

 

[blueone]

Reasons not to:
1. AMD would pay more for TSMC wafers due to reduced quantities.

With chiplet-based designs isn't this more difficult to predict?

2. AMD would no longer be a trusted partner for process development.

Are they really one now? Aren't they already in a league below Apple?

3. AMD would be required to have duplicate design teams since they cannot work on both TSMC and Samsung processes.

Not at the logic design levels, only in circuit design and below.

4. AMD would be required to have duplicate IP teams since they cannot work on both TSMC and Samsung processes.

Intel used to use very limited external IP in their CPUs. I wonder what AMD's current status with IP is? I suspect almost everything is an internal design.

5. AMD would have to use Samsung for packaging since TSMC will not package other foundry wafers.

No doubt.

6. It would upset TSMC.

Perhaps, but I suspect AMD is still upset/concerned about past supply issues.

 

[nghanayem]

With chiplet-based designs isn't this more difficult to predict?

Are they really one now? Aren't they already in a league below Apple?

Not at the logic design levels, only in circuit design and below.

Intel used to use very limited external IP in their CPUs. I wonder what AMD's current status with IP is? I suspect almost everything is an internal design.

Really? I would assume almost everything is off the shelf. After all why waste AMD’s limited resources on duplicating efforts that synopsis, cadence, et al already made? It also helps they’ve been fabless for decades at this point.

No doubt.

Packaging doesn’t seem like that big of a deal given AMD uses OSAT provided organic for everything CPU. As long as GPU stays all TSMC I would have to assume TSMC will gladly package those still.

Perhaps, but I suspect AMD is still upset/concerned about past supply issues.

To me that is an AMD problem not a TSMC problem. They forecasted low and when it came time to get in line for more there was none left. If AMD was willing to put up the pre pays I don’t have any reason to doubt TSMC would have built out fab-18 faster.

As for another point against, there is nothing to me that suggests SF4P will have better performance than N4P-HPC. Heck with how far behind the performance of 4LPE was I wouldn’t be shocked if they would be downgrading from what their EPYC chiplets currently use (N5P-HPC). Seems really weird to me to downgrade your node.

 

[blueone]

Really? I would assume almost everything is off the shelf. After all why waste AMD’s limited resources on duplicating efforts that synopsis, cadence, et al already made? It also helps they’ve been fabless for decades at this point.

What external IP do you think is interesting for CPU chiplets, beyond, say, I/Os, like PCIe and DDR?

Packaging doesn’t seem like that big of a deal given AMD uses OSAT provided organic for everything CPU. As long as GPU stays all TSMC I would have to assume TSMC will gladly package those still.

I agree that AMD already knows how to solve the multi-source chiplet packaging problem. As The Register article noted, AMD previously used an I/O chiplet made by GF.

To me that is an AMD problem not a TSMC problem. They forecasted low and when it came time to get in line for more there was none left. If AMD was willing to put up the pre pays I don’t have any reason to doubt TSMC would have built out fab-18 faster.

I don't know enough about the causes of the supply chain issue to dive deeper.

As for another point against, there is nothing to me that suggests SF4P will have better performance than N4P-HPC. Heck with how far behind the performance of 4LPE was I wouldn’t be shocked if they would be downgrading from what their EPYC chiplets currently use (N5P-HPC). Seems really weird to me to downgrade your node.

If this AMD-Samsung rumor comes to fruition I'd have to assume AMD knows something we don't about Samsung processes.

 

[Fred Chen]

This should be enough reason for AMD not to use Samsung: https://www.tomshardware.com/news/samsung-developing-in-house-cpus

 

[blueone]

This should be enough reason for AMD not to use Samsung: https://www.tomshardware.com/news/samsung-developing-in-house-cpus

I confess that I've never understood why internal chip design teams in the foundry company are an IP theft vulnerability for foundry customers. How would the theft occur? Making some extra samples and sending them to the foundry-internal design team? How would the foundry's team know what to do with them? Even if they examined a raw die with an electron microscope can they really see circuit innovations at a high level?

 

[Daniel Nenni]

Samsung competes with AMD

Reasons not to:
1. AMD would pay more for TSMC wafers due to reduced quantities.
2. AMD would no longer be a trusted partner for process development.
3. AMD would be required to have duplicate design teams since they cannot work on both TSMC and Samsung processes.
4. AMD would be required to have duplicate IP teams since they cannot work on both TSMC and Samsung processes.
5. AMD would have to use Samsung for packaging since TSMC will not package other foundry wafers.
6. It would upset TSMC.

7. Samsung directly competes with AMD.

 

[blueone]

7. Samsung directly competes with AMD.

Okay, I get it, you and Fred are convinced that AMD will be negatively impacted by another Samsung business unit while/if using Samsung as a foundry. Can either of you explain to me how this negative impact might occur?

 

[Daniel Nenni]

I confess that I've never understood why internal chip design teams in the foundry company are an IP theft vulnerability for foundry customers. How would the theft occur? Making some extra samples and sending them to the foundry-internal design team? How would the foundry's team know what to do with them? Even if they examined a raw die with an electron microscope can they really see circuit innovations at a high level?

Patent infringement mostly. Apple sued Samsung for billions of dollars for IP theft from their foundry relationship.

Apple Inc. v. Samsung Electronics Co. - Wikipedia

en.wikipedia.org

 

[blueone]

Patent infringement mostly. Apple sued Samsung for billions of dollars for IP theft from their foundry relationship.

Apple Inc. v. Samsung Electronics Co. - Wikipedia

en.wikipedia.org

I remember these patent battles, but none were over chip design issues. These disputes were over design and utility concepts like the shape of the phone case and the GUI Samsung deployed into Android. I don't remember any of them being related to ASIC design. So you must be contending that Samsung would knowingly hobble Apple's foundry work because their own mobile phone business unit was having a legal dispute with Apple. All I can say is wow. If that's true, and that culture still pervades Samsung's top line staff, Samsung should be on a "short" list.

 

[Daniel Nenni]

I remember these patent battles, but none were over chip design issues. These disputes were over design and utility concepts like the shape of the phone case and the GUI Samsung deployed into Android. I don't remember any of them being related to ASIC design. So you must be contending that Samsung would knowingly hobble Apple's foundry work because their own mobile phone business unit was having a legal dispute with Apple. All I can say is wow. If that's true, and that culture still pervades Samsung's top line staff, Samsung should be on a "short" list.

I do remember design issues with Apple and other companies.

At the time I worked for an SRAM company and I was the foundry liaison. Samsung used to outright steal our designs but we did not have the strength to sue them. We pushed hard on technology so they were always a generation or two behind but they absolutely stole design IP. It is harder to prove especially to a jury of average Americans who didn't even know what a semiconductor was.

Samsung is a well known IP thief to us insiders. If AMD did choose to use Samsung it would be a decision they would regret, my opinion.

 

[nghanayem]

Is IP theft as relevant/risky if AMD is doing x86 stuff and Samsung is doing ARM stuff? Doubly so if most of their low level IP comes from the rest of the fabless ecosystem? I had assumed the main competitive risk was knowing roughly what AMD’s product is (ie core count, cache, and any other easily visible things) and more or less knowing the product's schedule.

 

[akumar]

News outlets have been spreading this "rumor" as fact. Personally I do not believe it, not even close. Does anyone know different? I will be at an industry event next week and will find out.

According to TechNews’ report, there are recent rumors indicating that AMD’s next-generation chip, with the Zen5C architecture codenamed “Prometheus,” will adopt a “Dual Foundry Mode.” This means it will simultaneously utilize TSMC’s 3nm and Samsung’s 4nm processes. This move suggests that AMD aims to diversify chip manufacturing, avoiding reliance solely on TSMC for its upcoming products.

[News] Samsung Reportedly Lands a 4nm Mega Order – Why is AMD Switching to “Dual Foundry Mode” for Its Next-Gen Chips? | TrendForce Insights

According to TechNews’ report, there are recent rumors indicating that AMD's next-generation chip, with the Zen5C architecture codenamed "Prometheus,"...

www.trendforce.com

Rumors swirl around AMD’s upcoming chip architecture, codenamed “Prometheus,” featuring the Zen 5C core. As reported by TechNews, the chip is poised to leverage both TSMC’s 3nm and Samsung’s 4nm processes simultaneously, marking a shift in the competitive landscape from process nodes, yield, and cost to factors like capacity, ecosystem, and geopolitics, are all depends on customer considerations.

[News] A battle on 4nm: AMD Teams Up with Samsung, while Google Weighs Supplier Split | TrendForce Insights

Rumors swirl around AMD's upcoming chip architecture, codenamed "Prometheus," featuring the Zen 5C core. As reported by TechNews, the chip is poised t...

www.trendforce.com

According to a report from Wccftech, senior executives at Samsung have indicated that major players such as super-scale data centers, automotive original equipment manufacturers (OEMs), and other clients have been in contact with Samsung, considering the adoption of Samsung’s foundry services to manufacture their designed chips.

[News] Samsung Reportedly Secures AMD and Tesla Orders for 4/5 nm Chips | TrendForce Insights

According to TechNews' report, during a recent financial conference, Samsung revealed its plans to diversify its sales structure by expanding its clie...

www.trendforce.com

I read an article suggesting Groq is switching from Globalfoundries 14nm to Samsung 4nm. "Thinking ahead to that next-generation GroqChip, Ross says it will have a 15X to 20X improvement in power efficiency that comes from moving from 14 nanometer GlobalFoundries to 4 nanometer Samsung manufacturing processes. " https://www.nextplatform.com/2023/1...oy-1-million-ai-inference-chips-in-two-years/

 

[Paul2]

AMD inventory situation is really strange for the last 2 years.

AMD based products are frequently out of stock in the West, and the lion's share of their output is going to Asia only brands, like different gaming notebook makers like Mechrevo, or Acer's Asian market only models.

I believe they expected that there will be enough high bin low voltage SKUs for mainstream Western brands, but for some reason they got much more low bin, which they dump to Asian market brands, which buy them a lot.

 

[Xebec]

The "leaks" on this have consistently been a mix of TSMC N4 and N3 for Zen 5 and 5c respectively. No mention of Samsung.

 

[Daniel Nenni]

There were also reports of QCOM going back to Samsung for Snapdragon 8 Gen 4 at 3nm but I can assure you that is not the case. I doubt TSMC will let QCOM go back to Samsung for the big running products. From what I have heard QCOM is back in the TSMC inner circle and that is a very big moat to cross.

New report contradicts previous rumor: Snapdragon 8 Gen 4 SoC to be built by TSMC only

A new report out of Taiwan says that TSMC will be producing the Snapdragon 8 Gen 4 chipset using its 3nm process node.

www.phonearena.com

 

[lefty]

It seems unlikely that Samsung they would steal AMDs IP, considering they already are licensing it from AMD (RDNA 2).
That said I think there were AMD using Samsung foundry rumours before and they always turned out false.

 

[cliff]

I do remember design issues with Apple and other companies.

At the time I worked for an SRAM company and I was the foundry liaison. Samsung used to outright steal our designs but we did not have the strength to sue them. We pushed hard on technology so they were always a generation or two behind but they absolutely stole design IP. It is harder to prove es pecially to a jury of average Americans who didn't even know what a semiconductor was.

Samsung is a well known IP thief to us insiders. If AMD did choose to use Samsung it would be a decision they would regret, my opinion.

This is shocking.

Do thefts mostly occur from sending the GDS2 to the mask shop? Can a layer or two be sent to another mask making shop?