Semiwiki on LinkedIn Semiwiki on Facebook Semiwiki on Twitter Register / Log in 
Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/pat-gelsinger-18a-update.18681/
        )

    [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
)

Pat Gelsinger 18A Update

Daniel Nenni

Daniel Nenni

Admin
Staff member
1693518575776.png


Intel Corporation (INTC) Deutsche Bank's 2023 Technology Conference

My favorite clips:

"And I would say, and I was just in Oregon yesterday with our technology development teams, looking good. We feel like we're solidly on track with 18A. We're getting the -- and this -- for foundry customers, hey, they need good solid PDKs. They need confidence that we can do this. The amount of interest that we're getting in this area is progressing very well for it. The technology itself is progressing very well.

And as we said, we believe that, that will be manufacturing ready end of next year, giving us a leadership position in '25. We're making good progress for our internal products like Clearwater Forest, our next-generation client products are all in the latter design phases, but also foundry customers.

And we've now received a large customer prepay for 18A capacity. So now customers are getting confident enough that they're putting dollars on our balance sheet to accelerate our 18A capacity. So quite excited about that.

So overall, everything is coming together, as we've said. And this customer prepay really is a strong exclamation point to momentum for 18A and the manufacturing capacity for that.

Yes. And we -- as we look at it, TSMC has established a market, right? Super clear. Remember, I'm a customer of TSMC. So I know exactly what their wafer costs are, what their wafer ASP is, they're presenting to their N5 customers, to their N3 customers, their budgetary for N2. We know what the target is, right, for that."

It is a good read. TSMC was mentioned a dozen times so they are very focused on delivering better value.

seekingalpha.com

Intel Corporation (INTC) Deutsche Bank's 2023 Technology Conference (Transcript)

Intel Corporation (NASDAQ:NASDAQ:INTC) Deutsche Bank's 2023 Technology Conference August 31, 2023 1:15 PM ETCompany ParticipantsPat Gelsinger -...
seekingalpha.com seekingalpha.com
 
By listening to what Pat was talking, it seems to me Intel product/design division (a TSMC customer) is sharing confidential TSMC information, such as pricing, with Intel Foundry Service (a TSMC competitor).

TSMC probably won't sue Intel for this. But in the long run it will hurt Intel greatly. The trust and secrecy kept between foundries and fabless companies is part of the reasons why TSMC, Apple, Qualcomm, Nvidia, AMD, Broadcom, and Mediatek are prosperous.

On the other hand, potential IFS customers will be very nervous about how much confidential information IFS will casually leak to Intel product division.
 
hist78 said:
By listening to what Pat was talking, it seems to me Intel product/design division (a TSMC customer) is sharing confidential TSMC information, such as pricing, with Intel Foundry Service (a TSMC competitor).

TSMC probably won't sue Intel for this. But in the long run it will hurt Intel greatly. The trust and secrecy kept between foundries and fabless companies is part of the reasons why TSMC, Apple, Qualcomm, Nvidia, AMD, Broadcom, and Mediatek are prosperous.

On the other hand, potential IFS customers will be very nervous about how much confidential information IFS will casually leak to Intel product division.
Click to expand...
It certainly doesn't sound like the sort of thing he should be saying in public. Should at least maintain the pretence that his 18A customer is telling him what the targets are.

I struggle to see IFS being a real solution to industry needs unless it becomes a separate company independent of Intel. Perhaps I'm wrong. Perhaps that's actually in the plan. Managing the conflicts of interest (as demonstrated here) looks too difficult.
 
I doubt that Intel knows what TSMC's wafer costs are. Why would TSMC disclose that to anyone? TSMC's wafer prices to Intel are all they know. I can't believe Intel knows anything, for example, about what TSMC is charging Apple.
 
hist78 said:
By listening to what Pat was talking, it seems to me Intel product/design division (a TSMC customer) is sharing confidential TSMC information, such as pricing, with Intel Foundry Service (a TSMC competitor).

TSMC probably won't sue Intel for this. But in the long run it will hurt Intel greatly. The trust and secrecy kept between foundries and fabless companies is part of the reasons why TSMC, Apple, Qualcomm, Nvidia, AMD, Broadcom, and Mediatek are prosperous.

On the other hand, potential IFS customers will be very nervous about how much confidential information IFS will casually leak to Intel product division.
Click to expand...

and honestly, I don't think TSMC will ask the same as other fabless clients' wafer pricing. And I think TSMC has been or will charge Intel a higher price than others. PG may know that. But he's more passionate about having more leading edge fab than he has on hand
 
blueone said:
I doubt that Intel knows what TSMC's wafer costs are. Why would TSMC disclose that to anyone? TSMC's wafer prices to Intel are all they know. I can't believe Intel knows anything, for example, about what TSMC is charging Apple.
Click to expand...
I have to assume he meant the latter because there is simply no way to know the former unless you are TSMC. Not even Apple should be privy to wafer cost. Maybe he means intel can get a ballpark guess based on what their PPA/arch targets are, what the price is, and where TSMC's margins are? However for nodes with dissect-able parts on the market like N5, knowing the wafer costs should be a trivial reverse engineering exercise.
 
nghanayem said:
However for nodes with dissect-able parts on the market like N5, knowing the wafer costs should be a trivial reverse engineering exercise.
Click to expand...
My guess is that TSMC has the lowest wafer costs in the industry, and if that's true there are no comparable figures. You can only guess in a range.
 
Last edited:
TMSC has reasonable grounds to pull the rug on Intel and sue for breach of confidentiality based on Pat's statements.
 
count said:
TMSC has reasonable grounds to pull the rug on Intel and sue for breach of confidentiality based on Pat's statements.
Click to expand...
I'm reminded of what they used to call the "newspaper test" at TI - i.e. how would you feel if what you did were reported in a newspaper (which effectively this now is - he published it himself). It's not only whether something is strictly legal (or not explicitly illegal) that should bound our behaviour. Trust is one of these less tangible areas.
 
tooLongInEDA said:
I'm reminded of what they used to call the "newspaper test" at TI - i.e. how would you feel if what you did were reported in a newspaper (which effectively this now is - he published it himself). It's not only whether something is strictly legal (or not explicitly illegal) that should bound our behaviour. Trust is one of these less tangible areas.
Click to expand...
About 20 years ago or so, I got to personally experience the soundness of this advice in action. And that's all I have to say about it. :)
 
blueone said:
My guess is that TSMC has the lowest wafer costs in the industry, and if that's true there are no comparable figures.
Click to expand...
I don't think that is necessarily true. For example I would have to assume that 4LPE is cheaper to produce then N5 family. Less EUV MP, no SiGe PMOS, etc. Granted N5 family is a far superior node to 4LPE, so a cost and complexity delta is to be expected. Regardless, I think it gets the point across that process complexity/integration decisions should be the largest cost driver. If we want to get into more subjective measures; I think Samsung has some extra advantages over TSMC from a cost perspective. It would seem the ROK purse strings are more open to reimbursing semi makers than the ROC which to my knowledge has walked back most everything besides the investment tax credits. Another is labor seeming to be a bit cheaper in SK than TW (not that labor prices are a huge deal). Finally I think the greatest asset Samsung Foundry has is being able to leverage the floor space and a good amount of the equipment from their memory lines. I would have to think the amount of scale (both on the leading edge and just in general) that comes from being able to share capital with the largest memory IDM acts as a large cost saver for their logic tech. Given the commanding lead that TSMC has in logic (as well as the premiums that result from said leadership), and Samsung's culture as a memory maker; I wouldn't be surprised if Samsung is that little bit more cost conscious/efficient than TSMC when it comes to building/operating fabs. But as I previously mentioned these are all intangibles, and as a result it is much harder to measure their real world impact then simpler cost deltas like process complexity.

You can only guess in a range.
Click to expand...
Given how strongly process complexity and the resulting tool cost impact wafer cost, I would have to assume that if you know the flow and know how the various vendor offerings perform you can get pretty darn accurate with your guess. But for processes without a definitively known flow the error bars are for sure much wider. It is kind of like chopped in a sense. Same ingredients and the same kitchen but different food coming out the other end. With that said, I feel like BEOL is probably pretty easy to get relatively a narrow range.
 
Last edited:
Yes. And we -- as we look at it, TSMC has established a market, right? Super clear. Remember, I'm a customer of TSMC. So I know exactly what their wafer costs are, what their wafer ASP is, they're presenting to their N5 customers, to their N3 customers, their budgetary for N2. We know what the target is, right, for that."
Click to expand...

I can assure you TSMC would not reveal these types of cost numbers directly to Intel. The information Pat mentioned however is readily available through third parties. Scotten Jones does this type of cost modeling for a living (now at TechInsights). Scotten and I have compared costing numbers many times and I have found his numbers to be quite accurate. I would bet Intel got them from Scotten.

But, as I mentioned before, it is unwise to wake a sleeping giant (TSMC) and this type of speak does just that. Morris Chang may have given Intel a pass for this type of press but CC Wei will not, absolutely.
 
You must log in or register to reply here.
Back
Top