Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel%E2%80%99s-3-5-billion-boondoggle.19811/
        )

    [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
)

Intel’s $3.5 Billion Boondoggle

Daniel Nenni

Admin
Staff member
305yj347064eJ9xgnnnl111GN0CusM1x.webp

Intel has only recently moved to expand its domestic fabrication operations to a globally competitive scale, largely in response to expected CHIPS Act funding.

With Congress hurtling toward an end-of-week deadline to avert a partial government shutdown, lawmakers are racing to pass appropriations to keep the government open. The House passed its “minibus” legislation containing half of fiscal year 2024’s appropriations Wednesday, and the Senate is set to follow suit today. Another minibus is expected by March 22, the second partial shutdown deadline.

The rushed appropriations package, which was only released last Sunday night, has led to a familiar scenario on Capitol Hill: lawmakers rushing to review and vote on hundreds of pages of bill text riddled with obscure and often unexpected provisions. The scramble has created an opportunity for committee heads and appropriators to stash a number of controversial policies in the legislation. One such provision would slash a fifth of the funding for the Department of Justice’s Antitrust Division as the agency prepares to take on several major cases in the coming year.

But the cut to antitrust enforcement is not the only threat the appropriations package poses to President Biden’s pro-competition agenda. The package also includes $3.5 billion over three years for microchip giant Intel to create a “secure enclave” facility that would exclusively produce highly sensitive microchips for the U.S. military. Instead of providing new funding for the facility, the bill pulls the money from the $39 billion allocated for manufacturing grants by the 2022 CHIPS and Science Act (commonly referred to as the CHIPS Act). Experts and congressional sources say the move threatens to further strain a fund that is already much too small to meet the needs of U.S. semiconductor manufacturers.

“My first reaction is to emphasize that this is nearly 10 percent of the entire CHIPS Act fund,” Charles Wessner, a senior adviser at the Center for Strategic and International Studies and research professor at Georgetown University. “And by doing this, that will necessarily foreclose other opportunities.”

While the United States has several leading chip design companies—Intel first and foremost—the CHIPS Act’s primary purpose was to provide funding to onshore the actual fabrication of microchips, a part of the industry that is currently dominated by Taiwan’s TSMC and South Korea’s Samsung. (Both companies are angling for their own slice of the grant funds to expand their operations in the United States; TSMC is building a fab in Arizona.) Intel has only recently moved to expand its domestic fabrication operations to a globally competitive scale, largely in response to expected CHIPS Act funding.

Instead of providing new funding for the facility, the bill pulls the money from the $39 billion allocated for manufacturing grants by the 2022 CHIPS and Science Act.

The Department of Commerce, which oversees the funds, has begun slowly rolling out its initial awards to domestic chip manufacturers, naming BAE Systems, Microchip Technology, and GlobalFoundries as its first three confirmed awardees. Despite reports indicating Intel is set to receive the lion’s share of the available funding with a package worth over $10 billion, the corporation is reportedly still in tough negotiations with the Department of Commerce for more. In a sign the company is playing hardball, it indicated last month that it plans to slow its expansion of its Ohio manufacturing plant, citing dissatisfaction with the rollout of CHIPS funds as part of its reasoning. If Intel’s package expands any further, it is unlikely the department will be able to adequately fund several other promising proposals. Commerce Secretary Gina Raimondo has indicated that the funding requests the department has received are for twice the available funds—with many more expected to continue rolling in.

The secure enclave earmark for Intel in the appropriations package drains that available CHIPS funding even further. And it is far from clear that the facility is a worthwhile investment. Two congressional staffers, who were granted anonymity to speak openly about the jockeying over the funds that the Department of Defense—the sole customer for the chips Intel is set to produce, aside from a tiny amount meant for intelligence agencies—has not requested the creation of a secure enclave facility. In fact, there is reason to believe the Pentagon actively opposes the plan.

One of those congressional aides instead described the funding as a solution in search of a problem. “We have systems for getting the military the sensitive chips it needs,” they said, referring to an accreditation process already in place within the Department of Defense and emerging innovations that make it possible to assess the integrity of a chip at various stages of the production life cycle. “We could be investing more in modernizing technologies we already have to make it easier and more secure for any supplier to build these chips.”

The aide claims the push for a secure enclave facility began in 2021 with a report from U.S. intelligence agencies that described an unknown risk requiring heightened security at the fabrication stage—the stage of chip production the secure enclave method is primarily designed to protect. While the intelligence committees in both chambers are said to have communicated the report’s conclusion to other relevant committees, the justification for its findings has not been widely dispersed, even to members of the House and Senate Armed Services Committees, many of whom only became aware of the secure enclave project provisions mere hours before the appropriations package text was released this weekend.

Representatives of both chambers’ Intelligence Committee chairs, Rep. Mike Turner (R-OH) and Sen. Mark Warner (D-VA), did not respond to a request to comment for this article. (Conspicuously, the top two Republicans on the House Intelligence Committee, Turner and Rep. Brad Wenstrup, both hail from Ohio, where Intel has recently slowed its expansion plans.)

Further complicating matters, the alleged conclusion of the mysterious intelligence report was contradicted by a report on current and emerging chip security models from the Air Force last August, which claimed that fabrication is the stage in the production process where microchips face the lowest security risk. (The secure enclave will potentially bring testing and packaging in-house as well, production stages that are deemed higher-risk.) The report goes on to specifically say “we need to invest less [in fabrication protections],” and claims that the economics of the highly sensitive chip market—which accounts for an estimated 2 percent of the total U.S. semiconductor marketplace—make it so the Department of Defense “cannot maintain dedicated facilities.” It also emphasizes that it is in the department’s “interest to have access to multiple sources of microelectronics components for resiliency and cost competitiveness.”

The congressional aide who highlighted the conflicting reports indicated that, by siphoning such a large portion of the funds meant to boost domestic manufacturing capacity, the mandate to fund the secure enclave project stands to make the goal of diversifying the supply chain for sensitive chips much more difficult. The hefty price tag to maintain the facility over time is also likely to create an incentive structure for the Department of Defense to source as many chips as possible from the facility, so as to justify the expense. The aide suggested the ongoing dispute may have been a factor in why CHIPS grant announcements have been slower to materialize than many observers anticipated.

Siphoning such a large portion of funds meant to boost domestic manufacturing capacity makes diversifying the supply chain for sensitive chips much more difficult.

It is unclear which of the conflicting risk assessments the Department of Commerce finds more credible. The funding for the secure enclave project was set to be approved through Commerce’s grant disbursement process, which would have made specific direction from Congress unnecessary. The report also noted that Sens. Maria Cantwell (D-WA), Roger Wicker (R-MS), and Jack Reed (D-RI) sent a letter to Commerce questioning the Intel facility, on grounds similar to those laid out in the Department of the Air Force’s report. The sudden move by appropriators to mandate funding for the project indicates that Commerce, which has been consulting with the Pentagon in its distribution of grant funds, may have had second thoughts.

It is also unclear if or how the secure enclave funding might impact the grants Commerce has already announced but not finalized for Intel’s competitors BAE Systems and GlobalFoundries, both of which are key parts of the defense microchip supply chain. The Department of Commerce, which did not immediately respond to a request to comment for this article, said it is“still reviewing the effect of the appropriations text on the program.”

The second minibus package, which includes appropriations for defense and like the first is currently taking shape behind closed doors, presents an obvious opportunity for Congress to undo the mandate and return the decision over how to allocate grant money to the Department of Commerce. At the very least, the package could allocate new funds to support the project, or redirect money from other parts of the Pentagon’s hefty budget to backfill the hole it leaves in the CHIPS Act manufacturing grant pot. But if legislators truly want to advance their goals of building out domestic microchip manufacturing capacity, they need to go a step further and expand, not commandeer, the fund.

“Congress is going to learn very quickly that $39 billion isn’t a lot of money in the microchip world,” Wessner said, pointing out that China has spent many multiples of that amount in the last decade. “[The $39 billion] can’t be a one and done. We have to have a regular flow of resources, or we can’t compete. Period.

 
305yj347064eJ9xgnnnl111GN0CusM1x.webp

Intel has only recently moved to expand its domestic fabrication operations to a globally competitive scale, largely in response to expected CHIPS Act funding.

With Congress hurtling toward an end-of-week deadline to avert a partial government shutdown, lawmakers are racing to pass appropriations to keep the government open. The House passed its “minibus” legislation containing half of fiscal year 2024’s appropriations Wednesday, and the Senate is set to follow suit today. Another minibus is expected by March 22, the second partial shutdown deadline.

The rushed appropriations package, which was only released last Sunday night, has led to a familiar scenario on Capitol Hill: lawmakers rushing to review and vote on hundreds of pages of bill text riddled with obscure and often unexpected provisions. The scramble has created an opportunity for committee heads and appropriators to stash a number of controversial policies in the legislation. One such provision would slash a fifth of the funding for the Department of Justice’s Antitrust Division as the agency prepares to take on several major cases in the coming year.

But the cut to antitrust enforcement is not the only threat the appropriations package poses to President Biden’s pro-competition agenda. The package also includes $3.5 billion over three years for microchip giant Intel to create a “secure enclave” facility that would exclusively produce highly sensitive microchips for the U.S. military. Instead of providing new funding for the facility, the bill pulls the money from the $39 billion allocated for manufacturing grants by the 2022 CHIPS and Science Act (commonly referred to as the CHIPS Act). Experts and congressional sources say the move threatens to further strain a fund that is already much too small to meet the needs of U.S. semiconductor manufacturers.

“My first reaction is to emphasize that this is nearly 10 percent of the entire CHIPS Act fund,” Charles Wessner, a senior adviser at the Center for Strategic and International Studies and research professor at Georgetown University. “And by doing this, that will necessarily foreclose other opportunities.”

While the United States has several leading chip design companies—Intel first and foremost—the CHIPS Act’s primary purpose was to provide funding to onshore the actual fabrication of microchips, a part of the industry that is currently dominated by Taiwan’s TSMC and South Korea’s Samsung. (Both companies are angling for their own slice of the grant funds to expand their operations in the United States; TSMC is building a fab in Arizona.) Intel has only recently moved to expand its domestic fabrication operations to a globally competitive scale, largely in response to expected CHIPS Act funding.

Instead of providing new funding for the facility, the bill pulls the money from the $39 billion allocated for manufacturing grants by the 2022 CHIPS and Science Act.

The Department of Commerce, which oversees the funds, has begun slowly rolling out its initial awards to domestic chip manufacturers, naming BAE Systems, Microchip Technology, and GlobalFoundries as its first three confirmed awardees. Despite reports indicating Intel is set to receive the lion’s share of the available funding with a package worth over $10 billion, the corporation is reportedly still in tough negotiations with the Department of Commerce for more. In a sign the company is playing hardball, it indicated last month that it plans to slow its expansion of its Ohio manufacturing plant, citing dissatisfaction with the rollout of CHIPS funds as part of its reasoning. If Intel’s package expands any further, it is unlikely the department will be able to adequately fund several other promising proposals. Commerce Secretary Gina Raimondo has indicated that the funding requests the department has received are for twice the available funds—with many more expected to continue rolling in.

The secure enclave earmark for Intel in the appropriations package drains that available CHIPS funding even further. And it is far from clear that the facility is a worthwhile investment. Two congressional staffers, who were granted anonymity to speak openly about the jockeying over the funds that the Department of Defense—the sole customer for the chips Intel is set to produce, aside from a tiny amount meant for intelligence agencies—has not requested the creation of a secure enclave facility. In fact, there is reason to believe the Pentagon actively opposes the plan.

One of those congressional aides instead described the funding as a solution in search of a problem. “We have systems for getting the military the sensitive chips it needs,” they said, referring to an accreditation process already in place within the Department of Defense and emerging innovations that make it possible to assess the integrity of a chip at various stages of the production life cycle. “We could be investing more in modernizing technologies we already have to make it easier and more secure for any supplier to build these chips.”

The aide claims the push for a secure enclave facility began in 2021 with a report from U.S. intelligence agencies that described an unknown risk requiring heightened security at the fabrication stage—the stage of chip production the secure enclave method is primarily designed to protect. While the intelligence committees in both chambers are said to have communicated the report’s conclusion to other relevant committees, the justification for its findings has not been widely dispersed, even to members of the House and Senate Armed Services Committees, many of whom only became aware of the secure enclave project provisions mere hours before the appropriations package text was released this weekend.

Representatives of both chambers’ Intelligence Committee chairs, Rep. Mike Turner (R-OH) and Sen. Mark Warner (D-VA), did not respond to a request to comment for this article. (Conspicuously, the top two Republicans on the House Intelligence Committee, Turner and Rep. Brad Wenstrup, both hail from Ohio, where Intel has recently slowed its expansion plans.)

Further complicating matters, the alleged conclusion of the mysterious intelligence report was contradicted by a report on current and emerging chip security models from the Air Force last August, which claimed that fabrication is the stage in the production process where microchips face the lowest security risk. (The secure enclave will potentially bring testing and packaging in-house as well, production stages that are deemed higher-risk.) The report goes on to specifically say “we need to invest less [in fabrication protections],” and claims that the economics of the highly sensitive chip market—which accounts for an estimated 2 percent of the total U.S. semiconductor marketplace—make it so the Department of Defense “cannot maintain dedicated facilities.” It also emphasizes that it is in the department’s “interest to have access to multiple sources of microelectronics components for resiliency and cost competitiveness.”

The congressional aide who highlighted the conflicting reports indicated that, by siphoning such a large portion of the funds meant to boost domestic manufacturing capacity, the mandate to fund the secure enclave project stands to make the goal of diversifying the supply chain for sensitive chips much more difficult. The hefty price tag to maintain the facility over time is also likely to create an incentive structure for the Department of Defense to source as many chips as possible from the facility, so as to justify the expense. The aide suggested the ongoing dispute may have been a factor in why CHIPS grant announcements have been slower to materialize than many observers anticipated.

Siphoning such a large portion of funds meant to boost domestic manufacturing capacity makes diversifying the supply chain for sensitive chips much more difficult.

It is unclear which of the conflicting risk assessments the Department of Commerce finds more credible. The funding for the secure enclave project was set to be approved through Commerce’s grant disbursement process, which would have made specific direction from Congress unnecessary. The report also noted that Sens. Maria Cantwell (D-WA), Roger Wicker (R-MS), and Jack Reed (D-RI) sent a letter to Commerce questioning the Intel facility, on grounds similar to those laid out in the Department of the Air Force’s report. The sudden move by appropriators to mandate funding for the project indicates that Commerce, which has been consulting with the Pentagon in its distribution of grant funds, may have had second thoughts.

It is also unclear if or how the secure enclave funding might impact the grants Commerce has already announced but not finalized for Intel’s competitors BAE Systems and GlobalFoundries, both of which are key parts of the defense microchip supply chain. The Department of Commerce, which did not immediately respond to a request to comment for this article, said it is“still reviewing the effect of the appropriations text on the program.”

The second minibus package, which includes appropriations for defense and like the first is currently taking shape behind closed doors, presents an obvious opportunity for Congress to undo the mandate and return the decision over how to allocate grant money to the Department of Commerce. At the very least, the package could allocate new funds to support the project, or redirect money from other parts of the Pentagon’s hefty budget to backfill the hole it leaves in the CHIPS Act manufacturing grant pot. But if legislators truly want to advance their goals of building out domestic microchip manufacturing capacity, they need to go a step further and expand, not commandeer, the fund.

“Congress is going to learn very quickly that $39 billion isn’t a lot of money in the microchip world,” Wessner said, pointing out that China has spent many multiples of that amount in the last decade. “[The $39 billion] can’t be a one and done. We have to have a regular flow of resources, or we can’t compete. Period.


"The report goes on to specifically say “we need to invest less [in fabrication protections],” and claims that the economics of the highly sensitive chip market—which accounts for an estimated 2 percent of the total U.S. semiconductor marketplace—make it so the Department of Defense “cannot maintain dedicated facilities.” It also emphasizes that it is in the department’s “interest to have access to multiple sources of microelectronics components for resiliency and cost competitiveness.”"

I guess:

1. It means Intel wants the US government to pay a lot (much much more than Intel's competitors are going to receive from the Chips Act) to build a fab in Ohio.

Also Intel demands DoD to commit to buy as much as possible and as long as possible from that fab (implicitly or explicitly).

2. DoD is resisting such idea. They believe supply chain resilience and supply chain diversity cannot be achieved by relying on a single company, such as Intel.

3. As I mentioned in another thread, DoD, DoE, and several other three-letter federal agencies are more practical than people think.
 
I think the loop the DoD is stuck in is that there’s only one American company making bleeding edge. (..but) They want resiliency / a second fab source (located in the US presumably) but ‘foreign ownership’ that will have access to their IP/tech for fabrication is a show stopper for certain products.

(There’s a reason companies like Lockheed Martin get the lions share of DoD contracts.)

EDIT: I suspect the real requirements for DoD "advanced chips" go beyond the usual TAA Requirements: https://www.gsascheduleservices.com/what-is-taa-compliance/
 
Last edited:
What does this mean?

Intel has only recently moved to expand its domestic fabrication operations to a globally competitive scale, largely in response to expected CHIPS Act funding.

Assuming Intel were to build-out the Ohio facility, they estimate it would cost $100B. No matter what Intel gets from the wimpy CHIPS Act, it is difficult to see that Intel's Ohio plan is "largely in response to expected CHIPS Act funding".

I also don't believe that "congressional staffers" are going to be having off the record discussions with the HuNan Printed Circuit Association.

Two congressional staffers, who were granted anonymity to speak openly about the jockeying over the funds that the Department of Defense—the sole customer for the chips Intel is set to produce, aside from a tiny amount meant for intelligence agencies—has not requested the creation of a secure enclave facility. In fact, there is reason to believe the Pentagon actively opposes the plan.

Or am I misinterpreting that section?
 
3.5B is not really that much in the Intel Budget, congressional budget, overall semi market. and it is actually less than most people were thinking intel would get

Side note: the Chips act budget includes 200B in "unclear spending" so 3.5B is not that much even in the chips act. If you want a US company with US fab and leading edge, you have no choice.

and no, Intel will not spend 100B on Ohio. those are just fake PR numbers.
 
3.5B is not really that much in the Intel Budget, congressional budget, overall semi market. and it is actually less than most people were thinking intel would get
Side note: the Chips act budget includes 200B in "unclear spending" so 3.5B is not that much even in the chips act. If you want a US company with US fab and leading edge, you have no choice.
and no, Intel will not spend 100B on Ohio. those are just fake PR numbers.

I think Intel will be on TSMC N3 for several years. It is easy to design to and there is a ton of TSMC design experience in India at Intel and many other firms. I think TSMC N3 will go down in history as one of the biggest nodes. And with chiplets, N3 is plenty fast enough for supporting chiplets and much lower power than anything Intel will come up with. Not to mention economies of scale. There will be a lot of N3 capacity available after the leading edge customer move to N2 and N1 over the next 5 years.

Will Intel need a cluster of fabs in Ohio when TSMC can do the supporting N3 chiplets for cheaper? No.
 
Side note: the Chips act budget includes 200B in "unclear spending" so 3.5B is not that much even in the chips act. If you want a US company with US fab and leading edge, you have no choice.
The $200B is for "scientific R&D and commercialization", specifically not for manufacturing.
and no, Intel will not spend 100B on Ohio. those are just fake PR numbers.
If the geo-political situation as-is remains the status quo, I don't believe Intel will build out Ohio either. If things change so will the Ohio plan, and I suspect there will be a new CHIPS Act-like thing, hopefully without the silly research spending items. I'm not betting either way.
 
If the geo-political situation as-is remains the status quo, I don't believe Intel will build out Ohio either. If things change so will the Ohio plan, and I suspect there will be a new CHIPS Act-like thing, hopefully without the silly research spending items. I'm not betting either way.
and no, Intel will not spend 100B on Ohio. those are just fake PR numbers.
What makes you guys say that? All you have to do is look at Oregon, NM, AZ, and IR on a google earth to see they are running out of space to put new shells. As a result I have a hard time believing that the Ohio site will never have the extra 6 fab slots filled out. Even if intel "only" extends the volume tale of their nodes by a decade, then intel still needs to build out the extra mods in OH and Germany to have more than just D1 for post 14A HVM.

Intel in 2022 had said that they would build out the site over the next decade. If we assume that this is still the schedule they are planning and that they hit their 1 new base process every 2 years after 18A. Then Intel will have 14A, 10A, and 8A by 2033. If you can no longer convert old fabs for this nodes, then you have no choice but to fill out OH/Germany or make a more expensive greenfield site. To me the best choice is obvious unless someone cuts intel a huge check to make a greenfield site elsewhere.
 
Remember what David Zinser Intel CFO said:

"I think probably, we are a little bit heavier than we want to be in terms of external wafer manufacturing versus internal, but we're always going to use external foundries for wafers."

Chiplets makes 2nd sourcing so much easier.... Intel has used TSMC for many years. The last number I remember, before Pat Gelsinger, was 25% of wafers were from TSMC. Now it is more than 50% and growing with TSMC N3.

What is Intel's planned revenue growth for the next 3 years?
 
What makes you guys say that? All you have to do is look at Oregon, NM, AZ, and IR on a google earth to see they are running out of space to put new shells. As a result I have a hard time believing that the Ohio site will never have the extra 6 fab slots filled out. Even if intel "only" extends the volume tale of their nodes by a decade, then intel still needs to build out the extra mods in OH and Germany to have more than just D1 for post 14A HVM.
Desktop and laptop CPU sales and some client I/O chips (like for wireless), which are the major drivers of Intel fab wafer volumes, are the only markets likely to be dominated by merchant chips (meaning chips sold by design companies like Intel, AMD, and Qualcomm) to system OEMs like Dell, HP, Lenovo, etc. Desktop and laptop volumes are not growing exponentially, since they are mostly mature markets. Phones are most popular client choices in developing countries, and no one is predicting that will change much for years.

Phones and tablets are entirely OEM-designed chips and fab'd at foundries.

Any chips sold into data centers are much lower volume by comparison, and in the future increasingly likely be designed in-house by big cloud companies and some system OEMs and fabricated at foundries. Also, the cloud companies have so much volume clout they can negotiate CPU deals with Intel, AMD, Ampere, etc. where pricing looks more like client CPU pricing than enterprise OEM pricing. This will likely weaken the data center R&D investment strategy for Intel and AMD. Other types of chips, like data center networking NICs and switches, are expensive, but their volumes are an order of magnitude or more less than client chips. FPGAs are like data center networking chips; expensive but low volume. Nvidia GPUs are currently an exception, but lots of AI merchant chips and in-house designs are popping up, though their future is far from certain. IFS could get some of this AI chip fab chip business. I can't believe Nvidia will do anything with IFS except test chips for the foreseeable future. Anything memory probably doesn't count for this discussion.

So it looks to me like Intel's wafer volume future is probably IFS-driven, but IFS growth is unlikely to be huge soon. I suppose there's a bump to be had from bringing internal CPU designs to IFS, but if IFS capacity is limited for the next several years, I'd guess Intel will prioritize IFS relationships over in-house designs, because it's easier to bring in-house designs home when IFS capacity increases than it is to cultivate an IFS customer base. As least that's what I'd do.

Also, if Ohio fabs are mostly EUV-based facilities, it would seem to take a long time to buy enough litho machines from ASML to need that many new fab shells in the foreseeable future. And seeing anything beyond seven years is purely guessing.

So I'm not seeing any market-based evidence that Intel will be building out the Ohio site for a long time.

Is my reasoning flawed?
 
Is my reasoning flawed?
I think your logic tracks. I have stated before that disaggregated chip designs are definitely a big volume loss for any foundry's the most advanced node. I think intel's capacity long range plan, supports my thoughts on the matter.
1710266618311.png


As weird as it sounds the "problem" is in some ways IFS. For the purposes of Ohio I don't think it matters if the wafer demands on intel's most advanced process at any given time is lower than it would have historically been. Take 18A as an example. Intel has said that 18A will be made at D1, Fab 42-62, and has in the past said that 18A would also happen in Ohio (albeit they never said if both Ohio mods or just mod 1 will make 18A). But let's say worst case 18A long term looks like Fab42-62 and Ohio-1. Looking at the images for Fab42/D1X style mods they are smaller than a Fab52/62 mod. So conservatively, let us call it 3.5 18A fabs at peak ramp. That is smaller than intel 7 peak ramp of D1, Fabs 12-42, and Fabs 18/28. But the only "new" fabs built to do that were D1X_mod 1/2 and Fab 42. Those other fabs were over a decade old when 10nm ramped

Going back to the future, let us then say that Fab29/39 are both 14A and that Ohio 2 is also 14A. At this point the empty shells at Germany and Ohio are full and 14A would have significantly lower capacity than 18A (3 fabs with a presumably longer process flow vs 3.5 fabs of 18A). What if Ohio 2 is also 18A? Now you need to build 2-3 shells just to match 18A capacity. And that is just for 14A! 10A and 8A will need even more shells within this one decade timeframe. So unless internal+IFS demands progressively decrease for all post 18A process families, then intel will need to eventually fill out the Ohio and Germany sites.
 
What makes you guys say that? All you have to do is look at Oregon, NM, AZ, and IR on a google earth to see they are running out of space to put new shells. As a result I have a hard time believing that the Ohio site will never have the extra 6 fab slots filled out. Even if intel "only" extends the volume tale of their nodes by a decade, then intel still needs to build out the extra mods in OH and Germany to have more than just D1 for post 14A HVM.

Intel in 2022 had said that they would build out the site over the next decade. If we assume that this is still the schedule they are planning and that they hit their 1 new base process every 2 years after 18A. Then Intel will have 14A, 10A, and 8A by 2033. If you can no longer convert old fabs for this nodes, then you have no choice but to fill out OH/Germany or make a more expensive greenfield site. To me the best choice is obvious unless someone cuts intel a huge check to make a greenfield site elsewhere.
couple items: Intel has no plans to spend 100B in Ohio. the PLAN is 2 fabs. Intel currently is not fully loaded and is building F-52 which is not filled with orders yet. Fab 62 and Israel and Ireland expansions still to come. Intel needs to about double its current orders internally and externally to fill Arizona, Ireland, Israel plans. etc etc.
plan for Ohio is to build a shell to start with... same with Germany

I would see if Intel ramps Intel3, and 18A before we worry about 14A and beyond. Intel will be running <10,000 production wafers per month Intel 4 and 3 at end of 24. that is well under half a Fab. So Intel has tons of room to grow. You can also look at Capex and cash flow finances to see that Intel cannot afford 100B in Ohio.

Lets get Fab 52 running wafers in 2025 before we worry about Intel running out of capacity.

We will have capacity by Fab report coming soon. we can discuss more then.
 
So I'm not seeing any market-based evidence that Intel will be building out the Ohio site for a long time.

Is my reasoning flawed?
I think you are spot on. I think Keyvan's presentation supports you on that.

Again: i would suggest we get 18A wafers out of F52 (2025) before we worry about mythical 8A processes... just an opinion
 
I think you are spot on. I think Keyvan's presentation supports you on that.

Again: i would suggest we get 18A wafers out of F52 (2025) before we worry about mythical 8A processes... just an opinion

For a while I thought Intel had a "build fabs and customers will come" strategy versus the TSMC "customers pay upfront" business model. Probably somewhere in the middle? Either way we will not be lacking in fab capacity anytime soon.

Jensen will keynote next week at the Synopsys User Group Meeting. It will probably be a preview of GTC. Does Jensen see AI as a bubble? Like Crypto?
 
Going back to the future, let us then say that Fab29/39 are both 14A and that Ohio 2 is also 14A. At this point the empty shells at Germany and Ohio are full and 14A would have significantly lower capacity than 18A (3 fabs with a presumably longer process flow vs 3.5 fabs of 18A). What if Ohio 2 is also 18A? Now you need to build 2-3 shells just to match 18A capacity. And that is just for 14A! 10A and 8A will need even more shells within this one decade timeframe. So unless internal+IFS demands progressively decrease for all post 18A process families, then intel will need to eventually fill out the Ohio and Germany sites.
Very interesting logic. I'm not knowledgable enough to judge it. If true, this means 14A generation wafers will have to generate 150% of the revenue, very roughly, that 18A wafers will, assuming fabs costs are similar. What do you think will cause the longer process flow of 14A versus 18A, and why?
 
couple items: Intel has no plans to spend 100B in Ohio. the PLAN is 2 fabs.
1710278852525.png


1710278929026.png


"The scope and pace of Intel’s expansion in Ohio, however, will depend heavily on funding from the CHIPS Act.”.

Even if we ignore everything intel has said, it makes no financial sense to build a greenfield site that has no room for expansion. When RA, OTF, or KG were bought intel didn't only build D1B/C, F22, and F18. They opened those sites when their old sites nearby filled up and they bought properties that would last for many expansions to support many nodes. In an IDM 2.0 era intel now needs to build fabs more frequently than they used to because the old fabs will be too busy doing trailing edge chips for internal and external customers. In fact I would go a step further. If F52/62 is supposed to be 20B, I would bet that when Ohio is eventually fully built out Ohio will be way more than $100B given the increasing tool costs and intel's RISE goals.

Intel currently is not fully loaded and is building F-52 which is not filled with orders yet. Fab 62
They did say they were getting a pre-pay specifically because they wanted the customer to have skin in the game before they accelerated the AZ ramp for them. So presumably F52 will at least be full.
and Israel and Ireland expansions still to come. Intel needs to about double its current orders internally and externally to fill Arizona, Ireland, Israel plans. etc etc.
plan for Ohio is to build a shell to start with... same with Germany
Intel 3 looks to have a similar or larger peak ramp than 18A. Why would intel expand intel 3 capacity beyond F34/44 if they had no customers to fill it?
What do you think will cause the longer process flow of 14A versus 18A, and why?
That isn't a 14A specific thing. This is universally true. N7 has more process steps than 16/12FF, which had more steps than 20nm, and 20nm had more steps than 28nm. intel 32nm has more steps than 45nm which itself has more steps than 65nm. Smaller devices require more process steps to form. New architectural features also need more steps to form. As an example of say 22nm vs 180nm, you now have strain, Cu, interconnects, HKMG, and finFET. Even if pitches were unchanged you now add the RMG loop, EPI loop, damascene interconnects, fin loop, etc. The only exceptions to this I can think of are TSMC N5 and intel 4. In those cases they were going from lots of multipatterning to doing the hard stuff with EUV. Taking intel 4 as an example, even with a 20% mask count reduction vs intel 7, intel claims the number of process steps "only" decreased by 5%.
 
Last edited:
50% TSMC content is, effectively, a tipping point. Intel is fab lite (and shell heavy).

I think building fab shells has strategic optionality, that makes it worth it; think how much advertising and positive press Intel gets for some steel and concrete. And Intel has shown a talent for banking shells and then finding a use later. Fab 42 in AZ is an example.

Just don't confuse fab shells with manufacturing power. I don't see any signs of Intel or Samsung closing the gap with TSMC. It bears watching and a little speculation, but I see more signs of Intel rejuvenation being TSMC-powered.
 
50% TSMC content is, effectively, a tipping point. Intel is fab lite (and shell heavy).

I think building fab shells has strategic optionality, that makes it worth it; think how much advertising and positive press Intel gets for some steel and concrete. And Intel has shown a talent for banking shells and then finding a use later. Fab 42 in AZ is an example.

Just don't confuse fab shells with manufacturing power. I don't see any signs of Intel or Samsung closing the gap with TSMC. It bears watching and a little speculation, but I see more signs of Intel rejuvenation being TSMC-powered.

For the next several years, Intel's first priority is to increase its revenue significantly or at least stop the shrinking revenue trend. Intel desperately needs to deliver attractive products and a lot of them. Due to Intel's own financial constraints, outsourcing is the best option for Intel to bring more products to the market without incurring huge capital spending. Intel foundry's revenue is too small and too far away for the whole Intel enterprise.
 
Last edited:
N7 has more process steps than 16/12FF, which had more steps than 20nm, and 20nm had more steps than 28nm. intel 32nm has more steps than 45nm which itself has more steps than 65nm. Smaller devices require more process steps to form. New architectural features also need more steps to form. As an example of say 22nm vs 180nm, you now have strain, Cu, interconnects, HKMG, and finFET. Even if pitches were unchanged you now add the RMG loop, EPI loop, damascene interconnects, fin loop, etc. The only exceptions to this I can think of are TSMC N5 and intel 4. In those cases they were going from lots of multipatterning to doing the hard stuff with EUV. Taking intel 4 as an example, even with a 20% mask count reduction vs intel 7, intel claims the number of process steps "only" decreased by 5%.
They've been steadily adding more metal layers (mainly mid-stack).
 
Back
Top