Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/gina-raimondo-unveil-the-exclusive-nature-of-the-chips-act.17482/
        )

    [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
)

Gina Raimondo unveil the exclusive nature of the CHIPS act

tonyget

Active member

United States to be the only country in the world where every company capable of producing leading-edge chips will have a significant R&D and high-volume manufacturing presence.
 
Taiwan alone produces 92% of the world’s leading-edge chips, even though the majority of them are still based on technology created at UC Berkeley—with federal funding.

That alone indicates technology originated in US still too expensive to be produced in US.
 
It seems like the US gov has a good idea what they are getting into/the challenges ahead. I approve.

Taiwan alone produces 92% of the world’s leading-edge chips, even though the majority of them are still based on technology created at UC Berkeley—with federal funding.
I don't really get this claim. For one it ignores the equally important memory chips. Two to say TSMC technology originated in Berkeley is a dubious claim to make. It would be like saying IMEC does the R&D for the big three in logic and DRAM.

That alone indicates technology originated in US still too expensive to be produced in US.
As many on this forum know I am skeptical of these sorts of claims. I don't think the inherent cost of operating a fab in the US is at most a few percent more expensive than the ROC. If labor costs had such a drastic effect on cost then we would see a mass exodus from the ROC to SE Asia. I also believe that the US is likely cheaper on power (at least in the midwest/eastern seaboard) and MUCH cheaper on water (depending on the local) vs the ROC. Given the fixed cost nature of this industry, I think it was as simple as the government partnered with private industry and ensured that expansions at home were always incentivized with cheap loans and tax rebates. These expansions at home also encouraged citizens from these nations to come back after studies domestically and abroad. With the large swaths of Korean, Chinese, Taiwanese, and SE Asian students flowing through American universities, American fabs may convince more of them to stay in the US on the promise of greater opportunities.
 
As many on this forum know I am skeptical of these sorts of claims. I don't think the inherent cost of operating a fab in the US is at most a few percent more expensive than the ROC. If labor costs had such a drastic effect on cost then we would see a mass exodus from the ROC to SE Asia. I also believe that the US is likely cheaper on power (at least in the midwest/eastern seaboard) and MUCH cheaper on water (depending on the local) vs the ROC. Given the fixed cost nature of this industry, I think it was as simple as the government partnered with private industry and ensured that expansions at home were always incentivized with cheap loans and tax rebates. These expansions at home also encouraged citizens from these nations to come back after studies domestically and abroad. With the large swaths of Korean, Chinese, Taiwanese, and SE Asian students flowing through American universities, American fabs may convince more of them to stay in the US on the promise of greater opportunities.
I am skeptical too, of the magnitude of the cost differences, as I was when I questioned the statement of the CEO of TSMC saying it was 4-5x the cost to build and operate a fab in the US versus Taiwan. That still seems silly to me. Nonetheless, I think given our environmental regulations and permitting nonsense, and certainly higher construction costs, it is IMHO far more than a few percent more expensive to build and outfit a fab in the US than in Taiwan. I'm thinking 20-30%, based on statements I heard in Intel some years back. (30% is huge. That would mean TSMC's ~$40B investment in AZ would only cost ~$31B in Taiwan. Imagine 4x, where the $40B investment is only $10B in Taiwan.)

Unfortunately, even with publicly-traded US-listed companies you can't get this kind of detailed financial information to make a valid comparison.

I think we agree that the US labor costs in a fab, while probably 2-3x what the same labor would cost in the Taiwan, are not a huge factor in the international cost differences. There just aren't that many people working in a fab compared to the billions of dollars in equipment and maintenance. I also think the statements about not being able to find skilled people in the US based on TSMC previously putting a facility in an unpopulated area are also difficult to believe. Many of the offsetting US advantages you list are probably correct in that they exist, but I don't know how significant they are.
 
Last edited:
As many on this forum know I am skeptical of these sorts of claims. I don't think the inherent cost of operating a fab in the US is at most a few percent more expensive than the ROC. If labor costs had such a drastic effect on cost then we would see a mass exodus from the ROC to SE Asia. I also believe that the US is likely cheaper on power (at least in the midwest/eastern seaboard) and MUCH cheaper on water (depending on the local) vs the ROC. Given the fixed cost nature of this industry, I think it was as simple as the government partnered with private industry and ensured that expansions at home were always incentivized with cheap loans and tax rebates. These expansions at home also encouraged citizens from these nations to come back after studies domestically and abroad. With the large swaths of Korean, Chinese, Taiwanese, and SE Asian students flowing through American universities, American fabs may convince more of them to stay in the US on the promise of greater opportunities.
I think it mainly reflects the CHIP act subsidy should have been done earlier. The popularity of going fabless, especially with AMD, also speaks to the financial difficulty of maintaining the fab in the US. Perhaps yield is an aggravator.
 
The popularity of going fabless, especially with AMD, also speaks to the financial difficulty of maintaining the fab in the US. Perhaps yield is an aggravator.
I don't agree, Fred. Going fabless has become more popular because the foundry model has become so much more sophisticated than it was, say 20 years ago, and because viable chip design companies can now be so much smaller, and so much bigger (like Apple). It doesn't matter where you put a fab or at what cost, it takes a lot of volume now to justify owning a fab. And 20 years ago I think it took a lot more internal technical expertise for a fabless company to use a foundry (like LSI Logic was, for example) than you can now get away with by using TSMC or even GF. Would you agree?
 
"Taiwan alone produces 92% of the world’s leading-edge chips, even though the majority of them are still based on technology created at UC Berkeley—with federal funding."

Did I get a wrong impression that Secretary Gina Raimondo believes the current leading edge semiconductor industry is based on the technology developed by UC Berkeley?

Did she oversimplify many technologies that semiconductor companies are using today?

She might be talking about the immersion technology or FinFet. But there are many other important technologies (and not related to UC Berkeley) that make today's semiconductor industry possible.
 
Last edited:
That still seems silly to me. Nonetheless, I think given our environmental regulations and permitting nonsense, and certainly higher construction costs, it is IMHO far more than a few percent more expensive to build and outfit a fab in the US than in Taiwan. I'm thinking 20-30%, based on statements I heard in Intel some years back. (30% is huge. That would mean TSMC's ~$40B investment in AZ would only cost ~$31B in Taiwan. Imagine 4x, where the $40B investment is only $10B in Taiwan.)
I could certainly be wrong. However when I look at how huge the US chemicals industry is, I don't know if these are deal breakers. The chemicals business is more or less a commodity business, or put another way people wouldn't be buying gasoline from CP or fertilizer from DuPont unless the costs were lower than or equal to current market rates. Another important piece of context is that separations can often be 40+% of a chemical plant's operating costs. With like 90+% of that cost being for supplying the energy and water to the boilers needed for many separation operations. In short given how important utilities costs are to most chemical plants and that US chemical firms are world class for costs in spite of the higher labor/environmental compliance costs, I don't see how fabs that are even less dependent on these factors couldn't succeed. The one area that you are right on the money is that the permitting nonsense can be ridiculous (especially in CA where it is literally impossible to get approval no matter how clean your plant design runs). It is for this reason you see so many plants in the south or the midwest. Based on the sexiness of fabs/the recent announcements from local governments, it seems like the permitting for new fabs is getting fast-tracked.
 
For CPUs we currently have AMD using TSMC N5 against Intel's 7. Can Intel win a price war against AMD with its Intel 7? That would be a more optimistic scenario.

For DRAM (which Secretary was probably not referring to), Micron seems to be currently ahead of Samsung and SK Hynix.

For NAND, Micron, SK Hynix and Samsung seem neck-in-neck at 23x layers, although Micron got there first.

It seems Micron rather than Intel is more obviously a champion for the US. Yet notably it also has sites in Japan and Taiwan and even Singapore.
 
Last edited:
For CPUs we currently have AMD using TSMC N5 against Intel's 7. Can Intel win a price war against AMD with its Intel 7? That would be a more optimistic scenario.
I wouldn’t say that this status quo is necessarily indicative of the future. N5 delivers better cost per transistor than N7. Additionally with how complex the process flow for i7 is, and when intel also puts some of the blame of the crumbling margins on “increasing intel 7 product mix”; I think that tells you all you need to know about i7 wafer costs. Intel still had very high margins when they had to pump out huge 14nm dies to compete vs small N7 dies. I would even argue that this was an even bigger technology gap than we see today.

Given the 20% mask reduction, “large per transistor cost reduction”, and if memory serves from Ben’s vlsi presentation a 15% reduction in process steps; I would have to assume that i4/3 per transistor costs are much closer to N5 than i7 was and might even have similar wafer costs to N3E given how long and complicated it’s process flow is.

Given how fat the margins on intel processors used to be it wouldn’t surprise me if many deficiencies in fab efficiency were hidden. After all, who cares if that 14nm node costs x% more than it has to if those i3s are getting sold for more money than Qualcomm or Mediatek can sell their top chips for. I think a move to foundry/the tough times for intel will uncover these hidden inefficiencies. How good a job intel does at fixing these inefficiencies, time will tell.

For DRAM (which Secretary was probably not referring to), Micron seems to be currently ahead of Samsung and SK Hynix.

For NAND, Micron, SK Hynix and Samsung seem neck-in-neck at 23x layers, although Micron got there first.

It seems Micron rather than Intel is more obviously a champion for the US. Yet notably it also has sites in Japan and Taiwan and even Singapore.
If there are to be national champions I think you need both. Memory is worthless without logic as well as the inverse also being true. My opinion and hope is that CHIPs allocation should be evenly distributed across memory logic and to a lesser extent the analog/power ICs.
 
N5 had been diluting margins for many quarters, maybe it's a fair match against I7. Intel still needs to ride Raptor Lake for a while, hope the yields are no longer a problem.
 
I could certainly be wrong. However when I look at how huge the US chemicals industry is, I don't know if these are deal breakers. The chemicals business is more or less a commodity business, or put another way people wouldn't be buying gasoline from CP or fertilizer from DuPont unless the costs were lower than or equal to current market rates. Another important piece of context is that separations can often be 40+% of a chemical plant's operating costs. With like 90+% of that cost being for supplying the energy and water to the boilers needed for many separation operations. In short given how important utilities costs are to most chemical plants and that US chemical firms are world class for costs in spite of the higher labor/environmental compliance costs, I don't see how fabs that are even less dependent on these factors couldn't succeed. The one area that you are right on the money is that the permitting nonsense can be ridiculous (especially in CA where it is literally impossible to get approval no matter how clean your plant design runs). It is for this reason you see so many plants in the south or the midwest. Based on the sexiness of fabs/the recent announcements from local governments, it seems like the permitting for new fabs is getting fast-tracked.
US chemical manufacturing is a beneficiary of the lowest feedstock costs in the world in terms of natural gas. Chemical production had been moving offshore like everything else in the early 2000s, but came back due to feedstock costs.
 
Labor cost is around 15% only.
TSMC engineer compensation is very close to Intel R&D.

The direct labor in TSMC get $25K-$35K in Taiwan (including bonus) which should be around 50% of Intel direct labor.

TSMC just approved average US$61,921 performance-based bonuses to the company's employees (including engineers and direct labors).
So you know labor cost of TSMC can not be 1/3 of Intel.
 
Labor cost is around 15% only.
TSMC engineer compensation is very close to Intel R&D.

The direct labor in TSMC get $25K-$35K in Taiwan (including bonus) which should be around 50% of Intel direct labor.

TSMC just approved average US$61,921 performance-based bonuses to the company's employees (including engineers and direct labors).
So you know labor cost of TSMC can not be 1/3 of Intel.
Excellent data. Thanks.
 
Intel still had very high margins when they had to pump out huge 14nm dies to compete vs small N7 dies. I would even argue that this was an even bigger technology gap than we see today.
Yes the ultimate cost of the 5 year 14n stall at Intel was the train wreck to their design pipeline. They originally expected to have Icelake to outcompete AMD at the tail end of Naples, with Sapphire supposed to get out ahead of Milan. Instead they sat in a state of denial, apparently not even able internally to assess the real schedule, so they did not even retarget those designs to available processes and what we saw was a series of minor revisions of Skylake up through Cooper Lake.

The real costs in the USA are around can they deliver and yield. Water and salaries are in the noise compared to the value prop of supporting leading edge designs and geographical diversity of supply chain. That is exactly what the USA economy has evolved to need - high margin value and skilled, innovative workforce. Which Silicon Valley proved we have, and now needs to spread to other parts of the country.
 
Yes the ultimate cost of the 5 year 14n stall at Intel was the train wreck to their design pipeline. They originally expected to have Icelake to outcompete AMD at the tail end of Naples, with Sapphire supposed to get out ahead of Milan. Instead they sat in a state of denial, apparently not even able internally to assess the real schedule, so they did not even retarget those designs to available processes and what we saw was a series of minor revisions of Skylake up through Cooper Lake.
I partially agree, but don't think this is entirely true. For example take icelake server. Much larger die than SPR, and yet SPR got delayed by at least a year. To insinuate that 10nm yield problems caused SPR to be late is not supported by publicly available data nor the rumor mill talking about the large number of steppings/retapeouts. It is also hard to believe that intel design did not see the process (un)health of 10nm. From an outsider’s perspective looking in, my guess is that they really thought they had it fixed only to find more issues as they lowered DD. By the time we got to icelake laptops it seemed like DD was mostly solved, but because of how delayed 10nm was they needed to wait until SF to even have a node that was full stop better than 14nm. After all if memory serves from old roadmaps 10nm was supposed to be similar PPW to 14nm++. That would have been fine if icelake launched in 2017 or 2018, but when competing against 14nm++++ it only has a slight advantages for power and cost per transistor.

Either way it illuminates my point that intel was holding onto market share and high margins during the 14nm era in spite of the large node disadvantage and inferior architectures. To me that confirms the idea that i7 is a very expensive node. Looking at today I would say their architectures are at parity, packaging is at a slight disadvantage, and nodes are far behind (but less so than before). Granted this market share and margin lose is partially explained by AMD having the trust from oems to sell more than 0 units. But I don’t think customer trust in AMD can completely explain intel’s deteriorating financial situation in spite of the much more competitive products intel has been launching since late 2020.
 
Last edited:
So when do companies start cashing checks from Chips act? It is not clear who get the money, when and how much. Right now its like the lottery.... every thinks they are going to win big
 
US chemical manufacturing is a beneficiary of the lowest feedstock costs in the world in terms of natural gas. Chemical production had been moving offshore like everything else in the early 2000s, but came back due to feedstock costs.

You forgot Qatar which is on the way to becoming a petrochemical Godzilla.

Up until recently, they had a problem of a too pure (dry) gas, until more drilling found more "wet" gas with higher heavier molecular weight fraction.
 
Back
Top