[HN Gopher] TSMC's Arizona Plant to Start Making Advanced Chips
___________________________________________________________________
TSMC's Arizona Plant to Start Making Advanced Chips
Author : rbanffy
Score : 314 points
Date : 2024-12-27 19:43 UTC (1 days ago)
(HTM) web link (spectrum.ieee.org)
(TXT) w3m dump (spectrum.ieee.org)
| osnium123 wrote:
| Great news and arguably these are the most advanced
| semiconductors being produced in the United States today.
| teruakohatu wrote:
| Wikipedia lists Intel 3 is roughly the same tech level as TSMC
| 3nm [1], but without listing transistor density. Intel is
| producing the Xeon 6 using Intel 3 [2]. So arguably Intel has a
| more advanced process in the USA than TSMC, which is doing 4nm
| in the USA next year. Intel's production is probably not very
| high.
|
| [1] https://en.wikipedia.org/wiki/3_nm_process#cite_note-74
|
| [2]
| https://www.intel.com/content/www/us/en/products/details/pro...
| osnium123 wrote:
| Is Intel 3 manufactured in the US or Ireland?
| https://www.intel.com/content/www/us/en/newsroom/news/new-
| fa...
|
| Also, it depends on the metrics but TSMC's N4 is a mainstream
| foundry logic node. Who is using Intel 4/3 outside of Intel?
| kube-system wrote:
| Looks like both according to this quote:
|
| > Our Intel 3 is in high volume manufacturing in our Oregon
| and Ireland factories
|
| https://www.tomshardware.com/tech-industry/intel-3-3nm-
| class...
| georgeburdell wrote:
| TSMC 3nm is a double digit percentage denser. Intel 3 is
| closer to TSMC 5nm
| soganess wrote:
| Citation?
|
| (Not intended as a snipe. I honestly just don't know where
| to look for that kinda info.)
| georgeburdell wrote:
| Wikichip is my go to (which is down right now for me
| unfortunately). It's important to look at the latest data
| because Intel's internal nodes real specs have not met
| the stated expectations recently
| fooker wrote:
| IIRC it was the other way around.
| ryao wrote:
| That was before Intel renamed their process nodes. They
| went from being 1 node more dense to being 1 node less
| dense with their new naming scheme. You need Intel 4 to
| match TSMC 5nm.
| silisili wrote:
| I'm not refuting the statement, only pointing out that
| density is not the only factor.
|
| Unfortunately, these numbers are arbitrary and companies
| are guessing what performs about like what based on
| numerous factors. Often wrongly - Samsung's equivalents
| were so bad Qualcomm pretty much abandoned them, and for
| good reason. Anyone who used an Exynos or SD888 understands
| why.
|
| I feel like we should have landed on a better tracking
| system now, like perf/watt, but here we are.
| Dalewyn wrote:
| >these numbers are arbitrary
|
| Seeing as Intel 7 is formerly Intel 10nm, there is at
| least a reasonable argument in that Intel's number is one
| size(?) smaller than it should be.
| silisili wrote:
| It's equally likely Intel realized it performed as well
| as Samsung/TSMC "7". Which is the whole issue, we'll
| never really know.
| ryao wrote:
| These are 4nm facilities. Intel's 18A process is more advanced.
| Hopefully it will turn out well. If not, that is the end of
| Intel.
| Dalewyn wrote:
| >Intel's 18A process is more advanced.
|
| _Can be_ , not "is". I will believe them when I see it.
| clumsysmurf wrote:
| Maybe ...
|
| But for those living close to the plant, I'm not so sure:
|
| "Environmental, and public health groups, including the Sierra
| Club, are urging President Joe Biden to veto a controversial
| bill that exempts most semiconductor companies applying for
| federal CHIPS Act funding from having to complete essential
| environmental reviews, as required by the National
| Environmental Policy Act, or NEPA."
|
| "Exempting the semiconductor industry from NEPA is completely
| unwarranted, especially considering the projected significant
| increase use of PFAS and other toxic chemicals by the industry
| and their track record of releasing these dangerous chemicals
| into the air and water surrounding the facilities," said Tom
| Fox, Senior Legislative Counsel at the Center for Environmental
| Health"
|
| https://www.sierraclub.org/press-releases/2024/10/environmen...
| wbl wrote:
| NEPA doesn't govern any releases: that's done by the EPA
| under clean Air and clean water act.
| dpedu wrote:
| Is this fab on par with TSMC's fabs in Taiwan? I am not up to
| date with the various processes.
| wtallis wrote:
| No. TSMC's 4nm processes are part of the 5nm family. 3nm has
| been shipping for over a year, and is only fabbed in Taiwan for
| now and the next few years.
| hackernewds wrote:
| And purposefully so to keep the "silicon shield" intact for
| Taiwan. I did read that the yields in the US are just as good
| mdasen wrote:
| It's not on par with the best TSMC has in Taiwan, but most
| companies are still using 4nm. Yes, 3nm has been shipping for
| over a year - but only if your company is named Apple. Intel
| just launched a small portion of its products using 3nm two
| months ago.
|
| I think realistically it'd be more fair to say that 3nm is
| coming in 2025 and there's a huge distance between 2025 and
| 2028 (when they'll start doing 3nm and 2nm in the US). Right
| now, AMD, Nvidia, and Qualcomm aren't doing 3nm. If the world
| lost 3nm today, it'd basically be Apple's products that would
| get hit. It'd definitely screw over Apple and it'd mess up
| the future plans for AMD, Nvidia, and Qualcomm, but it's not
| like the industry has been using 3nm for over a year. No,
| only Apple.
|
| The big problem is that there's a big difference between
| "we'll be bringing 3nm to the US in early 2026" and "we'll be
| bringing 2nm and 3nm to the US in 2028". If they started
| making 3nm in the US in early 2026, that's going to be less
| than a year behind most companies using 3nm. Qualcomm and
| Nvidia will probably start shipping 3nm in February 2025 and
| AMD will probably start shipping 3nm in late 2025.
|
| If TSMC's US fab were 12-18 months behind their Taiwan fabs,
| it wouldn't really be a problem, except for Apple. Everyone
| else is waiting 18 months for TSMC's latest gen stuff anyway.
|
| The problem isn't that the US fab can't do 3nm today. TSMC's
| Taiwan fabs aren't doing 3nm at scale unless your name is
| Apple. The problem is that their US fabs won't be doing 3nm
| for around 3 years after the industry moves over to 3nm. If
| the US fab could satisfy 4nm demand and Taiwan disappeared
| today, it'd mostly hit Apple's product line. The issue is
| that in 2026 or 2027, every company will be relying on 3nm
| and if Taiwan disappeared then, it'd hit the whole industry's
| product lines.
|
| But it's possible that Intel's 18A will do amazing and Intel
| will be able to manufacture at scale and a lot of TSMC's
| business will move to Intel. Then the US (Intel) would be
| manufacturing more advanced chips than TSMC in Taiwan. TSMC
| isn't expected to make the move to High-NA EUV for a few more
| years so Intel has some time when it could overtake TSMC.
| numpy-thagoras wrote:
| It may not seem like much since it's only Apple right now,
| but their 3nm SoCs are stunning. I can only imagine what
| the industry is going to look like when this tech becomes
| the standard. The miniaturization potential alone can
| transform many other technologies, let alone its value for
| low-power edge compute.
|
| The difference isn't revolutionary, but noticeable. Whoever
| has it will have a competitive advantage.
| wtallis wrote:
| > Yes, 3nm has been shipping for over a year - but only if
| your company is named Apple. Intel just launched a small
| portion of its products using 3nm two months ago.
|
| > I think realistically it'd be more fair to say that 3nm
| is coming in 2025
|
| Almost everyone but Apple decided to skip N3B and wait for
| the later N3E. Intel decided to just be late with N3B,
| launching their laptop part in September and the desktop
| part in October. Apple, Qualcomm, and Mediatek all have N3E
| parts on shelves and in consumer's hands. 3nm is _here,
| now_. Two generations of TSMC 3nm have ramped to full
| production.
| ryao wrote:
| Taiwan has a law barring the export of technology more than one
| generation behind:
|
| https://www.taipeitimes.com/News/biz/archives/2024/11/08/200...
|
| The article mentions that Arizona was set to ramp 4nm, which is
| presumably what they have now.
| rayiner wrote:
| Why doesn't the U.S. have laws like that?
| jazzyjackson wrote:
| Because keeping the bleeding edge in manufacturing to
| ourselves is not vital to our survival as a sovereign state
| yieldcrv wrote:
| yeah but stopping China from invading China isn't either
| kayewiggin wrote:
| China is in the beginning of a 30 year Great Depression,
| in no shape to invade Taiwan. Consumer spending in
| Beijing and Shanghai fell 20% y/y in November. Real
| estate prices have collapsed 50%, even in some parts of
| Beijing and Shanghai. Trump has filled the cabinet with
| mostly anti-China hawks, indicating large tariffs coming
| next year. Capital outflow from China increased to $45B
| in November, largest monthly deficit ever. China is
| pretty fucked.
| Paradigma11 wrote:
| That does sound like a fabulous time to start a
| jingoistic war to flame the nationalistic sentiments and
| declare any dissenters traitors to the nation.
| azernik wrote:
| It is for Taiwan
| yieldcrv wrote:
| and not going to be our problem for much longer,
| 2028-2030 can't come fast enough
|
| the US is selectively getting involved in worldwide
| conflicts to deter China from invading China, and its
| awkward, with arduous contrived rationales to maintain
| its people's support
|
| and once we get stateside semiconductors at low enough
| nanometers we wont have to do any of that any more
|
| I cant wait
|
| good thing there are 185 _other_ countries that could
| care if they really did. this wont be controversial to
| point out, in the future. it will be a time period that
| made little sense.
| ebruchez wrote:
| I am not sure if it is worth answering but here it goes
| anyway:
|
| 1. Taiwan is not China, any more than Ukraine is Russia,
| except if you believe all the propaganda coming from the
| mainland (or Russia). Ask any Taiwanese, and while many
| consider and appreciate a solid Chinese cultural
| heritage, they consider themselves independent and want
| nothing to do with China (except business). Newer
| generations of Taiwanese are even more independently-
| minded and consider themselves even more Taiwanese than
| the previous generations.
|
| 2. Even if for some reason you truly think that it is the
| same country or should be the same country, it is immoral
| to wish that a peaceful, independent, democratic, and
| open society like Taiwan's should be brutally attacked
| and absorbed by a war-mongering,
| authoritarian/dictatorial, opaque country. (Things could
| be different if mainland China was democratic, but it
| isn't, and won't be for a long time.)
|
| 3. Even if for some reason you are ok with the above,
| odds are that the difficulty and complexity of an attack
| on Taiwan would end up being extraordinarily costly for
| China (and Taiwan of course). It could lead to all sorts
| of escalations in the region, sanctions, the collapse of
| trade with China from the US and other countries, nuclear
| proliferation (see Ukraine considering developing nuclear
| weapons if they don't get security guarantees), and who
| knows what else.
| squillion wrote:
| > Taiwan is not China
|
| Taiwan might be considered a de facto independent
| country, but according to most institutions it's
| officially part of China.
|
| 1. Taiwan's official name is _Republic of China_ (ROC):
| it regards itself as part of China, and the sole
| legitimate seat of China 's government. It's true however
| that "it has not formally renounced its claim to the
| mainland, but ROC government publications have
| increasingly downplayed this historical claim". [1]
|
| 2. In 1971, the UN General Assembly Resolution 2758
| "recognized the People's Republic of China (PRC) as 'the
| only legitimate representative of China to the United
| Nations'". [2]
|
| 3. Only 11 (tiny) countries officially recognize Taiwan
| as an independent country, i.e. maintain full diplomatic
| relations. [3]
|
| 4. The U.S. official position is that "The United States
| has a longstanding one China policy", and "we not support
| Taiwan independence". [4]
|
| [1] https://en.wikipedia.org/wiki/Taiwan#Foreign_relation
| s_and_i... [2] https://en.wikipedia.org/wiki/United_Natio
| ns_General_Assembl... [3] https://en.wikipedia.org/wiki/F
| oreign_relations_of_Taiwan#Fu... [4]
| https://www.state.gov/u-s-relations-with-taiwan/
| ebruchez wrote:
| There are several aspects that come into play:
|
| 1. How the PRC (mainland China) regards Taiwan (or ROC).
|
| 2. How Taiwan regards itself. This has changed over time.
|
| 3. How third-parties play that situation.
|
| Since Nixon's visit to China in the 1970s, the world
| recognized that it was pointless to deny that the CCP
| (Chinese Communist Party) ruled mainland China for good.
| From there, the PRC progressively got official
| recognition in institutions like the UN. In order not to
| inflame the PRC's leadership and keep access to mainland
| China, many countries state that they do not recognize or
| encourage Taiwan's independence. But note that they also
| maintain de facto diplomatic relationships, being careful
| not to use the name "embassy" or "consulate".
|
| In reality, Taiwan has been absolutely independent since
| the 1950s. It's just that it's not officially recognized
| by most institutions and countries for diplomatic
| reasons.
|
| I'll add that the "one China policy" is ambiguous by
| design. It doesn't mean that it must happen in the
| foreseeable future. It also doesn't mean that the PRC
| should be allowed to take over Taiwan through military
| might.
|
| In the end, no matter what the various parties' policies
| are, almost nobody in Taiwan at this point believes that
| a peaceful so-called "reunification" is desirable or
| possible. I put the word "reunification" in quotes in
| particular because the CCP never controlled Taiwan, and
| also because in general the historical argument doesn't
| make any sense. Personally, I think that the principle of
| self-determination is what should apply here, for moral
| reasons. [1]
|
| [1] https://en.wikipedia.org/wiki/Self-determination
| yieldcrv wrote:
| I'm aware, both entities have claims to the whole
| mainland and still have China in their name
|
| both entities would have territorial disputes with other
| regions, that we don't agree with
|
| some parts of the ROC have dropped claims to the mainland
|
| and its all so hilarious that it reminds me how we, the
| US, shouldn't be involved, and wont be after the
| semiconductor problem is hedged
| ThinkBeat wrote:
| I presume you are aware that Taiwan is in fact occupied
| by China now?
|
| The Chinese who ran from the communist revolution,
| invaded Taiwan, setup their own military dictatorship,
| and they were extremely brutal to the natives Taiwanese.
| (Sadly this has been their lot through several
| occupations by different entities.)
|
| During more recently history they have been polishing
| more democratic values and life for the natives has
| improved.
|
| But for Taiwan to be free, in any proper sense, the
| Chinese occupiers must leave.
| rayiner wrote:
| Why is "vital to our survival as a sovereign state" the
| criterion?
| gavindean90 wrote:
| Because it is for Taiwan
| alephnerd wrote:
| Semiconductor fabrication was viewed as a commodified cost
| center until COVID related supply chain instability.
|
| Furthermore, packaging and testing was largely outsourced
| and the domestic semiconductor industry imploded in the
| 2010s with IBM Micro and AMD's failures.
|
| The same thing happened to Japan when they began offshoring
| Memory Fabrication to South Korea and Taiwan in the
| 1990s-2000s.
|
| That said, from a NatSec perspective legacy processes
| (28nm, 48nm) and compound semiconductors would be much more
| critical (and a significant amount of funding has been
| devoted to that).
| ryao wrote:
| Japan is trying to rebuild its leading edge capability
| with Rapidus using IBM technology. Interestingly, IBM
| still does the research needed to make a fabrication
| plant. They just don't want to assume the risks from
| deploying it in production anymore as far as I can tell.
| alephnerd wrote:
| > Interestingly, IBM still does the research needed to
| make a fabrication plant
|
| Yep. They still own the IP from the IBM Microelectronics
| days.
|
| Much of the breakthroughs in EUV were done in Upstate NY
| (especially at SUNY Albany, SUNY Polytechnic, and RPI),
| and a lot of that was co-owned by IBM, ASML, and TEL.
|
| > They just don't want to assume the risks
|
| The capex - and pretty much.
|
| Semiconductor Fabrication is high cost, low margins, so
| it's difficult to spin up without industrial policy.
| ryao wrote:
| It is a shame that the 450mm transition did not occur. It
| would have been better for all parties as it should have
| lowered the cost of fabrication.
| anigbrowl wrote:
| Free market doctrine, plus the investor class wanting to be
| able to reap the benefits of outsourcing without being
| concerned about strategic issues. Occasional proposals to
| this effect have historically been denounced as
| protectionism, industrial policy (practically socialism!)
| and 'picking winners and losers'. I am surprised you're
| unaware of this.
| selimthegrim wrote:
| We had to give Japan something in the 90s to keep them on
| side.
| alephnerd wrote:
| Japan itself largely began offshoring fabrication in the
| 1990s.
|
| It was Japanese OSAT players like Hitachi that sparked
| the Penang packaging cluster in Malaysia in the 70s-90s
| and Japanese Memory firms like NEC+Hitachi that started
| South Korea and Taiwan's fabrication industries.
|
| Taiwan didn't truly become a leader in the cutting edge
| fab space until the 2010s when US, SK, and Japanese
| players dropped the ball, and Apple chose TSMC in the
| 2010s due to their patent litigation with Samsung (nixing
| South Korea).
| brcmthrowaway wrote:
| So much of TSMC's dominance now is due to the influx of
| Apple cash in the 2010s boosting R&D spending, which in
| turn is because millenials bought a shit tonne of Apple
| devices because they were convinced by marketing.
| wtallis wrote:
| TSMC's dominance is at least as much Intel's fault as it
| is Apple's. And even if Apple hadn't been funneling so
| much money to TSMC, the smartphone industry as a whole
| still would have been a cash cow for TSMC. Intel sure
| wasn't going to be in the running as a smartphone SoC
| designer _or_ as a foundry for somebody else 's
| smartphone SoCs. In an alternative history where Android
| thoroughly beat out iOS even for high-end/high-margin
| smartphones, Samsung's foundry business probably would
| have been a bit better off, but overall it would still be
| TSMC as the leading foundry, just with Qualcomm as the
| launch customer for new nodes rather than Apple.
| alephnerd wrote:
| > Intel sure wasn't going to be in the running as a
| smartphone SoC designer or as a foundry for somebody
| else's smartphone SoCs
|
| Intel did try doing this in the 2000s, but couldn't
| justify the resourcing needed for this due to x86 as well
| as their restrictive licensing of Intel Atom.
|
| Meanwhile, ARM was fabless and just licensed to anyone (a
| major reason why Chinese challenger brands exist in the
| Chips space today)
|
| Fundamentally, you cannot be both an IP creator (eg.
| Design) and chip fabricator, as both functions have
| different economics and competitive structures, and one
| BU inevitabely holds the other back.
|
| > Samsung's foundry business probably would have been a
| bit better off, but overall it would still be TSMC as the
| leading foundry
|
| Samsung, SK Hynix, and other Korean players dropped the
| ball due to the Apple lawsuit as well as the 2016-17 SK-
| China trade war (impacted SK exports to China - including
| intermediate parts) and the 2019-23 SK-Japan trade war (a
| number of critical components in fabrication are supplied
| by Japanese firms like Tokyo Electron and Nikon and were
| impacted by mutual tariffs)
| klooney wrote:
| > Fundamentally, you cannot be both an IP creator (eg.
| Design) and chip fabricator, as both functions have
| different economics and competitive structures, and one
| BU inevitabely holds the other back.
|
| Vertical integration can win too, it worked for Intel for
| decades.
| alephnerd wrote:
| Until it didn't.
|
| Most players in the hardware industry try to specialize
| in one function and do that very well, as this builds
| your competitive advantage AND allows you to leverage
| partnerships to further enhance your moat by building an
| ecosystem.
|
| For example, ARM is purely design driven - targeted
| specifically at low power compute usecases - and licensed
| it's IP out to just about any player, which allowed an
| ecosystem to develop.
|
| Nvidia did the same thing by remaining fabless and only
| concentrating on GPUs.
|
| TSMC concentrates only on fabrication and doesn't dare
| enter design because they know all their customers would
| leave overnight because they would not want to subsidize
| a potential competitor.
|
| Intel was in too many segments, which meant it was
| inevitably competing with everybody, which forced
| everyone to leverage partnerships to challenge the big
| baddie.
|
| A similar thing happened to Samsung to a certain extent
| as well.
| ryao wrote:
| Apple pays TSMC better than anyone else does since they
| want the best processes and are willing to pay a premium
| to cover much of the investment needed to achieve them.
| Losing them would really hurt TSMC. Not having them in
| the 10s would have slowed down TSMC's development of new
| process technology.
| _DeadFred_ wrote:
| Because we already have enough current/ex superfund sites.
|
| (see the Santa Clara section here: https://en.wikipedia.org
| /wiki/List_of_Superfund_sites_in_Cal...)
| nrp wrote:
| We do for a wide variety of products and IP:
| https://www.trade.gov/us-export-controls
|
| See also the US sanctions on SMIC.
| kube-system wrote:
| It does.
|
| https://en.wikipedia.org/wiki/Export_Administration_Regulat
| i...
|
| And in fact, the machines that make these chips are
| restricted by US export law:
|
| https://www.asml.com/en/news/press-releases/2024/asml-
| statem...
| dylan604 wrote:
| Regarding chips, if your best is generations behind someone
| else's best, nobody want's to buy your old and busted
| anyways.
| fooker wrote:
| The US had famously tried and failed to do this for
| software techniques like cryptography.
| ojbyrne wrote:
| I think you meant "less than one generation behind." Or as
| the article you linked to says: "Taiwanese law limits
| domestic chipmakers to producing chips abroad that are at
| least one generation less advanced than their fabs at home"
| ryao wrote:
| I did. My apologies for the typo.
| wincy wrote:
| I thought the machines that make this stuff come from ASML in
| The Netherlands? How does this work, couldn't we buy the
| machines from ASML?
| bri3d wrote:
| As I post every time this question gets asked: no. ASML
| build fancy printers. Buy an ASML machine and you can now
| etch nanometer-scale features into something. That's a
| great party trick. You still need to know what features to
| print and how to make the materials you print your design
| on. The ASML part (lithography) is a hard part but it's not
| even close to the biggest hard part. Thus, why
| semiconductor processes are differentiated in the first
| place.
| MangoCoffee wrote:
| Plus, ASML's EUV machines for TSMC are different from
| those for Samsung or Intel. Each order is tailored to the
| buyer's specifications.
|
| TSMC's manufacturing process using ASML EUV machines is
| different from Intel or Samsung.
|
| People think you just buy an EUV machine, and you can
| start printing money. Far from it.
| ryao wrote:
| Presumably, one could license IBM's 2nm research, buy the
| equipment and try doing 2nm fabrication. That is what
| Rapidus is doing.
| codedokode wrote:
| This is stupid. They should amend it to ban export of less
| than 3 generations behind.
| edgyquant wrote:
| This seems like the exact kind of law that would not standup
| to extreme pressure from a determined US president
| Spooky23 wrote:
| Taiwan has alot of leverage given the tension between the
| US and China. The upcoming admin will be even easier on
| play off.
| nxobject wrote:
| I wonder how the economics will end up - sure, American fabs
| won't have cutting-edge processes, but in the end there's a
| stable market for older processes that are critical to
| industrial capability (e.g. automotive and sensing, high-
| reliability processors, etc.) One node behind still remains
| very good value without the visicssitudes of relying on the
| unstable market for leading-edge products.
| yieldcrv wrote:
| > As more fabs open, the United States is also facing a shortage
| of engineers and technicians.
|
| levels.fyi says principle level engineers are making $86,000
| annually in Taiwan, with zero shares. $49,000 being the average
| for [software] engineers in Taiwan
|
| there will be a shortage at that compensation range, which they
| can solve with higher cash and amplify with shares and a
| competitively short cliff like Meta and others have, of 3 months
| or less.
| magic_smoke_ee wrote:
| Basically 25% of what American SWEs make. I can only surmise
| the cost of living is much lower in Taiwan.
| yieldcrv wrote:
| yeah of course, but many organizations have a rigid corporate
| ladder to overcome. this seems like one of them, and their
| many "cultural differences".
|
| they are incentivized to underpay americans, complain that
| they "can't" find talent, to ensure the relevancy of Taiwan
|
| but their arguments are weak and solved with compensation
| j_walter wrote:
| They are not solved with compensation...simply put
| Taiwanese in both the US or Taiwan will put in more hours
| and work harder regardless of pay. Will compensation get
| some US workers to work as hard...yes...but not enough for
| what is needed to expand the AZ plant and keep it running.
| The numbers in the OP are Taiwan salaries...AZ salaries are
| upwards of $140-150K (not including bonuses) for someone
| with <10 years experience. These are not SWE...these are
| mechanical, electrical and chemical engineers...not in
| competition with Google, Apple, Meta, etc.
| mook wrote:
| Yep, not that many top-tier talent in the US willing to
| be in the factory for the graveyard shift under high
| pressure. The lines run 24/7 and if anything is slightly
| wrong techs need to be already on site to go fix it,
| because it's crap tons of money for every second of
| downtime. That leads to a corporate culture where even
| R&D has similar pressures from your boss (because
| essentially you're always racing with the competing
| fabs).
| pm90 wrote:
| Ive never understood this culture. This kind of operation
| could be achieved by having several teams of folks
| working in shifts so noone is working crazy long, no? It
| seems like the company is unwilling to invest in the
| manpower required to achieve that SLA? fwiw ive heard
| similar things about the fruit company.
| j_walter wrote:
| They have 24/7 people on site, 4 shifts that cover the
| work...but those are generally techs in the US, which
| have associate degrees or less. In Taiwan they are
| generally engineers that work the 24/7 shifts.
| j_walter wrote:
| They don't have that many SWE...so be careful on your
| comparisons. 95% of their engineers are non-SWE...and those
| engineering disciplines do not make 4X those salaries listed
| above.
| lifeformed wrote:
| It is mostly, but real estate in Taipei is more expensive
| than SF.
| MangoCoffee wrote:
| Let's get real. A lot of talent has gone into ad tech in the
| U.S.
| ryao wrote:
| https://www.numbeo.com/cost-of-living/compare_cities.jsp?cou...
|
| > Cost of Living Including Rent in Taipei is 59.0% lower than
| in San Francisco, CA
|
| Salaries tend to scale with cost of living. The cost of living
| in Taiwan is lower than the US. The difference is particularly
| large if you compare Taipei, the capital where the cost of
| living is likely the highest, to San Francisco. Presumably, the
| salaries would be higher if they hire people from the US.
| siva7 wrote:
| Even if you adjust for cost of living, the pay is still
| significantly lower than US and somewhat lower than central
| europe.
| drtgh wrote:
| By _" we are facing a shortage of engineers and technicians"_
| they really mean _" we want to pay less to engineers and
| technicians"_.
|
| With the first type of statement, the industries are pursuing a
| call effect, an excess of demand, which translates into the
| second statement.
| magic_smoke_ee wrote:
| But not 2 nm node (N2{,P,X}) until about 2028. The delay is still
| indicative of protectionism. Until the US has a (or preferably
| more) American company with 2 nm capabilities with the whole
| process including diffusion and packaging, there's no real
| native, strategic capability.
| atty wrote:
| That is what Intel 18A is, no? In some ways it's worse than N2,
| and in some ways it's better. Overall seems comparable to me,
| and apparently it's still on track for next year.
| ryao wrote:
| What American company would even attempt this aside from Intel?
| IBM still does the relevant research, but quit the business of
| actually using it. They licensed their 2nm process technology
| research to Japan's Rapidus if I recall. I cannot think of
| anyone else in the US that would be willing to take the risk of
| trying to start a 2nm foundry service.
| oldpersonintx wrote:
| Samsung is going for 2nm in Taylor TX
| chasil wrote:
| I've read in the interview below that all attempts to
| implement IBM's copper interconnects failed, except for TSMC.
|
| At least for this particular technology, IBM did not deliver
| everything needed to do this.
|
| "So, when we went to .13u, .13u the people began to change
| from aluminum to copper. And IBM was the leader for the
| copper metal. They had the longest history of developing
| copper technology. They worked for more than ten years on
| copper. TSMC didn't have any experience in copper at all. So,
| when we decided we need to adopt copper, okay. So, the copper
| is one story and low-k material is another one. IBM decided
| kind of low-k material is a spin-on material called SILK. IBM
| had a Research Consortium that IBM-- Samsung joined them, I
| think, ST Micro joined them. Several companies joined the
| Consortium.
|
| "And UMC joined them. But we didn't join them. They all used
| that spin-on low-K material. But we decided to use CVD -
| instead of flourine-doped it's a carbon-doped made by Applied
| Materials. They're called Black Diamond. So, we choose Black
| Diamond. The reason we chose Black Diamond was very simple,
| because I suffer at .18 with a spin-on. I wouldn't touch
| spin-on again. <laughter> But they didn't go through that.
| So, we were very, very lucky. TSMC became the first company
| in the world which was able to ship a manufacturing wafers
| with the copper and low-k, because IBM failed... Later on
| they found reliability the problem."
|
| https://www.computerhistory.org/collections/catalog/10279267.
| ..
| brennanpeterson wrote:
| Spin in is an interesting tech history. As for cvd low-k,
| it is mostly how much C is in your silicon, and likewise
| how you setup the damascene etch stop. Intel was low-ish k
| in about 2002 on 130nm.
|
| I am not so sure tsm was first. Depends on how you define
| lowk.
| anon291 wrote:
| Awesome!!!!
| xenospn wrote:
| Is it just me, or did this behemoth get built in record time?
| Extremely impressive.
| sangnoir wrote:
| The CHIPs Act is a great piece of legislation.
| redserk wrote:
| CHIPS and the Inflation Reduction Act are two of the most
| underrated domestic policy bills in recent American history.
| Conservative-driven contrarian politicking aren't doing the
| country any favors.
| marxisttemp wrote:
| As much as I dislike Biden from a leftist perspective, I
| must commend him for the inflation reduction act. Felt
| funny seeing a president actually, you know, improve the
| country
| xenospn wrote:
| Do you dislike him because he's a leftist or because
| you're a leftist?
| jazzyjackson wrote:
| (username is relevant here)
| KerrAvon wrote:
| This is getting way off topic, but Biden is a centrist.
| The idea that he's a lefty socialist is a political
| cudgel the right wing has swung at every Democratic prez
| candidate since 1988; it's getting more traction in
| recent years as the media has been increasingly purchased
| by right wing billionaires. They called Obama a socialist
| even as he was praising Reagan and helping out the
| bankers who caused the 2008 economic meltdown.
|
| Everything I'm saying here is a documented. Biden has
| been in public service since 1973; look up his Senate
| voting record. Look up the ownership and political
| stances thereof for any given traditional media outlet;
| newspapers, websites, etc.
|
| "Biden is a lefty" is a false, lazy canard.
| xenospn wrote:
| That is correct.
| marxisttemp wrote:
| The latter
| jimt1234 wrote:
| Off topic: the Biden administration also made a commitment
| to passenger rail, including high-speed rail, something the
| rest of the world has had for years. Unfortunately, the
| incoming administration will most likely kill all this.
|
| https://www.whitehouse.gov/briefing-room/statements-
| releases...
| tester756 wrote:
| But this plant cant stand alone, I mean if something bad happens
| to TSMC on Taiwan, then they will no be able to move fab to the
| newer nodes, I think.
| iddan wrote:
| Rome was not built in a day. I think it's a stepping stone for
| that
| tester756 wrote:
| If there is no plan to make it standalone, even worse, for
| TSMC it would be illegal to make this plant the leading one,
|
| then still US should put more money into Intel
| cute_boi wrote:
| What will happen if US put more money into Intel? I believe
| they have money, but they aren't focusing on cutting edge
| technology.
| Hilift wrote:
| It's an awesome contingency. If the island falls, they can
| destroy/impair the local infrastructure, and reconstitute it in
| the US. Destroy in this context does not mean mass physical
| destruction. It is a combination of removal of keys and select
| components. The message being you can have the island but not
| the business.
| wbl wrote:
| And? Advanced nodes aren't really relevant to subtracting
| where it is from where it isn't. China is ideologically
| motivated to conquer Taiwan not economically.
| User23 wrote:
| I think it's probably a case of both?
|
| In any event it appears that reunification is essentially
| inevitable and the only question is when. China doesn't
| appear to feel especial urgency about it.
| phendrenad2 wrote:
| My understanding is it CAN stand alone, it's just not making
| the most cutting-edge node (but it'll come with time).
| WhereIsTheTruth wrote:
| A victory for sovereign Taiwan, protecting the industry they
| built from the ground up!
| yupyupyups wrote:
| Doesn't this remove the incentive for the US to protect Taiwan
| then?
|
| I'm speculating, but if China invades Taiwan, it's cheaper for
| the US to bomb the fab in Taiwan to not let it get into Chinese
| hands in case of an invasion. They could additionally offer
| generous asylums to Taiwaneese researchers and engineers. Then
| whatever happens to Taiwan happens?
| 93po wrote:
| The US is not anywhere close to replacing the outputs of
| Taiwan. The US will be dependent on Taiwan chips for a long
| time.
| raincole wrote:
| The parent comment was being sarcastic.
| ksec wrote:
| I just want to add the term "ADVANCED" in terms of foundry node
| now has an official meaning anything sub 7nm. With specific rules
| in place in terms of export especially to China. This was a
| reference from ASML presentation not so long ago.
|
| It is also important to point out, the achievement here is how
| fast TSMC manage to set things up and running even without the
| home ground advantage. Intel couldn't even replicate this time
| frame if it was their Intel 7nm Fab. And of course the greatest
| record was that with enough planning and permission done before
| hand TSMC manage to have the fab built and running within 18
| months in Taiwan. ( Arguably closer to 12 months )
|
| This also means unless a miracle happen or US Gov being unfair
| with certain things the chances of Intel catching up with its
| current team, management, board members and investors, against
| TSMC in terms of capacity, price, and lead time as a foundry is
| close to zero. ( I am sorry but I lost all faith and hope now Pat
| Gelsinger is out. )
|
| Once TSMC 2nm hits the ground later this year, TSMC US will also
| start their 3nm work if they haven't started now.
| necovek wrote:
| If you believe you can consistently predict future like that,
| it should clearly guide your investment in stocks.
|
| However, just like how quickly and suddenly Intel lost the
| lead, things may turn around for TSMC too: at some point, their
| research hits a dead end and somebody overtakes them too.
| ksec wrote:
| >If you believe you can consistently predict future like
| that, it should clearly guide your investment in stocks.
|
| Perhaps I should have written with Disclosures. For the
| record I did invest in AMD when it was below $3 and TSMC at
| below $400TWD. None of these are investment advices so take
| it what you will. ( You get much better return with Tesla and
| Nvidia in the same period of time but then investment isn't
| always about best returns. ) And I was waiting to invest into
| Intel, unfortunately Pat is gone. To my words I said this in
| April 2023 [1]
|
| " _I am just worried if Stock price continue to fall, Pat may
| be forced out again by those stupid Board. And if Pat is out,
| I won't invest in Intel at all._ "
|
| As you will read in my reply below, I have a very negative
| view on Intel's board for a very _very_ long time.
|
| >However, just like how quickly and suddenly Intel lost the
| lead
|
| It wasn't quick or even sudden. I wrote about it in 2014 and
| got a death threat from Intel Fan boys then. I have been
| questioning about Intel's management on GPU, Fab capacity
| allocation, CapEX, dividends etc for a very long time. For
| another point, TSMC never wanted to be the most advance
| manufacturing Fab. Them having leading node is purely
| accidental and Intel's slip up. They have been doing Intel -1
| node for most of their history and are doing just fine.
| Providing Pure Play Foundry Services with Industry wide
| support on Tools at a reasonable / acceptable price for
| Fabless players. And right now, they are firing on all
| cylinders.
|
| Again. None of these are investment advices and personal
| opinion only.
|
| [1] https://news.ycombinator.com/item?id=35722974
| necovek wrote:
| "Sudden" with big enterprises is still a span of multiple
| years: probably iX-4 series CPUs hit the wall on
| performance, with power efficiency continuing improvements
| into 2017 with iX-8*U CPUs -- so 2013 and 2017. And as soon
| as their first Tick-Tock blip hit, it was clear to everyone
| that they don't have a clear path forward.
|
| In that sense, I fully expect the incumbent top fab to
| maintain the lead for a number of years even when a
| "sudden" competitor enters the market with clearer path
| forward.
| samdjstephens wrote:
| It's about demand isn't it? TSMC have red hot demand, it's not
| hard to understand their urgency in setting up new fabs,
| wherever they may be. Intel don't have the same incentive -
| their incentive is to take the money (because, why wouldn't
| you), build newer fabs and hope for some breakthrough in
| demand. The urgency is not there: being complete before there
| is demand could be detrimental
| ksec wrote:
| >It's about demand isn't it?
|
| Yes. There used to be a saying the most expensive Fab ( or
| factory ) isn't the most advance Fab, but an _empty_ Fab.
|
| You cant built without first ensuring you can fill it, you
| cant fill it without first ensuring you can deliver. And
| Intel has failed to deliver _twice_ with their custom
| foundry. Both times with Nokia and Ericsson. How the two fall
| for it twice is completely beyond me, but then Intel are
| known to have very good sales teams.
|
| Intel will need another Apple moment that has huge demand,
| little margin, but willing to pay up front. On the assumption
| that Intel is even price competitive. The Apple modem may be
| it. But given the current situation with Intel as they want
| to lower Capital spending I am not even sure if betting on
| Intel is a risk Apple is willing to make. Comparing to a
| stable consistent relationship with TSMC.
| donavanm wrote:
| > On the assumption that Intel is even price competitive.
| The Apple modem may be it.
|
| Which is super interesting/ironic with the entire reason
| for an "apple modem" is due to Intels failure there a
| decade ago. Bonus irony for the subsequent acquisition.
| zitterbewegung wrote:
| Intel wasn't able to ship a competitive modem to Qualcomm
| and the whole point of the acquisition was to get rid of
| Qualcomm and even apple hasn't gotten a shipping version
| of a 5g modem for six years since the first intel modem
| started in 2018. This was really to vertically integrate
| the modem in all of the relevant Apple Silicon devices
| and it keeps going on...
| causality0 wrote:
| At this point I'm starting to wonder if Intel's corporate
| strategy is "pray all of the fabs in Taiwan are destroyed
| during a Chinese invasion".
| kayewiggin wrote:
| Then Intel is going to have to wait for a very long time.
| At best, China is currently in a scenario similar to
| Japan's lost decade of 30 years or US's Great Depression.
| At worst, China's current deflation + massive debt seems
| eerily similar to Weimar Germany's early internal
| devaluation. China is pretty fucked.
| philipov wrote:
| It's unwise to forget that the thing that pulled both the
| US and Germany out of the Depression was war.
| kayewiggin wrote:
| US fully recovered from Great Depression in 1939, 2 years
| before entering ww2. Weimar Germany started in 1918 and
| ended in 1933 at the beginning of nazi Germany, 15 years
| later.
|
| You can't start a war when you are truly broke, much like
| China is today. And China is aging super fast, unlike
| Germany or US during the 30s.
| Hikikomori wrote:
| IIRC you can add LG to the list of intel failures.
| amelius wrote:
| I don't get it. If TSMC has demand, then so could Intel. What
| am I missing?
| jazzyjackson wrote:
| TSMC makes nvidia GPUs and iPhone chips among other things,
| intel doesn't
| ninkendo wrote:
| There was some discussion awhile back about Intel
| potentially fabbing ARM chips (or any other custom
| non-x86 chip) as a viable business in the future. I don't
| know how serious they were but it sounded plausible when
| you think about how important it is to have an American
| leading edge fab, independent of the market future of the
| x86 ISA.
|
| Basically what would it take for Intel to still have
| Apple as a customer even if Apple made their own ARM
| designs...
| dehugger wrote:
| The missing bit is "TSMC makes better chips than Intel" and
| thus they have higher demand.
| amelius wrote:
| Yes, but then there should be a higher level of urgency?
| dehugger wrote:
| Urgency with what? You asked why TSMC has higher demand
| then Intel...
| amelius wrote:
| No, you have to read more of the thread to understand why
| I asked it.
|
| > TSMC have red hot demand, it's not hard to understand
| their urgency in setting up new fabs, wherever they may
| be. Intel don't have the same incentive (...)
| dehugger wrote:
| They set up a 3nm fab in the US in less than two years.
| That seems pretty urgent on TSMCs part...
| guipsp wrote:
| You might be missing that you cannot just "port" across
| fabs.
| amelius wrote:
| I honestly don't believe that e.g. Apple couldn't
| relatively easily base their designs on a different
| underlying technology.
|
| They do it all the time when they change nodes.
| wbl wrote:
| Drop another billion is sort of the name of the game
| here.
| IshKebab wrote:
| Why not? You might have to redo lots of phys work but
| essentially all of the RTL will be the same and that's
| the vast majority of the work.
|
| Intel doesn't have demand because they only make Intel
| chips, and they haven't been doing too well lately.
| wbl wrote:
| They feed into each other especially for anything that
| isn't a vanilla gate. Got a deeply ported SRAM with
| bypasses? That might fail synthesis if it is too choked
| by wire rules for the size of the cells so now it's
| banking time.
| IshKebab wrote:
| Right, you might get a different PPA...
|
| I think realistically you wouldn't port the exact same
| design between manufacturers. That would be a waste of
| money unless one manufacturer is _really_ rinsing you.
|
| More likely you'd switch manufacturers when you planned
| to switch process nodes anyway, in which case the
| increase in workload probably wouldn't be too bad.
| amelius wrote:
| I thought Xnm was just a marketing term and not related to any
| physical measurements? How are they going to legally enforce
| this if foundries can just change the naming convention?
| icf80 wrote:
| they are making "dies", they have to export them to china/taiwan
| to make the finals chips... as far as I understand it.
| ThinkBeat wrote:
| This proves that with sufficient political and military pressure
| and the ability to give away nearly unlimited amounts of money
| you can get production moved to the US in a way that works,
|
| (Any deals the US has with Taiwan, will always have a military
| backdrop, they just recently took deliver of some nice new
| military hardware). Stuff you will never see in Ukraine. )
| elzbardico wrote:
| What really baffles me is how Taiwans leadership can't see the US
| endgame with the CHIPs act and the Chinese sanctions. The US
| government wants to steal TSMC by using subterfuge, sheer force
| and malice, while making Taiwan paying for it by refusing the
| revenue of selling advanced chips to China. Not even TSMC should
| feel safe even if they successfully relocate themselves to the
| US. Buccaneering has a long tradition in Anglo-Saxon countries
| and as TikTok shows the US has no qualms in preaching free
| commerce, stable legal rules and all that bullshit to everyone
| else, while doing the most egregious mercantilist stuff without
| even an once of shame.
___________________________________________________________________
(page generated 2024-12-28 23:01 UTC)