How Huawei made a cutting-edge chip in China and surprised the US
- Thread starter JournalBot
- Start date
+1
It's the same everywhere. It's impossible to build any cutting edge infrastructure without huge state subsidies. One can always invent weird and wonderful things in their dad's garage but scaling to a tech giant is imposible without direct or indirect subsidies. And there's no state out there that won't protect whatever company they consider too big to fail or of strategic significance.
Hwawei is not just a very important tech giant. The chinese sate also wants to prove that sanctions don't work and are counterproductive. So they 're pouring in money for a strategic goal, exactly in the same way that any other state would do.
I 've only been to China twice and it's not what most westerners think it is. It's obviously a form of dictatorship (especially to minorities) but a) it's a benign one to the majority of the citizens and b) it's laser focused on developing the economy and tech. This second feature is actually impressive, even compared to the US. Chose any sector of their economy and you 'll find a throng of startups crowdfunded by VCs, municipal (regional goverments play a huge role in financing and attracting new business) and state sponsors. People in the west are often in awe by the chinese investments in Africa or S.America. They 're pennies to the amount of cash invested inside China. And most of it ain't state money.
The collective west is still underestimating China to our own peril.
edit typo
Last edited:
Upvote
6
(6
/
0)
Starting with the leadership here at the site, trickling down to the members there is a distinct "china bad" mentality. It doesn't matter that we have military bases surrounding china, it doesn't matter that we have 3 times the spending china does for our military it does not matter if we have 20 aircraft carrier to china's 5 china will always be bad here and usa good. Because somehow their abhorrent treatment of their minorities is worse than our own abhorrent treatment of our minorities and poor...
Just another distraction used by oligarchs in this country to diffuse any kind of opposition to their stranglehold on power/wealth.
Upvote
2
(5
/
-3)
macosandlinux
Ars Praefectus
This was a multi-page article when it was posted. Just got around to read it now, and it's been truncated. Was this (mistakenly) the full FT we are not allowed to see?
Upvote
0
(0
/
0)
China had made its goals known in 2015. The EU followed suit a couple of years later with the EU Processor Initiative. In both cases a reduction in strategic dependencies was the reason so your points are valid.
Sanctions, however, simply forced China and its tech industry to accelerate their efforts. Now it has put the pedal to the metal.
The knock-on effect was that sanctions cut off US suppliers from their biggest customers (Chinese companies). They even stopped US citizens working for many Chinese companies. They also impacted non-US interests around the world.
Sanctions blew a gigantic revenue hole into the semiconductor industry worldwide (excepting China).
Revenues are the lifeblood of R&D and future technologies. Without them, progress stalls.
Sanctions were never going to work. They could only stall things until solutions were put in place.
So, immediately, Huawei began a massive de-Americanisation effort in 2019. It is very, very likely that non-US companies around the globe also began de-Americanisation initiatives too.
At the time, 3-5 years was the estimated timeframe for newly designed products to reach the market.
The company has opened up new revenue streams (especially automotive and cloud)
This year Huawei has stated that it has successfully de-Americanised over 13,000 components.
It delivered MetaERP on schedule. A gargantuan effort. It shipped and developed HarmonyOS/Open Harmony and EulerOS.
In 2019 Huawei scaled up its investment projects in the semi-conductor toolchain, investing in over 40 companies as well as strengthening recruitment at HiSilicon.
It has invested in literally every step of the semiconductor design and manufacturing process.
SMIC went from 14nm to 7nm in a couple of years. No one thought it possible.
Not only that, the biggest wakeup call for the US administration was that no US company had 7nm fabrication capacity.
An analyst report stated that if China were to repurpose its existing production capacity to 7nm it would far exceed the capacity of Samsung and TSMC - combined.
The upshot is that Chinese companies have mopped up billions in investment (the revenues originally destined for US companies plus Chinese state investment) and have improved existing technologies and capacity.
Investment in research has also ballooned and there are even reports of a Chinese alternative to EUV already under construction (at least the structures to house the project). It seems that it has already moved beyond the theoretical stage.
If you look at everything that has happened since sanctions were imposed, I don't think 'jumpstarting' is out of place.
This article doesn't even mention Huawei's progress in supposedly making chipstacking a far more economical venture or the fact that Qualcomm stated in its last earnings call that it would not see any further material revenue from Huawei. Nor RISC-V and chiplet advance or photonics.
It seems clear that Huawei thinks it has 2024 pretty well covered in terms of capacity. Rumours also point to Kirins appearing in tablets, so yields don't seem to be an issue and on the Ascend front, reports stated that 60% of the Baidu order had already been delivered.
2024 will be interesting.
Upvote
5
(6
/
-1)
I'd love to read an actually Ars-quality story on this, one that digs into more detail and doesn't spend half the article explaining the basics of chip making.
I'd also like to see some well-researched insight related to the comments around theft versus development. As a number of folks have pointed out, there's no question that China has the ability and power to throw money at the problem and truly catch up to TSMC. The question in my mind is how fast. It seems to me a lot of this equipment is incremental improvement over years, combined with occasional leaps to new methods (DUV, EUV, etc). The new methods presumably take years of experiments to get things set up right, find the right materials, build tooling to precisely align mirrors/lenses, and numerous other factors I don't know about since I'm not in the space.
Sure, throwing money at the problem can speed things up to some degree (e.g., the Apollo program), but there are going to be diminishing returns, and even the Chinese goverment can only get so much speedup. My gut is that it they could be truly competitive with TSMC within 10-20 years, and somewhat sooner if they take the hit on yield. But if they can steal the results of someone else's years of experiments, that would let them skip ahead and save a few years.
I'd also like to see some well-researched insight related to the comments around theft versus development. As a number of folks have pointed out, there's no question that China has the ability and power to throw money at the problem and truly catch up to TSMC. The question in my mind is how fast. It seems to me a lot of this equipment is incremental improvement over years, combined with occasional leaps to new methods (DUV, EUV, etc). The new methods presumably take years of experiments to get things set up right, find the right materials, build tooling to precisely align mirrors/lenses, and numerous other factors I don't know about since I'm not in the space.
Sure, throwing money at the problem can speed things up to some degree (e.g., the Apollo program), but there are going to be diminishing returns, and even the Chinese goverment can only get so much speedup. My gut is that it they could be truly competitive with TSMC within 10-20 years, and somewhat sooner if they take the hit on yield. But if they can steal the results of someone else's years of experiments, that would let them skip ahead and save a few years.
Upvote
4
(4
/
0)
Firstly, China was already spending billions on semi companies long under China's MADE-IN-CHINA 2025. China can't do much without outside experts or equipments.
Upvote
0
(0
/
0)
Right. The US's sanction was too little too late. But considering that China had virtually no chip manufacturing prior to 2010, it would have taken China much longer to jumpstart their own organically.
There is no such home-grown alternative to ASML in China. China's most advanced litho maker is SMEE and they are still trying, after having announced their first 28-nm in 2020, to release their first litho system. Now that their key collaborators from Japan are no longer in it, I think it's safe to say it's almost as good as dead.
Last edited:
Upvote
1
(1
/
0)
Yep, China has been the largest buyer of new, used DUVs in recent years as China rushed to stockpile them before next sanction. Mong Sang Liang, former head of TSMC R&D and now co-CEO of SMIC, stated that SMIC's 7/5nm were complete back in 2020 and that their 3nm was just waiting on EUVS from ASML.
It's probably not that horrible. After all, MSL and his 200+ former TSMC engineers now working at SMIC had plenty of time to improve the yield.
Last edited:
Upvote
1
(1
/
0)
OrangeCream
Ars Legatus Legionis
I’m not negating anything, but I am trying to weed out racists.
Upvote
3
(3
/
0)
OrangeCream
Ars Legatus Legionis
Nope. I’m not saying we shouldn’t criticize China. They oppress their peoples, control their media, and kill minorities, just like us, and are just as worthy of criticism as we are. I just don’t want people to be racist. Don’t criticize China as being something we are not.
Upvote
3
(3
/
0)
There was probably some of that, but the key problem here is the US's lax export control. As late as 2022, BIS, the US Dept of Commerce, granted over $100+B in export licenses to China, 1/3 of them classified and to those blacklisted in foreign entity of concern for instance (and I'm not making this up).
China's numerical advantage means jack. As Wilde said, education is an admirable thing, but nothing worth knowing can be really taught.
Last edited:
Upvote
-4
(0
/
-4)
Why? The US's sanction against Tsinghua-Unigroup worked quite well. China's multi-billion DRAM aspiration collapsed.
it's more difficult to rein in on China's foundry ambition though, because unlike DRAM business which is largely dominated, guarded by the Korean memory chip makers, the foundry business us now dominated by Taiwan and there are already thousands of ex-TSMC engineers leaking tech to China -- eg, Mong Sang Liang, former head of TSMC R&D, now co-CEO of SMIC.
Last edited:
Upvote
-1
(0
/
-1)
My understanding is that there is no "figuring out" anything -- SMIC's 7nm is essentially a replica of TSMC's 7nm N+2 developed over a decade ago. When Mong Sang Liang, former head of TSMC's R&D's, went to SMIC, he also brough along 200+ engineers from Taiwan.
Upvote
-1
(0
/
-1)
There is hardly any evidence of "homegrown" stuff here. It's not to say that China can't learn, but that SMIC's 7nm is largely made possible by Mong Sang Liang, a former TSMC R&D head and 200+ ex-TSMC engineers he brought along years ago. SMIC's factory is also full of chip-making equipments imported from Japan, the US and the Netherland -- there is hardly any signs of "homegrown" stuff.
Sure, most key executive and/or engineers at TSMC have advanced degrees from top US universities and spent decades at various US tech companies learning their trade before returning to Taiwan to build what they now have. And TSMC was still really nobody until it became geopolitically important a few years ago. You don't just short-cut decades of cumulative learning in a few years and pretend you did something on your own.
Last edited:
Upvote
-1
(0
/
-1)
I see several comments about China catching up.
That is more complex than people think. Catching up is not about one technology it's about catching up with all kinds of very specialized components and techniques.
Take the machines from ASML, it's not one machine you got to figure out. There is a whole supply chain surrounding ASML with highly specialized and unique components and knowledge and know-how. The lenses in the EU machines for instance, only one supplier in the world was technically able to make those (they tried different ones).
Maybe China can catch up, in the long run, they have vast resources, but it won't be quick.
Upvote
1
(1
/
0)
It's not a matter of expertise, DUV cannot economically produce chips at 7nm and lower. There's a claim further up the thread of 30% but I'd trust EDN over that.
https://www.edn.com/the-truth-about-smics-7-nm-chip-fabrication-ordeal/
Upvote
0
(0
/
0)
Crying Croc
Ars Praetorian
`
My sentiments: China will catch up - in the long run. But no, it won't be quick.
Upvote
1
(1
/
0)
OrangeCream
Ars Legatus Legionis
Only if we let them, honestly. Which, given my previous comments about how we have our own fascist autocratic authoritarian orthodox extremist individuals in government, isn’t out of the question. They’re entirely too focused on controlling the population and culture to see the external threat and a meaningful way to address it.
Upvote
-1
(0
/
-1)
Anyone commenting that China could just find an alternative to EUV for 5nm and below nodes:
EUV has been in active development since the 90s.
Nanoprint lithography for nearly the same length of time, and it isn't remotely close to commercial viability in a mass production environment.
DUV was used at 7nm by every other foundry, so SMIC pulling it off is not surprising from a technical standpoint. It's surprising they had the DESIGN knowledge to do it, given the gap that has existed on that front. So I tend to think the illegal export of either parts or IP is a given in order for this chip to materialize
EUV has been in active development since the 90s.
Nanoprint lithography for nearly the same length of time, and it isn't remotely close to commercial viability in a mass production environment.
DUV was used at 7nm by every other foundry, so SMIC pulling it off is not surprising from a technical standpoint. It's surprising they had the DESIGN knowledge to do it, given the gap that has existed on that front. So I tend to think the illegal export of either parts or IP is a given in order for this chip to materialize
Upvote
0
(1
/
-1)
Catch up to where the rest of the world is now? Sure, in 10 or more years. Actually catch up to the moving target of technological progress? Never. Doing so would require cooperation with the rest of the world.
Upvote
-4
(0
/
-4)
Why would you assume that there aren't backdoors already? It would be far too easy to hide them in the massive complexity of modern equipment and far too advantageous to not be doing so for any national power with the resources. Also, it's something the U.S. was already doing way back in like the 1970s, even to allied nations:
https://www.washingtonpost.com/grap...ity/cia-crypto-encryption-machines-espionage/
(Mysteriously, this is a very hard article to bring up in search engine results!)
Upvote
4
(4
/
0)
OrangeCream
Ars Legatus Legionis
The secret of divorce is that the richer partner takes less of a hit than the less rich partner. Both need to 'replicate' what the other provided, but the richer partner already has more resources to begin with.
When you look at the outcome decades later, taking into account how compound interest works, the poorer partner (usually women) end up financially ruined:
https://slate.com/business/2016/03/...ement-crisis-especially-if-youre-a-woman.htmlThen the divorce happened. Prince says her ex-husband got almost all the retirement money. She got the family’s home—which she ultimately had to sell to pay her legal bills. What retirement money she had she ultimately cashed out to “pay off debt and buy groceries,” she told me recently.
Or you force them to spend all their money and time desperately trying to stay afloat:
Today, Prince, 49, is a clinical social worker—for which she had to return to school, using student loans to fund part of it. Frantic to make up for her losses, she puts $150 a month into her workplace retirement plan. It’s more than she can comfortably afford some months, she says. It’s also nowhere near enough. Her account balance is $5,000.
I think it's foolishly simplistic to think you've come to the correct conclusion. The point of the sanctions, embargoes, and export controls is to force China to overspend in order to stay relevant, let alone catch up, and per the article that's exactly what happened; chip yields of 30% means each chip by definition costs 3x what a similar chip would cost otherwise since 70% are useless. Likewise they're spending far more just to maintain 7nm production, let alone 5nm and 3nm.
Upvote
0
(0
/
0)
A balanced presentation is the death of sensation. Our media, being in the business of creating anxiety will never give us a balanced view. Nor any other media - even historians seldom present a balanced view, falling to their own political or social preferences.
We have to live in the world as it is.
Upvote
3
(3
/
0)
Sanctions worked. Just not the way they were intended to work.
Upvote
2
(3
/
-1)
OrangeCream
Ars Legatus Legionis
No, I think they've worked exactly as intended.
Remember how we caused the USSR to collapse by spending on our military, which impoverished them in an attempt to catch up?
Sanctions are just another form of economic warfare. If we can get NVIDIA 7nm AI chips for $1k, China has to spend $3k for 7nm chips. If the cost of 5nm doubles the cost again then if a 5nm AI chip costs us $2k for 3x the performance, it will cost China $12k.
Meaning for every $1b we are spending to simulate weather, China has to spend $12b.
Given how NVIDIA booked $18b in revenue, then attempting to create an equivalent value of GPU compute would be akin to spending $216b.
That's $200b not spent on education, healthcare, housing, or the military. That's $1t a year delta, 1/18th of their GDP not being used to grow their country. To put that in comparison, we spent 5% of our federal budget on Medicare or Education. Imagine having to slash both of those in half in order to invest into our semiconductor industry. It would be akin to us pumping $1.2t into Arizona to bring up a 5nm chip foundry, wafer manufacturing, and packaging facility. We're not even spending a tenth of that in Arizona, and we're going to end up with something (hopefully) of higher quality, yield, and performance at a fraction of the cost.
Upvote
4
(4
/
0)
Thanks for your info. Yet it is quite amusing that if these events are used as reasons to condemn China by the US government and some major international media while religious conflicts in other places of the world are killing people everyday.
And you've mentioned an important thing. Do average Chinese people complains about some policies made by the Chinese government? Of course. But for most of the time, they are just complaints, not hatreds. The majority of the Chinese people clearly knows that the overall benefits they received from the government in the past few decades far exceed the inconvenience brought by it when compared with almost all other developing countries. Yet there're anti-China propagandas that keep exaggerating such complaints and make people believe that the Chinese people hate their government all the time and ready for an uprising whenever there's a chance. What a joke.
Upvote
2
(2
/
0)
See? There are still idiots who believe China kill minorities. Where are the photos or videos of people getting killed? Or piles of dead bodies? Or mass graves?
Upvote
2
(2
/
0)
OrangeCream
Ars Legatus Legionis
Is this a joke? Sarcasm? Are we ignoring the Uyghurs? Tibet? The difference is we quiet-kill our black and brown people under the guise of police jurisdiction.
Upvote
0
(2
/
-2)
The advantage of having a leadership model for a long period of time, with a set unwavering core values is the ability to take a vision for decades and centuries.
Hence in the west leaders/political parties only have about 3 years where they can do anything. The 1st year is spent unwinding the previous administrations policies (usually the ones that were better ones too). 3 years implementing election pledges/pandering to there base. 1 year of election mode.
It's not really a model for progress for countries which exist potentially for millennia, like China has.
So yes, its definitely changed the trajectory of development for chip manufacture in China. But the goals are not going to change.
Western economies are probably going to have to get back to war probably the way its going. It's what worked for them in the past.
Hence in the west leaders/political parties only have about 3 years where they can do anything. The 1st year is spent unwinding the previous administrations policies (usually the ones that were better ones too). 3 years implementing election pledges/pandering to there base. 1 year of election mode.
It's not really a model for progress for countries which exist potentially for millennia, like China has.
So yes, its definitely changed the trajectory of development for chip manufacture in China. But the goals are not going to change.
Western economies are probably going to have to get back to war probably the way its going. It's what worked for them in the past.
Upvote
1
(1
/
0)
OrangeCream
Ars Legatus Legionis
I think you’ve misunderstood how ‘the western economies’ work. You’re coming from the assumption that the government dictates progress, which is false. Biden doesn’t dictate how Apple or NVIDIA develop, nor how AMD and Intel compete. Biden, hopefully, sets policies in place where others choose what the future looks like. It wasn’t a US President that decided NVIDIA would focus on AI and that OpenAI would develop an amazing LLM, or that Apple would design some of the world’s most competitive CPUs, smartphones, and wearables. Nor was it a US President that led to Tesla jump starting the EV industry, either. They didn’t create SpaceX, or the drive towards self driving cars.
The US system largely rewards corporations who come up with good implementations of good ideas to collect lots of money. They can set policy, but don’t go out of their way to prevent, for example, AT&T, T-Mobile, Sprint, and Verizon from building separate wireless networks. They allow for the parallel and simultaneous development of geothermal, hydropower, solar, wind, natural gas, coal, nuclear, and biomass plants, as well as letting them fail when it turns out they aren’t good enough.
So we designed the system that the long term thinkers are the CEOs of corporations, who get to lead their companies for decades, and not the president. We set it up so power is split up between the church, the government, and the corporations, accepting the inefficiencies because it apparently works better.
Now if China’s system is even better, then great. It might very well be a better system. We will see, eventually.
Upvote
0
(1
/
-1)
I think it's pretty apparent that the US system is better overall, unless Trump gets the chance to destroy it.
Upvote
-1
(0
/
-1)
OrangeCream
Ars Legatus Legionis
I think so too, but we’ve only existed for a little over 200 years. I’m talking about giving it another 500 years.
Upvote
3
(3
/
0)
I think the big impact is going to be access to higher education. That's going to be the determining factor.
as it stands now the US access to education is restricted by cost. Not the case in the PRC.
As always what is visible and what they have under the covers is going to be 2 very different things.
It's a good thing to have more competition to spur innovation. Well I think it is.
as it stands now the US access to education is restricted by cost. Not the case in the PRC.
As always what is visible and what they have under the covers is going to be 2 very different things.
It's a good thing to have more competition to spur innovation. Well I think it is.
Upvote
1
(1
/
0)
Many commenters want to declare success or failure for the sanctions -- they are not an end in themselves, but are part of a process. Review the cold war with the Soviet Union (which was even more insular than after-Mao China), and you find them figuring it out themselves, stealing it via espionage, or buying it from non-aligned nations...and typically all 3 at once. If they want it bad enough, they get there. But sanctions and other economic restrictions slow progress, impose extra costs, separate friends from the others, and highlight weaknesses. Ultimately they are a negotiating tool.
Upvote
1
(1
/
0)
Worked once at a company that worked with lots of foundries. When we did designs for SMIC, we'd run the designs through the TSMC designrule checkers at least once. The processes were almost identical. Except TSMC's design checks were more torough. Similar for simulations. TSMC models were more accurately predicting the behavior of SMIC devices. Rumors said that a lot of Chinese people worked at TSMC for a while and then moved over to SMIC. SMIC argued that their design rules and process matched that of TSMC so much so that you are not locked in to a single foundry. It was not uncommen back then. (Decade ago?)
Upvote
0
(0
/
0)
But what about post silicon? How far away are we from a generational leap? And can an underdog leapfrog to the front?
That is relevant to the cutting edge, but well over 90% of the world's current silicon production is on much older and mature nodes, and it is those nodes which make the world go round.
Last edited:
Upvote
0
(0
/
0)
And curiously, Cuba even managed to carve out some very impressive results in a very technologically advanced area. Bio and Pharma tech.
Upvote
0
(0
/
0)
there is nothing in the pipeline that will replace silicon anytime soon - they keep working around the issues withe new processes, plus enhanced dopants
by the time it does happen, it will take 20x the Chinese gdp to even keep-up with the rest of the worlds way-advancement research facilities -if you think euv is hard, just wait and see how much harder it is to upgrade all these fab, and redesign tools with thousands of prebuilt-premodeled gates for thousands of design houses dependent on them.
Last edited:
Upvote
0
(0
/
0)
The 5nm chip in the Huawei laptop appears to be TSMC, right before the embargo.
Huawei Teardown Shows 5nm Chip Made in Taiwan, Not China
Huawei Teardown Shows 5nm Chip Made in Taiwan, Not China
Upvote
0
(0
/
0)
China makes all our electronic devices almost thanks to greed of our multinationals they knowhow is getting better but i don't see any USA/UK/European companies getting out of China for god sake some still doing business Russia
Upvote
0
(0
/
0)