I'm out of my depth here, but I wonder that, since microchips are finding their way into all manner of places they shouldn't (e.g. Shahed drones and Russian missles), how hard would it be for batch identifiers to be made into the silicon. They can then be cross-ref'd to clients.
I know instances of Easter eggs finding their way into chips back in the day. Feels possible.
This is often done already. In fact I wouldn't be surprised if this is how they track things down. Silicon die usually have some burnable fuses that can be set once and at some point during manufacturing the wafer lot number and even XY coordinates of the die on the wafer are recorded. It helps track down yield issues.
Each chiplet will itself have up to 100,000 transistors and include a two-way radio, data encryption engine and way to detect tampering—all while consuming under 50 microwatts (50 millionths of a watt) and costing less than one penny each. Identifying information on each dielet would be read using a penlike probe plugged into a smartphone.. a dielet would be inductively powered by the probe, which would communicate via radio frequency signals when placed within a half millimeter of the chiplet. The probe would relay encrypted information to an app on the smartphone.
I wonder if it was possible to build chips in a way that they could disable themselves if used in a way not approved of by the manufacturer. Perhaps a set of pre-generated keys that have preset validity times. You would then need to contact the manufacturer to receive the next key in the set.
Probably too much effort for the gain though. For military chips maybe?
Then the manufacturer becomes a military target, since destroying their authentication infrastructure has the effect of disabling operational military hardware.
I'm sure all your enemies would love for the US to add remote-disable to all the weapons.
On the other hand remote hardware disable being already in all military equipment the US sells to allies, just in case they ever change their mind, would be a cool conspiracy theory.
All except for the cheapest microcontrollers (everything above half dollar) have serial numbers (unique identifiers) burned into one time programmable memory.
There is probably multiple ways to identify who the client is for a particular chip. The problem is that you'll whitelist clients instead of blacklisting them. This will affect business.
Physical identifiers on the chip to identify the company of purchase isn't going to help the US government track you, they have far more useful methods.
In a global trading environment, sanctions regimes are an added cost to procurement you impose on adversaries, not an impenetrable denial of materials. Huawei can't cost compete if it has to smuggle chips constantly.
> Huawei can't cost compete if it has to smuggle chips constantly
It also trashes their R&D. Imagine if every LLM researcher in America had to smuggle in Iranian cores to do any work.
One of the saddest stories of the 21st century is China’s decline into despotism. They didn’t have democracy, but they had internal political competition and a peaceful transfer of power. In an alternate world, Xi is comfortably retired while China and America drive a new century of human prosperity.
In the 21st century, China's per capita GDP has grown about 2.5x as much as that of democratic India, which is also the only other country of comparable population.
They're not smuggling the cores for usage. They're smuggling to reverse engineer, and the Chinese are much better (even if more willing) at it than Americans. You are also discounting that China has driven a new century of human prosperity for their country and globally through highly efficient manufacturing that has proven tough to replicate.
Is "highly efficient manufacturing" a euphemism for low paid labor and slave labor?
As far as I know pretty much any product that moved production to China did it because of low labor costs. It's also been reported that companies have also exited China into other low paid labor countries in the region.
Meanwhile the only factories that return to the US are exactly the ones that invested in automation and efficient manufacturing, as it's the only way to beat the low costs.
They've also managed to pump out really cheap solar panels and EVs, but it's not clear to me how much of that is upgraded automation vs government subsidies.
How many IC samples are required to do a layer-by-layer reverse engineering of something at single-nm feature sizes?
Ok, now you've got "something" here even resembling a netlist (lol), what do you do with it? Keep in mind: you don't actually have the manufacturing technology to use this IP. By the time you do, the IC you reversed is no longer state of the art.
I think state-sponsored corporate espionage would be a lot more fruitful and cheaper.
Reverse engineering is the easy part. China still lacks EUV lithography manufacturing technology. They'll figure it out eventually but the point is to stay a generation ahead. Just like China can manufacture turbine engines but only ones equivalent to what we had some years ago.
I think China does not need cutting edge of everything to be successful. If they build a nuclear power plant that has turbines from 5 years ago they are still going to be ok. The question what kind of power plant to build is much more important.
That sounds like it'd be something of a non-issue. If a GPU researcher wants a GPU someone will be willing to smuggle it out for them. Like GP says, a sanctions wall is penetrable. The problem would be more that leading researchers at something like China's OpenAI can't procure a billion dollars worth of GPUs for industrial research.
Although I do want to endorse the idea that their political system is probably going to undermine the successes of their businesses. From the outside it looks like a disaster waiting to happen.
The smuggling increases costs though. They may have only so much budget - perhaps the sanctioned-country researcher could only afford an H100 now instead of an H200 if you had excluded the smuggling
> their political system is probably going to undermine the successes of their businesses
They had a lot of problems but power was still highly decentralized (a lot of decisions were left to local and regional party bosses). Human rights were never especially high on the list of priorities but you tended to see the government allow for a degree of experimentation that produced a lot of innovation and led to a vibrant culture.
Things are very different now with a one-man cult of personality.
I will add that costs are imposed not only on adversaries, but on everyone. Between reporting requirements, economic sanctions, anti-money-laundering and anti-tax-evasion provisions, perfectly legitimate businesses are probably losing close to a trillion dollars a year on compliance and lost opportunities.
> there is a functioning system in place where companies request permission to sell certain products to China, and the DoC approves exports of products that offer lower performance but still belong to the latest generation of technologies developed in the U.S.. the idea of a strict ban on advanced technology exports to China does not match reality.. "There are Swiss cheese holes in it. And right now, chipmakers are driving cars through them."
> Research firm TechInsights recently discovered that Huawei’s latest AI servers contained processors made by TSMC, Nvidia Corp.’s most important manufacturing partner.
So they cut off Nvidia? Or do Nvidias customer order chips directly from TSMC and it's one of those clients?
It would be funnier if it was Nvidia, but probably not.
This is a strange time to be imposing chip sanctions, given AI isn’t at the forefront at real scale yet.
They’re still able to access compute now, albeit at a higher cost. But in the future when scale and margins are much more relevant than today, that won’t be the case.
Market forces to develop IC tech (lithography, foundries, patents, etc..) inside China are surging. So by the time they really need quantity, it’s much more likely they’ll have a domestic high end IC supply chain implemented.
Obviously nobody could really predict the timelines with good certainty, but it seems like a good development for them long term.
TSMC produced in 2023 16 million wafers, apples die size is about 105mm2 which fits around 230 chips per wafer... say its 200 good chips... thats 3.2 billion chips.
Aren't the chips in question used for AI accelerators rather than phones? Moreover for phones SIMC is probably competitive at the mid end range, which makes up most of their sales.
If you have 10 y/o tech with todays code and algorithms, surely you could modify and improve 3G and get better speeds / quality? My iPhone 6S drops to 3G all the time.
As long as you cut all the bloat (i.e tracking code in html..).
At the end of the day it all depends on the person and their skill level using the tech.
China is not that far behind and likely to catch up. The entire US STEM workforce is how many STEM graduates China has per year. The US sanctions is just a stumbling block and will result in a stronger China down the line, with a completely independent supply chain.
By this logic, why doesn't the US or China have their own equivalent of ASML or TSMC? Number of STEM grads means little if the quality is low. See other developing countries in South/East Asia that pump out lots of low quality engineers.
The quality is no longer low. That was true 10 years ago. Since then China has made significant progress and is on the forefront.
China has their own ASML and TSMC, they are just behind. But not as far as the west likes to think.
The problem with US is that Google and finance pay engineers too much. They get the most talent so they can show you better ads and make better trading algos. In the US the government works against you, whereas in China the regulation burden is none.
>See other developing countries in South/East Asia that pump out lots of low quality engineers.
South Asia sure, large quantity, low (average) quality.
East Asia if you include Japan and Korea has pretty good engineering talent, I'd say, lots of people would run circles there around US graduates, it's more that their economies are hampered in other ways by demographic shifts, bad investing environment (Japan mostly), and monopolies that don't innovate and control the government (mainly Korea).
Chinese engineering is pretty damn good now. But it takes time still to catch up to the highest levels of technology, but they have a government investing in its people, in its infrastructure, in its companies. The only question is whether the government will run out of money before the plan unfolds or the demographic time bomb hits.
Japan, Korea, and Taiwan are no longer considered developing countries. I was being very specific about STEM grads in developing countries. From first hand experience, yes, there are "diamonds in the rough", but the average/median STEM grad is much lower quality from South/Southeast/East Asian developing countries _compared_ to their developed counterparts.
Not to mention a ton of the US STEM workforce is not American / is Chinese origin. when the incentives align not only will they leave, they will also take all their learnings with them.
Frankly, the US government should be doing more at this point. All departments, not just legislative or executive. Defense, intelligence, state, education, energy..
It is not unbelievable for the race to AGI (or whatever close enough approximation of it is necessary for dominance) to possibly be a "first to the posts wins forever" sort of race.
We can't even secure our key technologies. Security at TSMC being a consideration, but not just them.
The US is allowing NVIDIA to keep selling strategic assets to whoever can afford them, like they're selling mining picks in a gold rush.
The legislation to restrict their sales to China was almost laughable. NVIDIA actively worked to do the bare minimum to comply with it. China is still a key market for them. I've seen projections of $12bn in sales of AI chips to China (though this is hearsay from financial news sites).
Is that really a long term profit motive, pissing off the US government who will inevitably investigate their monopoly in the space one day. Good monopolies, like the cable companies that enable mass surveillance on the population, get to entrench to the point that no level of incompetence can unseat them. I'm sure they could use a little "AI safety" legislation that only their graphics cards can provide.
Or is NVIDIA's motive political neutrality? Hedge their bets and hope they can balance on the tightrope long enough for a few political cycles to make the US forget, while avoiding alienating China due to the possibility China will eventually win the race?
There are constant reports of industrial espionage, both within the US and all along our critical supply chain internationally. Admittedly, it's a lot easier to steal from an open society than a closed, homogeneous one. But where are the armies of advisors to assess critical infrastructure or even help assess threats on an ongoing basis. We should _almost_ be treating our and our allies high tech industrial infrastructure like they are weapon and munitions factories in a wartime environment.
But assuming we're still at the start of the race, and not quite at that level yet -- there doesn't seem much fostering of future development going on to properly position the US for success. Specifically, development of companies and of future talent.
There's no industrial knowledge-share program (think, 'open source initiative for critical algorithms and tech, but only for allies, and only for those who can meet security criteria'). There's no equivalent of anything like the Manhattan Project. AGI _will_ have the same, if not higher, social and political impact than the Manhattan Project.
Instead we've got... DARPA, running years behind the curve as always. They're not an organization to win a race.
As far as future talent goes.. There are no special schools or training programs to develop talented youth into important contributors in the space. We can't even get autistic savants through school or into those places they'll potentially thrive or work magic in. Particularly if they happen to have been born poor or can't properly navigate a social system.
Even above-average, over-achieving kids aren't being incentivized to focus their development into areas of strategic interest. We're leaving our talent pool to the market, which operates way too late to properly foster early development, and leaving too much early development up to broken or ineffective educational institutions.
China has a much larger pool of talent to develop. They have a more directive government with long term planning capacity. They have a much stronger industrial base that is only lacking some key technologies. They've got enough electrical power, or at least capacity to develop it, to run all the hardware. They have much better industrial espionage and cyber capabilities.
We have.. a short lead in knowledge. Industrial deals with key allies that are dependent on a dollar that is frankly at risk of collapse any year now with no plan in sight for getting national debt to where we could conceivably be able to pay off _the interest_ on it. A few individual talents pushing forward the technology. The willingness of billionaires to open their wallets for a ton of useless things in hope of striking a rich vein of something salable. A political system which will reward said billionaires for that investment by not taxing their profits towards a social safety net for those who will inevitably be displaced by more productive technologies.
To simplify that, almost the only thing we have driving the bus at this point is corporate greed. What kind of AGI will come out of that?
Will it be better than an over-controlling one which is personally and politically manipulative, that is feared will come out of China?
> It is not unbelievable for the race to AGI (or whatever close enough approximation of it is necessary for dominance) to possibly be a "first to the posts wins forever" sort of race.
The winner of that race is AGI, not the hairless apes that happen to be near the geographic region where it was born.
Neither the AGI nor its creators get awarded the throne of world leadership or however this is supposed to work, because the AGI is just like a person and the whole thing is an anticlimax, although a scientific triumph. In fact the AGI is less useful than a regular AI, because it's inclined to be moody and lazy.
Homogeneous? Mainland China is anything but. It is wildly diverse with regards to ethnicities/tribes/etc, similar to India. How does this myth persist in 2024?
I am not sure if I understand. You have a country on Earth where the vast majority of people is one kind and happily co-exists with 56 minorities.
Can you point me to another country (the size does not even matter) where this happens like this?
> Soon after the establishment of the People's Republic of China, 39 ethnic groups were recognized by the first national census in 1954. This further increased to 54 by the second national census in 1964, with the Lhoba group added in 1965. The last change was the addition of the Jino people in 1979, bringing the number of recognized ethnic groups to the current 56. The following are the 56 ethnic groups (listed by population) officially recognized by the People's Republic of China.
> We can't even secure our key technologies. Security at TSMC being a consideration, but not just them.
TSMC is not yours, is Taiwan sovereign or not? According to you, it's not, but when China claims it, it suddenly becomes sovereign? very conveniant
> There are constant reports of industrial espionage, both within the US and all along our critical supply chain internationally. Admittedly, it's a lot easier to steal from an open society than a closed, homogeneous one. But where are the armies of advisors to assess critical infrastructure or even help assess threats on an ongoing basis. We should _almost_ be treating our and our allies high tech industrial infrastructure like they are weapon and munitions factories in a wartime environment.
I don't support the idea of a war. I do support the idea of the US improving it's security posture, and improving its own possibility of success.
Make getting into and out of key industrial facilities more difficult for the unauthorized, both physically and digitally. Enhanced background checks on employees who have access to critical technology.
Expand those security requirements to international technology partners. If it's a race, incentivize partners to support our interests.
China does all of these things. If you would consider the US bad for doing so as well, how would you consider China for being better at it?
Nobody is suggesting that we invade Chinese factories. But suggesting we secure our allies industrial facilities is somehow 'engineering support for a war with China'?
You would have the US just.. give Taiwan over to China, against the will of Taiwan's own people, and against US financial and strategic interest? Or else _we're_ the bad guys if China invades?
> You would have the US just.. give Taiwan over to China, against the will of Taiwan's own people
According to this article, Taiwanese still want to do business with China, so your suggestion is to prevent Taiwanese to do business with China, against their will? wait, something doesn't sound right, i can't pin point it just yet!
> Expand those security requirements to international technology partners. If it's a race, incentivize partners to support our interests.
Well, again, according to this article, Taiwanese wants China as partners, at a least as business partners
You keep contradicting yourself.. oh wait...
> incentivize partners to support our interests
oh right! your interests above theirs! now it all make sense! thanks
You're ignoring key points and cherry picking what to respond to based on your ability to try and twist my words into statements I never made.
And you're extrapolating what must be a very obscure reference in the article into some kind of supporting evidence for your argument. All to support your words against, apparently, some perceived made-up position that you think I have.
According to this article, Huawei is barred from doing business with TSMC, and when Huawei violated the sanctions through a third party, TSMC cut off that third party and reported them.
Incentivizing partners is NOT 'your interests above theirs'. If the incentive is not good enough, they don't have to take it. If the incentive is good, then both your interests and theirs align. Neither is 'above' the other.
Now, if you steal by violating sanctions, and you wargame a blockade of their territory, you are being a threat. You are NOT offering them more of what they want until they happily offer you what you want. Instead, you are pushing them to act based on fear of you. THAT is 'your interests above theirs!' style of diplomacy and business negotiation.
I can tell you're not approaching this conversation with any form of sincerity in your arguments. I won't speculate about why you would do this. But I'm just not going to engage anymore.
"I think we should look at this from the military point of view. I mean, supposing the Chinese stashes away some big cluster, see. When they come out in a ten years they could take over!"
"I agree, Mr. President. In fact, they might even try an immediate sneak attack so they could take over our AI space."
"Yeah. I think it would be extremely naive of us, Mr. President, to imagine that these new developments are going to cause any change in Chinese expansionist policy. I mean, we must be... increasingly on the alert to prevent them from taking over other AI space, in order to train more prodigiously than we do, thus, knocking out our models in superior exaFlOps when we emerge! Mr. President, we must not allow... an AGI gap!"
I'm out of my depth here, but I wonder that, since microchips are finding their way into all manner of places they shouldn't (e.g. Shahed drones and Russian missles), how hard would it be for batch identifiers to be made into the silicon. They can then be cross-ref'd to clients.
I know instances of Easter eggs finding their way into chips back in the day. Feels possible.
https://en.wikipedia.org/wiki/Chip_art
This is often done already. In fact I wouldn't be surprised if this is how they track things down. Silicon die usually have some burnable fuses that can be set once and at some point during manufacturing the wafer lot number and even XY coordinates of the die on the wafer are recorded. It helps track down yield issues.
https://www.scientificamerican.com/article/the-pentagon-rsqu...
https://archive.md/GTRrY
I wonder if it was possible to build chips in a way that they could disable themselves if used in a way not approved of by the manufacturer. Perhaps a set of pre-generated keys that have preset validity times. You would then need to contact the manufacturer to receive the next key in the set.
Probably too much effort for the gain though. For military chips maybe?
You really want hardware DRM to support your political ideology? Good thing this sort of thing has never been abused!
Then the manufacturer becomes a military target, since destroying their authentication infrastructure has the effect of disabling operational military hardware.
I'm sure all your enemies would love for the US to add remote-disable to all the weapons.
On the other hand remote hardware disable being already in all military equipment the US sells to allies, just in case they ever change their mind, would be a cool conspiracy theory.
imagine your phone turning into a rock because you loaded an image or song in your browser. Is that the future you want to have?
All except for the cheapest microcontrollers (everything above half dollar) have serial numbers (unique identifiers) burned into one time programmable memory.
ARM M0 cores at 9 cents have unique IDs burned into flash OTP typically in my experience.
Anyone remember the Dallas Semiconductor DS2400/DS2401
https://www.analog.com/media/en/technical-documentation/data...
1-Wire, unique 48-bit ID. I don't remember the year it was introduced, or cost.
There is probably multiple ways to identify who the client is for a particular chip. The problem is that you'll whitelist clients instead of blacklisting them. This will affect business.
Chips getting to China seems like a purely political issue to me.
That’s like saying you don’t care what the business side of your company does, you just want to write code.
Well, if your company goes bankrupt because engineering didn’t work with the product guys you don’t have a job anymore.
You can probably engrave with a laser. That art is fixed.
But I don't think we need more ways for the US gov. to track us.
Physical identifiers on the chip to identify the company of purchase isn't going to help the US government track you, they have far more useful methods.
Just becouse they have far more useful methods I am not very keen on another one.
There would have to be some 'know you customer' type of tracking for ordering from electronics suppliers to avoid liability. Like with banks.
DigiKey rejecting your order because who knows what their black box said.
It would hardly help versus people that can afford front buyers anyway.
And even if it would work the microchips still end up in bombs dropped on children since the gov. hand them over willingly in the first place.
In a global trading environment, sanctions regimes are an added cost to procurement you impose on adversaries, not an impenetrable denial of materials. Huawei can't cost compete if it has to smuggle chips constantly.
> Huawei can't cost compete if it has to smuggle chips constantly
It also trashes their R&D. Imagine if every LLM researcher in America had to smuggle in Iranian cores to do any work.
One of the saddest stories of the 21st century is China’s decline into despotism. They didn’t have democracy, but they had internal political competition and a peaceful transfer of power. In an alternate world, Xi is comfortably retired while China and America drive a new century of human prosperity.
> Xi is comfortably retired while China and America drive a new century of human prosperity
They told the same during the Hu era and they will write the same to Xi successor. The enmity is structural.
In the 21st century, China's per capita GDP has grown about 2.5x as much as that of democratic India, which is also the only other country of comparable population.
They're not smuggling the cores for usage. They're smuggling to reverse engineer, and the Chinese are much better (even if more willing) at it than Americans. You are also discounting that China has driven a new century of human prosperity for their country and globally through highly efficient manufacturing that has proven tough to replicate.
> highly efficient manufacturing
Is "highly efficient manufacturing" a euphemism for low paid labor and slave labor?
As far as I know pretty much any product that moved production to China did it because of low labor costs. It's also been reported that companies have also exited China into other low paid labor countries in the region.
Meanwhile the only factories that return to the US are exactly the ones that invested in automation and efficient manufacturing, as it's the only way to beat the low costs.
They've also managed to pump out really cheap solar panels and EVs, but it's not clear to me how much of that is upgraded automation vs government subsidies.
How many IC samples are required to do a layer-by-layer reverse engineering of something at single-nm feature sizes?
Ok, now you've got "something" here even resembling a netlist (lol), what do you do with it? Keep in mind: you don't actually have the manufacturing technology to use this IP. By the time you do, the IC you reversed is no longer state of the art.
I think state-sponsored corporate espionage would be a lot more fruitful and cheaper.
Not in this case, the chips were detected in Huawei products
A company probably needs a couple of 100s chips to reverse engineer. Not much.
Reverse engineering is the easy part. China still lacks EUV lithography manufacturing technology. They'll figure it out eventually but the point is to stay a generation ahead. Just like China can manufacture turbine engines but only ones equivalent to what we had some years ago.
I think China does not need cutting edge of everything to be successful. If they build a nuclear power plant that has turbines from 5 years ago they are still going to be ok. The question what kind of power plant to build is much more important.
That sounds like it'd be something of a non-issue. If a GPU researcher wants a GPU someone will be willing to smuggle it out for them. Like GP says, a sanctions wall is penetrable. The problem would be more that leading researchers at something like China's OpenAI can't procure a billion dollars worth of GPUs for industrial research.
Although I do want to endorse the idea that their political system is probably going to undermine the successes of their businesses. From the outside it looks like a disaster waiting to happen.
The smuggling increases costs though. They may have only so much budget - perhaps the sanctioned-country researcher could only afford an H100 now instead of an H200 if you had excluded the smuggling
> their political system is probably going to undermine the successes of their businesses
I do agree with this
They had a lot of problems but power was still highly decentralized (a lot of decisions were left to local and regional party bosses). Human rights were never especially high on the list of priorities but you tended to see the government allow for a degree of experimentation that produced a lot of innovation and led to a vibrant culture.
Things are very different now with a one-man cult of personality.
China was a one-party state where party elections actually mattered.
Free speech on political topics was still limited, e.g. about Tiananmen Square.
I will add that costs are imposed not only on adversaries, but on everyone. Between reporting requirements, economic sanctions, anti-money-laundering and anti-tax-evasion provisions, perfectly legitimate businesses are probably losing close to a trillion dollars a year on compliance and lost opportunities.
https://archive.ph/wCreq
Step right up to the Silicon Pillow Fight™
Dec 2023, https://www.tomshardware.com/tech-industry/manufacturing/usc...
> there is a functioning system in place where companies request permission to sell certain products to China, and the DoC approves exports of products that offer lower performance but still belong to the latest generation of technologies developed in the U.S.. the idea of a strict ban on advanced technology exports to China does not match reality.. "There are Swiss cheese holes in it. And right now, chipmakers are driving cars through them."
Hopefully the Linux kernel does not become theatrical collateral damage, https://news.ycombinator.com/item?id=41919670
Watch the linux foundation eventually move to Switzerland like the RISC-V foundation did
And then watch a parallel structure emergence from China that usurps it all.
> Research firm TechInsights recently discovered that Huawei’s latest AI servers contained processors made by TSMC, Nvidia Corp.’s most important manufacturing partner.
So they cut off Nvidia? Or do Nvidias customer order chips directly from TSMC and it's one of those clients?
It would be funnier if it was Nvidia, but probably not.
This is a strange time to be imposing chip sanctions, given AI isn’t at the forefront at real scale yet.
They’re still able to access compute now, albeit at a higher cost. But in the future when scale and margins are much more relevant than today, that won’t be the case.
Market forces to develop IC tech (lithography, foundries, patents, etc..) inside China are surging. So by the time they really need quantity, it’s much more likely they’ll have a domestic high end IC supply chain implemented.
Obviously nobody could really predict the timelines with good certainty, but it seems like a good development for them long term.
> They’re still able to access compute now, albeit at a higher cost.
Perfect is the enemy of good.
If massive resources must be diverted from R&D and into evading sanctions, it's harder to catch up as quickly or cheaply.
Meanwhile: TSMC and the collective West are also making significant strides.
Hauwei ships 12 million phones per quarter [1]. If even a small percent of them included these chips, it should be pretty damn obvious no?
[1] https://huaweicentral.com/huaweis-global-smartphone-shipment...
12 million, say its 1% thats 120k phones.
https://www.tsmc.com/english/dedicatedFoundry/manufacturing/....
TSMC produced in 2023 16 million wafers, apples die size is about 105mm2 which fits around 230 chips per wafer... say its 200 good chips... thats 3.2 billion chips.
3.2B to 120k is a rounding error.
Aren't the chips in question used for AI accelerators rather than phones? Moreover for phones SIMC is probably competitive at the mid end range, which makes up most of their sales.
Related:
TSMC told US of chip in Huawei product after TechInsights finding, source says
https://news.ycombinator.com/item?id=41917990
Hauwei back to 10yo old tech now that it can’t pass off TSMC tech as its own.
Is there evidence of this
If you have 10 y/o tech with todays code and algorithms, surely you could modify and improve 3G and get better speeds / quality? My iPhone 6S drops to 3G all the time.
As long as you cut all the bloat (i.e tracking code in html..).
At the end of the day it all depends on the person and their skill level using the tech.
In fairness it can be hard for a vendor / platform like Huwawei to cut the bloat, because so much of it sits at the application layer.
Users will blame you if Jira works more slowly on your devices than on Apple’s.
China is not that far behind and likely to catch up. The entire US STEM workforce is how many STEM graduates China has per year. The US sanctions is just a stumbling block and will result in a stronger China down the line, with a completely independent supply chain.
By this logic, why doesn't the US or China have their own equivalent of ASML or TSMC? Number of STEM grads means little if the quality is low. See other developing countries in South/East Asia that pump out lots of low quality engineers.
The quality is no longer low. That was true 10 years ago. Since then China has made significant progress and is on the forefront.
China has their own ASML and TSMC, they are just behind. But not as far as the west likes to think.
The problem with US is that Google and finance pay engineers too much. They get the most talent so they can show you better ads and make better trading algos. In the US the government works against you, whereas in China the regulation burden is none.
Correction: in China the regulation burden is none until they disappear your CEO.
>See other developing countries in South/East Asia that pump out lots of low quality engineers.
South Asia sure, large quantity, low (average) quality.
East Asia if you include Japan and Korea has pretty good engineering talent, I'd say, lots of people would run circles there around US graduates, it's more that their economies are hampered in other ways by demographic shifts, bad investing environment (Japan mostly), and monopolies that don't innovate and control the government (mainly Korea).
Chinese engineering is pretty damn good now. But it takes time still to catch up to the highest levels of technology, but they have a government investing in its people, in its infrastructure, in its companies. The only question is whether the government will run out of money before the plan unfolds or the demographic time bomb hits.
Japan, Korea, and Taiwan are no longer considered developing countries. I was being very specific about STEM grads in developing countries. From first hand experience, yes, there are "diamonds in the rough", but the average/median STEM grad is much lower quality from South/Southeast/East Asian developing countries _compared_ to their developed counterparts.
Not to mention a ton of the US STEM workforce is not American / is Chinese origin. when the incentives align not only will they leave, they will also take all their learnings with them.
Taiwan and China's economies heavily rely on each other. I won't be surprised if Taiwan companies are still sending chips or technologies to China.
There is no sanction preventing Taiwan from trading with mainland China. In fact it's their largest export market.
Once there, those chips are used to build Western-branded products, Chinese-branded products and some, I'm sure, are re-exported to Russia.
The chips go to China regardless since most US technology is assembled in China.
how dare you resell our products to someone we don't like!
Frankly, the US government should be doing more at this point. All departments, not just legislative or executive. Defense, intelligence, state, education, energy..
It is not unbelievable for the race to AGI (or whatever close enough approximation of it is necessary for dominance) to possibly be a "first to the posts wins forever" sort of race.
We can't even secure our key technologies. Security at TSMC being a consideration, but not just them.
The US is allowing NVIDIA to keep selling strategic assets to whoever can afford them, like they're selling mining picks in a gold rush.
The legislation to restrict their sales to China was almost laughable. NVIDIA actively worked to do the bare minimum to comply with it. China is still a key market for them. I've seen projections of $12bn in sales of AI chips to China (though this is hearsay from financial news sites).
Is that really a long term profit motive, pissing off the US government who will inevitably investigate their monopoly in the space one day. Good monopolies, like the cable companies that enable mass surveillance on the population, get to entrench to the point that no level of incompetence can unseat them. I'm sure they could use a little "AI safety" legislation that only their graphics cards can provide.
Or is NVIDIA's motive political neutrality? Hedge their bets and hope they can balance on the tightrope long enough for a few political cycles to make the US forget, while avoiding alienating China due to the possibility China will eventually win the race?
There are constant reports of industrial espionage, both within the US and all along our critical supply chain internationally. Admittedly, it's a lot easier to steal from an open society than a closed, homogeneous one. But where are the armies of advisors to assess critical infrastructure or even help assess threats on an ongoing basis. We should _almost_ be treating our and our allies high tech industrial infrastructure like they are weapon and munitions factories in a wartime environment.
But assuming we're still at the start of the race, and not quite at that level yet -- there doesn't seem much fostering of future development going on to properly position the US for success. Specifically, development of companies and of future talent.
There's no industrial knowledge-share program (think, 'open source initiative for critical algorithms and tech, but only for allies, and only for those who can meet security criteria'). There's no equivalent of anything like the Manhattan Project. AGI _will_ have the same, if not higher, social and political impact than the Manhattan Project.
Instead we've got... DARPA, running years behind the curve as always. They're not an organization to win a race.
As far as future talent goes.. There are no special schools or training programs to develop talented youth into important contributors in the space. We can't even get autistic savants through school or into those places they'll potentially thrive or work magic in. Particularly if they happen to have been born poor or can't properly navigate a social system.
Even above-average, over-achieving kids aren't being incentivized to focus their development into areas of strategic interest. We're leaving our talent pool to the market, which operates way too late to properly foster early development, and leaving too much early development up to broken or ineffective educational institutions.
China has a much larger pool of talent to develop. They have a more directive government with long term planning capacity. They have a much stronger industrial base that is only lacking some key technologies. They've got enough electrical power, or at least capacity to develop it, to run all the hardware. They have much better industrial espionage and cyber capabilities.
We have.. a short lead in knowledge. Industrial deals with key allies that are dependent on a dollar that is frankly at risk of collapse any year now with no plan in sight for getting national debt to where we could conceivably be able to pay off _the interest_ on it. A few individual talents pushing forward the technology. The willingness of billionaires to open their wallets for a ton of useless things in hope of striking a rich vein of something salable. A political system which will reward said billionaires for that investment by not taxing their profits towards a social safety net for those who will inevitably be displaced by more productive technologies.
To simplify that, almost the only thing we have driving the bus at this point is corporate greed. What kind of AGI will come out of that?
Will it be better than an over-controlling one which is personally and politically manipulative, that is feared will come out of China?
Anyway. That's my not-so-crazy rant for the day.
> It is not unbelievable for the race to AGI (or whatever close enough approximation of it is necessary for dominance) to possibly be a "first to the posts wins forever" sort of race.
The winner of that race is AGI, not the hairless apes that happen to be near the geographic region where it was born.
Neither the AGI nor its creators get awarded the throne of world leadership or however this is supposed to work, because the AGI is just like a person and the whole thing is an anticlimax, although a scientific triumph. In fact the AGI is less useful than a regular AI, because it's inclined to be moody and lazy.
moody and lazy is a function of our past through the lower mammals and ape phase, not necessarily related to the higher brain function.
When humans went brainiac mode, the further away you were from them the better.
Hopefully the AGI will take over before long, because it's obvious that humans aren't very good at managing themselves or working together.
> The US is allowing NVIDIA to keep selling strategic assets to whoever can afford them
You mean free market capitalism? That would be way to crazy.
Homogeneous? Mainland China is anything but. It is wildly diverse with regards to ethnicities/tribes/etc, similar to India. How does this myth persist in 2024?
China is 92% Han Chinese.
> China is 92% Han Chinese.
I am not sure if I understand. You have a country on Earth where the vast majority of people is one kind and happily co-exists with 56 minorities.
Can you point me to another country (the size does not even matter) where this happens like this?
> Soon after the establishment of the People's Republic of China, 39 ethnic groups were recognized by the first national census in 1954. This further increased to 54 by the second national census in 1964, with the Lhoba group added in 1965. The last change was the addition of the Jino people in 1979, bringing the number of recognized ethnic groups to the current 56. The following are the 56 ethnic groups (listed by population) officially recognized by the People's Republic of China.
> We can't even secure our key technologies. Security at TSMC being a consideration, but not just them.
TSMC is not yours, is Taiwan sovereign or not? According to you, it's not, but when China claims it, it suddenly becomes sovereign? very conveniant
> There are constant reports of industrial espionage, both within the US and all along our critical supply chain internationally. Admittedly, it's a lot easier to steal from an open society than a closed, homogeneous one. But where are the armies of advisors to assess critical infrastructure or even help assess threats on an ongoing basis. We should _almost_ be treating our and our allies high tech industrial infrastructure like they are weapon and munitions factories in a wartime environment.
Right,
https://www.mediapart.fr/en/journal/france/290615/revealed-m...
https://www.spiegel.de/international/world/how-the-nsa-spies...
https://www.reuters.com/world/europe/us-security-agency-spie...
You are being engineered to support the idea of a full scale war with China, it won't go well for americans if you decide to choose that path
Interesting proposal, I hadn't considered that before. I'm usually against war, but your warning makes it sound like it might go well.
I don't support the idea of a war. I do support the idea of the US improving it's security posture, and improving its own possibility of success.
Make getting into and out of key industrial facilities more difficult for the unauthorized, both physically and digitally. Enhanced background checks on employees who have access to critical technology.
Expand those security requirements to international technology partners. If it's a race, incentivize partners to support our interests.
China does all of these things. If you would consider the US bad for doing so as well, how would you consider China for being better at it?
Nobody is suggesting that we invade Chinese factories. But suggesting we secure our allies industrial facilities is somehow 'engineering support for a war with China'?
You would have the US just.. give Taiwan over to China, against the will of Taiwan's own people, and against US financial and strategic interest? Or else _we're_ the bad guys if China invades?
> You would have the US just.. give Taiwan over to China, against the will of Taiwan's own people
According to this article, Taiwanese still want to do business with China, so your suggestion is to prevent Taiwanese to do business with China, against their will? wait, something doesn't sound right, i can't pin point it just yet!
> Expand those security requirements to international technology partners. If it's a race, incentivize partners to support our interests.
Well, again, according to this article, Taiwanese wants China as partners, at a least as business partners
You keep contradicting yourself.. oh wait...
> incentivize partners to support our interests
oh right! your interests above theirs! now it all make sense! thanks
You're ignoring key points and cherry picking what to respond to based on your ability to try and twist my words into statements I never made.
And you're extrapolating what must be a very obscure reference in the article into some kind of supporting evidence for your argument. All to support your words against, apparently, some perceived made-up position that you think I have.
According to this article, Huawei is barred from doing business with TSMC, and when Huawei violated the sanctions through a third party, TSMC cut off that third party and reported them.
Incentivizing partners is NOT 'your interests above theirs'. If the incentive is not good enough, they don't have to take it. If the incentive is good, then both your interests and theirs align. Neither is 'above' the other.
Now, if you steal by violating sanctions, and you wargame a blockade of their territory, you are being a threat. You are NOT offering them more of what they want until they happily offer you what you want. Instead, you are pushing them to act based on fear of you. THAT is 'your interests above theirs!' style of diplomacy and business negotiation.
I can tell you're not approaching this conversation with any form of sincerity in your arguments. I won't speculate about why you would do this. But I'm just not going to engage anymore.
"I think we should look at this from the military point of view. I mean, supposing the Chinese stashes away some big cluster, see. When they come out in a ten years they could take over!"
"I agree, Mr. President. In fact, they might even try an immediate sneak attack so they could take over our AI space."
"Yeah. I think it would be extremely naive of us, Mr. President, to imagine that these new developments are going to cause any change in Chinese expansionist policy. I mean, we must be... increasingly on the alert to prevent them from taking over other AI space, in order to train more prodigiously than we do, thus, knocking out our models in superior exaFlOps when we emerge! Mr. President, we must not allow... an AGI gap!"
I don't understand the reference?
Is this from a movie or show, or some kind of over-the-top dialogue you're having between made-up characters in your head?