China's Tsinghua Unigroup to invest $47B to build chip empire | Hacker News Reader

China's Tsinghua Unigroup to invest $47B to build chip empire

reuters.com

Raise your hand if you saw this coming when the US government blocked Intel from supplying chips for China's supercomputing initiative. Way to go, ITAR: once again you've successfully spurred a country to move away from its dependence on American technology, while doing nothing to meaningfully improve our security.

Michael Hayden, former NSA chief, also said zero-sum games in the computing industry would only hurt the US (because Intel/etc gets to make less money and therefore less money for research and investments):

https://www.washingtonpost.com/opinions/dont-let-america-be-...

I am actually glad this is happening, as I think competition is great. For the same reason, I hope China eventually sees some success in displacing Windows with Linux, as it could only help Linux's popularity and ecosystem of apps if there's a much bigger Linux installed base, even if I would personally never want to use a Chinese chip or a Chinese app.

I would second that competition in the server/desktop/laptop cpu space would be salutary. Intel has had a near monopoly for a couple decades or more, and this has possibly stifled innovation.

I was hoping that ARM would provide that competition, but server side ARM is still a niche market.

Yep. Matter of fact, I thought it would happen anyway at a smaller amount when they started laying down lots of fabs. Figured they'd want vertical integration plus market-takeovers with their labor advantages.

It would have always happened. As China gets bigger, it wants do more technical things.

Sure, but why accelerate it?

Because we're all human, and you should be proud of your neighbour's success?

"Proud of your neighbor's success" generally doesn't present as "forbidding trade of strategic goods with them".

It's an epic fuckup of the US, but a win for humanity.

That great news. Why are you complaining? India and China need to both step up and innovate more. We'll get 2-3x the R&D at some point. Forget about call centers and assembling iPhones. Move up the food chain. Drugs research, electronics, planes, etc.

> once again you've successfully spurred a country to move away from its dependence on American technology, while doing nothing to meaningfully improve our security.

That's only if they are successful. And success, especially in that industry, is far from guaranteed. There are some critical breakthroughs they would need to make just to compete with the likes of Intel on even their low-mid range processors. Most of their designs are propriety which means China's computer initiative is going to be starting from scratch. That means they're starting at a point which puts them behind everyone else by at least 10-13 years or more. And still, there's no guarantee they'll ever come close to catching up. It's a huge $47B gamble on their part, which is almost certainly why they haven't done it sooner. That's a lot of money to lose, even for China, if they bust.

And I know some of you are thinking, "Well, they'll just steal the tech! Espionage baby!" Sorry, but that too, is unlikely. Due to China's recent aggressive tech grabs over the last decade or two, companies have "wised up" to IP theft and go to extreme measures to protect their assets. Some of those protections rival or even exceed the U.S governments. China's wild west days of stealing everyone else's IPs are winding down.

I understand that making chips is hard, but this thinking reminds me of the Palm CEO's quote "PC guys are not going to just figure this out. They’re not going to just walk in."

Just look at how far Apple and TSMC have come since 2009. Sure, Apple "cheated" by acquiring PA Semi (the DEC Alpha people). But China has the rest of the ingredients: vertical integration of hardware/software (I assume these will be used for defense-related HPC), strong political incentives to be independent, and massive amounts of cash.

Apple does not fab its own chips, it just designs them from IP licenced from British firm ARM. This article is referring to constructing a fab plant to produce nand chips.

China has near clones of the F-35 Lightning, the Comanche stealth helicopter, the best of Russia's newest generation of tanks, etc.

I purchased a Trijicon ACOG clone from China for $136. It's just as good as, if not better than (doesn't require a battery for the chevron, and a battery powered chevron is, IMHO, not a good feature), the real thing, which costs civilians $1,400, and our military something like $800 each.

I can't speak for the highest of the high tech, in terms of a "wind down", but I can assure you that $47B in China will get you a LOT more than $47B will in the US, and the lack of R&D costs is a big part of that.

> China has near clones of the F-35 Lightning, the Comanche stealth helicopter, the best of Russia's newest generation of tanks, etc.

So they say. There's no way I'm taking China at their word. They're only slightly less full of it than North Korea.

And according to the wikipedia page on the Shenyang J-31, it (at best) matches up with 4th generation fighters. Not the F35. They tried to copy it but their copy is only (maybe) as good as the previous 4th generation. Which is of little to no concern.

Which is kinda that point. Their knock-offs are just that. Knock-offs. And none of those IP theft acquisitions have been recent. Their F35 theft supposedly happened nearly a decade ago.

It was easy to predict this would happen when the US gov made this silly and very pointless move of blocking Intel Xeon chips in China, with an even sillier excuse that China is using its supercomputers to do nuclear research (shocking!):

http://www.pcworld.com/article/2908692/us-blocks-intel-from-...

Wow, they're investing 300 billion yuan over 5 years, and they only have ~12 billion yuan in annual revenue. There must be some monster loans being made available...

It seems clear who's footing the bill:

China's Tsinghua Unigroup Ltd plans to invest 300 billion yuan ($47 billion) over the next five years in a bid to become the world's third-biggest chipmaker, the chairman of the state-backed technology conglomerate said on Monday.

Chinese government wants to be in CPUs. (Edit: Oops, the article talks about NAND. Well, they want to be in memory chips then, I guess.)

Yeah, maybe s/loans/government grants/g

I'm sure they do want to be in CPUs, and getting into NAND is a stepping stone towards that by building up a skills base in semiconductor manufacturing processes.

I proposed doing it differently by investing billions in I.P. development for standardized interfaces and constructions at every half of the process node. As in, most of the work is already done with process shrinks easy too. On top of that, platform ASIC's or S-ASIC's with mapping to FPGA's. Anyone that used the fabs would get the I.P. free or dirt cheap with that used to make more. All the stuff that isn't a competitive advantage is gratis.

The other one that would kick ass is heavy investment into EDA tooling. China itself may or may not need it with the rampant piracy. However, driving the prices of EDA synthesis, verification, etc tools down to four to five digits might make the market really interesting. Especially with lots of reference I.P. and flows. That plus inexpensive, maskless prototyping like eASIC does.

Someone might steal my idea with me posting it all over the place but I don't have the capital to pull it off. So, steal away and get me EDA that's as cheap as Windows. :)

gEDA and KiCad are pretty low-cost!

Correct me if I'm wrong, but I don't think you can design silicon in kicad.

You're not wrong. Neither gEDA, they're both schematic capture and layout tools. Maybe the OP thinks they can output masks from schematics or something like that but that has absolutely nothing to do with each other.

Fabmasks have about as much to do with schematics as architecture has to do with city utility planning.

Lol well said...

You have to understand the ASIC or FPGA design flow to understand why that doesn't make sense. For some reason, I struggled to find a Google result explaining it without being a tutorial on specific EDA tools. This one from IBM mentions many details, though:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.92....

The analog components were largely hand-made by pro's, simulated, and then prototyped. Today, there's lots of so-called mixed-signal blocks that combine analog and digital with extra complications:

https://www.cadence.com/rl/Resources/conference_papers/1.5%2...

So, I'm focusing on digital design with pre-made analog blocks at interfaces and minimal integration work. Getting to the rest with OSS will be... an uphill battle on top of an uphill battle. For now, we actually can use something like Qflow at least for larger nodes:

http://opencircuitdesign.com/qflow/

Still need standard, cell libraries for the target process. There's an open set or two...

http://www.vlsitechnology.org/

...but best is to have pro's (or top academics) to straight up design a set for you for specific nodes then open-source it once silicon-proven. Need common analog components which pro's will have to do, as well. Need to integrate the digital and analog logic plus make sure whole chip is working. Gotta do the design rule checks to make sure physical properties of each layer or between them is right.

There's a lot to it. I have one or more academic papers for almost every step in synthesis, re-synthesis, formal verification, gate-level testing... you name it. Plenty for FOSS-loving academics to build on and integrate. That will take time.

Meanwhile, my main recommendation is to finalize an open-source FPGA w/ plenty of I/O interfaces on advanced process node like 45nm. Then, target ABC synthesis and/or VPR to it. Let's us get open hardware while ignoring a large part of the design process. Archipelago FPGA's are already quite a ways there:

http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-43...

Can create a metal-programmed S-ASIC that matches it for easier porting. If it takes off, do one for 28nm. Those should cover most use-cases with standard cell model for others.

Personally, though, I think there will be a mix of proprietary and FOSS tools. Best model is getting vendors to make theirs verifiable by FOSS tools so we can ignore the complexity of how they do it to focus on what they did and its equivalence/correctness. Especially good to get a smaller vendor like Mentor Graphics on it as a differentiator in post-Snowden market. They keep secret sauce while we get verification. Win, win.

I'm waiting for China to do for the US chip industry what Japan did to the US auto industry: come in and eat it from the bottom up with cheaper products that are initially dismissed as inferior toys until they move up the chain and eat the entire market.

https://en.wikipedia.org/wiki/Loongson

I'm not sure it's comparable, the tool supply chain for one is very different. A Dutch company is the lithography tool supplier for both American and non American chip companies.

If Moore's law is over that could be less of an issue (old tooling will still be useful).

Unlike GM, Ford, Chrysler, the current big semiconductor producers, Intel, Samsung, and TSMC are already excessively cost sensitive. There isn't a lot of extra overhead to suck out.

China can subsidize things at a loss (look at the solar cell manufacturers) until the volumes get big, but they can't maintain that too long once things take off.

That's a good point, but I wonder if coming up from below with a cleaner architecture might be able to accomplish something similar. Is MIPS64 cleaner than X86, and could this let them create lower power higher performance chips or manufacture equivalent performance chips at a lower cost?

http://blog.imgtec.com/mips-processors/loongson-mips64-proce...

The benefit of a lack of a decode step (for arm vs x86, but I assume the analysis is roughly correct for mips) was estimated at RealWorldTech as about 5% of a core's energy. Given that the amount of energy the cores take up as a portion of the total energy budget, I doubt it would matter for high performance chips - particularly in the face of Intel's 1-2 process generation lead over everyone else.

Admittedly, phone chips are a bit lower power than desktop CPUs, but they are pretty beefy these days.

Dunno... 5% at data center scale could make a difference in the highly competitive cloud services market where margins are already fairly small.

It's incredible how they can plan to create an entire industry by investing so much. It's slightly different from how India approaches such problems by forcing companies to manufacture stuff locally. I wonder what approach will work out best in the long run.

China does the manufacture locally thing too by jacking up the tariff and then requiring all local companies be majority Chinese owned. It has done wonders for them because that's how they've learned modern manufacturing techniques. Over the past three decades, Chinese manufacturing has improved a great deal now that the premium consumer products are made in China.

These massive investments is another tool they have and the track record is somewhat mixed. For example, they completely dominate the solar industry now because of their massive investments in that industry. In addition, they also have the world's largest solar capacity in the world. The problem is that a lot of the capacity is unused because of the lack of transmission lines. So at least their solar investment have generated results but done so with a lot of waste.

"They" can get these almost unlimited monies basically for free from the Chinese banking system while similar funding in the US and Europe ends up with so called investment banking or similar where it will never arrive with new entrants or is allotted to a small group of established players and special interest groups.

China is acting similar to what successful (new) raising economies have been doing in the past when they disrupted established (industry) players - my best historic comparison on this would be the transition of Germany from a farm based economy to one of the key industrial nations after the first industrial revolution. While the leading industrial player - the UK - was blocked from progress by overstretched copyright / patent / other protection mechanism of the established special interest groups, Germany provided top down support and access to all necessary means to forward technological progress.

Then - while most people in Germany still gained only limited benefits and just a few became very influential / rich, Germany as an overall entity / nation gained massively. At the same time the situation / influence of the UK overall deteriorated - these are of course all longer term developments often spanning more than one generation (30 years).

Anyone familiar with the industry know what companies might be targets for acquisition?

ARM would seem to me to be an obvious target. ASML another.

But ARM is UK based. The article seems to suggest a US based company.

Reply on news.ycombinator.com