Every time I cover the vast technological gulf between China’s semiconductor ambitions and their actual native technological knowledge, some commenter blithely states that it’s easy for China to reverse engineer all Western semiconductor technology, which is why they’re already successfully producing chips at the [insert latest CCP propaganda here]nm node, etc. In this they’re engaged in the time-honored rhetorical device known as “talking out their ass.”
Semiconductors are hard. Not only do you have to exactly machine the thousands of painstakingly precise parts in the equipment itself, you need to possess the deep institutional knowledge necessary necessary to tweak the thousands of differing process parameters for different types of chips. Steppers, the lithography machines that actually project the patterns necessary to make each layer of the chip, are at the very top of the mountain in terms of technological complexity, and ASML dominates the stepper market. If it was easy to build steppers, Applied Materials, LAM, or Tokyo Electron would have come out with their own steppers long ago, and they haven’t.
But China would love to get their hands on that technology, which is why they tried to disassemble and backward engineering an ASML DUV stepper and ended up ruining it in the process.
A Chinese firm reportedly has sought technical support from ASML, the world’s largest chipmaking equipment supplier, after it failed to reassemble a deep-ultraviolet (DUV) lithography machine following an internal teardown for alleged reverse engineering.
Note that a DUV stepper is not ASML’s top of the line model. Their Extreme Ultraviolet (EUV) NXE and EXE steppers are far more complex and selling them to China is embargoed.
“An ASML DUV machine that China has used to make their chips recently broke down. They called the Dutch company for help repairing it,” Brandon Weichert, a senior national security editor at The National Interest, says in a X post. “ASML sent some techs. They discovered that the Chinese broke the machine when they disassembled it and tried to put it back together.”
“The reason Chinese technicians took apart their older ASML DUV system is simple. They are trying to find a way around US sanctions on the newest machines,” Weichert says. “By taking apart the older model and attempting to rebuild it, they hope to learn how to produce their own advanced versions. But it seems they still can’t figure it out.”
Weichert says he was unsure whether ASML had repaired the system. He adds that, in his view, although China maintains service agreements with the Dutch company, ASML would be unlikely to honor them given what he characterized as apparent foul play by the customer.
The fact that they couldn’t get the machine back together correctly rather belies the idea that China has world-class semiconductor technical knowledge.
Some Chinese commentators have noted that reverse-engineering ASML’s immersion DUV lithography machines is an exceptionally complex challenge.
A Guangdong-based columnist writing under the pen name “Chengwa” highlighted four key difficulties:
Extreme precision: DUV systems use 193-nm argon fluoride (ArF) lasers and a thin water layer beneath the lens. Even the tiniest misalignment can cause a chain reaction of system errors. Complex mechanics: Inside modern DUV immersion tools, twin wafer stages move rapidly under the lens with sub-nanometer accuracy and process around 330 wafers per hour. Removing one stage without factory calibration can destroy the delicate alignment that field engineers cannot easily restore. Highly integrated technology: ASML’s equipment depends on intricate optical systems, motion platforms and control software perfected in Europe over decades. Replicating all these technologies from scratch is extraordinarily difficult. Precision calibration: The system’s accuracy depends on closed-loop calibration linking optics, sensors and motion control. Dismantling the tool can lead to particle contamination, interferometer drift and loss of key reference points. These problems require vendor-level software keys and procedures to correct.
Like I noted, down below 10nm, everything is exceptionally hard.
After the US banned sales of ASML’s extreme ultraviolet (EUV) lithography machines to China in 2019, Beijing began pouring substantial investment into homegrown lithography development. However, much of that funding has been marred by inefficiency and corruption scandals, limiting technological progress.
Snip.
Last month, the Financial Times reported that China’s leading chipmaker, Semiconductor Manufacturing International Corporation (SMIC), was testing a DUV immersion lithography machine made by Yuliangsheng.
The machine is understood to be an immersion DUV scanner targeting 28-nm chip production, roughly matching the performance of ASML’s Twinscan designs in 2008. Yuliangsheng planned to deploy it on production lines by 2027.
China using their IP-stealing and backward engineering skills to finally replicate a 19-year old design sounds about right…
Tags: ASML, China, Communism, Semiconductors, technology
Back during the 60s-80s timeframe, IBM leased mainframes to Russia. It was said that these had purposely reconfigured internal setups, that made them unreliable and failure prone. Because they knew the Russians were going to take them apart and copy them.
Reverse engineering can be used to understand how to make compatible products (classic example: the BIOS ROM in the IBM PC) or to understand a competitive threat (e.g. the MiG-25 flown by a defector to Japan in 1976), but reverse engineering to *catch up* works only in certain narrow circumstances, when the underlying component technologies are understood, but the system-level trade-offs are what you need.
Take an automobile design. If you want 0 to 60 in 8 seconds, you can have a more powerful engine, or lighter weight. When BMW did it first to mass-market, they needed to project the power of a not-yet-invented engine and the weight of a not-invented chassis. And so on for a hundred other trade-offs: bigger radiator/more airflow, more displacement/4 valves per cyliinder, more gear ratios/wider powerband, blah blah blah. When Japanese car makers copied it, they could get the trade offs by looking at what BMW did.
Of course, you copy the system-level mistakes, too, as the Soviets did with the Space Shuttle and Tu-144 Concordski.
Reverse engineering a *manufacturing process* to catch up doesn’t work. I can examine a plastic bottle until Kingdom Come and and not learn a thing about the 20-year struggle to develop a process to make them cheap, reliable, and (the haaaaard part) not smell bad. Having the machine that makes the bottles gives me few clues, fast cycle time vs. more molding cavities for example, but no help with on the hard parts.
I worked for ASML. Making a chip requires two main pieces of equipment (Stepper scanner and Light Source). Both are produced by ASML. The Stepper Scanner is the easy one. The Light Source is the DUV or EUV part of the chain and it is insanely more complex. China started with the easiest part, and failed.
They better hope they have some Quantum trick in their hat because they are two decades behind and losing ground fast.
This is a major example of why IP (in this case patents) doesn’t really help innovation, and even hinders it. It is based on the notion that every discovery would be easily reproduceable, so that all but the “obvious” things need patent protection — but the funny thing about that is if it’s obvious, it’s not going to be hard for an expert in the field to duplicate, and there’s a very strong chance that the same expert would have come up with the same solution on their own. On the other hand, if it’s *not* obvious, then experts in the field trying to reverse engineer something will be spending time and money they need to research their own competition, and if successful in their reverse engineering, will only find themselves experts in obsolete technology — because the competitor they are trying to reverse engineer will have moved on to selling something better.
The best that I can say of reverse engineering is that it can ba stepping stone for someone to *become* an expert in their field — but they are still on their own if they want to create something better.
This is the part where I’d like to say “I don’t fear China gobbling up IP when they can get it and ignore it when they can’t” but I cannot: America tends to respect IP, even when owned by China, and thus won’t knowingly use the IP China manages to acquire. Where’s the sense in that?!? If we’re not going to ignore IP, then why the heck are we selling it to someone who does, thus tying our own hands in the process???
Casinos protect their systems against gaming (manipulating) the gaming machines. Vehicle manufacturers restrict access to high-level diagnostic and repair tools. US GOV/DoW has anti-tamper (https://at.dod.mil/What-Is-Anti-Tamper/)
It works, whether designed into the product or based on complexity or sophisication of the system.
Re: M. Rad.
What USAF learn about the Mig-25 that flown to Japan is that the preconception Western analysists have about the plane was absolutely wrong. They looked at the shape and the kind of engines it uses and thought it was a interceptor capable of doing all sort of crazy maneuvers. As it turns out, it had one and only 1 job, to fly very fast and high up to fire missile at B-70 Valkyrie, which would’ve penetrate all of the known Soviet countermeasure due to how high it flown and being supersonic. The problem is that USAF overestimated the capability of Soviet SAM and cancelled the program. In response to Mig-25, USAF commissioned what would be come F-15 that have all of the capability that they thought Mig-25 have.