Export Controlled Gates
Fun With Regulations
Some news came across my feed recently about new export controls being implemented for quantum computing technology among a few nations. You can find the article here, but be warned it’s paywalled and the site itself is filled with intrusive garbage ads. I will supply the relevant links below.
The TL;DR- A few EU nations (Britain, France, Spain, the Netherlands) and Canada appear to have implemented nearly identical new export controls for quantum computing and adjacent technology, with zero explanation or rationale supplied for the reasoning/source of the new restrictions. It’s national security, you see.
You can find the text of the new export controls linked here1, as well as the general EU export controls.
Canada; UK; France; Spain; Netherlands; EU2
Evidently, these controls have been in the works for a while per an EU compilation of member-state regulations published in 2021. It seems to imply that the Netherlands and Spain were the drivers of these new regulations. Unsurprisingly, the Dutch are the source of the fabrication controls on cryoCMOS technology. While, somewhat surprisingly, the Spanish were the first to issue the (nonsensical) qubit controls I’ll list below.
You’ll note that two more items are added on the UK restrictions: parametric (quantum-limited) amplifiers like TWPAs3 and dilution refrigerators4 with cooling power greater than 600 uW at 100 mK.
The Export Controls
Since this is a QC blog, let’s start with the qubits. Here’s a quote of the regs.
(I) quantum computers supporting 345 or more, but fewer than 100, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 10-4,
(II) quantum computers supporting 100 or more, but fewer than 200, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 10-3,
(III) quantum computers supporting 200 or more, but fewer than 350, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 2 x 10-3,
(IV) quantum computers supporting 350 or more, but fewer than 500, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 3 x 10-3,
(V) quantum computers supporting 500 or more, but fewer than 700, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 4 x 10-3,
(VI) quantum computers supporting 700 or more, but fewer than 1,100, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 5 x 10-3,
(VII) quantum computers supporting 1,100 or more, but fewer than 2,000, fully controlled, connected and working physical qubits, and having a C-NOT error of less than or equal to 6 x 10-3, and
(VIII) quantum computers supporting 2,000 or more fully controlled, connected and working physical qubits,
That’s right, these numbers are for physical qubits and CNOT error rates across nearest-neighbor pairs of qubits. In this case fully-connected means that it is possible to perform a 2Q gate between any two pairs of qubits, not that they’re all-to-all connected. These export controls DON’T apply to adiabatic QCs (annealers). This is explicitly called out in the regulations. Additionally, quantum systems & devices that support the operation of the quantum computing systems listed above6 are also subject to these controls.
TWPA and Fridge Restrictions
If you take the time to also go look at the British export restrictions, you will notice some differences from the Canadian version. Specifically with respect to quantum limited amplifiers and cryocooling systems.
If I understand correctly, the Oxford Instruments ProteoxLX is now export controlled, as its cooling power exceeds 600 uW at 100 mK. Finland, as far as I can tell, has made no changes to its export control policies and continues to follow the EU law. Additionally, this text does not appear in the Netherlands’ version of the controls, but I haven’t been able to find the official .nl government page to confirm. Since Leiden Cryogenics offers dilution refrigerators that exceed the limits above, they would have been impacted by these restrictions.
cryoCMOS Restrictions
Someone who knows much more about fab should be writing this, so I’ll just content myself to say that the semiconductor restrictions are specifically targeted toward the manufacture of GaAs semiconductors intended for use at temperatures below ~4 K. So, cryoCMOS. Clearly someone thinks that cryoCMOS is the most likely avenue for the large-scale control of QC systems.
Paired with the cryoCMOS restrictions are some restrictions on SEM systems and software designed to reconstruct layout files (.gds format) from SEM images of quantum computing or cryoCMOS chips. Presumably to prevent, or hamper, reverse engineering of these systems.
There is another option for QC control, of course. It’s single flux quantum (SFQ) pulse control. No! Don’t laugh! There is a significant amount of SFQ logic research going on, and at least one startup predicated on the idea (SeeQC). In fact, SFQ digital logic has been in the news lately.
I can’t imagine that the people responsible for these export controls are ignorant of this work, so I assume they don’t believe it can plausible work on any time scale. Fair enough.
Why?
When I look through these qubit restrictions, combined with the gate error restrictions I am perplexed. The Canadian document states:
Quantum computing and its related technology and equipment
The emerging field of quantum computing can have a significant impact in many commercial and military areas. Quantum computers are powerful computers that take advantage of quantum physics to solve mathematical problems that traditional devices would take a very long time to solve, and sometimes not be able to solve at all. While the technology could bring important advances in chemistry, simulation, medicine and many other civilian applications, its potential for use in cryptanalysis has wide-ranging implications in the national and international security arenas. A quantum computer of sufficient power and scale could have the ability to break virtually all forms of public-key cryptography in current use and compromise the most secure communications and transactions conducted over encrypted networks, as well as the integrity of the software used on such networks.
Many quantum computers must operate at extremely low temperatures to be accurate. The currently available and conventional devices called Complementary Metal Oxide Semiconductor devices (CMOS devices) that power these computers are not optimal, as they operate at higher temperatures. For this reason, quantum computers must be equipped with Cryogenic CMOS devices (CryoCMOS device), which are integrated circuits capable of functioning at cryogenic temperatures.).
The best estimate I know of for RSA integer factoring is from Gidney & Ekera in 2021, in which the authors estimate the cost of factoring a 2048 bit RSA integer to be about 8 hr with 20,000,000 physical qubits. They assume an error rate of 0.1%.
You’ll note that 20,000,000 qubits is much greater than 2000 qubits, the maximum listed in the Canadian (and European) export restrictions. Additionally, for qubit counts above 200, the error rates that trigger restrictions are greater than 0.1%.
The best estimate I know of for non-trivial quantum chemistry (the chromium dimer) is between 100,000 - 1,000,000 physical qubits.
At the surface level, there is a vast distance between where we are now (roughly on par with these restrictions) and where the literature suggests we’ll need to be to do anything plausibly dangerous with quantum computers, especially with respect to cybersecurity.
So, why all these new restrictions? Some options.
The bureaucrats responsible are morons and easily manipulated by $YOUR_FAVORITE_VILLAIN.
This is not a particularly satisfying explanation for me, particularly because it short circuits all further thought.The intelligence agencies involved have discovered a major breakthrough in quantum error correction and algorithms that would make a 34Q, 1e-4 error QPU plausibly dangerous.
I don’t believe this is likely, since the rollout of these new restrictions is not EU wide, nor is it mirrored by similar restrictions in the largest QC power, the United States. Additionally, to telegraph something like this by immediately moving to export control regs with so few qubits suggests that the insight necessary is fairly obvious and might be discovered by academic or industry researchers at any time. The NSAs and GCHQs of the world are reported to have some very bright people, but do I think the QEC researchers at Google, IBM, and the universities just missed 100,000x savings in qubit count at just slightly lower error rates?The bureaucrats responsible want to be seen Taking Quantum Seriously, but cynically enacted regulations that largely don’t affect their local quantum industries.
I’m lukewarm on this option as well, for reasons similar to those stated in 1. I will note, though, that the Oxford Quantum Circuits 32Q system deployed to Japan (OQC Toshiko) falls far short of the qubit restrictions, for now. I will also note that, according to the Quantum Computing Report, Spain’s quantum industry is largely software and consulting. Their best known quantum computing hardware company, Qilimanjaro, is pursuing high-coherence quantum annealing, which is specifically exempt from these restrictions. On the other hand, Oxford Instruments is a well known dilution refrigerator manufacturer which offers systems that seem to be impacted by the export controls. This is odd to me, as Finland7 has not made any changes to its export regime for dil fridges as far as I can see. Nor has the Netherlands8.
However, ASML is THE fabrication tooling manufacturer for semi-conductors, so these restrictions on cryoCMOS equipment seem like they will be meaningful for the Netherlands. On the other hand, perhaps these restrictions are basically OK, since ASML is not a chip manufacturer and presumably they’d get an export license to sell machines to the USA. Maybe not to China, though?A fourth, mysterious option.
Maybe I’ll think of something better.
Final Considerations
In one respect, this post has been a great success, since it forced me to find and collect information about quantum computing export controls. Sadly, all of this collating and synthesizing does not leave me with any blistering hot takes. A Friend of the Blog notes that the United States has been considering some kind of export control for a few years now. I have no faith that the eventual restrictions won’t be as arbitrary and pointless as what we see coming out of the EU and Canada.
One concern I have is that a conversation can be deemed to be an export, at least in the US export control regime. What does that mean for the stream of publications, blog posts, conference talks, etc coming from the teams at the forefront of industrial QC? Even if the new quantum export control rules end up being pretty loose, there’s no way corporate ITAR lawyers are going to be anything but paranoid maniacs, especially given the potential penalties for unlicensed export. In general, university research is considered exempt from most of this stuff, since it is considered basic research, so academic work will probably continue unimpeded.
One final question I have is simply ‘why now?’ If you give the bureaucrats the benefit of the doubt, you might argue that they simply believe these restrictions will be met by quantum computing systems in the nearish future. But what do actual quantum computing companies believe? Their incentive is to sell a vision of QC to investors and the public, and, crucially, the vision should be at least plausibly achievable. To find out, we can take a look at a few roadmaps. On the high side of things, Diraq published a roadmap that targets ‘1M++’ qubits and error correction within 10 years. While the most recent iteration of the IBM roadmap suggests they’re looking to have ~100,0009 qubits and 1B gates by 2033ish10. QuEra thinks they’ll get to 10,000 qubits by 2026. The Google roadmap is comically11 vague, suggesting ~1,000 qubits sometime after 2025. I won’t link to the Microsoft “roadmap” which has no numbers of any kind on it at all. Infleqtion doesn’t provide a nice graphic, but implies they are targeting 100 logical qubits and 1M gates within 5 years. Pasqal aims to have 10,000 physical qubits by 2026.
So, I guess if you take these plans at face value, there’s a decent chance someone will have 10 - 100 logical qubits by 2030. On one hand, most of these efforts will fail. On the other hand, our civilization has been built by doing seemingly impossible things, eventually. Knowing that government takes a long time to do anything, a true believer in the necessity of regulation might look at all of these roadmaps and the huge amount of money sloshing around for QC investment and conclude that there’s a high enough chance that we’ll have hundreds of logical qubits by 2035 that it’s worth making the regulations now. I still think they’re dumb regs, but I can see now how this could have happened.
I tried, but navigating government websites is already difficult when you speak the language. I couldn’t find the specific Dutch and Spanish pages that list these restrictions, but did find the EU document that claimed that the Netherlands and Spain were the sources of much of the new regulations.
Go to Annex I for the list.
A Friend of the Blog (AFotB) notes that the smallest distance 3 logical qubit is composed of 17 physical qubits, and that 34 physical qubits gets you two such logical qubits. Still a ridiculously small limit, but may hint at some ‘reasoning’ that went into these restrictions.
Readout, calibration, initialization, plus things like ion traps, interconnections, etc.
BlueFors
Leiden
An earlier version of this post put the number 2000 here. It’s actually 100,000 physical qubits defining 2,000 logical qubits. I apologize for the error. Big thanks to AFotB for catching it.
We’ll need 100Bs or trillions of gates to do things like attack RSA, but that number may come down substantially as we learn more.
But I respect it.