pull down to refresh

This article by Justin Thaler is a nice counterpoint to the interview with Scott Aaronson that I posted yesterday (#1312936).
The real challenge in navigating a successful migration to post-quantum cryptography is matching urgency to actual threats.
We are nowhere near a cryptographically relevant quantum computer by any reasonable reading of public milestones and resource estimates. Companies sometimes claim a CRQC is likely before 2030 or well before 2035, but publicly known progress doesn’t support those claims.
By a “cryptographically relevant quantum computer” I mean a fault-tolerant, error-corrected quantum computer capable of running Shor’s algorithm at scales sufficient to attack elliptic curve cryptography or RSA within a reasonable timeframe (e.g., breaking secp256k1 or RSA-2048 with at most, say, one month of sustained computation).
Recent systems approach the physical error rates where quantum error correction begins to work, but no one has demonstrated more than a handful of logical qubits with sustained error-corrected circuit depth… much less the thousands of high-fidelity, deep-circuit, fault-tolerant logical qubits actually required to run Shor’s algorithm. The gap between demonstrating that quantum error correction works in principle, and achieving the scale needed for cryptanalysis, remains vast.
  • Demos claiming “quantum advantage”, which currently target contrived tasks.
  • Companies claiming to have achieved many thousands of physical qubits.
  • Companies making liberal use of the term “logical qubit”.
Thaler presents 7 recommendations for "What we should do now" at the end of the piece. They seem pretty reasonable to me. I also really like how he makes a distinction between the vulnerability of encryption and signatures -- encrypted data that is treated as safe to expose to the public can be harvested now for later decryption, while signatures are mostly only vulnerable once a cryptographically relevant quantum computer finally comes along.
116 sats \ 1 reply \ @freetx 9h
Great writeup.
physical / logical qubit.
To solve the decoherence problem, QC uses QEC (Quantum Error Correction). That is they use multiple qubits in parallel to discover and correct errors.
The result of all these "physical qubits" error correcting each other produces one "logical qubit" that is error free. (Logical qubits are the things that do the work)
With current error rates of roughly 0.3%, maintaining a single error-corrected logical qubit requires approximately 10,000 physical qubits.
So when people say "Shor's will take 10,000 qubits" - they are talking about LOGICAL qubits, so that may be something like 1,000,000 physical qubits to do that.
But there is a cruel paradox at the heart of this: Each time you bring another physical qubit online, that introduces more noise and more interference. Quite quickly your error rates start to rise faster than you are improving fidelity...so it sort of cascades into a negative feedback loop.
IBM's has a chip with 1200 physical qubits they achieved in 2023. They promised 4000 qubits in 2025 and missed their goal, but released a press release saying "100,000 qubits by 2033" (notice the bait-and-switch).
At this point most of QC is pie-in-the-sky snakeoil investor pumping. However, I think so much money has been spent that no one is ever going to admit defeat. So get used to yearly barrages of "we reached 8000 qubits" (fine print: physical).
reply
"Cruel paradox" is a nice line ✍️
reply
Thanks for presenting representatives from both sides of the issue—that's helpful!
About this:
*Companies sometimes claim a CRQC is likely before 2030 or well before 2035, but publicly known progress doesn’t support those claims.
There isn't incentive everywhere to keep QC developments public; and, presumably, great incentive to keep it private enough to execute Zero-Day attacks on vulnerabilities people don't know are vulnerable. That seems to be the major fear (and it is a fear, though always written in sanitized language) behind recent QC articles.
But is that just an irrational superstition, based on completely ignorant views of the technology? Or—in the spirit of December 7—is it more like someone in 1936 saying, "You know, the Japanese could attack Pearl Harbor." 🧐
reply