Surrey's Helium Qubit Is Calculated, Not Built
The sentence that travels is “100 times less error-prone.” The sentence that should travel with it is “calculated.” ScienceDaily’s Oct. 1 writeup, using material from the University of Surrey, describes a proposed device called the Superfluid Helium Oscillator Quantum, or SHOQ. The researchers’ calculations suggest error rates about 100 times lower than conventional superconducting qubits. A securities.io piece on Oct. 5 prints the lifetimes those calculations produced. They are estimates. Nobody in these writeups is reporting a coherence time measured on a finished helium qubit this week.
We wrote an ion experiment that did run, thirteen ions and a string that broke, in Duke’s 13 ions showed string breaking. We wrote software aimed at logical qubits that are not in the box yet in NVIDIA’s CUDA-Q Logical. We wrote a decoder that sits on a laptop chip while the quantum part still does not, in IonQ’s decoder runs on a laptop chip. SHOQ is closer to the second and third than to the first. It is a design paper with a table. It is not a machine you can book time on.
The paper is a design, and the October dates are writeups
ScienceDaily dates its page October 1 and points at npj Quantum Information. The authors are Priya Sharma, Jens Koch, and Eran Ginossar. The title is “Towards a micromechanical qubit based on quantized oscillations in superfluid helium.” The DOI it prints is 10.1038/s41534-026-01355-3. Inkorr’s Oct. 3 brief repeats the journal and the “nearly 100 times lower than current superconducting qubits” claim. I am using October pages because that is when these explainers landed in front of readers who do not open npj. I am not going to call October 1 the day the qubit was switched on. ScienceDaily’s own verbs are “proposed” and “would.”
The group is Surrey’s Quantum Sciences Group, with Koch at Northwestern. ScienceDaily notes that Koch was among the researchers who developed the transmon, the superconducting qubit design now widely used. That credit matters as context, not as a product launch. The person associated with the qubit a lot of labs already run is co-authoring a paper about a different physical object. It does not retire the transmon. It says the people who know why superconducting qubits are noisy are looking at a fluid that carries no charge.
SHOQ, in the ScienceDaily account, uses charge-neutral superfluid helium-3, a liquid helium that can flow without friction. Because the helium carries no electric charge, the design could be naturally protected from some kinds of electromagnetic noise. “Could” is doing the work in that sentence. Securities.io makes the same split in plainer engineering language: the attraction is less exposure to certain electrical noise, and the cost is cooling, fabrication, and control that are still demanding. The study, in that account, establishes a theoretical route and leaves experimental performance and practical scaling unresolved.
If a slide this week says Surrey built a qubit with 100 times fewer errors, the slide has changed the tense. The sources say a design, calculations, and a prototype still ahead. Inkorr’s headline is closer to “develops” than the paper’s “towards.” I am following the paper’s verb as the writeups quote it, not the headline verb.
Charge neutrality is the reason to bother with helium at all
Securities.io’s description of the mechanism is the one with moving parts. The device stores quantum information in the motion of superfluid helium-3. The central mechanism is not an electrical current. Neutral fluid flow meets a flexible mechanical plate. The fluid and the plate form an oscillator. Under the right conditions its energy is quantized, so it sits in discrete levels rather than any energy you like. Two of those levels could be the states of a qubit.
Discrete levels are not enough, and the piece says so. The spacing has to be uneven enough that you can drive one transition without kicking the system into another. That uneven spacing is anharmonicity. Without it, an oscillator is a bad qubit because your control pulse does not know which rung it hit. The authors, as securities.io summarizes them, show how aperture dimensions, plate size, and mechanical stiffness could supply that nonlinearity. Those are design knobs. They are not a fabrication report with yield numbers. I do not have a foundry name, a wafer size, or a photo of a device in the October pieces. I will not add them.
The noise argument is why someone would accept the cooling bill. Superconducting qubits are electrical. Electrical things couple to electrical noise. A charge-neutral fluid does not couple the same way. ScienceDaily’s phrase is protection from certain types of electromagnetic noise, not from every error a quantum computer has. A reader who turns “immune to electromagnetic noise” into “error-free” has left the paper. The 100-times figure, in ScienceDaily, is a comparison of predicted error rates against conventional superconducting qubits. It is not a claim that the remaining errors are zero, and it is not a claim that the comparison was measured on two running chips.
Koch’s transmon history is the right comparison class. The transmon became useful because it was buildable and because its noise was understood well enough to engineer around. SHOQ is a proposal in that tradition, from one of the people who did the previous engineering, aimed at a noise channel the transmon still has. Useful, if the calculations survive a lab. Not a drop-in replacement announced on October 1.
The table is three designs, and the lifetimes are not stopwatch times
Securities.io says the researchers evaluated six example designs and then prints three. The lifetimes are from single quasiparticle tunneling, which the authors identify as the dominant low-temperature loss in their analysis. I am copying the three rows they printed, not estimating the other three.
Design 1: plate radius 8 micrometres, oscillation frequency 17.60 MHz, estimated lifetime 0.57 milliseconds at 0.4 millikelvin and 23.35 milliseconds at 0.3 millikelvin. Design 5: radius 5 micrometres, 14.25 MHz, 0.64 ms at 0.4 mK, 25.96 ms at 0.3 mK. Design 6: radius 2 micrometres, 28.16 MHz, 0.45 ms at 0.4 mK, 18.47 ms at 0.3 mK.
The authors, in that account, single out design 5 as especially promising. Longer lifetime is not the whole score. Gate speed, leakage into unwanted levels, initialization, measurement accuracy, and coupling between devices all change how much computation you get. A 25.96 millisecond calculated lifetime that you cannot initialize is a number in a table. Design 5 wins the table they showed. It does not win a lab that has not been built.
The temperature step is the part I would not let a summary delete. For design 5, going from 0.4 to 0.3 millikelvin raises the calculated lifetime from 0.64 to 25.96 milliseconds, about fortyfold, securities.io says. Those temperatures are 400 and 300 microkelvin above absolute zero. A tenth of a millikelvin is not a rounding error in this model. It is the difference between a lifetime under a millisecond and one around 26 milliseconds, on paper, for that one loss channel.
Securities.io is also careful about what the column means. These are theoretical estimates, not measured coherence times. An estimated energy-relaxation lifetime should not be treated as a complete experimental demonstration of how long a qubit keeps phase information. Energy relaxation and dephasing are different failures. A press line that says “coherence of 26 milliseconds” has merged them. The table does not say that. If you cite 25.96 ms, cite the temperature, the design number, and the word estimated.
The fridge is the product, even if the plate is tiny
A plate a few micrometres across is a small object. Securities.io’s point is that device size and system size are not the same. Cooling equipment, connections, shielding, and measurement apparatus may dominate the installation. A qubit you can only see under a microscope can still require a room. People who have stood next to a dilution refrigerator already know this. The October pieces are saying the helium design does not get a waiver.
ScienceDaily says the SHOQ device would need extremely low temperatures, and that researchers have already reached the necessary conditions in previous superfluid helium-3 experiments. Securities.io puts a number on that precedent: the paper notes temperatures around 0.4 millikelvin have been reached in earlier helium-3 work. That supports physical feasibility. It does not mean a controlled quantum device has been operated there as a qubit. “The fridge can get that cold” and “a qubit has run at that cold” are different sentences. The writeups keep them apart. A recap should too.
The fortyfold swing between 0.4 and 0.3 millikelvin is also an operations problem, not just a physics plot. If the calculated lifetime collapses when the mixing chamber is a tenth of a millikelvin warmer, the machine is not “a helium qubit in a fridge.” It is a fridge stability spec with a qubit attached. I do not have the authors’ required stability in microkelvin of drift. I have the two columns. The columns say the second temperature is doing a lot of the work. Anyone planning a facility around SHOQ, which would be early, should budget for holding 0.3 millikelvin, not for “cold enough that helium is superfluid” as a slogan.
None of this is a stock note. The October 5 piece lives on a markets site and still, in the sections used here, sticks to the paper’s table. I am not adding a ticker, a price target, or a “who benefits” list. The commercial question the table actually raises is whether a calculated fortyfold gain is worth a colder fridge than the transmon crowd already complains about. That question is open because the device is not built. Duke’s ion result was a run. This is not that kind of week.
What a serious reader can say on Monday, and what they cannot
You can say Surrey and Northwestern co-authors proposed a micromechanical qubit that stores information in the motion of superfluid helium-3, in npj Quantum Information, DOI 10.1038/s41534-026-01355-3. You can say the design is charge-neutral on purpose, to dodge some electromagnetic noise. You can say the calculations put error rates about 100 times below conventional superconducting qubits, and that ScienceDaily and Inkorr both carry that figure as the researchers’ claim. You can say design 5, 5 micrometres, 14.25 MHz, is the row the Oct. 5 table calls especially promising, at 0.64 ms and 25.96 ms for the two temperatures, from quasiparticle tunneling loss, estimated.
You cannot say a prototype ran. You cannot say the 100-times figure was measured against a transmon in the same lab this week. You cannot turn 25.96 milliseconds into a coherence time a cloud provider will sell. You cannot skip anharmonicity and still claim the oscillator is a qubit. Securities.io spent a section on that gap. It is the gap between “the energy is quantized” and “we can address one transition.”
The next sentence in a real program is a prototype, which ScienceDaily’s framing leaves in front of the group, not behind them. Conditions for the temperature have been reached in earlier helium-3 experiments. Conditions for a qubit are a controlled device: initialization, a gate, a measurement, and a lifetime you timed rather than modeled. Until those exist, SHOQ belongs in the same mental drawer as other 2026 papers that describe a qubit the hardware does not contain yet. The drawer is getting full. The table is still worth keeping, because a fortyfold calculated change across a tenth of a millikelvin is a specific prediction. Specific predictions are what a later experiment can embarrass. A headline that only says “100 times better” cannot be embarrassed. It can only be repeated.
