"A Steam Engine Smaller Than a Grain of Sand": Quantum Computing's Bottleneck Moves from Qubit Count to Cold and Heat
One-line summary: Aalto University (Finland) reported in *Nature Communications* (Nat. Commun. 17, 6054) the world's first cyclic quantum heat engine built inside a superconducting circuit, using a flux-tunable transmon qubit as the working medium and a quantum circuit refrigerator (QCR) as both hot and cold baths. It ran a complete quantum Otto cycle with a measured average output power of ~0.039 eV and efficiency of ~0.0055 (27% of the ideal Otto limit at that operating point). The same week, Zhongke Liangyi in Huairou, Beijing won third prize at HICOOL 2026 for a "stackable modular helium-free dilution refrigerator," with its large-capacity XXL3000 completing full-machine testing; meanwhile, Quantum Machines' CTO put forward another number — 3.3 microseconds, the feedback speed quantum error correction must catch up with. All three stories point to the same thing: the scaling bottleneck of quantum computing has shifted from "how many qubits can you build" to "can you cool it down, extract its heat, and close the feedback loop in time."
An Engine That Cannot Push a Speck of Dust — Why It Shook the Industry
The working medium is not steam but a flux-tunable transmon qubit — the most common "artificial atom" in modern superconducting quantum computers, with two main energy levels. Its hot and cold reservoirs are not combustion chambers and radiators but a quantum circuit refrigerator (QCR), which can seamlessly switch between "extracting heat" and "injecting heat" via electrical pulses.
The four strokes work as follows:
1. Adiabatic expansion: quickly lower the qubit frequency → level spacing shrinks → the system does work on the outside; 2. Isochoric cooling: connect the QCR in cold mode → excited-state population falls → waste heat is dumped into the cold reservoir; 3. Adiabatic compression: raise the frequency again → work is done on the qubit (fewer excited particles, so this costs far less work than expansion released); 4. Isochoric heating: QCR switches to hot mode → the qubit re-absorbs heat and gets excited → ready for the next cycle.
After each cycle, the net balance is positive. The team tracked three complete cycles, repeated the first cycle eight times, and monitored the state evolution in real time via single-shot qubit readout.
Result: average output power of about 0.039 electron-volts, measured efficiency of about 0.0055 — 27% of the ideal Otto limit at that operating point.
For perspective: a single AA battery stores about 10²² eV. One cycle of this heat engine doesn't produce enough energy to lift a speck of dust by a micrometer.
So why is the entire quantum computing industry watching? The answer lies in its location.
The Otto Cycle: Replacing Watt's Piston with Energy Levels
The Otto cycle, devised by Nikolaus Otto in 1876, is the four-stroke cycle used in most gasoline engines today. Its efficiency ceiling has a closed-form expression:
where \(r\) is the compression ratio and \(\gamma\) the adiabatic index. Gasoline engines have \(r\) around 8–12 and \(\gamma \approx 1.4\), giving a theoretical ceiling of about 55%–60%; actual engines reach just over 30%.
In the quantum version, the "compression ratio" becomes the ratio of maximum to minimum level spacing. But the quantum version has a loss no gasoline engine has — and this is the most fascinating part of the paper.
> Tip: Quantum coherence acts as "internal friction" here. In a classical engine, running fast mainly costs mechanical friction. But at the quantum scale, switching strokes too quickly excites quantum coherence inside the qubit, which manifests as additional heat dissipation and drags efficiency down. This is a speed–efficiency trade-off that no classical thermodynamics textbook contains.
Flooring the accelerator on a quantum engine means paying a quantum tax.
How to Read the Number 27%
The measured efficiency of 0.0055 is 27% of the ideal Otto limit. A modern gasoline engine reaches roughly 55%–70% of its own theoretical ceiling. So this quantum heat engine achieves about half the relative efficiency of a mature combustion engine — not bad for a first proof-of-principle.
Beyond coherence-induced "internal friction," two more loss sources:
- Limited tuning range of level spacing: the transmon's flux-tunable range caps the achievable "compression ratio."
- Non-instantaneous QCR switching: establishing thermal contact takes time, generating irreversible entropy — analogous to incomplete combustion and exhaust losses.
- Aalto's engine addresses whether the cold stage can do work on its own;
- Zhongke Liangyi's refrigerators address whether external cooling can scale elastically;
- Quantum Machines' 3.3 µs addresses whether the quantum and classical sides can talk in time.
Why bother? Because the engine was never meant to supply power. It proves that a controllable quantum system can convert heat to work stably and repeatably across cycles — and you cannot do qubit reset with a device that only works once.
Millions of Cables: The Real Achilles' Heel of Quantum Computing
Today's superconducting quantum computers are giant golden "chandeliers." The chip at the bottom sits at about 15 millikelvin (colder than deep space), while control and readout electronics remain at room temperature, connected by gold-plated coaxial cables.
Corresponding author Mikko Möttönen, professor at Aalto University, ran a grim calculation: Finland's national quantum strategy targets roughly 1,000 error-corrected logical qubits around 2035, implying hundreds of thousands of physical qubits. With the current control architecture, that means millions of specialty microwave coaxial cables at thousands of euros each.
And money is only part of it. Every cable penetrating the cryostat carries in room-temperature heat and electromagnetic noise. Space fills up, cooling capacity is exhausted, noise floods in — the superconducting quantum computer would be strangled by its own cables before its compute takes off.
The quantum heat engine points toward cold-stage energy autonomy: if tiny local temperature differences can run on-chip heat engines inside the cryogenic chip — autonomously performing qubit reset, state readout, even on-chip microwave pulse generation — the need for millions of expensive cables plummets.
To be clear: this engine still relies on the dilution refrigerator's extreme low temperatures and cannot replace that infrastructure. The current prototype remains an "externally driven" engine — the flux ramps and QCR pulses all come from external instruments.
The team has already laid out the next step in *SciPost Physics*: a fully autonomous quantum heat engine requiring no external control pulses, generating detectable coherent microwave power spontaneously from heat flows through superconducting circuits combined with nonlinear quantum electrodynamics.
The Other Half of the Answer: Refrigerators as Stackable Racks
The same week (August 27–28), in Huairou, Beijing: Zhongke Liangyi (Beijing) Technology Co. won third prize at the HICOOL 2026 global startup competition for its "modular quantum computing ultra-cold platform" — a stackable, helium-free dilution refrigerator.
Traditional dilution refrigerators are "cylinder-style": scaling up means buying a bigger cylinder. Zhongke Liangyi's design is rack-based and standardized, supporting multi-unit stacking, so cooling capacity scales alongside qubit count. Its layered cooling architecture drastically reduces cable heat leakage and allows cable maintenance without warming up the whole machine — saving users days per maintenance cycle. The newly developed large-capacity XXL3000 dilution refrigerator passed full-machine testing with all key metrics met or exceeded — the largest-capacity dilution refrigerator tested in China. The full technology stack is 100% independently developed and manufactured. Founded in Huairou Science City in February 2024 with founder Ji Zhongqing, the company collaborates with the Beijing Academy of Quantum Information Sciences and the Institute of Physics, CAS.
| Route | Representative | Approach | Timescale | |---|---|---|---| | Scale external cooling | Zhongke Liangyi modular fridges | Make cooling itself elastically scalable | Delivering now | | On-chip self-cooling | Aalto quantum heat engine | Let the chip manage its own heat | Proof of principle → theory roadmap | | Bypass cryogenics | Japan's "Shunkai" neutral-atom machine | Take a path that needs no mK temperatures | Online |
That third route also happened this week: the full-stack neutral-atom quantum computer "Shunkai" by Kenji Omori's team at Japan's Institute for Molecular Science went live. It runs its outer systems at ambient temperature and pressure, with the QPU supplied by US-based Infleqtion, using optical tweezers and Rydberg states for gates, targeting ten thousand qubits with error correction by 2031.
3.3 Microseconds: The Bottleneck in the Time Dimension
If "cold and heat" is the thermodynamic bottleneck, 3.3 microseconds is the temporal one.
Quantum Machines co-founder and CTO Yonatan Cohen (PhD in physics, Weizmann Institute; co-founded the company in 2018) gave this number: 3.3 microseconds for a complete quantum–classical feedback loop — read the qubit state, send data to a GPU or other classical accelerator, complete decoding, return the result to the controller, and decide the next quantum operation.
His message is blunt: for some modalities — especially superconducting qubits — sufficiently low latency is essential, otherwise large-scale quantum error correction simply cannot happen, because classical data volumes accumulate too fast.
He also flagged something often overlooked: calibration is another form of error correction. Qubit parameters drift over time and need continuous monitoring and adjustment — again dependent on fast communication between quantum controllers and classical accelerators.
He predicts a "highly heterogeneous" future architecture: quantum controllers coordinate operations on the QPU while streaming information to HPC resources; classical components handle error decoding, system stabilization, and logical operation coordination; instructions flow back to the controller for execution.
On Majorana qubits — his PhD topic — he notes the appeal is building part of the error correction into the hardware itself, eliminating continuous measure-decode-correct cycles. But he is clear: building a single component and integrating all components into a large-scale fault-tolerant system are two very different things.
A Counterintuitive Observation: The Bottleneck Is Leaving the Qubit Itself
Put the three stories side by side:
In Cohen's words: "More and more bottlenecks are no longer confined to the quantum processor itself… many performance limits people encounter stem from these interfaces, not just the qubits."
For the past decade, quantum computing progress has been defined by a single metric — qubit count. But as the industry shifts from "proving it can run" to "making it reliable and valuable," qubit count still matters, but what determines whether those qubits become a real computer is how fast the whole system responds, how stably it stays cold, and how cleanly it sheds heat.
Epilogue: An Echo Across 250 Years
In 1769, James Watt patented his improved steam engine. To compute the limits of cylinders and coal, physicists founded classical thermodynamics — a science born from engineering pressure.
Two hundred and fifty-seven years later, on a superconducting chip smaller than a grain of sand, pistons have become energy levels, coal fires have become voltage biases, and steam has become a single qubit. Thermodynamics has put on a quantum-mechanical uniform.
Near absolute zero, it is trying to solve an deeply unromantic but critical problem on the road to universal quantum computing: how to remove heat, how to keep the cold, and how to send correction instructions back before errors spread.
These three things may determine whether the 1,000-logical-qubit machine of 2035 materializes better than another thousand physical qubits would.
References
1. Uusnäkki, T., Mörstedt, T., Teixeira, W., Rasola, M. & Möttönen, M. *Initial demonstration of a quantum heat engine based on dissipation-engineered superconducting circuits*. Nature Communications 17, 6054 (2026). https://www.nature.com/articles/s41467-026-6054 2. Aalto University, "World's first superconducting quantum heat engine opens the path to larger quantum computers," 2026-07-13 3. Huairou Media, "HICOOL 2026: Fully independently developed — Huairou Science City company builds quantum computing 'super refrigerator'," 2026-08-28, https://www.toutiao.com/article/7678869329349214735 4. Photon Box QUANTUMCHINA, "3.3 microseconds: the feedback speed quantum error correction must catch up with — a conversation with Quantum Machines CTO Yonatan Cohen," 2026-08-27 5. Quantum Computing Daily (Tencent News), 2026-08-27 / 2026-08-28 issues, including the Japan "Shunkai" neutral-atom quantum computer launch item