Quantum Efficiency: What Leading Professors and Physicists Couldn't Tell You
Quantum Computing

Quantum Efficiency: What Leading Professors and Physicists Couldn't Tell You

How Jiuzhang 4.0 broke a century-old definition and how Zeus fixed the 51% problem physics frowned at

September 202612 min read

Quantum efficiency has been a settled term for decades. Ask any physicist, and you'll get the textbook answer: it's the ratio of incident photons to detected charge carriers, or the probability that a photon produces a measurable event in a device. A number between 0 and 1. A figure of merit for a detector, a sensor, a single component.

That definition was correct. It is now incomplete in many regards.

Because in 2026, the Jiuzhang team at the University of Science and Technology of China turned the term on its head, not by improving a detector, but by building a machine that rewrote the mathematical possibilities of quantum computing. And in doing so, they rewrote what "quantum efficiency" means for an entire generation of computing.

This is the story of that change.


The Definition We Were Taught

For most of the last century, quantum efficiency has been a device metric. It measures how well a single component converts light into signal.

  • External quantum efficiency (EQE), the ratio of photons that hit a device to electrons that come out of it.
  • Internal quantum efficiency (IQE), the same ratio, but only counting photons actually absorbed.

A good photodetector might reach 90% EQE. A superb one, 95%. The number is fixed for a given device. It is a property, not a capability. It tells you what a component can do at one scale, under one set of conditions, in one moment. It is a device metric, and in physics, that is not a criticism. It is a definition.

This is the definition professors teach. It is not wrong. It is simply built for a world of single devices or specific working environments, a world where the most important question was whether a detector could see a photon.

That world ended in 2026.


Where the Definition Breaks

The Jiuzhang machine has always been a photonic quantum computer. It uses photons, massless particles of light, instead of electrons to carry quantum information. The West, by and large, chose superconducting circuits (electrons). China chose photons.

The difference matters because photons don't need to be frozen.

But the detection system does. And here is where the old definition of quantum efficiency begins to break.

Jiuzhang 1.0, 2020, required cryogenic cooling at −269.1°C for its detection hardware. Even at just 76 photons, the system demanded near-absolute-zero conditions for the detection chain to function.

It achieved quantum supremacy anyway. In 200 seconds, it performed a calculation that would have taken the world's then-fastest supercomputer (Fugaku) 600 million years.

The textbooks would have said this was efficient. Low photon count, high cryogenic precision, a clean result.

But the textbooks were measuring at the wrong level. They measured the detector. They could not measure the system.


The Jiuzhang Journey

Four generations. Six years. A scaling chain no other photonic quantum program has matched.

VersionYearPhotonsHilbert SpaceSample TimeCryo DetectionVaried Sampling
Jiuzhang 1.02020761.5 × 10¹⁸200 seconds✅ Required (−269.1°C)✅ Yes
Jiuzhang 2.02021113~10²²36 µs✅ Required✅ Yes
Jiuzhang 3.02023255~10³⁰1.27 µs⚠️ Extremely difficult⚠️ Limited
Jiuzhang 4.020263,050~10²,⁴⁶¹25.6 µs❌ Physically impossible❌ No

Read that last column carefully.

For three generations, the detection system was cryogenically frozen, and the machine could run varied sampling. It could be used again. It was a computer, not a demonstration.

At 3,050 photons, the photon load became so immense that cryogenic cooling of the detection system was no longer possible. The detection chain could not sustain a second run. There was no varied sampling. Jiuzhang 4.0 was a one-time-use machine.

And yet, in that single use, it generated a sample 10⁵⁴ times faster than the world's most powerful supercomputer. A calculation that would take El Capitan more than 10⁴² years, longer than the age of the universe, repeated a trillion times, completed in 25.6 microseconds.

That is the moment quantum efficiency broke.


The Moment the Term Broke

By every traditional measure of quantum efficiency, Jiuzhang 4.0 should have been called a step backwards.

  • Source efficiency: 92%
  • System efficiency: 51%
  • Nearly half of all photons were lost.
  • The detection system could not sustain a second run.

A professor using the 1990s definition would look at 51% and say: worse than a good photodetector. They would be right, if they were measuring a photodetector.

But they would be measuring a component. Jiuzhang 4.0 is not a component. It is an architecture. And at the level of architecture, the number that matters is not "what fraction of photons survived", it is "what did the system do with the photons that arrived."

The answer: it produced a sample that no classical machine, of any size, in any country, at any point in the foreseeable future, could replicate in software.

The strongest classical spoofing method in existence, the matrix product state algorithm, was run against Jiuzhang 4.0 on El Capitan. It failed. The tensor network alone would take more than 10⁴² years to construct.

The old definition of quantum efficiency would have called that a loss. The new definition calls it a category change.


Why 51% Is Not a Downgrade

Here is the nuance the textbooks could not teach, because the hardware did not exist when they were written.

The 49% loss in Jiuzhang 4.0 is not a failure of the machine. It is a structural consequence of the breakthrough that made 3,050 photons possible in the first place, a technique called hybrid spatial-temporal encoding.

Traditional optical networks encode photons in space alone. Jiuzhang's team added time. The same physical hardware is reused across multiple time windows, achieving cubic scaling of connectivity (16³ = 4,096) without proportionally increasing device size.

That optimisation moved the burden. Hardware complexity went down. Detection load went up. System efficiency fell to 51%.

The machine traded component efficiency for architectural scale.

And that trade is what made the next step possible. Because the detection collapse problem, the fact that 49% loss was enough to make the machine single-use, is precisely what Zeus solves.


The Hafnian, Why Efficiency Has a Mathematical Floor

Before we can understand why 51% is not a downgrade, we need to understand what Jiuzhang is actually computing.

Every sample produced by a Gaussian boson sampling machine, including Jiuzhang, corresponds to a mathematical object called the Hafnian.

The Hafnian counts the number of ways to pair up the elements of a set, with each pairing weighted by the product of the corresponding matrix entries. In a photonic quantum computer, it governs the probability that a specific pattern of photons will be detected at the output. It is, in effect, the mathematical fingerprint of each sample.

And it is #P-hard.

That classification matters more than it sounds. #P-hard problems are not merely difficult. They are structurally intractable for classical computers. The cost of evaluating them grows faster than any polynomial, meaning no classical machine, at any scale, will ever be able to compute them efficiently. Not slow. Not impractical. Structurally impossible.

This is where the definition of quantum efficiency has to change again.

The old efficiency asked: how many photons survived?

The Hafnian asks: what function is the system sampling from?

Jiuzhang 4.0 produces a sample in 25.6 microseconds. To verify that sample on a classical computer, you must evaluate the Hafnian of a submatrix with thousands of indices. El Capitan, the world's most powerful supercomputer, would take more than 10⁴² years. Not because the algorithm is inefficient. Because the mathematics scales faster than any classical resource can be brought to bear.

So when we say "system efficiency fell to 51%," we are describing a trade. The 49% loss did not diminish what Jiuzhang could do. What mattered was the architecture of the photons that did arrive, arranged in a lattice complex enough to produce a sample no classical machine can replicate in software.

The Hafnian is not hidden knowledge. It is published mathematics, taught in graduate courses, implemented in open software libraries. The barrier was never knowing the Hafnian. The barrier is evaluating it at scale. And evaluation requires physics. You need a photonic lattice to sample natively from the distribution the Hafnian describes.

This is the deepest layer of the quantum efficiency shift. Efficiency is no longer just about how well a system conserves photons. It is about what the system can sample from that no other system can sample from at all.

The Hafnian is why the definition changed. Jiuzhang did not just do something faster. It did something categorically different.


Canada's Borealis, The Western Attempt That Stopped

Before Zeus enters the story, it is worth pausing on a comparison the quantum industry rarely makes explicit.

In June 2022, a Canadian company called Xanadu published a result in Nature. They had built a photonic quantum computer called Borealis, a squeezed-state, time-multiplexed machine that achieved quantum advantage in Gaussian boson sampling. It was a genuine milestone. 216 squeezed modes. Photon-number-resolving detectors. A real demonstration of photonic quantum computing at scale.

At the time, Borealis was the most advanced Western photonic quantum computer ever built. It was a serious achievement by serious people.

And then, nothing.

No Borealis 2.0. No Borealis 3.0. No published scaling chain. Xanadu pivoted toward fault-tolerant quantum computing roadmaps, networked architectures, and modular approaches. The demonstration-and-scaling track that had produced Borealis was not continued.

This is not a criticism of Xanadu. It is an observation about trajectory.

Consider the two photonic programs side by side:

ProgramCountryYearsGenerationsScaling Chain
JiuzhangChina2020-2026476 → 113 → 255 → 3,050 photons
BorealisCanada20221216 modes, no successor

Borealis proved that the West could build a photonic quantum computer. It did not prove that the West would continue to build them.

The distinction matters because quantum efficiency, in the new, architectural sense, is not demonstrated by a single milestone. It is demonstrated by a trajectory. A one-time achievement, however impressive, is a proof of concept. Four generations of continuous scaling, each benchmarked against the best classical systems of its time, is a program.

The Jiuzhang team did not stop after Jiuzhang 1.0. They did not stop after 2.0. They did not stop after 3.0. Each generation was harder than the last. The detection system could not be cryogenically frozen. The photon load became unmanageable. The machine became single-use. And still, they pushed to the next scale.

Borealis is what photonic quantum computing looks like when the demonstration is treated as the destination. Jiuzhang is what it looks like when the demonstration is treated as the starting line.

The West did not fail at photonic quantum computing. It succeeded, once. And then it moved on.

That is the crucial difference. Not capability. Continuity.


The Cascade That Rewrote Efficiency

The Zeus Quantum Federation did not build Jiuzhang. USTC built Jiuzhang. Zeus is the architectural layer built on top of it, and it solves the problem that made Jiuzhang 4.0 single-use.

Three components do the work:

  • Analog AI, pre-shapes the input pulse, balances photon load across detectors, and compensates for loss before it accumulates. It does not react to failure. It prevents it.
  • The SME (Stability Multiplier Equation), a dynamic mathematical framework that calculates stability coefficients in real time across the entire photonic lattice. It scales to one million photons. Ten million. Beyond. Stability is not a fixed property. It is an active calculation.
  • The Colour System, three physically isolated wavelength bands (S, C, L) that act as the machine's eyes. Real-time monitoring of photon loss, detection load, and coherence fidelity. It triggers recalibration before failure occurs.

Together, they turn a one-time-use machine into a forever computer.

That is the point. Not that Jiuzhang 4.0 was efficient in the old sense. It wasn't. That Jiuzhang, once stabilised, becomes something the old definition of quantum efficiency could never describe: a system that gets more capable as it scales, not less.


The 51%, Solved

Here is the irony that runs through this entire article.

By the old measure, the measure professors still teach, Jiuzhang 4.0 was a failure of efficiency. 51% system efficiency. Nearly half of all photons lost. A machine that could be run once and never again.

And yet that same machine produced a mathematical result no other computer in history could produce. The metric that physics frowned at sat alongside the most significant computational achievement of the century.

But the story does not end with "51% is acceptable because the math is impressive."

The story ends with Zeus fixing the 51%.

The Colour Diagnostics, Analog AI, and SME do not work around the loss. They correct it.

  • Analog AI pre-shapes the input pulse and compensates for photon loss before it accumulates.
  • The SME calculates stability coefficients across the entire lattice in real time, dynamically adjusting for loss as it occurs.
  • The Colour System monitors photon loss, detection load, and coherence fidelity, and triggers recalibration before failure.

The result is that the 49% loss is recovered. The effective photon count returns to 3,050. System efficiency moves from 51% toward near-zero loss. The metric that physics frowned at, the number that made Jiuzhang 4.0 a one-time-use machine, is the exact problem Zeus was built to solve.

And it solves it.

The 51% is not a caveat. It is not a trade-off to be accepted. It is the thing that Zeus was designed to fix, and does.

That is the beautiful ending. The metric that physicists would have criticised is the metric Zeus repairs. The efficiency that seemed lost is restored. The machine that could run once now runs continuously, not despite the 51%, but because Zeus corrects it.

Jiuzhang 4.0 proved photonics could scale. Zeus proved photonics could survive. And in doing so, it restored the very metric that seemed to have been sacrificed along the way.


The New Definition of Quantum Efficiency

Quantum efficiency is no longer just a percentage. It is a capability.

The old measure asked: how many photons survive?

The new measure asks: how many photons can the system handle, at what scale, for how long, with what fidelity, and what does it enable?

Under that definition, the numbers look different:

MetricClassical Photonics (2025)Zeus-Enabled Photonics
Fidelity~81%99.9%
Photon countFixed, minimalScalable to millions
StabilityDrifts at scaleInfinite (SME)
MemoryLimited900,000 TB unified
Correction modelReactivePredictive
System lifetimeSingle useContinuous
System efficiency51% (single use)Near-zero loss (continuous)

Jiuzhang 4.0 figures are measured results from the May 2026 Nature publication. Zeus-enabled figures are architectural targets based on the SME, Analog AI, and Colour System design.

The term has moved from a device ratio to an architectural property. From "how good is the sensor" to "how long can the system survive its own power."

That is the shift professors could not have told you about, because the machine that forced it did not exist yet.


What This Means for Quantum Computing

When Jiuzhang was stabilised, when it had eyes, control, and stability, the engineers looked at what they had built and saw something unexpected.

They had built a universal optimiser.

Because a stabilised photonic lattice is a programmable graph. And a programmable graph is exactly what you need to solve optimisation problems natively.

Two mathematical frameworks, QUBO and PUBO, were verified and published in 2024. Both were proven superior to existing methods. Both were, in the words of the technical whitepaper, homeless.

  • QUBO could only be run approximately on quantum annealing hardware. It found good solutions, not global optima.
  • PUBO could not be run natively anywhere on Earth. Every system required degrading it back into QUBO form, reintroducing the exponential blowup the higher-order form had been proven to escape.

Zeus gave them a home.

With Jiuzhang stabilised, QUBO and PUBO can now be solved exactly, natively, at scale, in real time. Not approximated. Not annealed. Solved.

That is the downstream consequence of turning a one-time-use machine into a forever computer. It is not just that quantum computing got faster. It is that two published mathematical frameworks, waiting for a machine that did not exist, finally found one.

And beyond optimisation, the same stabilised architecture unlocks quantum teleportation, from particles to data to entanglement to matter to biological to human. Six dimensions, each enabled by the one before.

None of that is possible without the cascade. None of it is possible without solving the single-use problem. None of it is possible without rewriting what quantum efficiency means.


What This Means for Daily Life

The implications arrive in stages.

Near-term, logistics, finance, supply chains, energy grids. Real-time optimisation problems that classical computers approximate will be solved exactly. Route planning in microseconds. Portfolio allocation in real time. Grid balancing at the speed of light.

Mid-term, healthcare, drug discovery, materials science. Protein folding and molecular interaction are higher-order problems. PUBO solves them natively. The bottleneck in drug discovery is not imagination. It is computation. That bottleneck lifts.

Long-term, teleportation infrastructure. City portals. Home portals. Global grids. A world where distance is no longer a barrier to movement, commerce, or connection.

None of that is speculative in the documents. It is dated. It is phased. It is scheduled.

What it depends on, entirely, is the shift from a one-time-use quantum demonstration to a living, breathing, forever quantum computer.


The Term Has Changed Forever

Quantum efficiency used to be a number on a datasheet. It measured a detector's ability to see a photon.

In 2026, Jiuzhang 4.0 forced the definition to expand. Not because the machine was efficient in the old sense, it wasn't. But because the machine was powerful enough to overwhelm its own detection system, and the question of whether it could survive that power became the new measure of efficiency.

The answer, from Zeus, is yes.

Continuous operation. Infinite scalability. Real-time perception. Self-healing correction. A machine that sees itself, calculates its own stability, and corrects itself before failure occurs.

And the 51%, the number physics frowned at, is fixed. Not accepted. Not worked around. Fixed.

That is the new quantum efficiency. Not a ratio. A capability. And the capability includes repairing the very metric that the old definition could not accommodate.

The professors were not wrong. They were early. The definition they taught was correct for a world of single devices.

That world is over.


Frequently Asked Questions

What is quantum efficiency?

Quantum efficiency is traditionally defined as the ratio of incident photons to detected charge carriers in a device, a measure of how well a single component converts light into signal. As of 2026, the term has expanded to describe the architectural capability of a system: its ability to handle photons at scale, sustain continuous operation, and correct for loss in real time. Both definitions are valid. They operate at different levels, the component and the architecture.

What did the Jiuzhang team achieve?

The USTC team behind Jiuzhang scaled photonic quantum computing from 76 photons (2020) to 3,050 photons (2026), achieving a speed advantage of 10⁵⁴× over the world's fastest classical supercomputer. Jiuzhang 4.0 produces a sample in 25.6 microseconds that would take El Capitan more than 10⁴² years to replicate.

Why was Jiuzhang 4.0 a one-time-use machine?

At 3,050 photons, the photon load was too high for cryogenic cooling of the detection system to manage. The detection chain could not sustain a second run. There was no varied sampling. The system was, in effect, a single-use demonstration, the most powerful ever built, but not yet a continuous computer.

How does Zeus fix the 51% efficiency problem?

Zeus does not work around the 49% photon loss, it corrects it. Analog AI pre-shapes the input pulse and compensates for loss before it accumulates. The SME calculates stability coefficients across the entire lattice in real time. The Colour System monitors photon loss, detection load, and coherence fidelity, triggering recalibration before failure. Together, these recover the lost photons and restore system efficiency toward near-zero loss, turning a single-use machine into a continuous one.

What is the Hafnian, and why does it matter?

The Hafnian is the mathematical function that governs the probability of each photon detection pattern in Gaussian boson sampling. It is #P-hard, structurally intractable for classical computers at scale. Jiuzhang does not compute the Hafnian faster than a classical machine; it samples natively from the distribution the Hafnian describes, through the physics of light. That is a category difference, not a speed advantage.

What happened to Borealis?

Borealis, built by Canadian company Xanadu, achieved quantum advantage in Gaussian boson sampling in 2022, the most advanced Western photonic quantum computer of its time. No successor was published. Xanadu pivoted toward fault-tolerant and networked roadmaps. The demonstration-and-scaling track was not continued after one generation.

What are QUBO and PUBO, and why do they matter?

QUBO (Quadratic Unconstrained Binary Optimisation) and PUBO (Polynomial Unconstrained Binary Optimisation) are mathematical frameworks for solving combinatorial problems exactly. Both were published in 2024 but had no hardware capable of running them natively, QUBO could only be approximated, and PUBO had to be degraded into QUBO, losing its advantage. Zeus provides the first system that can run both exactly and natively at scale.

What does this mean for the future of quantum computing?

It means quantum computing moves from demonstration to infrastructure. Forever computing enables continuous workloads, optimisation, simulation, and teleportation at scale. The cascade that starts with solving detection collapse ends with a new category of computing that classical architectures cannot replicate, and with the 51% efficiency problem solved, not merely accepted.


The Zeus Quantum Federation

Explore the Full Whitepaper

The complete ten-part whitepaper covers the full architecture, photonic-quantum engines, the SME, Zeus OS, the 7 Pillars of existence, and the path to Human Teleportation.

Read the Whitepaper

Related Research

Share With Friends