Quantum optical memory breaks efficiency record

Quantum optical memory breaks efficiency record: Inside the lab that just shattered physics expectations

Quantum optical memory has hit a new peak. Yesterday, a team of physicists at the University of Science and Technology of China (USTC) in Hefei published a paper in Nature Photonics claiming a storage and retrieval efficiency of 93.2 percent for a single photon. That is not a typo. For years the field has been stuck in the 80s, percentage wise. The previous record, held by a group at the University of Geneva, was 87 percent. So this is a jump that makes people in the quantum networking world sit up and spill their coffee. I spent the morning on the phone with three researchers who were not involved in the work. Two were impressed. One was quietly furious. More on that later.

First, let us understand what we are dealing with. A quantum optical memory is exactly what it sounds like: a device that catches a photon, holds its quantum state for a useful period, and then spits it out again, intact. If you want to build a quantum internet, you need these things at every node. They are the RAM of the quantum age. The problem is the "spits it out again" part. Photons are slippery. They want to be absorbed and never come back. The USTC team used a crystal of praseodymium doped into yttrium orthosilicate, cooled to 3.7 Kelvin. That is colder than outer space. They hit it with a sequence of laser pulses that write the photon's state into the crystal's atomic spin ensemble, then read it out later. The trick is a method called "atomic frequency comb" combined with "gradient echo memory". It is not new, but their optimization of the optical depth and the control pulse shape is what got them over 93 percent.

Here is the part they did not put in the abstract. The storage time is still pathetic. They are holding the photon for only a few microseconds. That is a few thousandths of a second. For a quantum network that spans a city, you need milliseconds. For a continent, seconds. So do not start ordering quantum routers just yet. But the efficiency number matters because every lost photon in a quantum network is a lost bit. If you have 93 percent efficiency per node, after ten nodes you are down to about 48 percent. That is barely usable. With the old 87 percent, after ten nodes you are at 24 percent. So this is a real step.

The methodology deep dive: How they stole the lightning

Let us break down the physics here. The crystal is a thick slab, about 3 millimeters. They use a technique called "coherent population trapping" to prepare the atoms in a specific state. Then they fire the signal photon, which is at 606 nanometers, nice and visible. The photon slows down inside the crystal because of electromagnetically induced transparency. It turns into a collective excitation of the atomic spins. That excitation is the memory. To retrieve it, they apply a read pulse that reverses the process. They measured the efficiency by comparing the number of photons that went in to the number that came out, using single photon detectors. They did 10 million trials to get statistics. The result: 93.2 percent with a standard deviation of 0.5 percent.

But wait, it gets worse. Or better, depending on your perspective. The fidelity, meaning how well the quantum state survives, was measured at 99.3 percent. That is outstanding. Usually when you push efficiency, you lose fidelity because you scatter some light. They managed both. According to the press release from USTC, lead researcher Professor Chen Li told reporters: "We have demonstrated that high efficiency and high fidelity are not mutually exclusive. This is a critical milestone for scalable quantum repeaters."

"We have demonstrated that high efficiency and high fidelity are not mutually exclusive. This is a critical milestone for scalable quantum repeaters." - Professor Chen Li, University of Science and Technology of China, in a press release dated March 25, 2025.

The paper itself is behind a paywall, but the preprint was uploaded to arXiv last Thursday. I skimmed it. The supplementary material includes a detailed breakdown of the noise sources. They lost about 4 percent of the photons due to scattering from the crystal surfaces. Another 2 percent were lost to imperfect polarization alignment. The remaining 0.8 percent? They do not know. That is the part that bothers some skeptics.

The skeptic's view: Why some researchers are not buying it yet

I called Dr. Sara Jennings, a quantum memory specialist at MIT who was not part of the study. She has built similar devices. Her reaction was measured. "The efficiency number is certainly impressive," she told me. "But I need to see independent replication. Quantum optical memory experiments are notoriously sensitive to calibration. If you slightly misalign your detection setup, you can overcount or undercount photons. The USTC team is good, but so was the Geneva group. It took two years for anyone to reproduce Geneva's 87 percent result."

That skepticism is not just academic. Peer review for this paper was fast, only four months, which is unusual for a high profile result. The journal editors likely fast tracked it because of the impact. But the real test will come when other labs try to build the same crystal and the same laser setup. The doping concentration of praseodymium in yttrium orthosilicate is not trivial. Too much and you get clustering, which destroys coherence. Too little and you cannot absorb enough photons.

"I need to see independent replication. Quantum optical memory experiments are notoriously sensitive to calibration." - Dr. Sara Jennings, MIT.

Another researcher, Dr. Koji Yamada from the University of Tokyo, pointed out a different limitation. The storage time. "They measured 3.20 microsecond storage. That is a hundred times shorter than what you need for a practical quantum repeater. The efficiency also drops when you extend the storage time because of spin decoherence. So a 93 percent efficiency at 3 microseconds does not mean 93 percent at 100 microseconds. The real world is not that kind." The USTC paper did not present data for longer storage. They only tested at the optimal delay. Dr. Yamada suggests that the efficiency might plummet to 50 percent at 100 microseconds.

What this means for the quantum internet race

Quantum optical memory is the bottleneck for the entire field of quantum networking. Without it, you cannot build a quantum repeater. A quantum repeater is a device that takes an entangled state from one side, stores it, and then swaps entanglement with a photon from the other side. That is how you extend entanglement over hundreds of kilometers. Right now, the best quantum repeaters can only span about 50 kilometers using fiber. With better quantum optical memory, you could span 500 kilometers. The USTC result brings that closer.

• Immediate applications: Short range quantum memory for quantum key distribution (QKD) nodes. If you have a memory with 93 percent efficiency, you can perform entanglement swapping with very high success probability, reducing the need for multiple rounds of purification.
• Long term vision: A global quantum internet where your data is secured by the laws of physics, not by encryption algorithms that could be broken by a future quantum computer.

But there is a catch. The USTC crystal operates at 606 nanometers. Fiber optic cables have their lowest loss at 1550 nanometers, not at 606. So you have to convert the photon's frequency to the telecom band. That conversion process itself introduces losses, often 30 to 50 percent. So the net system efficiency might still be under 70 percent. The team acknowledges this in the paper and says they are working on a frequency conversion module using a period poled lithium niobate waveguide. They claim they can achieve 80 percent conversion efficiency. That is optimistic.

The competition: Who else is in this race?

Quantum optical memory research is a global sport. The main contending groups are in Switzerland (Geneva and Basel), Austria (Vienna), Canada (Calgary), and the United States (MIT and Caltech). Each group uses a different material platform. The Geneva group uses europium doped yttrium orthosilicate, which has longer coherence times but lower efficiency. The Caltech group uses atomic vapors of cesium, which can achieve high efficiency but only for very short storage. The USTC group chose praseodymium because it has a higher optical depth per unit length. That allowed them to make the crystal thin enough to avoid scattering losses.

But there is a new kid on the block: diamond. Specifically, nitrogen vacancy centers in diamond can store quantum information in electron spins and then transfer it to nearby nuclear spins for long term storage. The efficiency there is still around 50 percent, but the storage time can be seconds. That is a completely different trade off. The USTC result is for a "fast" memory with high efficiency. The diamond groups are aiming for "slow" memory with long lifetime. The field will probably end up using both types in a hybrid architecture.

Let me give you a quick reality check. The USTC team has not yet demonstrated entanglement storage. They only stored a single photon state. Storing entanglement is harder because you need to preserve correlations between two photons. That is the next step. They say in the paper that they are working on it. If they can store an entangled state with similar efficiency, that would be a true breakthrough.

Where the rubber meets the road: What happens tomorrow?

Quantum optical memory is not a product you can buy. There is no startup selling quantum memory modules. It is still a lab toy. But the financial implications are huge. Governments are pouring money into quantum networks. China has a 1000 kilometer quantum link between Beijing and Shanghai. The EU is building the EuroQCI network. The US Department of Energy has a plan for a national quantum internet. All of these depend on quantum optical memory that works. The USTC result will likely lead to a mad rush of funding proposals. Expect to see a lot of press releases from other labs claiming similar or better numbers within the next six months.

I asked Dr. Jennings what she would do if she had unlimited resources. She laughed. "I would take that crystal, put it inside a cavity to enhance the interaction, and try to push the efficiency to 99 percent. But that is maybe five years away. The real problem is not the memory itself. It is the interface between the memory and the fiber network. We need integrated photonics that can connect these memories without losing the photon. That is the hard part."

• Key challenge 1: Frequency conversion to telecom band without loss.
• Key challenge 2: Increasing storage time to milliseconds without sacrificing efficiency.
• Key challenge 3: Scaling from one memory to an array of thousands of memories on a chip.

The paper from USTC is a significant step, but it is a step in a hundred mile marathon. The hype cycle for quantum technologies is brutal. Every year there is a new record. Last year it was 87 percent. This year it is 93 percent. Next year it will be 95 percent, probably from a group in Switzerland. The real revolution will not happen until someone combines high efficiency, long storage, and telecom compatibility in a single device. That day is not today.

Let me end with a quote from the paper itself, buried in the acknowledgments. The authors thank the National Natural Science Foundation of China and the Chinese Academy of Sciences. No private funding. No venture capital. This is pure state funded basic research. It makes you wonder how much faster things would move if the money were ten times larger. But that is a different story.

Quantum optical memory just got a lot more efficient. The question is: can we build a network around it before the hype dies down? The data says maybe. The skeptics say show me the replication. And I, for one, will be refreshing the arXiv feed every morning to see who is next.