On October 9, 2026, IonQ announced that it had connected a trapped-ion qubit to a diamond quantum memory with light and generated entanglement at a rate of 1,032 times per second. Entanglement is a phenomenon in which separate qubits share correlated quantum states, and it is a key technology for linking multiple quantum computers so they can work together on a computation. In this experiment, the researchers combined an ion, which performs the computation, with a solid-state quantum memory that interacts well with light, significantly increasing the connection rate. Challenges remain, however, including the effective rate once quantum-state readout is included and the accuracy of the states produced. Connecting qubits of different types with a single photon raises a question: which constraints on distributed quantum computing does this technique ease, and what does it leave for future work?
1,032 Entangled Pairs per Second: What Is the Actual Throughput?
In the experiment, the entanglement generation rate reached 1,032 ± 11 per second, and the fidelity with the target quantum state was 87.9 ± 0.7%. Fidelity indicates how close the quantum state actually produced is to the target state. It is different from the probability of successfully generating entanglement.
The results are summarized in a paper titled "Quantum Networking at the Speed of Quantum Computation," by Lukas Hartung and colleagues, including researchers from IonQ and Duke University. It was posted to arXiv as a preprint, not yet peer-reviewed, on October 7. The results are not simulation-based predictions. They were measured by actually connecting a single barium-138 ion to a single quantum memory.
To understand the figure of 1,032 per second, however, it is necessary to check which processing time it includes.
According to the full text of the paper, the rate of 1,032 per second is calculated from the time between the start of initialization and the point at which entanglement is successfully generated and readout of that state begins. It includes re-preparation after failed attempts, and it took an average of 969 ± 10 microseconds to generate one entangled pair.
Adding the 664-microsecond readout time used to evaluate the generated quantum state brings the overall rate, from generation through measurement, to 612 ± 4 per second.
| Measurement | Result | Scope / meaning |
|---|---|---|
| Entanglement generation rate | 1,032 ± 11 per second | From initialization and re-preparation to the start of readout |
| Average time to successful generation | 969 ± 10 microseconds | Generation time per pair, including initialization |
| Quantum-state readout time | 664 microseconds per pair | Time to measure the state after generation |
| Rate from generation through measurement | 612 ± 4 per second | Effective rate including readout |
| Fidelity with target state | 87.9 ± 0.7% | Evaluated by measuring and reconstructing the generated state |
Even within the same experiment, the 612 per second that includes readout is about 40.7% lower than the 1,032 per second that measures generation alone. This figure is calculated from the central values in the paper as (1,032 − 612) ÷ 1,032 × 100.
The readout process, however, was designed to evaluate the quantum state and was not optimized for speed. In future distributed quantum computing, gate operations that use the generated entanglement will take the place of this measurement step. The figure of 612 per second therefore does not represent an upper limit on the connection speed between quantum computers.
IonQ cites as a point of comparison the ion-to-ion optical connection at 250 per second that O'Reilly and colleagues reported in Physical Review Letters in 2024.
The new generation rate is more than four times that figure. Because the type of qubits being connected and the photon detection methods differ, however, this does not mean the performance of a quantum computer as a whole has quadrupled under the same conditions. What has improved is the entanglement generation rate.
From a Two-Photon Scheme to a Single-Photon Connection
To understand the technical advance, it helps to compare it with earlier methods of generating entanglement.
Conventional high-speed ion-to-ion connections have used a scheme in which one photon is extracted from each ion, and the two photons are interfered and detected.
In this method, entanglement is not generated unless both photons reach the detector. If only one photon arrives, it is not enough.
If the efficiency of extracting a photon from each ion is the same value p, the success probability therefore depends on p squared. Because photon losses occur on both sides, this is a major constraint on improving connection efficiency.
In the new method, by contrast, a photon emitted by the ion is reflected off the quantum memory, and the single photon that returns is detected.
This changes the dependence on the ion-side photon extraction efficiency from p squared in the conventional scheme to p.
Photon losses in the interaction with the quantum memory and in the detection process still occur. However, because two photons no longer need to be aligned and interfered at the same time, the constraint of the conventional method is eased.
The photon emitted by the ion cannot simply be sent straight to the quantum memory, however.
The barium ion used here emits photons at a wavelength of 493.5 nanometers, while the photons used by the quantum memory have a wavelength of 737.45 nanometers.
The research team therefore used a technique that converts the wavelength of a photon while preserving its quantum information. They also converted information encoded in polarization, the direction of the light's oscillation, into two time components: whether the photon arrives early or late.
On the quantum memory side, a resonator is provided whose reflection properties for light change depending on the state of the electron spin.
The researchers flipped the memory's spin between the early and late time components of the photon, and made the photon interact with the quantum memory using both components.
By then detecting the returning photon in a way that does not distinguish between the two time components, they can confirm that entanglement has been generated between the ion and the quantum memory.
Furthermore, they applied real-time corrections depending on factors such as the time at which the photon was detected, so that the same target state was obtained in the end.
This series of processes is explained in detail in Figures 1 and 2 of the paper.
Ion Handles Computation, Diamond Quantum Memory Handles the Optical Interface
The quantum memory used in this experiment is a structure called a silicon-vacancy center (SiV).
This is a special crystal defect containing a silicon atom, formed within a diamond crystal. Although the name includes silicon, it does not refer to a memory chip made of silicon.
By using the electron spin of the SiV as a qubit and combining it with a tiny resonator that confines light, incoming photons can be made to interact with it efficiently.
In the trapped-ion approach, ions are confined by electromagnetic fields, and computation is performed by controlling their quantum states. The SiV memory in this work, by contrast, plays the role of passing quantum information via light.
On the ion side, emitted photons are collected by a lens placed outside the vacuum chamber, whereas on the memory side, coupling to light is strengthened by a resonator.
Rather than building both sides from the same type of qubit, this configuration combines the functions each approach does best.
The ability of the quantum memory to store states will matter when connecting multiple quantum computers in the future.
For example, if entanglement generated between two nodes is stored, that state can be held until the next entanglement is generated between another pair of nodes.
The paper also notes that earlier research has shown a technique for generating new connections while storing entanglement in the nuclear spin of silicon-29 contained in the SiV.
What this experiment demonstrated, however, is a single quantum link connecting the ion and the electron spin of the SiV.
It has not demonstrated the operation of an entire network that incorporates storage in nuclear spins, nor distributed computation by connecting two quantum processors.
The idea of combining different types of qubits is also in line with the direction of the HARQ (Heterogeneous Architectures for Quantum) program run by the U.S. Defense Advanced Research Projects Agency (DARPA).
HARQ aims to choose qubits suited to different roles, such as computation, storage, and communication, and to develop both the technology to connect them and the software to assign operations to each approach.
IonQ also expects this connection method to be applicable to neutral-atom systems and to superconducting-qubit systems via devices that convert between microwaves and light.
At present, however, the only combination for which a connection has actually been demonstrated is a trapped-ion qubit and an SiV quantum memory.
Higher Generation Rates Lower Fidelity: What Is the Cause?
The experiment confirmed a trade-off between the entanglement generation rate and fidelity.
Increasing the number of attempts before re-preparing the state reduces the share of time spent on initialization, which raises the entanglement generation rate.
On the other hand, the fidelity of the generated quantum state decreased.
One cause is that, after interacting with the quantum memory, a photon is sometimes lost on its way back and goes undetected.
In that case, even if the photon was not detected, the spin of the quantum memory may already have been affected.
Specifically, the spin's phase is disturbed, and the effect carries over to the next attempt. In other words, the fact that a photon was not detected does not mean that nothing happened to the quantum memory.
In Table 1 of the paper, assuming that each component is an independent source of error, the fidelity loss is estimated at 4.3% for the ion side and 3.9% for the quantum memory side.
The effect of wavelength conversion is evaluated at 2.9%, and the effect of undetected photons at 1.9%.
These values are estimates obtained by combining correlation-measurement results, however, and the overall fidelity cannot be obtained simply by adding them up.
In wavelength conversion, noise light may be mistakenly detected as a genuine signal. On the ion side, light leaking from the optical modulator and similar sources cause errors.
The research team believes these errors can be reduced by improving the filters that remove noise light and the reflection characteristics of the resonator, and by collecting the light returning from the quantum memory to the detector more efficiently.
Raising both the entanglement generation rate and the fidelity at the same time will require improving the entire optical system, from wavelength conversion through to detection, not just the part that collects light from the ion.
Cooling the device is another challenge.
According to the apparatus configuration shown in Figure 3 of the paper, this quantum memory operates at about 100 millikelvin, a temperature extremely close to absolute zero.
The authors cite earlier research in which a strained SiV operated at 1.5 kelvin, and mention the possibility of using a compact cryocooler.
Whether the generation rate and fidelity achieved here can be maintained at higher temperatures will have to be verified separately.
Distributed Quantum Computing Needs 10,000 to 100,000 Pairs per Second: Challenges for Commercialization
The generation rate IonQ demonstrated this time was about 1 kHz, or roughly 1,000 per second.
The paper, however, anticipates that practical distributed quantum computing will require entanglement generation rates of 10 to 100 kHz, that is, 10,000 to 100,000 pairs per second.
This is a value estimated as necessary for continuously running logical operations across multiple modules, assuming known quantum algorithms along with the qubit fidelities and error-correction schemes expected in the near future.
Although the rate of about 1 kHz achieved here approaches the operation cycle of about 1 millisecond for ion-based quantum computers under development, it has not yet reached the level required for distributed quantum computing.
The research team says that improving the coupling efficiency of light from the ion into optical fiber and the signal path, and adopting a different gate scheme that allows light to be reflected from either spin state of the quantum memory, could be expected to yield generation rates above 5 kHz.
This is a figure expected to be achievable through future improvements, however, and was not demonstrated in this experiment.
Furthermore, at the stage of running multiple quantum links in parallel or using stored entanglement, it will be important whether the fidelity obtained on a single link can be maintained across multiple nodes.
IonQ is also pursuing research on manufacturing quantum memories.
The company describes its research on diamond thin films, carried out jointly with Element Six and Amazon Web Services, as a technical path toward incorporating quantum memories and quantum interconnect devices into semiconductor manufacturing processes.
If diamond thin films can be bonded to substrates such as silicon, existing semiconductor processing technology can be used, making it easier to combine them with optical switches, modulators, and other components.
This experiment did not, however, demonstrate manufacturing costs or yields in volume production.
Deployment of research systems is also moving forward.
The agreement includes deployment of SiV quantum memory nodes, along with expanded access to quantum computers, joint research, and upgrades to experimental equipment. The $7.5 million is the value of the entire agreement and does not indicate the sale price of a quantum memory by itself.
In addition, a partnership with South Korea's SDT announced on September 21 outlined plans to supply the Superion 256 quantum computer and SiV quantum memories, and to establish a manufacturing and system integration site in Gumi, South Korea.
Such contracts and partnerships do not mean, however, that a commercial quantum data center is already operating at the connection speed demonstrated here.
What IonQ has shown is the possibility of combining the strengths of different types of qubits to achieve fast quantum connections while easing the constraints imposed by photon loss.
If the entanglement generation rate and fidelity can be raised further, and if the storage of quantum states and the control of multiple links can also be demonstrated, the vision of expanding computing power by interconnecting small quantum computers will move a step closer to practical use.