The world would look radically different without rulers and measuring tapes that fit into a pocket. Carpenters, fashion designers and engineers rely on these trusty tools to check the size of everything from a wooden board to a fabric swatch. But physicists who work with light lack that same convenience for one of the basic measurements of their craft. They routinely need to measure and compare the frequencies—colors—of the light waves they are using. To date, they have lacked a similarly pocket-sized tool for routine measurements, and they make do with cumbersome equipment that crowds their lab space and is far from portable.An artistic visualization of a novel process that forms an optical frequency comb, which researchers can use to perform a variety of precision measurements. During the formation process, two lasers circulate in a small resonator and spread out, forming the blue and red curves, while also interacting to produce the purple curve. All three curves merge into a comb that provides a pristine ruler for measuring light, represented by the purple lines at the top of the image. (Credit: Carl De Torres at the Optics Lab)

The standard tool for measuring the frequency of light is called an optical frequency comb. This device produces a rainbow of different frequencies of light, all spaced at regular intervals like the tick marks on a measuring tape. Frequency combs let researchers measure the difference between distinct frequencies and are crucial for many experiments and measurements that use light. Large lab setups have proven the usefulness of frequency combs by enabling the creation of the most precise clocks in the world—atomic clocks—as well as many other applications. Researchers have been seeking smaller optical frequency combs both to make their lives easier and to provide the basis for new light-based technologies. For instance, portable atomic clocks could help map underground variations in mineral deposits and enable navigation systems that don’t rely on GPS satellite signals.

JQI researchers have worked with an international collaboration to develop and demonstrate a new type of frequency comb. The new design eliminates the need for bulky equipment while also making it simple to adapt a single device to a variety of practical measurement tasks. The researchers described the advances behind their frequency comb and its performance on common tasks in an article published Sept. 30, 2026, in the journal Nature. To demonstrate the adaptability of their compact device, they used it with a variety of light sources to test its capability at tasks that are the bread and butter of optical frequency combs.

The result builds on a decade of research that JQI research scientist Grégory Moille and JQI Fellow and Co-Director Kartik Srinivasan have put into miniature optical frequency combs that fit on a portable chip. Their new approach grew out of a partnership with a team led by Miro Erkintalo, who is a researcher at the University of Auckland (UoA) in New Zealand and the Dodd-Walls Centre for Photonic and Quantum Technologies, to explore a new way to make optical frequency combs. Using the new approach, the team, including additional colleagues at the University of Maryland at Baltimore County (UMBC), the University of California at Santa Barbara (UCSB), AV Incorporated and the Air Force Research Laboratory (AFRL), has made an optical frequency comb that performs as well as the older behemoth tabletop versions but takes up a fraction of the lab space.

“Though we have been working on chip-integrated optical frequency combs for many years, their control and stabilization—essential for many applications—has often been complicated and difficult,” says Moille, the first author on the paper who is also an associate of the National Institute of Standards and Technology (NIST). “With this new approach, we finally see a viable path for their use in deployable atomic timekeeping, which is one of their most demanding and important applications.”

Two Lasers Are Better Than One

The new optical frequency comb relies on a phenomenon called parametrically driven cavity solitons (PDCSs), first predicted in 2023 by a team led by Erkintalo. PDCSs involve circulating light from two lasers around a tiny ring, called a microresonator. If the ring is the right shape and researchers inject light into it in just the right way, then the circling light interacts with itself via the material comprising the ring and generates a string of pulses that researchers can use as a frequency comb. PDCS-based combs themselves build on nearly two decades of research into frequency combs using a single laser injected into a microresonator on a chip, but prior attempts haven’t succeeded at freeing chip-based techniques from bulky equipment and getting it reliably deployed outside labs.

The collaboration began after Moille and Srinivasan, who is also a NIST Fellow, learned about Erkintalo and his colleagues’ prediction, and the two groups teamed up to make it a reality. They reported on their experiments producing PDCSs in 2024 and showed that they had unlocked a way to measure a range of frequencies that were previously inaccessible. 

That first PDCS demonstration proved the concept and explored the underlying physics but left much of the research into how it performed at practical tasks for future experiments. In particular, the frequency comb produced multiple overlapping sets of frequency lines. The overlap made the comb almost impossible to use, like a ruler misprinted with multiple sets of tick marks.

In the new paper, the researchers combined the PDCS approach with a synchronization technique that Srinivasan, Moille and their colleagues previously demonstrated can stabilize optical frequency combs. The combination of techniques locked all the frequency lines into alignment—a process called self alignment—and provided a pristine ruler with just a single set of tick marks for measuring frequencies.

Moille and Srinivasan nicknamed this particular way of driving light in a resonator a SParCS (self-aligned parametrically-driven cavity soliton). The new SParCS-based combs have unique advantages derived from the way they distribute the power carried by light at different frequencies throughout the comb. Past approaches used a single laser to create a comb that spread out on either side of the laser’s frequency. Those additional frequencies—the comb teeth—have less power the further they are from the central frequency, and they become less sharply defined towards the edges. 

In contrast, researchers can select two lasers at distinct frequencies to create a comb for a particular task, and the two lasers used in the PDCS technique serve as bookends for the comb with the additional teeth forming in the space between. This produces more precise teeth throughout the comb and concentrates the power toward the edges, which are often the critical comb teeth for many practical measurements.

In particular, measuring the exact value of a frequency isn’t possible if you don’t know how far the comb is from zero frequency—a value called the zero-frequency offset. Physicists must find the offset for each comb, since small idiosyncrasies in the fabrication of every resonator makes each comb a little different.

The edge teeth are generally essential for determining the zero-frequency offset for a given comb. There is a standard method for finding it, which earned its creators the 2005 Nobel Prize in physics, but combs must meet a crucial requirement for it to work: One of the teeth must be at approximately twice the frequency of a lower-frequency tooth. Physicists use the same language as musicians and call such a span of frequencies an octave. In practice, teeth at the edge of an octave are going to be at or near the comb’s edge, and if they are too weak or unstable, researchers need additional bulky equipment to find the offset.

The team’s ability to select their two laser frequencies, and thus the edge teeth, let them ensure that the comb spans an octave and that the critical edge teeth are easy to work with—making the process of setting up experiments considerably easier and quicker.

Even though the new approach uses twice as many lasers to create a comb, it still makes the overall setup smaller than prior attempts. Other approaches generally have to add a second laser anyway, along with a host of other equipment, to amplify the edge teeth to be strong enough to use in experiments. The new approach eliminates the need for complex, lab-scale equipment and provides a clear path toward devices that are easy to deploy outside a lab.

Running the Comb Through Its Paces

After producing a pristine comb, the team turned toward demonstrating the convenience and versatility of their approach by performing a variety of common measurement tasks. Frequency combs enable diverse measurement techniques by allowing researchers to link two frequencies, to detect subtle fluctuations of frequencies, and to produce light at a very stable frequency. 

In particular, the team wanted to show that SParCS-based combs can act as a perfect frequency gearbox that serves as a link between two frequencies to keep them locked together, similar to how gears coordinate rotations even when they are at different speeds. Essentially, the comb can not only help researchers measure the difference between two frequencies but can also serve as a bridge that locks the spacing in place. Researchers can lock a laser to a particular tooth, establishing a fixed relationship between their laser and the other teeth in the comb. This lets teeth at other frequencies reflect the stability, or conversely the fluctuations, in that laser. If you lock the comb to a very stable laser, it provides stability to teeth at other frequencies, or if your attached laser fluctuates, you can detect that at another frequency that is easier to measure.

The group partnered the comb with various frequencies of light from different sources to demonstrate that a simple swap of lasers could allow the single device to perform tasks that are useful for performing different types of measurements. The tasks drew on the unique hardware and expertise provided by the members of the collaboration.

First, they used the comb to link microwaves, which oscillate at billions of cycles per second, to optical light waves, which oscillate at hundreds of trillions of cycles per second. This sort of portable comb could improve frequency measurements in many devices that measure light frequencies or use light to measure distances, including the lidar used in self-driving cars. The microwaves partnered with the comb let them better pin down the frequency of the teeth, which can allow the comb to make precise frequency measurements or produce stable optical frequency sources.

They then turned their attention to performing the opposite process—linking optical light to microwaves—using the same device. When the comb works in this direction, it has the extra advantage that it naturally decreases the amount of random fluctuations—noise—in the microwaves.

They demonstrated links in this direction through two tasks that frequency combs are used for in different measurements—monitoring light in atomic clocks and the production of microwaves with little noise.

In an atomic clock, the waves in laser light serve as rapid ticks marking time. The laser is made incredibly stable by keeping it tied to the behavior of atoms, but the waves that interact with the atoms oscillate so fast that no electronics can track their fluctuations. A comb is used to transfer the stability of the optical light to a microwave signal that carries the same long-term stability. Those microwaves can then be used by the clock’s electronic systems to track time.

Moille used a device called a “stable atomic clock reference” that was supplied by his colleagues at AFRL, to show how the comb performed at this essential role in an atomic clock. The experiment demonstrated that using the optical light with their comb produced stable microwaves, which are needed to run an atomic clock. They showed that they could use these microwaves to measure the original light frequency, which was oscillating several hundred trillion times per second, to within about a hundred thousand oscillations per second of the value the atoms should produce.

Producing microwaves without much noise is also useful for other measurements outside of atomic clocks. For instance, lowering the noise of microwaves can improve measurements of distance by radar systems. So Moille also swapped in a laser, called a “low-noise laser reference,” that was produced using a specialized chip supplied by their colleagues at UCSB. The experiment demonstrated that using light from a low-noise laser with their comb produced ultra-pure microwave frequencies.

To judge the results, the group also attached all three sources—the two optical sources supplied by colleagues and the microwave source—to their standard tabletop comb system for comparison. They found, to the levels they could check in the experiment, that their small chip produced the same frequencies and generally lived up to its large predecessor in terms of its stability and noise.

“With SParCS, we have a substantially different comb generation process than has been shown previously, and it was important to verify that regardless of how the comb is generated, it can perform its essential functions well,” Srinivasan says. “We're always basically saying, how well are we doing relative to the existing technology? We were very happy to find that our SParCS comb is indeed working well.”

The SParCS approach did more than live up to its massive forebears; it also made it easy for the team to adapt the frequency comb to different applications. Moille used the same device in all the demonstrations. This contrasts with previous on-chip demonstrations, which required a specially tailored microcomb for a given application.

Moille says that performing one of these tasks with a small frequency comb would previously require a team of several people working for weeks or months to get practical results. By comparison, the new experiments were much easier, with him easily swapping in the different light sources supplied by his collaborators. 

“As experimentalists, this new optical frequency comb has simplified much of our work,” Moille says. “The system made it so easy that you actually have only one operator at a time doing each application.” 

The adaptability of the comb also allows the team to compensate for variations, decreasing the demand for accuracy when fabricating combs. This flexibility could make the combs more practical for mass production and integration into products. The convenience also lets the team spend more time performing experiments instead of needing to sift out functional devices from several fabrication attempts and then spend additional time fine-tuning the working ones to get experimental results.

The group demonstrated this robustness by performing measurements using devices made with several different layouts and showed that they could still produce useful results with simple adjustments to their experiments. Moving forward, the team wants to further refine the devices and explore their properties to see how far they can push their performance, such as by making combs that cover larger frequency ranges. Eventually, they hope they may even be able to make the approach work with a single laser feeding into both sides of the comb. Meanwhile, they are using the combs as tools in other experiments.

“Our lab, as of 18 months ago, was 90% dedicated to the typical approach of pumping in the center and then extending out to the edges,” Srinivasan says. “We've now switched all these experiments to focus on the new SParCS approach. Within the last year, we’ve been able to stabilize more than ten times the number of microcombs than we had across all the preceding years combined, and that’s why we feel so strongly that this approach has a lot of potential going forward.”

Original story by Bailey Bedford: New Chip-Based Frequency Combs Demonstrate Potential for Portable Atomic Clocks | Joint Quantum Institute

In addition to Moille, Erkintalo, and Srinivasan, co-authors of the paper include UMBC graduate student Pradyoth Shandilya; UMBC professor Curtis Menyuk; NIST research scientist Jordan Stone; JQI graduate student Shao-Chien Ou; UoA professor Zongda , UCSB graduate students Mark Harrintong and Kaikai Liu; UCSB professor Daniel Blumenthal; River Beard from AFRL and AV Incorporated; and AFRL research physicists Robert Rockmore and Sean Krzyzewski. 

A team led by faculty at the Duke Quantum Center (DQC), in collaboration with researchers at the JQI, has used a small number of atoms to simulate an aspect of the extreme physics at play in modern particle colliders and in the chaotic environment that existed shortly after the big bang.

This approach, described in a paper published in the journal Nature Physics on Sept. 23, 2026, demonstrates the viability of trapped-ion quantum computers to begin probing fundamental questions about the universe. The experiment emulates a phenomenon called string breaking in which two connected fundamental building blocks of matter stretch apart, eventually creating so much energy that new particles “pop into existence” when the connection snaps.  

“Quantum computer simulations provide the best platform to investigate complex questions like matter formation, short of having witnessed the big bang itself,” says Christopher Monroe, a professor of electrical and computer engineering and physics at Duke and a College Park Professor of Physics at the University of Maryland (UMD), who led this research. “These findings signal a marked development in the quantum science field and open new avenues for us to understand string-breaking dynamics.”

This research was conducted by an international collaboration that also included researchers working at Oxford University, the California Institute of Technology, Cornell University and KU Leuven. The results join two similar published findings, led by other research teams in the field, which simulated the same phenomenon on different quantum computer platforms.

"This beautiful experiment builds on an earlier collaboration with Chris Monroe, in which we demonstrated the closely related phenomenon of confinement,” says JQI Fellow Alexey Gorshkov, who is also a theoretical physicist at the National Institute for Standards and Technology, a Fellow of the Joint Center for Quantum Information and Computer Science (QuICS) and a Senior Investigator at the National Science Foundation Quantum Leap Challenge Institute for Robust Quantum Simulation (RQS). “My graduate student Fangli Liu was the one who first got me interested in simulating high-energy physics with trapped-ion chains. Bringing together experimentalists and theorists with different areas of expertise has been incredibly rewarding."

The Building Blocks of Matter

The fundamental building blocks of matter, quarks, only exist when bound together inside particles such as protons and neutrons. They are about a billion times smaller than an atom and can’t currently be observed directly. Pairs of these tiny, charged particles are held together by a force that acts like a taut string; quarks want to stick together, and it takes quite a bit of energy to pull them apart. 

But once they are forced apart, the energy built up in their connection can be enough to create more charged particles. When this happens, the string snaps, leaving two or more pairs of particles rather than one. This process requires so much energy, however, that it only happens in extreme environments like the Large Hadron Collider or the aftermath of the big bang. 

In the new study, the team successfully observed analogous string-breaking dynamics on a trapped-ion quantum platform. Quantum simulators, with their high degree of controllability, can be programmed to recreate the real-world processes occurring at the atomic or even subatomic quantum scales.

“Working at the intersection of quantum simulation and high-energy physics is incredibly exciting,” said Arinjoy De, the first author on the paper and a former JQI and Duke graduate student who now works at QuEra Computing. “By simulating quark confinement and string-breaking phenomena in a controlled lab environment, we're opening up new pathways for experimental investigations into the behavior of matter at its most fundamental level.”

How the Simulation Worked

To perform the simulation, the team encoded a string-breaking model into a chain of 13 trapped ions. Using precisely controlled laser beams, researchers were able to tune the interactions among the ions. These interactions effectively control the energy to the system in a way that mimics the stretching and eventual breaking of a string. 

By preparing the system out of equilibrium and tracking its evolution over time, the researchers observed the emergence of effective charges and reconstructed the resulting string dynamics.

The team also simulated the process on a classical computer and confirmed that their experimental results were accurate. As the problem size grows in future experiments, however, only quantum computers will be able to solve these problems. 

The string-breaking process in other models was also recreated by teams led by Google and QuEra Computing on platforms built using superconducting circuits and neutral atoms, respectively, which each have their own advantages and challenges.

“These are the three platforms leading the charge in quantum computing, so it’s a nice benchmark and comparison for the quantum community,” Monroe says.

The authors say that the trapped-ion platform results mark an exciting step forward in building quantum simulations complex enough to exceed the capabilities of even the largest supercomputers, which will eventually allow researchers to explore the most fundamental questions of the universe, like matter evolution after the big bang. 

“As a physicist, it is incredibly exciting to investigate the conditions of the early universe in an atomic-level computing machine,” says Zohreh Davoudi, an associate professor of physics at UMD, who was part of the research team and is also a QuICS Fellow and a Senior Investigator at RQS. “Even the slightest insights from an out-of-equilibrium physics model will guide us in the future.”

This story was written by Andrew Tie and originally published by the Duke Pratt School of Engineering. It has been adapted here with minor changes.

This work was supported by the Department of Energy (DE-SC0020312, DE-SC0025341, DESC0019040, DE-SC0024220, DE-SC0020271), National Science Foundation (OMA-2120757), Air Force Office of Scientific Research, Defense Advanced Research Projects Agency and Amazon Web Services.

By rapidly twisting a microscopically small tip back and forth, researchers at the University of Maryland (UMD) have unlocked a new way to detect subtle changes on the surface of a material flexing in response to infrared light.

In a paper published Aug. 5, 2026 in the journal Nature Communications, the researchers describe a new way to take high-resolution images that they call infrared torsional force microscopy, or TFM-IR for short. It allows them to measure the surface of a material with near-nanometer precision as it stretches and warps in response to infrared light—invisible light readily absorbed by many kinds of chemical bonds that causes them to vibrate. It’s the first technique that can measure both the vertical and horizontal vibrations induced by light with such high precision.Caption: A schematic of a novel infrared torsional force microscopy (TFM-IR) experiment conducted at the University of Maryland. A material sample (located on the dark gray disc) is illuminated with pulses from an infrared laser. A microscopic tip, twisted rapidly back and forth on a long arm, scans the surface of the material to sense its response to the light. A second laser is bounced off the arm to measure small changes to its twisting motion, allowing researchers to capture a detailed image of the sample's surface. (Schematic courtesy of the authors.)

“We have overcome a long-standing limitation in optical imaging,” says Min Ouyang, a professor of physics at UMD and a member of the Quantum Materials Center and the Maryland NanoCenter who led the new project. “Modern quantum materials derive many remarkable properties from variation that happens over distances of only a few nanometers, or even less. Conventional optical microscopes cannot really resolve this.”

The method Ouyang and his colleagues developed builds on atomic force microscopy (AFM), which was first developed in the 1980s and has matured into a standard technique for measuring the surfaces of material samples with extreme precision. Unlike traditional microscopes, AFM works more like your fingertips than your eyes. Instead of creating an image by collecting light with a lens, AFM drags a tiny tip across a material to feel the forces from its bumps and ridges. The technique can spot features that are smaller than a nanometer and, under the right conditions, can even resolve individual atoms.

By itself, AFM only sees the topography of a surface: It can sense that it’s higher over here and lower over there, but it doesn’t provide any details about the chemical composition of a sample. To learn about the makeup of a material, researchers often excite vibrations in a sample with infrared light and use the AFM tip to sense how the surface changes. Because chemical bonds respond in predictable ways to infrared light, the combination of AFM and infrared light can identify the signatures of particular molecules, either to verify the composition of a sample or to detect the presence of unwanted contaminants.

“If you shine some light on your sample, you're going to get some very slight thermal expansion,” says Yonatan Gazit, a graduate student in physics at UMD who is also the lead author of the new paper. “AFM is essentially measuring that thermal expansion to understand how well your sample is absorbing or interacting with the light.”

Modern AFM devices feature a sharp tip suspended above a sample on the underside of a skinny arm. Electronics rapidly drive the arm up and down, which gently taps the tip against the sample at a regular rhythm. The tapping helps the tip avoid getting stuck on a ridge, which would potentially damage material being studied, and enhances the sensitivity to infrared vibrations by closely matching their frequency. As the tip traverses the surface, the rhythmic tapping is altered as the material flexes and pushes against it. Researchers measure these subtle changes in tapping frequency by bouncing laser light off the top of the vibrating arm to monitor its motion.

The tapping technique is excellent for making sensitive measurements of height, but it doesn’t do a good job sensing how a material expands or contracts horizontally. In a paper published in 2024 in the Proceedings of the National Academy of Sciences, a team from Stanford University and their colleagues showed that twisting the AFM tip at a regular frequency instead of tapping it could measure previously undetectable variations along the surface of a double-decker stack of graphene, formed from two stacked layers of carbon atoms each arranged in a honeycomb pattern of repeating hexagons. They named their technique torsional force microscopy (TFM).

Caption: Two torsional force microscopy (TFM) images of bilayer graphene. On the left, standard TFM captures the material's signature honeycomb lattice. On the right, the new method (TFM-IR) captures much more detail about how the chemistry of the surface responds to infrared light. (Images courtesy of the authors.)Inspired by this result, Gazit, Ouyang and their colleagues designed an experiment that combined infrared illumination with the twisting technique. As the tip twists back and forth, pulses from an infrared laser periodically wash over a small sample of a material. Some energy from each pulse gets absorbed by the material, causing it to swell and vibrate. Choosing the frequency of the pulses—that is, how many pulses arrive at the sample per second— enables the tip to pick out either the vertical changes or the horizontal changes, similar to how a strobe light can selectively pick out or freeze certain kinds of motion.

As a proof of concept, the researchers tested the new technique by studying the surface of a small piece of mica, a shiny and flaky mineral used in manufacturing everything from drywall to tires to fireproof material for industrial ovens. They chose mica both because it’s already well-understood and because the chemical bonds that hold it together point along different directions, making it a good candidate for measuring both the vertical and horizontal vibrations induced by infrared light.

The team showed that they could detect four vibration patterns in mica and demonstrated that they could distinguish the horizontal and vertical movement by using different infrared laser pulse rates. They zeroed in on a small bump—a mica nanobubble on the surface just a few nanometers tall—and carefully dragged the tip from the center of the bump to its edge. They compared the results of their measurements with simulations of the horizontal and vertical vibrations expected from the way the bubble strained and bulged, and they found that the locations of the strongest horizontal and vertical responses to infrared light lined up between theory and experiment.

The team next turned their attention toward a double layer of graphene, the same material studied in the paper that first introduced the torsional technique. On its own, graphene has intrigued scientists for more than two decades because of its unique electrical and mechanical properties. When it’s stacked into two layers, with one layer rotated by a small amount, it gets even more interesting. The two layers form what’s called a moiré material, and in 2018, researchers found that a very particular angle turned a moiré stacking of graphene into a perfect electrical conductor—a quantum effect that made the material a superconductor.

Researchers remained in the dark about the microscopic origins of the effect. Because a moiré material is only a couple of atoms thick, the tiny changes in the lattice that give rise to its remarkable properties cannot be revealed by simply scanning its height. The torsional trick introduced in 2024 pointed toward a new way to image these atomically thin materials.

In the new paper, the team examined a sample comprising two layers of graphene stacked together at a small angle. They compared a standard TFM image of the sample with an image taken using their TFM-IR approach—both taken of the same exact sample at the exact same spot. The standard TFM image clearly showed the material’s signature lattice of hexagons, but the TFM-IR image revealed a wealth of additional details. Instead of merely showing the shape of the lattice, TFM-IR showed for the first time how different chemical bonds in a single hexagon—including bonds within a single sheet of graphene and bonds between the two sheets—react to infrared light, revealing a unique vibrational fingerprint of the underlying material. Understanding this fingerprint and the way that it changes when the stacking angle changes could prove crucial to gaining a better understanding of moiré materials and their properties.

“Our technique combines three capabilities that are rarely available in a single measurement,” Ouyang says. “First, it brings optical imaging and spectroscopy to the nanoscale, providing spatial resolution down to nearly one nanometer. Second, it can distinguish directional responses within a material, allowing us to uncover anisotropic properties that conventional techniques cannot resolve, Third, it provides each material’s unique spectroscopic fingerprint. In other words, our technique doesn’t just show what a material looks like; it also identifies what it is and reveals the hidden physical processes that govern its behavior by mapping how it responds to light with nanometer-scale precision.”

Ouyang and the team hope that the technique will be a key tool in characterizing and even designing materials going forward, and they emphasize that it has the added benefit of working at room temperature. In particular, TFM-IR might be useful for semiconductor companies, who are on the hunt for techniques to spot defects in their chips. The authors say that a technique capable of mapping the mechanical signatures of local chemistry could guide the development of new advanced manufacturing processes and might even help researchers optimize next-generation nanoscale devices, including quantum sensors and photonic quantum computers.

Story by Chris Cesare

In addition to Ouyang and Gazit, the paper had three other authors: Son T. Le, an associate research scientist at the Laboratory for Physical Sciences (LPS) and in the Department of Electrical and Computer Engineering at UMD; Aubrey T. Hanbicki, a research physicist at LPS; and Adam L. Friedman, a physicist and technical director at LPS.

Quantum computing and AI are among the most rapidly developing modern technologies. AI, in the form of machine learning, has been deployed for decades to recommend movies and TV shows and make it easier to search for images. Over the past several years, large language models have permeated even more facets of daily life, from writing emails to producing images, videos and songs following requests expressed in a few written lines.

Quantum computers, on the other hand, have remained almost exclusively in labs at universities and a handful of companies. Nevertheless, many researchers and engineers developing them are already looking for the earliest applications and predict a bright future in which quantum computers excel at certain tasks, like drug development and enabling new cryptographic techniques.

Despite machine learning and quantum computing both being heralded as revolutionary technologies, neither is a magic solution to every problem. They are each the products of a long line of research advances and are both still under active study.An artistic representation of a quantum neural network identifying a handwritten digit. Each ball containing an arrow represents a qubit that is serving as a neuron in the network. The neurons are organized in layers, and connections between neurons in adjacent layers control how the information is processed as it moves through the network. The final layer identifies the most likely digit. (Credit: Chris Cesare/JQI)

As the two technologies continue to mature, researchers are beginning to investigate ways to utilize them together. A collaboration between JQI Fellows Alaina Green, Norbert Linke, Victor Galitski and their colleagues recently reported on new experiments that explore how quantum computing influences—and might improve—machine learning. They used a variety of quantum computers to run a simple neural network­—a type of machine learning inspired by the structures in real brains.

Neural networks are the foundation of many prominent AI technologies, including large language models like ChatGPT and Claude. They have been around for decades and have been used for many applications, including image recognition. For instance, banks used the neural network LeNet-5, which was developed in 1998, to identify handwritten zip codes. But making neural networks run on quantum computers is unexplored territory.

In an article published as an Editors’ Suggestion in the journal Physical Review Letters on July 22, 2026, the JQI researchers and their colleagues describe experiments running neural networks on several different quantum computing platforms. Their results demonstrated that a certain amount of intentional quantum randomness, arising from measurements, can be beneficial to a neural network. They also saw hints that the errors that plague current quantum computers might be able to play a useful role when running a neural network on a quantum computer.

“Measurement outputs on quantum systems are inherently random, and this can improve the performance of a neural net,” says Linke, who is also the IonQ Endowed Associate Professor of Physics at UMD, a senior investigator at the National Science Foundation Quantum Leap Challenge Institute for Robust Quantum Simulation (RQS) and the director of the National Quantum Laboratory (QLab) at UMD. “Additionally, operations on current quantum hardware are often imprecise. This limited control produces more randomness that, if it's not too large, can further boost the neural net performance.”

Designing a Quantum Neural Network

The new neural network experiment implemented an approach that Galitski, who is also a Chesapeake Chair Professor of Theoretical Physics in the Department of Physics at UMD, and two graduate students working in his group proposed in 2025. Their goal wasn’t to use a quantum computer to outperform existing neural networks. Instead, they wanted to use quantum computers to investigate when incorporating quantum features is helpful or harmful to a neural network.

Neural networks operate by transforming an input through a sequence of calculations to accomplish a goal. The information contained in the input—usually represented as a list of numbers—flows through a connected network of artificial “neurons” that systematically breaks apart the information and recombines it using calculations associated with each of the connections. The particular calculations that the network performs are tailored by a process called training, which involves testing many example inputs and adjusting the connections between neurons until the network reliably produces a desired result.

The network developed by Galitski and his colleagues was trained using traditional computers to perform a task that is straightforward for a neural network: identifying handwritten numbers. They focused on recognizing digits because it is an established exercise used to test image processing and machine learning models. The well-worn task allowed them to train and test the neural network using a set of images called the Modified National Institute of Standards and Technology database (MNIST), which provides examples of handwritten digits in a convenient standard format.

Examples of digits included in the MNIST dataset. (Credit: Suvanjanprasai, CC BY-SA 4.0, via Wikimedia Commons)

Their neural network takes an image as an input and at the end predicts the number it most likely depicts. The ready-made database and well-understood task provided a convenient starting point for the group to explore how quantum ingredients influence a neural network.

Galitski and the graduate students in his group described a way to make qubits—the basic building blocks of quantum computers—function as neurons. Their method lets them use quantum measurements to control the amount of quantum randomness intentionally included in the neural network’s operation on each run. The group designed their neural network architecture to be compatible with any of the various types of quantum computers being developed. There are many ways to make a quantum computer by starting with different qubits. However, all quantum computers rely on shared quantum principles.

The team took advantage of the way all qubits store information in their quantum states to incorporate different amounts of quantum randomness into their proposed model. A qubit will always be observed in one of two states when measured, but in between measurements it can be in a mixture of the two—called a superposition—where it is only possible to know the probability of finding it in one state or the other. A qubit can be put in a superposition where one or the other outcome is more likely to occur, but it’s not definitively in one state or the other until a measurement destroys the superposition and one of the states is observed.

In the proposal, the final state of the neurons is one of the two unmixed states of the qubits, and superposition and quantum measurements are used to inject randomness into the neural network. The approach allows different experimental runs of the neural network to use different superpositions. Each superposition provides a different chance that, at each step, the results might randomly flip between states and potentially alter the identification the neural network makes. Or researchers can perform a test without randomness by leaving the qubits exclusively in one state instead of a superposition. The approach provides a way to compare the success of neural networks featuring different amounts of quantum influence as well as to judge how the neural network operated on different quantum computers.

Testing a Quantum Neural Network

To test the proposal with experiments, Galitski shared it with experts at IBM as well as Linke and Green, who is also a physicist at the National Institute of Standards and Technology and a senior investigator at RQS. Galitski requested their help testing the neural network on real quantum computers, and they all agreed the idea was worth pursuing.

IBM’s quantum computer is built using superconducting circuits, while Green and Linke have built their quantum computers using trapped ions—electrically charged atoms contained and manipulated with light waves. Linke and Green can operate their quantum computers by manipulating ions in two different ways—one using microwave light and the other using laser light. The opportunity to test the neural network on these three distinct systems gave them the chance to look for differences in how the neural network operated on each platform and to get a more complete picture of how the quantum neural network performed.

“We showed that for this architecture, you can see improvement in classification of images when you tune the quantumness from zero to some golden spot,” says Djamil Lakhdar-Hamina, a JQI graduate student and the first author of the paper.

The neural network performed slightly differently on each quantum computer. On all the platforms, the results showed that a limited amount of randomness from the quantum measurements improved the neural network’s performance compared to cases with no intentional randomness, but at some point, additional randomness made the neural network perform worse.

This outcome wasn’t completely surprising. A little bit of randomness is known to be a useful ingredient in machine learning—although it is normally incorporated in a different way and at a different time. This is because perfectly following a single fixed path inevitably gets the same answer every time, which can be a problem. Without any randomness to provide wiggle room in exploring all the options, a computer program or the training of a neural network is more likely to get stuck at a wrong answer that is almost correct. For instance, attempts to identify handwritten digits may mistake a sloppy seven as a one and lock onto that wrong answer. By contrast, a similar approach with some randomness to shake things up can often break free of the almost right answer and move on to the correct one.

When researchers develop neural networks, they often incorporate randomness into the training. In this experiment, the team looked for a benefit from using quantum measurements to inject randomness during the actual identification. The team saw that there were certain images that the neural network falsely identified when there was no quantum randomness, but it identified more images correctly when just the right amount of randomness was added to the quantum measurements. The group focused on these troublesome cases and did repeated experiments using a particular image that the neural network identified incorrectly when they didn’t add randomness. They showed that with the optimal amount of randomness added, the trapped-ion quantum computer could almost always identify the number and IBM’s quantum computer succeeded around 90% of the time.

Development Opportunities

The researchers also considered sources of randomness beyond what they introduced using quantum measurements. Quantum computers all experience noise—random errors from things like heat fluctuations that throw off their operations, and the group looked for signs of these errors influencing their results. The errors in quantum computers are currently a significant impediment to putting them to useful work, and researchers are looking for the most practical problems that can be tackled with imperfect devices that we have or expect to build soon.

“We're living in the era of what's called NISQ—noisy intermediate scale quantum computing,” Lakhdar-Hamina says. “So the question is, rather than this being a bad thing, can we harness the noise to positive ends?”

Based on the experiment, the researchers hope that machine learning is an area where quantum noise can be a benefit and not just a hindrance. They saw a couple of signs of the extra randomness being beneficial. The fact that the results varied a little when the team ran identical versions of the neural network on each computer suggests the imperfections of each are playing a noticeable role. Additionally, the researchers saw better results in the experiments both with and without randomness from measurements than they expected from their simulations of how the neural network would perform. The team attributed these results to the different imperfections and noise of each computer and concluded that the noise helped by contributing some amount of useful randomness.

The team’s results suggest that both intentional quantum randomness from measurements and the noise of quantum computers might serve a useful role in some neural networks. Even with these encouraging results, the researchers don’t propose that their current neural network design will produce the best possible neural network or even the best quantum neural network. However, the ability to tune the quantum randomness provides a tool to study the effect it plays and to learn how quantum computers might effectively be used with future neural networks.

Moving forward, they plan to develop a neural network that lets them explore even more quantum effects. In particular, they’re interested in exploring the potential benefit of utilizing entanglement, which links quantum particles together independent of their separation in space. They hope these neural networks will help them identify cases where quantum computing and machine learning can boost each other’s potential and demonstrate their combined power to solve challenging problems.

“Machine learning is not only socially and economically important, but is a fascinating part of computer science,” Lakhdar-Hamina says. “I think that it is one of the rare instances where we might actually find quantum advantage. And so this was a first step in what I think is going to be a big project for many, many people trying to find that quantum advantage within machine learning.”

Original story by Bailey Bedford: https://jqi.umd.edu/news/researchers-explore-how-quantum-computers-and-their-errors-may-enhance-ai

In addition to Galitski, Linke, Green and Lakhdar-Hamina, co-authors of the paper include former JQI graduate students Richard Barney and Xingxin Liu and Sarah Miller, who is a Research Scientist at UMD's Applied Research Laboratory for Intelligence & Security and an expert on machine learning and artificial intelligence.

In a quest to build the most accurate sensors in the world, scientists are constantly improving their performance. Making them more precise, stable and reliable. Photon exchange through an optical cavity links two atomic ensembles, creating a shared entangled state. This entanglement is designed to be insensitive to common noise while remaining highly sensitive to differential signals. (Credit: Raphael Kaubruegger, JILA)

But eventually, physical constraints will prevent further improvements. 

“By fully embracing the laws of quantum physics, one can expand the performance limits imposed by these constraints,” says JQI Fellow Alexey Gorshkov, who is also a Physicist at the National Institute of Standards and Technology (NIST), a Fellow of the Joint Center for Quantum Information and Computer Science and an Associate Professor in the Department of Physics at the University of Maryland. “And it's very exciting to come up with protocols that come as close as possible to saturating these limits for different sensing tasks.”

Even the most precise sensors in the world are not fully isolated and are limited by noise—subtle disturbances from the environment like vibrations, electromagnetic fields or temperature changes. 

So, Gorshkov, JILA Fellows Ana Maria Rey and James K. Thompson and their colleagues from the Niels Bohr Institute and the Indian Institute of Technology Madras, asked, how can we improve the next generation of sensors despite these limitations? 

One promising idea is to use quantum entanglement, so atoms are connected to each other and working together as a system to form a quantum sensor. When atoms are entangled, they share properties even when separated by distance. In principle, this allows for more precise measurements. But entangled atoms are still subject to noise. “Entangled states are well understood for estimating a single parameter, but our goal was to create an entangled state that is highly sensitive to a parameter difference between two nodes of a sensor network,” says Raphael Kaubruegger, a research associate at JILA and the lead author of the article. 

The researchers set out to identify a new class of entangled states that could filter out noise affecting both sensors. They then developed two ways to create these states inside an optical cavity, a pair of mirrors about one inch apart that bounce photons back and forth. They describe the state and two methods to create it in a recent paper published in Physical Review X. 

The entangled state they identified uses decoherence-free subspaces which are protected from certain types of disturbances to quiet noise affecting both sensors. 

Lasers are used to create coherent superposition between two internal states of an atom, but to accomplish that, the laser’s frequency needs to exactly match the atomic transition. 

The challenge, as Rey explains, is that even the most precise lasers cannot maintain a stable frequency for long enough. These laser frequency instabilities generate noise which is equally experienced by both sensors and is currently one of the most detrimental errors in state-of-the-art clocks. “Ideally, one would like to prepare the atoms in a state that is insensitive to this type of noise,” says Rey, who is also a NIST fellow and professor adjoint of physics at the University of Colorado Boulder. 

“The state we create is entanglement between these atoms, but in a way that you cannot distinguish which atom is in which ensemble,” Rey says. “They are fully symmetrized.” 

“After the fact, we realized this was the same kind of state people were thinking about to describe antiferromagnets, or quantum magnets,” says Thompson, who is also NIST fellow and professor adjoint of physics at the University of Colorado Boulder. 

In condensed matter physics, the Lieb-Mattis state describes a quantum version of an antiferromagnet, where two groups of atoms act like they point in opposite directions, but without the system picking one fixed direction in space. 

One method the team developed to prepare the desired state involves entangling two nodes of a sensor network by engineering a “spin exchange,” by having the atoms send photons back and forth through an optical cavity. This leads to a state where each atom in one node is perfectly anticorrelated with an atom in the other. If one atom is “up,” the other atom is “down.” 

Thompson likens this approach to baseball, where each ensemble is a baseball team. The teams are throwing balls, or in this case photons, to each other. Every time a ball is thrown, the other team catches it. Thompson adds that it’s important that we don’t know which player threw the ball or who caught it. 

“That’s what builds these links,” Thompson says. “If a ball is thrown, it is definitely caught.” 

The approach produces Heisenberg scaling, or the best possible precision scaling where all the atoms act as one quantum object. 

Optical cavities are not perfect. As Rey explains, sometimes you may lose a photon. The team’s second approach takes this into account. 

Inside the optical cavity, photons can bounce back and forth between very reflective mirrors about 100,000 times before they accidentally slip through to the other side. 

“We are losing photons, but the important part is that the photons are lost in a collective way,” Rey says. 

Because it’s impossible to tell which atom is to blame, this can create entanglement—driving them into a state where they cannot lose more photons. 

“At some point they get really good at not dropping the ball anymore,” Thompson says. 

“They go into a ‘dark state,’ or a state where the phases of the emitted photons completely cancel out, leading to what it is known as destructive interference,” Rey adds. 

The team was initially trying to understand the detrimental effect of losing those photons. But as Rey explains, ultimately this type of dissipation actually led them to a state they wanted. 

“The state we initially wanted to prepare was one in which half the atoms are excited, but the system cannot collectively emit a photon,” Kaubruegger adds. 

The team’s proposed states can be created quickly, and more importantly, faster as the system gets larger, making them practical for scaling quantum sensors. 

“People have thought about this kind of state when you only have two atoms, which is cool, but you’d like to use more,” Thompson says. “It turns out, the more atoms you have, the better!” 

By making quantum sensors more precise, these entangled states could one day help guide navigation when GPS is unavailable or reveal hidden underground resources such as minerals, oil or gas. 

Close collaborations between theorists and experimentalists have been key to this work. The groups inspire each other—and keep each other in check. Because they work so closely together, Kaubruegger says they have a deeper understanding of the challenges experimentalists face. 

And now, the ball, so to speak, is in Thompson’s group’s hands; to demonstrate the state in experiment.

This text has been adapted with permission from a story written by Kirsten Apodaca and originally published by JILA. It has been adapted with minor changes here.