Electrical engineers at the University of Wisconsin-Madison have developed a new digital memory technology that’s more efficient than any other currently on the market.
The technology is a step toward integrating memory and processing on one computer chip—a design that could significantly improve the performance of AI networks and high-powered computing. Importantly, the patented memory device could be produced using current cleanroom technology, making it easier to scale up commercially.
The researchers’ design uses nanoscale devices called sliding ferroelectric tunnel junctions, created using an ultrathin two-dimensional or quantum material. PhD students Haolin You and Chen Shao led the work under the guidance of Ying Wang, an assistant professor of electrical and computer engineering at UW-Madison. The research appears in the Oct. 20, 2026, issue of the journal ACS Nano.
Most conventional computer memory uses billions of tiny silicon transistors, often together with capacitors, to store binary data as electrical charge or stable circuit states representing 0s and 1s.
After decades of improvement, this system is reaching its limits: It’s getting difficult to cram more transistors onto memory chips, which also use a lot of energy. And, because memory modules are a separate component from the processor, computers waste a lot of time and energy shuttling data between the two.
Now researchers are looking into new memory paradigms that allow co-locating memory and processing on a single chip, which would boost computing speed and energy efficiency.
One of the most promising designs is based on nanoscale electronic devices called ferroelectric tunnel junctions, which could replace transistors. The junctions are comprised of an atomically thin ferroelectric layer sandwiched between two conductive materials. Ferroelectrics are a special class of materials that change and maintain a polarized state, either positive or negative, when electricity passes through them, unlike most materials, which revert to their unpolarized state once the electricity stops flowing. This means an electric charge can flip the tunnel junctions into “on” and “off” positions, making them ideal for recording digital data.
The thinness of the ferroelectric material, just one or two atoms thick, also allows a quantum phenomenon called quantum tunnelling to take effect, which allows electrons to “jump” the ferroelectric layer and flip the polarization without actually passing through that layer. That makes switching happen in just billionths of a second. It’s also incredibly energy efficient.
But ferroelectric tunnel junctions have one major flaw: Each time the polarization flips, it displaces an ion, breaking covalent bonds in the material—which in turn produces defects over time and shortens the material’s life.
That’s why Wang and her students decided to investigate a new version of the devices called sliding ferroelectric tunnel junctions in which a slight sliding between layers of the materials flips the charge, no bond-breaking required.
“Imagine you have two pieces of paper, and they can slide from one polarization to another,” explains Wang. “This sliding behavior is governed by Van der Waals forces, which are three orders of magnitude smaller than covalent bonds. That means this switching process is very energy efficient and quite fast. It can be as fast as one picosecond, or one trillionth of a second—about three orders of magnitude faster than nanosecond-scale volatile-memory switching.”
The challenge, however, is that the energy produced by this sliding is so faint that many researchers believed it was not possible to detect the signal, meaning it was not actually possible to read information from the memory.
Wang and her students believed that making the sliding tunnel junctions work was just a matter of engineering. They started with a promising sliding ferroelectric material called bilayer rhombohedral molybdenum disulfide. The trick was finding the correct metal contacts at the top and bottom of the ferroelectric sandwich that could boost the readout signal of the sliding tunnel junctions.
“This material is very sensitive to the metal contact you choose; we tried different combinations and studied their work function and interface,” says Wang. “We found that a heavily doped 2D silicon on one side and gold on the other is the best combination to get good performance from these tunnel junctions.”
In fact, the combo boosted the electrical contrast level by 105, or 100,000 times, which is within the range of conventional memory devices and enough to make the sliding tunnel junction device usable.
At this point, the team has fabricated and tested a few dozen of the devices. Now, the researchers are assessing how to scale up the memory. In particular, they’re looking at how elements in the fabrication process could affect the devices when scaled up to the larger wafers used in chip making. They also plan to connect the devices together to see how well they crosstalk. That’s the next step toward demonstrating their suitability for neuromorphic computing, a very intensive computing paradigm that mimics the interconnectedness of the human brain.
To enable that work, Yang and her students are looking for industry partners who can help them move the technology out of the lab and into the next generation of electronics.
Featured image caption: From left, PhD students Chen Shao, Haolin You and Assistant Professor Ying Wang. Credit: Joel Hallberg.
Ying Wang is the Dugald C. Jackson Assistant Professor.
Other UW-Madison authors include Adrian Jonathan Lim, Robert Boyd, Fan Fei, Carter Fox, Jun Xiao and Daniel Rhodes.
Other authors include Saagar Kolachina, Chengdong Wang and Xiaofeng Qian of Texas A&M University and Sina Najmaei of the DEVCOM Army Research Laboratory, Adelphi Laboratory Center, Adelphi, Maryland.
The UW-Madison authors acknowledge support from the National Science Foundation through grant number ECCS-2339093; the National Science Foundation CAREER program grant number DMR-2237761; and the NSF through the University of Wisconsin Materials Research Science and Engineering Center (DMR-2309000).