The standard remedy is to pair each memory element with a selector, a two-terminal device that conducts strongly only when a sufficient voltage is applied across it and remains nearly insulating otherwise. The quality of a selector is judged by its nonlinearity, the ratio between the current it passes in the on-state and the current it leaks in the off-state at the same operating conditions. High nonlinearity suppresses sneak paths, but achieving it without sacrificing drive current, switching speed or endurance has proved remarkably difficult. Threshold switches, ovonic devices, mixed-ionic-electronic-conduction devices and transistor-based access elements each bring trade-offs in variability, fabrication complexity or compatibility with the memory stack above them.

Against this backdrop, a News and Views commentary by Yuting Sun and Min Zhu of Shanghai Jiao Tong University, published in Nature Electronics, highlights recent progress that may reshape the selector landscape. The commentary discusses work demonstrating tunnel junction selectors built from five vertically stacked layers of two-dimensional materials, a device architecture that delivers high nonlinearity, robust endurance, fast switching speeds and low device-to-device variability. The authors suggest that such devices could potentially function as a common access element across multiple memory technologies, a prospect that would simplify the design of hybrid memory systems considerably.

Two-dimensional materials, the family of crystals that can be isolated in layers only a few atoms thick, have long been touted for electronic applications because their surfaces are free of the dangling bonds that plague conventional semiconductors at scaled dimensions. In a vertical stack, dissimilar two-dimensional layers can be assembled atom by atom, allowing engineers to engineer the band alignment across the junction with a precision that bulk heterostructures cannot easily match. A tunnel junction exploits this control directly: the barrier formed between the layers governs quantum-mechanical tunnelling, so a modest change in the voltage across the stack can swing the current through many orders of magnitude. That voltage-dependent tunnelling is precisely the behaviour a selector needs.

The five-layer architecture is central to the performance reported. By sandwiching tunnelling barriers between conducting and semiconducting two-dimensional sheets, the device creates an energy landscape in which current transport is strongly suppressed at low bias but rises steeply once the applied voltage exceeds a threshold. The steepness of this transition determines the nonlinearity, and the atomically thin layers mean the effective barrier width is set with atomic precision rather than by the statistical roughness that afflicts amorphous oxide films. Because the layers are crystalline and their interfaces are abrupt, device-to-device variation, historically the Achilles heel of selector technology, is kept low. Variability matters enormously in practice: a crossbar array may contain millions or billions of cells, and the array can only be read reliably if the statistical spread of selector behaviour remains narrow across the entire population.

Endurance and speed are the other pillars of the achievement. A selector in a working memory must withstand billions of switching cycles without degradation, since it is cycled every time the adjacent memory cell is written or read. It must also switch quickly enough not to become the bottleneck of the memory operation. The commentary notes that the two-dimensional tunnel junction selectors exhibit robust endurance and fast switching speeds, indicating that the physical mechanism underlying the switching, governed by electronic tunnelling rather than by the movement of ions or the formation and dissolution of conductive filaments, is inherently less prone to the wear mechanisms that limit other selector types. Filamentary devices, for example, rely on stochastic ionic processes that both slow the switch and gradually rearrange the material, whereas a tunnelling barrier responds on the timescale of electronic transport itself.

The significance of the work extends beyond any single memory technology. Resistive random-access memory, phase-change memory, conductive-bridge memory and ferroelectric memory all require selectors, but each has historically demanded a selector tailored to its particular voltage window, current range and integration scheme. A universal selector, one device that can serve as the access element across several memory platforms, would allow foundries to standardise a single back-end-of-line module and would give memory architects freedom to mix and match storage elements within a system without redesigning the array infrastructure. Sun and Zhu point to this cross-technology compatibility as the potentially transformative aspect of the two-dimensional tunnel junction approach, framing it as a common access element for multiple memory technologies.

The commentary situates the new devices within a broader arc of two-dimensional materials research. Since the isolation of graphene opened the field, researchers have assembled van der Waals heterostructures in which layers are stacked without the lattice-matching constraints of epitaxial growth. Reviews of the field have emphasised that this assembly freedom, combined with the diversity of available two-dimensional crystals, from semiconducting transition-metal dichalcogenides to insulating hexagonal boron nitride, enables device concepts that are difficult or impossible in conventional material systems. The tunnel junction selector is a compelling example of that promise translated into a device metric, nonlinearity, that the memory industry can measure and value directly.

Challenges remain before such selectors can leave the laboratory. Integrating two-dimensional layers into large-area manufacturing remains an unsolved problem at scale, since the growth and transfer of atomically thin films over full wafers must achieve uniformity comparable to that of deposited oxides. Contact resistance between the two-dimensional stack and conventional metal interconnects, thermal stability during back-end processing, and the yield of defect-free multilayer stacks are all active areas of investigation. The commentary’s authors, who are themselves affiliated with a school of integrated circuits and a national key laboratory of micro and nano manufacture technology, are well placed to appreciate the distance between a demonstrated device and a manufacturable module.

Even so, the demonstration marks a notable convergence of two research communities. Memory engineers have spent two decades searching for a selector that combines the nonlinearity of a diode with the bidirectional operation and simple two-terminal structure that crossbar arrays require, while two-dimensional materials researchers have sought applications in which atomically thin layers provide a decisive advantage rather than an incremental one. The tunnel junction selector, with its high nonlinearity, robust endurance, fast switching and low variability, offers evidence that this convergence can produce devices that matter to the industry. If the approach survives the rigours of large-area integration, the humble selector, long an afterthought in memory design, may become the element that finally unlocks the full density promise of three-dimensional crossbar memories.

Subject of Research: Two-dimensional-material tunnel junction selectors for crossbar memory arrays

Article Title: Universal selectors from two-dimensional materials

Article References: Sun, Y., & Zhu, M. (2026). Universal selectors from two-dimensional materials. Nature Electronics. https://doi.org/10.1038/s41928-026-01723-z

Image Credits: AI Generated

DOI: 10.1038/s41928-026-01723-z

Keywords: two-dimensional materials, tunnel junction, selector device, crossbar arrays, memory technology, nonlinearity, endurance, switching speed, device variability, van der Waals heterostructures, Nature Electronics, resistive memory

The underlying problem is well known to memory engineers. Crossbar arrays, in which memory elements sit at the intersections of perpendicular word lines and bit lines, offer the most compact possible memory layout and can be stacked in three dimensions. But in a passive crossbar without a selector, current sneaks through neighboring unselected cells along paths known as sneak currents, corrupting the readout of the intended cell. Selectors are two-terminal devices placed in series with each memory element—forming one-selector-one-resistor or one-selector-one-capacitor cells—to suppress these parasitic currents. An ideal selector must be extremely nonlinear, passing large current only at its selected operating voltage while blocking current at half or a fraction of that voltage, yet it must also survive trillions of switching cycles, respond in nanoseconds, remain stable across temperatures and operate identically from device to device. No single selector technology has satisfied all of these requirements simultaneously.

The USC-led team’s answer is a graded tunnel barrier realized entirely within a van der Waals heterostructure. Instead of a single uniform insulating barrier, the device stacks materials of different band alignments so that the effective barrier profile changes with applied voltage. At low voltages, the barrier remains thick and high, strangling leakage current to negligible levels. As the voltage across the device rises toward the read or write condition, the barrier is thinned and lowered in a controlled fashion, allowing electrons to tunnel through with high efficiency. The result is an exponential increase in current over a small voltage window—precisely the nonlinearity that crossbar selectors demand.

To move beyond trial and error, the researchers developed a theoretical model that predicts the electrical characteristics of selectors with graded tunnel barriers. Guided by this model, they designed and fabricated two specific device structures. In the first, the primary tunnel barrier is hexagonal boron nitride, the wide-bandgap insulator often called white graphene. In the second, the barrier is gallium sulfide, a layered semiconductor with a smaller bandgap. Both variants sandwich the barrier between two layers of molybdenum disulfide, with graphene sheets serving as the top and bottom electrodes, forming the complete five-layer stack graphene/molybdenum disulfide/barrier/molybdenum disulfide/graphene.

The performance of the hexagonal boron nitride device is striking. It exhibits a nonlinearity exceeding 10 million, meaning the current at the operating voltage is more than ten million times larger than the leakage current at reduced voltages—among the highest values reported for any two-terminal selector. Perhaps more impressive is its endurance: the device survived more than a trillion switching cycles without failure, a figure that dwarfs the lifetimes of many competing selector technologies and approaches what commercial memory products require. Switching occurs in less than 20 nanoseconds, the current-voltage characteristics show minimal temperature dependence, and the variation from one device to another is low—critical attributes for manufacturing, where billions of cells must behave nearly identically.

The gallium sulfide variant trades some nonlinearity for a dramatically reduced operating voltage. Because gallium sulfide has a smaller barrier height than hexagonal boron nitride, the device can deliver a nonlinearity above one million while operating at only 2.5 volts. Low-voltage operation matters enormously for modern memory, which must integrate with silicon circuitry whose supply voltages continue to shrink. A selector that requires high voltages to turn on forces the surrounding periphery circuitry to handle elevated stress; a selector that switches at 2.5 volts eases that burden and reduces overall energy consumption during write operations.

Beyond single devices, the team demonstrated that their selectors can be integrated into functional memory cells. They built one-selector-one-resistor cells by pairing the selectors with memristive devices, including a hafnium-oxide-based memristor stack with palladium electrodes, and they also demonstrated one-selector-one-capacitor configurations. This dual compatibility with both resistive nonvolatile memory and capacitive volatile memory suggests the selector technology is genuinely universal, applicable across different memory families rather than tied to a single cell type. The researchers benchmarked their devices against the broad landscape of existing selector technologies—including ovonic threshold switching chalcogenides, niobium oxide threshold devices, metal-insulator-metal tunnel diodes and mixed-ionic-electronic-conduction access devices—and found that the van der Waals approach uniquely combines high selectivity with trillion-cycle endurance, nanosecond speed, temperature stability and low variability.

Choosing van der Waals materials is central to the achievement. These layered crystals are held together by weak interlayer forces, so each atomic plane can be exfoliated and restacked like molecular building blocks without the dangling bonds and interfacial defects that plague conventional three-dimensional semiconductors. Atomically sharp interfaces mean the tunnel barrier thickness is controlled layer by layer, giving designers reproducible, deterministic barrier profiles. Graphene electrodes contribute their own advantages, providing chemically inert, highly conductive contacts that do not interdiffuse with the underlying layers. The same properties that have made van der Waals heterostructures a playground for condensed matter physics here translate directly into manufacturable device metrics: negligible cycling variation, minimal temperature dependence and good interdevice uniformity all flow from the crystalline perfection of the interfaces.

The implications extend toward the long-sought goal of three-dimensional memory stacking. Because the entire selector is built from vertically stacked two-dimensional layers, it is inherently stackable, unlike selectors that rely on complex oxide growth or electroforming processes that are difficult to repeat layer upon layer. Paired with scalable nonvolatile memories such as memristors and phase-change or resistive cells, these selectors could enable dense three-dimensional crossbar memories for storage-class memory, in-memory computing and neuromorphic artificial intelligence hardware, where analog crossbar arrays perform vector-matrix multiplications directly in hardware. The research team, whose work was supported by the Army Research Office, the Air Force Office of Scientific Research and the National Science Foundation, has also released the experimental and simulation data through the Harvard Dataverse, giving the broader community the tools to build on the design framework. As the demand for data storage and energy-efficient computation continues its relentless climb, devices that tame sneak currents with atomic precision may prove to be one of the quiet enablers of the next memory generation.

Subject of Research: Van der Waals tunnel-junction selector devices with graded tunnel barriers for high-density memory crossbar arrays

Article Title: High-performance tunnel-junction selectors with graded tunnel barriers based on five-layer van der Waals heterostructures

Article References: Liu, H., Liao, H.-T., Li, M., Yang, N., Hsu, T.-H., Zhao, R., Xu, Y., Kim, S. J., Ni, Y., Chenna, V., Mahalingam, K., Bowers, C., Ganguli, S., Roy, A. K., Watanabe, K., Taniguchi, T., Hashemi, H., Guo, J., Dubey, M., … Joshua Yang, J. (2026). High-performance tunnel-junction selectors with graded tunnel barriers based on five-layer van der Waals heterostructures. Nature Electronics. https://doi.org/10.1038/s41928-026-01704-2

Image Credits: AI Generated

DOI: 10.1038/s41928-026-01704-2

Keywords: van der Waals heterostructures, tunnel-junction selector, memory crossbar arrays, molybdenum disulfide, hexagonal boron nitride, gallium sulfide, memristors, nonlinearity, endurance, three-dimensional memory, nanoelectronics, Nature Electronics