Researchers from Nanjing University of Aeronautics and Astronautics have demonstrated an atom‑thin sliding ferroelectric transistor that can reliably store 3,024 distinct, non‑volatile polarization states at room temperature - a record for ferroelectric neuromorphic hardware. The device is built from a well‑aligned monolayer graphene channel on hexagonal boron nitride (hBN), forming a moiré superlattice that enables fine, electrical control over ferroelectric polarization and charge localization within just a few atomic layers.
Performance comparison of Gr/hBN device and schematic diagram of its working mechanism. Credit: Nature Electronics (2026)
The transistor consists of an aligned graphene monolayer atop ferroelectric hBN, with source, drain and gate electrodes defined by standard nanofabrication, including electron‑beam evaporation for the metal contacts. Graphene serves as a high‑mobility, atomically thin channel whose Fermi level can be tuned electrostatically, while the underlying hBN provides sliding‑induced ferroelectricity and an atomically flat, low‑disorder interface. The lattice mismatch of about 1.8% between graphene and hBN generates a long‑wavelength moiré potential, which plays a central role in localizing injected carriers and stabilizing multiple polarization configurations.
When only source–drain voltage pulses are applied, the sliding ferroelectric channel exhibits more than 36 quasi‑continuous, stable polarization states at a fixed doping level, already exceeding the typical ≤32 levels reported for earlier ferroelectric neuromorphic elements at room temperature. By superimposing a gate voltage during these source–drain pulses, the team can reversibly tune the graphene Fermi energy across 84 discrete doping levels, effectively multiplying the accessible states to 36 × 84 = 3,024 physically distinct polarization states in a single transistor. These states are non‑volatile: their polarization remains stable for more than 100,000 seconds in retention tests, and device modelling indicates they could persist for up to 10 years under normal operating conditions.
The rich state space arises from the interplay between sliding ferroelectric domain walls and the moiré potential at the graphene/hBN interface. In this van der Waals heterostructure, small relative sliding displacements between atomically thin layers modulate interfacial charge transfer and invert or tune the out‑of‑plane polarization, a mechanism known as sliding ferroelectricity that has been established in other 2D systems such as multilayer hBN and 3R‑MoS₂. Here, the moiré superlattice created by graphene on hBN introduces a periodic potential landscape that spatially localizes carriers injected by the source–drain pulses, effectively pinning different polar domain configurations and enabling many metastable, yet robust, polarization states. Graphene’s gate‑tunable Fermi level then provides an additional continuous knob, allowing precise adjustment of carrier density on top of each ferroelectric configuration without disrupting the underlying polarization pattern.
To evaluate neuromorphic capability, the researchers mapped the 3,024 polarization states onto synaptic weights in simulations of a deep residual neural network performing image classification on a fashion dataset. Using the experimentally derived multi‑level characteristics for during‑training quantization, the hardware‑constrained network achieved a recognition accuracy of around 93.5%, essentially comparable to a floating‑point implementation on the same task. The high number of stable levels per device directly enhances effective weight resolution, which can reduce the number of physical synapses and overall energy per operation in neuromorphic accelerators.
Because the sliding ferroelectric transistor is only a few atomic layers thick and relies on relatively simple, planar Gr/hBN heterostructures, it is well suited to dense integration in future neuromorphic chips. The authors note that their approach expands the accessible polarization state space by roughly two orders of magnitude over conventional ferroelectric systems, while retaining room‑temperature stability and potential scalability to larger wafers. With continued optimization of response speed, endurance and state density, graphene/hBN sliding ferroelectric transistors could form the basis of compact, non‑volatile synaptic arrays for brain‑inspired computing architectures.
