In MATBG devices, the interaction between itinerant electrons in graphene and localized electrons at the graphene/hBN interface induces unconventional electronic ferroelectricity. This phenomenon is driven by the alignment of the graphene crystalline orientation with that of hBN, creating a moiré superlattice at the interface. This superlattice is essential for the emergence of layer-specific localized electronic states and, consequently, unconventional electronic ferroelectricity. Unlike traditional ionic-displacement mechanisms, electronic ferroelectricity in this system does not have a fixed coercive field but instead exhibits continuously tunable remnant polarization within a single domain. The team discovered that the remnant polarization is linearly correlated with the applied electric field, indicating the possibility of achieving multiple ferroelectric states with adjustable remnant polarization through an external electric field. Moreover, when an in-plane magnetic field is applied, the remnant polarization of MATBG varies linearly with the magnitude of the magnetic field. This phenomenon, known as the orthogonal magnetoelectric effect, occurs due to the coupling between the out-of-plane remnant polarization and the in-plane magnetic field. The extracted magnetoelectric coupling coefficient shows an anisotropic dependence on the magnetic field’s azimuthal angle and is two orders of magnitude higher than that of traditional multiferroic materials. This remarkable effect challenges current theoretical models and warrants further experimental and theoretical exploration.

Recently, a research team led by Professors Shijun Liang and Feng Miao from Nanjing University, in collaboration with Professor Bin Cheng from Nanjing University of Science and Technology, has successfully achieved the coexistence of electronic ferroelectricity and Chern insulators in doubly aligned magic-angle twisted bilayer graphene (MATBG) devices^[5]. The team also introduced a noise-resistant neuromorphic computing scheme. By precisely controlling the amplitude of gate voltage pulses, they realized selective and non-volatile switching between different Chern insulator states, achieving >1200 quasi-continuous ferroelectric states in a single device. Moreover, they leveraged the quantized conductance of Chern insulator states as weights in convolutional neural networks (CNNs), making the first demonstration of ferroelectric Chern insulator devices’ potential in noise-resistant neuromorphic computation, as shown in Fig. 1. This work paves the pathway for the development of new low-power electronic devices based on topological edge states.

This experimental work represents a critical step toward proof-of-concept neuromorphic computing applications based on the quantum Hall effect and marks a significant milestone in the development of dissipationless topotronics. However, to fully realize the practical potential of moiré quantum materials in neuromorphic computing, further research is needed, particularly in eliminating the reliance on external magnetic fields at required temperatures and large-scale device integration.

In 1980, scientist Klaus von Klitzing discovered the quantum Hall effect^[1], a groundbreaking achievement that earned him the Nobel Prize in Physics in 1985. This discovery was a significant milestone in condensed matter physics, representing the first identification of topological quantum states. Following this, the discoveries of the quantum anomalous Hall effect^[2], the quantum spin Hall effect^[3], and Chern insulators^[4] further revealed the unique transport properties of topologically protected edge states. These states are characterized by their immunity to scattering from external perturbations, making them highly promising for developing low-power, noise-resistant quantum electronic devices—capabilities that traditional charge-based devices cannot offer. As a result, leveraging topologically protected edge states in quantum materials to create low-power quantum electronic devices has become a key area of focus in condensed matter physics and information technology.

The ability to continuously tune polarization in electronic ferroelectricity enables non-volatile adjustment of the electron concentration, which facilitates the selective switching of carrier-density-dependent quantum states. In multi-Chern band systems like MATBG, variations in electron concentration influence the Chern band filling factor and the total Chern number, allowing for selective switching of Chern insulator states. The team demonstrated this by applying gate voltage pulses of varying amplitudes to selectively control the magnitude of ferroelectric polarization, successfully achieving selective switching of quantized platforms corresponding to different Chern insulator states. This achievement represents the first demonstration of non-volatile switching between multiple topological edge states in quantum materials, paving the way for next-generation ultralow-power electronics.

The unique electronic ferroelectricity in these devices facilitates the achievement of an exceptionally high number of quasi-continuous ferroelectric states. By applying precisely controlled gate pulse amplitudes, the team achieved a record of ~1280 quasi-continuous non-volatile ferroelectric states within a single device. More importantly, they demonstrated the ability to perform arbitrary switching of ferroelectric polarization through delicately controlled gate pulses, marking the first quasi-continuous selective switching of ferroelectric states. This breakthrough offers new opportunities for ultra-high-density memory and robust neuromorphic computing. Finally, the research team successfully trained a convolutional neural network (CNN) using the Hall conductance of ferroelectric Chern insulator devices, showcasing noise-resistant neuromorphic computing applications. The CNN maintained high recognition accuracy even under significant noise levels, a feature attributed to the topologically protected edge states of the ferroelectric Chern insulator, which preserve precise Hall conductance despite external noise interference. This work sets the stage for the development of next-generation memory and computational devices with unprecedented efficiency and robustness.