Deep Learning-Based Quantum Transport Simulations Enable Efficient Analysis of Two-Dimensional Materials

The quest to unlock the potential of two-dimensional materials for future nanoelectronics demands accurate prediction of how electrons flow through them, a process known as quantum transport. Jijie Zou from AI for Science Institute and Peking University, alongside Zhanghao Zhouyin from McGill University, and colleagues, now present a significantly faster and more efficient method for simulating this behaviour. Their new framework, DeePTB-NEGF, combines the power of deep learning with established quantum transport techniques, allowing researchers to model electron flow with unprecedented speed and accuracy. Validating their approach on materials including graphene, hexagonal boron nitride and molybdenum disulphide, the team demonstrates excellent agreement with conventional simulations, while achieving substantial gains in computational efficiency, paving the way for large-scale exploration and innovative device design in the field of two-dimensional materials. Predicting 2D Material Electronic Properties with Machine Learning Two-dimensional (2D) materials hold immense promise for next-generation nanoelectronic devices, exhibiting a diverse range of electronic properties. Accurately predicting these properties is crucial for designing and optimising devices, yet remains a significant challenge due to the complex quantum mechanical interactions within these materials.

This research presents a novel computational approach to predict the electronic properties of 2D materials with both high accuracy and efficiency. The method combines first-principles density functional theory calculations with machine learning techniques, specifically utilising a convolutional neural network to accelerate the prediction of electronic band structures. By training this network on a comprehensive dataset of calculated band structures, researchers established a predictive model capable of rapidly generating accurate band structures for new and unexplored materials. This significantly reduces the computational cost associated with materials discovery and design, enabling efficient screening of a vast chemical space for materials with desired electronic properties.

The team demonstrated the model’s performance and applied it to predict the electronic properties of several previously unstudied 2D materials, paving the way for innovative technologies in areas such as flexible electronics, energy storage, and quantum computing.,.

Deep Learning Accelerates Quantum Transport Simulations Scientists have developed DeePTB-NEGF, a novel framework that combines deep learning with established quantum transport simulations to efficiently and accurately model the electronic properties of two-dimensional materials. The study addresses the computational limitations of conventional density functional theory combined with the non-equilibrium Green’s function formalism, which hinders large-scale investigations of nanoscale systems. DeePTB-NEGF leverages a deep learning model, DeePTB, to predict tight-binding Hamiltonians directly from first-principles data, significantly reducing computational demands while maintaining accuracy. This approach learns the relationship between local atomic environments and Hamiltonian matrix elements, generating corrections to conventional tight-binding models and preserving Hamiltonian sparsity. The methodology begins with atomic structures and employs DeePTB to both train and infer the tight-binding Hamiltonian, which is then integrated with the open-source quantum transport package, DPNEGF, to simulate electronic transport properties such as transmission spectra. DeePTB predicts Hamiltonian parameters by learning from diverse local atomic environments, enabling reliable predictions for previously unseen structures. Within the DPNEGF implementation of the non-equilibrium Green’s function formalism, the retarded Green’s function is calculated using the Hamiltonian predicted by DeePTB and self-energies representing semi-infinite electrodes. The transmission spectrum, a key indicator of electronic transport, is then evaluated using the calculated Green’s function and electrode self-energies, providing a measure of how easily electrons can pass through the material. Researchers validated the DeePTB-NEGF framework by applying it to graphene, hexagonal boron nitride, and molybdenum disulfide, materials exhibiting semimetallic, insulating, and semiconducting behaviours respectively, providing a comprehensive test of the method’s versatility and reliability.

Results demonstrate excellent agreement with conventional calculations of both band structures and transmission spectra, highlighting the potential of deep learning to accelerate and enhance quantum transport studies in two-dimensional materials.,.

Deep Learning Accelerates Quantum Transport Simulations This research presents a new computational framework, DeePTB-NEGF, capable of accurately and efficiently simulating quantum transport in two-dimensional materials. By combining deep learning with established methods, the researchers have achieved a significant speed-up in calculations while maintaining accuracy comparable to conventional techniques. The method was successfully benchmarked against graphene, hexagonal boron nitride, and molybdenum disulfide, demonstrating its reliability across a range of materials. The key achievement lies in the use of deep learning to predict tight-binding Hamiltonians, which dramatically reduces the computational cost associated with large-scale simulations. This allows for the exploration of larger and more complex structures than previously possible, opening avenues for the rapid prototyping of novel two-dimensional electronic devices. The researchers acknowledge that the efficiency gains become even more pronounced when simulating larger systems, where traditional methods become particularly demanding. Future work will focus on extending the method to even more complex materials and exploring its application to a wider range of nanoscale devices.

The team also intends to develop the deep learning component further, aiming to improve its accuracy and efficiency.,.

Deep Learning Accelerates Quantum Transport Simulations This work demonstrates a new approach to quantum transport simulations in 2D materials by combining deep learning with the Non-Equilibrium Green’s Function (NEGF) method. The authors developed a framework called DeePTB-NEGF that leverages deep learning to predict tight-binding (TB) Hamiltonians, significantly accelerating the simulation process. Key findings include a speed-up in simulations exceeding an order of magnitude compared to traditional Density Functional Theory (DFT) based NEGF simulations for graphene, hBN, and MoS2, while maintaining comparable accuracy. The method is well-suited for simulating large and previously unseen structures, enabling high-throughput materials exploration and rapid prototyping of 2D electronic devices. The researchers trained a deep neural network to learn the mapping between local atomic environments and corresponding TB Hamiltonian parameters, allowing the network to predict the Hamiltonian for new structures without requiring computationally expensive DFT calculations. The predicted Hamiltonian is then used within the NEGF formalism to calculate the quantum transport properties of the material. This work demonstrates the potential of deep learning to overcome the computational bottleneck in quantum transport simulations, facilitating the discovery and design of novel 2D materials and devices with tailored electronic properties. 👉 More information 🗞 Deep Learning-Based Quantum Transport Simulations in Two-Dimensional Materials 🧠 ArXiv: https://arxiv.org/abs/2512.11291 Tags: