Periodic Reporting for period 1 - AI4SPIN (Artificial Intelligence–Driven Materials Design for Spintronic Applications)
Reporting period: 2023-04-01 to 2025-09-30
AI4SPIN aims to revolutionize the discovery of two-dimensional (2D) materials for spintronic applications by developing an artificial intelligence–driven platform to design and optimize van der Waals (vdW) heterostructures with enhanced spin-orbit torque (SOT) performance. As the demand for faster, energy-efficient electronic devices continues to grow, AI4SPIN addresses the urgent need for scalable and intelligent methods to navigate the enormous design space of 2D materials and their combinations.
The project integrates advanced quantum transport simulations, density functional theory, and machine learning into a single automated workflow. It will first identify promising 2D building blocks through high-throughput screening and quantum modeling. Then, a deep-learning–based engine (AUTOMATA) will autonomously assess the spin transport properties of candidate materials. Finally, an evolutionary algorithm (COMPASS) will generate and optimize heterostructures targeting maximum SOT efficiency.
By bridging physical theory and artificial intelligence, AI4SPIN provides a paradigm shift in material design — enabling the rapid discovery of functional vdW structures for next-generation spintronic devices. The framework is generalizable and can be extended beyond SOTs to other quantum or electronic phenomena, supporting Europe’s leadership in the development of low-power, post-CMOS technologies.
The project integrates advanced quantum transport simulations, density functional theory, and machine learning into a single automated workflow. It will first identify promising 2D building blocks through high-throughput screening and quantum modeling. Then, a deep-learning–based engine (AUTOMATA) will autonomously assess the spin transport properties of candidate materials. Finally, an evolutionary algorithm (COMPASS) will generate and optimize heterostructures targeting maximum SOT efficiency.
By bridging physical theory and artificial intelligence, AI4SPIN provides a paradigm shift in material design — enabling the rapid discovery of functional vdW structures for next-generation spintronic devices. The framework is generalizable and can be extended beyond SOTs to other quantum or electronic phenomena, supporting Europe’s leadership in the development of low-power, post-CMOS technologies.
1. Development of Scalable Quantum Transport Frameworks
A major technical milestone has been the extension of real-space quantum transport simulations to enable high-throughput assessments of SOTs. Traditional tight-binding approaches, while accurate, suffer from poor scaling due to the long-range nature of spin-orbit interactions. To address this, we developed a novel mixed-representation Chebyshev polynomial expansion technique that retains physical accuracy while drastically improving computational efficiency.
In this method, periodic Hamiltonian components are treated in momentum space while electrostatic disorder is modeled in real space. The implementation leverages fast Fourier transforms at each recursion step of the Chebyshev expansion, enabling us to simulate systems of realistic device dimensions without compromising resolution.
2. Spin-Orbit Torques in Systems with Hidden Spin Textures
Using our new framework, we investigated the layered material PtSe2, known for hosting hidden spin textures. We combined ab initio simulations via Quantum ESPRESSO with tight-binding models generated using PAOFLOW. Our analysis revealed the orbital origin and spatial structure of the spin-orbital textures, and we demonstrated that these textures are strong candidates for efficient SOT generation.
Importantly, we showed that the nonequilibrium spin density responsible for torque can be tuned by electric fields, offering a practical knob for device operation. These results establish the feasibility of electric-field-driven SOT control in centrosymmetric materials by exploiting internal dipolar fields and layer-resolved Rashba effects.
3. Microscopic Theory of Damping-like Spin-Orbit Torques
A fundamental outcome of the project was the development of a microscopic theory for the damping-like component of the SOT, which had remained conceptually elusive. Our approach describes how a nonequilibrium spin polarization emerges in the Fermi sea via semiclassical dynamics under an applied electric field. This framework allowed us to formally connect band symmetries, broken inversion, and time-reversal properties to the strength and direction of the damping-like torque.
In this context, we uncovered a novel mechanism for torque generation inside topological band gaps, enabled by the interplay between spin and pseudospin angular momentum. These results represent a theoretical breakthrough and extend the range of systems and conditions under which SOTs can be realized and controlled.
Key publications stemming from this work include:
Phys. Rev. Lett. 132, 266301 (2024)
arXiv:2408.16359 (2024)
4. Automation and Acceleration of DFT-Based Magnetic Materials Simulations
On the materials discovery side, we developed a new computational automation layer that integrates density functional theory (DFT) tools with modular programming patterns. This framework enhances reproducibility, simplifies cross-code workflows, and supports automated exploration of electronic, magnetic, and structural properties across multiple simulation packages.
These tools were used to investigate critical classes of 2D magnetic materials, including CrSBr and CrGeTe₃, in collaboration with experimental partners. The simulations provided deep insights into magnetic anisotropy, electronic structure, and stacking-dependent properties, identifying promising candidates for future device integration.
These results culminated in a review article on ab initio 2D magnetism, which outlines a path forward for materials design and prediction in this emerging area.
A major technical milestone has been the extension of real-space quantum transport simulations to enable high-throughput assessments of SOTs. Traditional tight-binding approaches, while accurate, suffer from poor scaling due to the long-range nature of spin-orbit interactions. To address this, we developed a novel mixed-representation Chebyshev polynomial expansion technique that retains physical accuracy while drastically improving computational efficiency.
In this method, periodic Hamiltonian components are treated in momentum space while electrostatic disorder is modeled in real space. The implementation leverages fast Fourier transforms at each recursion step of the Chebyshev expansion, enabling us to simulate systems of realistic device dimensions without compromising resolution.
2. Spin-Orbit Torques in Systems with Hidden Spin Textures
Using our new framework, we investigated the layered material PtSe2, known for hosting hidden spin textures. We combined ab initio simulations via Quantum ESPRESSO with tight-binding models generated using PAOFLOW. Our analysis revealed the orbital origin and spatial structure of the spin-orbital textures, and we demonstrated that these textures are strong candidates for efficient SOT generation.
Importantly, we showed that the nonequilibrium spin density responsible for torque can be tuned by electric fields, offering a practical knob for device operation. These results establish the feasibility of electric-field-driven SOT control in centrosymmetric materials by exploiting internal dipolar fields and layer-resolved Rashba effects.
3. Microscopic Theory of Damping-like Spin-Orbit Torques
A fundamental outcome of the project was the development of a microscopic theory for the damping-like component of the SOT, which had remained conceptually elusive. Our approach describes how a nonequilibrium spin polarization emerges in the Fermi sea via semiclassical dynamics under an applied electric field. This framework allowed us to formally connect band symmetries, broken inversion, and time-reversal properties to the strength and direction of the damping-like torque.
In this context, we uncovered a novel mechanism for torque generation inside topological band gaps, enabled by the interplay between spin and pseudospin angular momentum. These results represent a theoretical breakthrough and extend the range of systems and conditions under which SOTs can be realized and controlled.
Key publications stemming from this work include:
Phys. Rev. Lett. 132, 266301 (2024)
arXiv:2408.16359 (2024)
4. Automation and Acceleration of DFT-Based Magnetic Materials Simulations
On the materials discovery side, we developed a new computational automation layer that integrates density functional theory (DFT) tools with modular programming patterns. This framework enhances reproducibility, simplifies cross-code workflows, and supports automated exploration of electronic, magnetic, and structural properties across multiple simulation packages.
These tools were used to investigate critical classes of 2D magnetic materials, including CrSBr and CrGeTe₃, in collaboration with experimental partners. The simulations provided deep insights into magnetic anisotropy, electronic structure, and stacking-dependent properties, identifying promising candidates for future device integration.
These results culminated in a review article on ab initio 2D magnetism, which outlines a path forward for materials design and prediction in this emerging area.
AI4SPINS has delivered a complete, functional framework that automates the design, simulation, and optimization of two-dimensional (2D) van der Waals (vdW) heterostructures for spin-orbit torque (SOT) applications. The main results include:
(a) A scalable quantum transport algorithm capable of simulating realistic system sizes with disorder and strong spin-orbit coupling.
(b) Identification and in-depth study of PtSe2 as a platform for electrically tunable spin textures and efficient SOT generation.
(c) A microscopic theory explaining the elusive damping-like spin-orbit torque mechanism, including a new torque generation process inside topological gaps.
(d) A robust, automated simulation framework for 2D magnetic materials based on density functional theory (DFT), enabling faster and more reproducible discoveries.
(e) A curated database of 2D magnetic candidates enriched with spintronic descriptors (e.g. spin-momentum locking, NESD strength).
(f) Foundational publications, including a review on ab initio 2D magnetism and high-impact articles on novel torque mechanisms.
(g) Together, these results demonstrate a shift from manual, expert-driven simulations to autonomous, scalable platforms that can accelerate discovery pipelines in materials science.
(a) A scalable quantum transport algorithm capable of simulating realistic system sizes with disorder and strong spin-orbit coupling.
(b) Identification and in-depth study of PtSe2 as a platform for electrically tunable spin textures and efficient SOT generation.
(c) A microscopic theory explaining the elusive damping-like spin-orbit torque mechanism, including a new torque generation process inside topological gaps.
(d) A robust, automated simulation framework for 2D magnetic materials based on density functional theory (DFT), enabling faster and more reproducible discoveries.
(e) A curated database of 2D magnetic candidates enriched with spintronic descriptors (e.g. spin-momentum locking, NESD strength).
(f) Foundational publications, including a review on ab initio 2D magnetism and high-impact articles on novel torque mechanisms.
(g) Together, these results demonstrate a shift from manual, expert-driven simulations to autonomous, scalable platforms that can accelerate discovery pipelines in materials science.