Practical No. 2

Current Developments and Future of AI in Solid State Physics

by Akeem Babatunde Agboola

Introduction

Solid-state physics deals with the physical properties of condensed matter systems, particularly crystalline solids, which underpin modern technologies such as semiconductors, sensors, and energy devices. Theoretical and computational approaches like density functional theory (DFT) have traditionally guided this field, but increasing system complexity and data volume have created major computational challenges. Artificial Intelligence (AI), especially machine learning (ML), has emerged as a powerful tool to overcome these limitations. AI now plays a critical role in materials discovery, electronic structure prediction, atomistic simulations, and experimental automation, reshaping research methodologies in solid state physics.

AI in Materials Discovery

Figure 1. Graph neural network (GNN) architecture for predicting solid-state material properties directly from crystal structures. Source: Ong et al. (2023).

One of the most transformative applications of AI in solid state physics is materials discovery. ML models trained on large databases such as the Materials Project can rapidly predict band gaps, formation energies, elastic constants, and thermal properties. Graph neural networks are particularly effective because they encode atomic environments and bonding relationships in crystalline lattices. These AI-driven approaches allow researchers to computationally screen thousands to millions of materials, significantly reducing experimental trial-and-error. As a result, AI has accelerated the identification of new semiconductors, thermoelectric materials, and functional oxides with tailored electronic and magnetic properties.

Electronic structure calculations lie at the heart of solid state physics but are computationally expensive for large or strongly correlated systems. AI surrogate models now approximate DFT-level accuracy with drastically reduced computational cost. Neural networks have been successfully applied to predict band structures, density of states, and Fermi surfaces. Electronic structure calculations lie at the heart of solid state physics but are computationally expensive for large or strongly correlated systems. AI surrogate models now approximate DFT-level accuracy with drastically reduced computational cost. Neural networks have been successfully applied to predict band structures, density of states, and Fermi surfaces.

Electronic Structure and Phase Prediction

Figure 2. Machine-learning-assisted materials discovery workflow combining DFT calculations with predictive AI models. Source: Butler et al. (2018).

AI has also proven effective in predicting solid solid phase transitions, including temperature-driven structural changes. By learning from first-principles datasets, ML models can identify phase boundaries and metastable structures that are difficult to capture using conventional methods alone.

Machine-Learning Interatomic Potentials

Figure 3. Comparison between classical interatomic potentials and machine-learning interatomic potentials (MLIPs), highlighting improved accuracy and scalability. Source: Behler (2016).

Machine-learning interatomic potentials represent a major advance in atomistic simulations. These models reproduce quantum-mechanical forces while allowing molecular dynamics simulations of large systems over long time scales. MLIPs are widely used to study phonon transport, defect dynamics, diffusion processes, and mechanical deformation in crystalline solids.

Autonomous Experiments and AI Laboratories

Figure 4. AI-controlled experimental platform performing autonomous nanoscale measurements and real-time optimization. Source: Nature News (2025).

A recent frontier in AI-assisted solid state physics is autonomous experimentation. AI systems have been demonstrated to control scanning probe and atomic force microscopes, design experiments, and analyze results in real time. These self-driving laboratories significantly reduce human intervention, minimize bias, and accelerate discovery cycles.

Future Outlook

The future of AI in solid state physics lies in physics-informed and explainable AI, where models incorporate physical laws, symmetries, and conservation principles. This improves interpretability and reliability while preserving computational efficiency. Integration with quantum computing may further enable solutions to strongly correlated electron systems beyond classical limits.

Conclusion

AI has fundamentally transformed solid state physics by accelerating materials discovery, improving electronic structure calculations, enabling large-scale atomistic simulations, and introducing autonomous experimentation. Rather than replacing traditional physics, AI complements it, allowing researchers to explore increasingly complex solid-state systems. Continued integration of AI with physical theory and experimentation will define the future of the field.

References

Behler, J. (2016). Perspective: Machine learning potentials for atomistic simulations. The Journal of Chemical Physics, 145(17), 170901. https://doi.org/10.1063/1.4966192

Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018). Machine learning for molecular and materials science. Nature, 559(7715), 547–555. https://doi.org/10.1038/s41586-018-0337-2

Nature News. (2025). Faster and more reliable autonomous materials experiments powered by AI. Nature.

Ong, S. P., et al. (2023). Accelerating materials science with graph neural networks. Computational Materials Science, 221, 111987. https://doi.org/10.1016/j.commatsci.2023.111987

Yoshioka, N., Mizukami, W., & Nori, F. (2020). Solving quasiparticle band spectra using neural-network quantum states. Physical Review B, 102(20), 205105. https://doi.org/10.1103/PhysRevB.102.205105