Authors: N. K. Karn^1,2 and V.P.S. Awana^1,2,#

¹CSIR-National Physical Laboratory, Dr. K. S. Krishnan Marg, New Delhi-110012, India

²Academy of Scientific and Innovative Research (AcSIR), Ghaziabad 201002, India

The quest for atmospheric and room-temperature superconductors, often termed the "Holy Grail" of physics. The long-sought quest is hindered by the requirement of extreme conditions such as low temperatures and high pressures for their properties to manifest [1,2]. This pursuit is further complicated by the vast, unexplored chemical space and a history of uncorroborated claims, underscoring the critical need for rigorous validation in this high-stakes field [1-4]. 

Artificial intelligence (AI) is fundamentally transforming the search for these elusive materials by enabling rapid exploration and "design-by-prediction" of superconducting candidates [5]. AI-driven automated pipelines significantly reduce discovery timelines and costs, accelerating the identification of promising new compounds. Specific AI methodologies include predictive models, such as Hierarchical Neural Networks (HNN), for accurate critical temperature (Tc) forecasting, achieving high accuracy even with limited datasets [6]. Generative models like Crystal Diffusion Variational Autoencoders (CD-VAE) and ScGAN can create novel superconducting material candidates with remarkable success rates, with ScGAN predicting many novel materials and a multi-fold increase in discovery rate compared to manual methods [7]. Active learning iteratively refines predictions, while physics-informed AI integrates fundamental scientific principles for enhanced accuracy in predicting superconducting properties [8].  

Despite this immense potential, significant challenges persist. The inherent pressure-temperature conundrum for high-Tc superconductors remains a core scientific hurdle [1]. Issues with data quality, availability, and consistency in superconductor databases continue to pose problems for AI models, and the "black box" nature of some AI impedes understanding of underlying superconducting mechanisms. A crucial "recipe problem" also exists, as AI currently does not provide reliable detailed synthetic procedures for novel superconducting materials [1-4]. Overcoming these hurdles necessitates deep interdisciplinary collaboration and the development of tailored AI architectures that can effectively integrate scientific knowledge. Ultimately, while AI is an indispensable accelerator, achieving practical room-temperature superconductivity demands a continued synergy between advanced AI, automated experimentation, and profound human scientific expertise.

# e-mail: awana@nplindia.org

References

1. E. Zurek, Pushing towards room-temperature superconductivity, Physics, 12, 1 (2019).
2. A. P. Drozdov, M. I. Eremets, I. A. Troyan, V. Ksenofontov, and S. I. Shylin, Conventional superconductivity at 203 kelvin at high pressures in the sulfur hydride system, Nature, 525, 73 (2015).
3. S. Lee, J. Kim, H. T. Kim, S. Im, S. An, and K. H. Auh, Superconductor Pb_10-x Cu_x (PO₄)₆O showing levitation at room temperature and atmospheric pressure and mechanism. arXiv:2307.12037 (2023).
4. Kumar, K., Kumar Karn, N., Kumar, Y. and Awana, V.P.S., Absence of superconductivity in LK-99 at ambient conditions. ACS omega, 8, 41737 (2023).
5. D. Nematov, and M. Hojamberdiev, Machine Learning-Driven Materials Discovery: Unlocking Next-Generation Functional Materials--A minireview. arXiv:2503.18975 (2025).
6. https://www.yalescientific.org/2024/02/room-temperature-superconductors-not-so-fast/
7. E. Kim, and S. V. Dordevic, Scgan: A generative adversarial network to predict hypothetical superconductors. J. Phys. Cond. Mat., 36, 025702 (2023).
8. S. Xu, P. Chen, M. Qin, K. Jin, and X. D. Xiang, Predicting superconducting temperatures with new hierarchical neural network AI model. Frontiers of Physics, 20, 014205 (2025).