Page 58 - Read Online

Wen et al. J. Mater. Inf. 2025, 5, 30 https://dx.doi.org/10.20517/jmi.2024.102 Page 3 of 21

mobility, promoting rapid hole transport to the electrode; (iii) it should have good solubility and film-
forming properties; (iv) it should possess good hydrophobicity, protecting the perovskite layer from
moisture-induced degradation and improving device stability; (v) it should demonstrate high light
transmittance, avoiding competition and absorption of light by the HTL that would otherwise be absorbed
by the perovskite layer; (vi) it should be low-cost and easy to synthesize.

SM-HTMs with methoxyaniline as the terminal group have garnered significant attention from researchers
due to their advantages of easy synthesis, straightforward purification, tunable energy levels, and strong
structural functionality [12,19] . These materials typically consist of a central group and two terminal groups.
Common central units include benzene rings, naphthalene, pyrrole, furan, thiophene, carbazole, and
dibenzothiophene, while the terminal groups are often composed of dimethoxydiphenylamine and
dimethoxytriphenylamine [20-23] . Studies have demonstrated that methoxyaniline groups play a critical role in
influencing the electronic properties of HTMs. Additionally, the incorporation of methoxyaniline groups
enhances the solubility of these materials, leading to improved film morphology [24-27] .

Currently, research on new SM-HTMs primarily relies on traditional laboratory synthesis and trial-and-
error methods, which are inefficient and costly. However, with the rapid advancement of artificial
intelligence technology, high-throughput computational screening and machine learning (ML) methods
have become increasingly valuable in materials science [28-30] . These approaches are now widely applied in the
development of new materials for various fields, including photovoltaics, optoelectronics, and
photocatalysis [31-34] . Therefore, combining high-throughput computational screening with ML to design new
SM-HTMs and predict their properties represents a highly promising direction for future research and
development of SM-HTMs. In recent years, significant progress has been made in the structural design and
performance investigation of HTMs for solar cells from the perspective of theoretical chemistry [35-39] .
Building on this foundation, the integration of high-throughput screening and ML has emerged as a pivotal
driving force in accelerating the discovery of novel SM-HTMs. Wu et al. pioneered a closed-loop workflow
combining high-throughput organic synthesis with Bayesian optimization (BO) to discover SM-HTMs
[40]
tailored for perovskite devices . By training predictive models on 149 synthesized molecules and screening
a virtual library of 1 million candidates, they achieved a certified PCE of 25.9%, demonstrating the power of
data-driven approaches in navigating complex material landscapes with limited datasets. Complementing
this, Faruque et al. employed a translational dimer model for high-throughput screening of 74
diacenaphtho-extended heterocycles, coupled with ML-guided crystal structure prediction (CSP) and
[41]
carrier mobility calculations . Their workflow identified candidates with hole mobilities exceeding
10 cm /V·s, validated by semiclassical Marcus theory, highlighting the role of computational screening in
2
optimizing molecular packing and charge transport. These studies exemplify a paradigm shift toward
autonomous, ML-enhanced material discovery, offering scalable strategies for designing next-generation
HTMs with superior optoelectronic properties. However, the success of high-throughput screening and ML
hinges on the availability of a robust library of candidate structures for HTMs. Consequently, the
development of innovative design methodologies and algorithms for the structure of SM-HTMs is essential.

In this study, an efficient design strategy for SM-HTMs is proposed, integrating molecular assembly
algorithms, high-throughput screening, and ML model predictions. A self-developed molecular splicing
algorithm (MSA) was employed to construct a database of potential SM-HTMs for PSCs. By integrating
density functional theory (DFT), high-throughput computations were conducted to identify six high-
performance candidate SM-HTMs for subsequent synthesis and performance evaluation. Furthermore, the
molecular structure and property datasets obtained through high-throughput calculations were utilized to
develop property prediction models for SM-HTMs using three ML approaches: random forest (RF),