Topological constraint theory (TCT) establishes quantitative predictive models linking the microscopic structure and macroscopic properties of glass via quantifying atomic-scale bonding constraints. Molecular dynamics (MD) simulations, utilizing force field models, enable the exploration of dynamic structural evolution across temporal scales from nanosecond-level structural relaxation to microsecond-level phase separation dynamics. Machine learning (ML) algorithms construct high-dimensional composition-property mappings, thus opening a paradigm for inverse glass design based on targeted performance requirements.

This review first elaborates on the fundamental principles of TCT and its pivotal role in predicting glass transition temperature, analyzing thermal expansion behavior, evaluating hardness, and uncovering the mechanisms of the mixed-alkali effect. Subsequently, it highlights the innovative research conducted by using MD simulations, i.e., structural optimization of encapsulation glasses for high-pressure power devices, mechanical reinforcement mechanisms of high-strength glass fibers, and dielectric property modulation of glass substrates for high-frequency electronic applications as well as the applications in glass ceramics. Finally, this review discusses the emerging paradigms of ML in glass property prediction and envisions the synergistic integration of TCT, MD, and ML in the development of next-generation glass materials.

TCT via quantifying the types and numbers of atomic constraints within the glass network effectively reveals the intrinsic correlation between glass structure and macroscopic properties. It provides a solid theoretical foundation for understanding and tailoring glass performance, offering a significant potential for the development of high-performance glass materials. Under the guidance of this theoretical framework, MD simulation serves as a powerful tool for investigating the atomic-scale structure and dynamic behavior of glasses, thereby offering an effective pathway to establish structure-property relationships. However, TCT is often limited to specific systems, which can introduce errors when applied to complex compositions. Meanwhile, MD simulations are computationally expensive and sometimes suffer from the absence of accurate potential functions. Several limitations still hinder their broader application i.e., a) insufficient temporal resolution. Femtosecond-level time steps are inadequate for resolving high-frequency transient polarization responses; b) force fields often simplify quantum effects-current models, and fail to accurately describe local charge fluctuations and dynamic polarizability; and c) Limited spatial scales. Nano-sized models cannot fully capture structural heterogeneity, and statistical convergence under high-frequency electric fields is constrained by available computational power. MD simulations remain inadequate for directly investigating glass performance under high-frequency applications. To overcome these challenges, multiscale coupling models are needed, such as integrating ML algorithms to enhance the accuracy of polarization dynamics through deep learning-based potential functions, and employing materials informatics to accelerate the screening of high-performance glass compositions. These strategies are expected to significantly improve the efficiency of rational glass design.

Glass-ceramics, which evolve from glasses, are widely used in applications such as encapsulation materials, printed circuit boards, microwave components, sealing glasses, and low-temperature co-fired ceramic (LTCC) substrates, having a considerable value in high-frequency communications, microelectronic packaging, and power devices. The existing research on the crystallization phenomena in glass-ceramics mainly follows two technical pathways, i.e., a) employing structural characterization methods in combination with diffusion kinetics simulations and experimental validation to indirectly infer crystal precipitation behavior, and b) constructing glass-ceramic models in MD systems by manipulation strategies such as “dig-insert” or “cut-combine” approach. However, these approaches remain inherently limited to either indirect representations of crystalline formation or manually constructed models. Overcoming the existing technological bottlenecks to enable real-time visualization of crystal nucleation and growth mechanisms during dynamic simulations remains a critical challenge.

Finally, in the context of advanced packaging and heterogeneous integration, glass substrates play a crucial role in 3D integration, but face multiple challenges in interfacial reaction dynamics with silicon/metal substrates. These include atomic-scale interdiffusion leading to dielectric degradation, cross-scale coupling between nano scale chemical bond reconstruction and macroscopic stress evolution, as well as non-equilibrium thermodynamic effects induced by laser-assisted processing. There is an urgent need to develop simulation frameworks that integrate co-evolution of multiple properties across scales, enabling quantitative prediction of atomic interdiffusion coefficients, chemical bond reconstruction energy barriers, and residual stress distributions. Such efforts will provide the theoretical foundation for the design and process optimization of high-reliability glass substrates.