Year: 2026
Type: article
Abstract: ABSTRACT The rapid growth of data‐intensive technologies, including artificial intelligence and 5G/6G communications, is driving demand for advanced IC packaging platforms with high electrical performance and thermo‐mechanical reliability. Glass substrates are promising candidates due to their high strength, low dielectric constant, and tunable coefficient of thermal expansion (CTE). However, composition design remains challenging since the amorphous structure obscures composition–property relationships and often requires extensive trial‐and‐error development. In this study, machine learning (ML) is used to accelerate oxide glass development by predicting key IC packaging properties, including elastic moduli, hardness, fracture toughness, dielectric constant and loss, CTE, glass transition temperature, and thermal conductivity. SHAP (SHapley Additive exPlanations) analysis provides model interpretability by quantifying the contributions of individual oxides to each predicted property. A grid‐based screening is then performed to identify candidate compositions satisfying multi‐objective targets benchmarked against commercial packaging glasses, followed by particle swarm optimization (PSO) to further refine the retained candidates. This integrated workflow enables efficient exploration of feasible compositional spaces and provides data‐driven guidance for the design of oxide glasses for advanced IC packaging.
DOI: 10.1111/ijag.70073
Authors: Shu‐Wei Chang, Po‐Wei Huang, Yueh‐Ting Shih
Institution: National Cheng Kung University
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Keywords: glass substrates, integrated circuit packaging, machine learning property prediction, SHAP, particle swarm optimization, elastic modulus, fracture toughness, dielectric constant, coefficient of thermal expansion, glass transition temperature, thermal conductivity, hardness, amorphous materials
Open Access status: bronze
OpenAlex ID: W7226934568
