Machine learning design of electronic packaging materials: A phonon spectrum continuity matching framework for high thermal conductivity

The rapid advancement of artificial intelligence and high-performance computing chips has posed significant challenges to thermal management in advanced electronic packaging. Conventional epoxy packaging materials struggle to meet the thermal dissipation demands of high-power chips due to their low thermal conductivity (<0.5 W m −1 K −1 ). This study introduces a phonon spectrum continuity matching strategy for multiple functional fillers integrated with a machine learning framework to develop high thermal conductivity composites. Through high-throughput screening, we identified three high performance composites. Experiments demonstrate that the composite achieves through-plane and in-plane thermal conductivities of 2.22 and 11.18 W m −1 K −1 , while maintaining excellent process adaptability (viscosity: 16.1 Pa s). Device-level thermal management validation using an Allwinner H133 platform demonstrated a 5.1 °C reduction in chip temperature. The Arrhenius analysis further suggests that the improved thermal management capability may contribute to enhanced reliability and extended service lifetime of electronic devices. The synergistic interaction between the high frequency phonon characteristics of diamond, the mid-to-high frequency characteristics of expanded graphite, and the mid-to-low frequency characteristics of SiO 2 enhances the thermal conductivity of the composite. Machine learning effectively explores the multi-component formulation space, overcoming the limitations of trial-and-error methods. This study establishes a systematic paradigm for the development and validation of high-performance composites in advanced packaging.