Enhanced heat conduction in diamond/copper composites via interconnected structures: Machine learning molecular dynamics simulation

Diamond/Cu composites have attracted considerable attention for thermal management applications due to their outstanding thermal conductivity. Recent studies have demonstrated that interconnected diamond network structures can significantly enhance the heat conduction of diamond/Cu composites, while the underlying microscopic heat transfer mechanisms remain to be fully elucidated. This study employed machine learning molecular dynamics simulations to validate and elucidate the heat conduction enhancement mechanisms of interconnected network structures. We developed a machine learning potential for diamond/Cu heterostructures based on the neuroevolution potential model and calculated the lattice thermal conductivity (LTC) of two series of network structures: diamond network/Cu (DN/Cu) and Cu network/diamond (CuN/D). The results demonstrate that DN/Cu exhibits monotonically increasing LTC with diamond content due to continuous heat transfer channels, while CuN/D shows non-monotonic behavior with a minimum LTC at intermediate diamond fractions, originating from the competitive mechanism between contributions from low-frequency Cu phonons (0–8 THz) and high-frequency diamond phonons (10–40 THz) to thermal transport. Through spectral LTC decomposition and wavelike phonon transmission analysis, we elucidated the suppressive effects of multiple heat transfer mechanisms on LTC, including phonon interfacial scattering, coherent interference effects, and total internal reflection in nano-network structures. This work establishes quantitative design thresholds for lattice thermal transport in DN/Cu, revealing a critical diamond volume fraction of ∼ 30% above which DN/Cu consistently outperforms CuN/D, and an unexpected LTC minimum at ∼ 50% diamond in CuN/D structures driven by phonon frequency competition. These findings explain why diamond network structures, despite demonstrating significant heat conduction enhancement compared to dispersed particles, still exhibit LTC values substantially lower than pure diamond, thus revealing substantial room for design optimization of network architectures.