Data-Driven Flipping Engineering for High Thermal Anisotropy in Janus Two-Dimensional Materials

The vertical integration of heterogeneous dies in three-dimensional integrated circuits demands materials with extreme thermal anisotropy to dissipate heat in-plane while suppressing cross-plane leakage. Randomly twisted van der Waals stacks can enhance anisotropy, but their stochastic nature lacks precise control. Here, we report flipping engineering in Janus MoSSe stacks, exploiting the intrinsic asymmetry between the S-Mo-Se and Se-Mo-S terminations as a programmable structural degree of freedom orthogonal to twisting. By integrating machine-learned potential molecular dynamics with Bayesian optimization for a 14-layer MoSSe stack, we identify a structure with strongly suppressed effective cross-plane thermal conductivity of 0.082 ± 0.002 W m-1 K-1 at 300 K while maintaining robust in-plane conduction, yielding an anisotropy ratio of about 760 and 35% higher than the crystalline reference. This effect arises from flipping-induced breaking of stacking periodicity, which perturbs midfrequency phonons (2-8 THz) and leads to band splitting and nearly flat, layer-localized modes. These changes reduce the cross-plane propagating channels and thus weaken their contribution to thermal transport, while their impact on in-plane transport dominated by low-frequency acoustic phonons is negligible. Flipping engineering thus provides programmable stacking-sequence strategy and a versatile route for designing anisotropic properties in layered materials.