Rheology-guided multiscale engineering of microvoid-suppressed thermally conductive adhesives.

Vertically stacked electronics require thermal interface adhesives that conduct heat through confined bondlines while retaining adhesion after placement. Yet, highly filled ceramic adhesives are still commonly designed as bulk composites, with performance pursued by increasing filler loading or filler conductivity. This view overlooks how heat transport and adhesion are determined within the formed bondline, where filler connectivity, shear-dependent processability, wetting, and trapped air collectively govern performance. Here, we introduce a rheology-guided bondline densification strategy for multiscale thermally conductive adhesives composed of spherical Al2O3, plate-like hBN, fine AlN, and heat-activated polycaprolactone (PCL). The Al2O3/hBN architecture was optimized to balance shear-thinning flow, shape retention, and cured-state integrity, while fine AlN reinforced the conductive network. PCL was used as a thermoplastic densification mediator that softens during heat-primed mixing and suppresses microvoids within the ceramic-rich network. This strategy reduced microvoid formation, drove the skeletal density toward the composition-dependent theoretical density, and increased the surface free energy of the cured adhesive. The adhesive showed scale-bridging adaptability, conformally filling microscale roughness while remaining processable into large-area sheets and stamped macroscopic forms. The optimized adhesive achieved a through-plane thermal conductivity of 3.57 W m-1 K-1, representing a 2280% enhancement over neat epoxy and a 35% increase over the Al2O3/hBN baseline, along with a lap shear strength of 3.704 MPa, corresponding to a 113% improvement over the 20/40 baseline. Mobile-device heat-dissipation and load-bearing demonstrations further confirmed its practical applicability. These results establish bondline densification as a design principle for dense, processable, conformable, and adhesive thermal interfaces.