Localized hotspot mitigation using a hybrid microchannel heat sink with alternating-orientation arched-bridge pin-fins

Abstract

Localized hotspots induced by non-uniform heat generation in AI accelerators and data-center processors necessitate microchannel heat sinks that can enhance local heat removal and improve substrate temperature uniformity. In this context, hybrid microchannel heat sink with alternating-orientation arched-bridge pin-fins, denoted as AO-ABPF, for localized hotspot mitigation is proposed in this paper. Three-dimensional steady-state conjugate heat transfer simulations are conducted to compare the proposed design with a straight parallel microchannel heat sink (SPMC) and a hybrid circular pin-fin heat sink (HCP). The effects of the Reynolds number ( Re ), pin-fin gap width ( W fingap ), pin-fin material, and local hotspot heat flux ( q hs ′ ′ ) are investigated. The thermal–hydraulic performance is evaluated in terms of the maximum base temperature ( T max , base ), thermal resistance ( R th ), temperature non-uniformity ( δ T ), and total pumping power ( P total ). When silicon is used for the central pin-fins, the AO-ABPF(Si) shows improved hotspot mitigation compared with the SPMC and HCP over Re = 200–1000. At Re = 200, the AO-ABPF(Si) reduces T max , base by 22.7 K and 8.3 K relative to SPMC and HCP, respectively. At Re = 400, it reduces R th by 53.1% and 28.0%, respectively. Across Re = 200–1000, replacing silicon with diamond reduces T max , base by 1.4–2.3 K and R th by 6.8%–12.1%, while the difference in P total between the AO-ABPF(Si) and AO-ABPF(Di) remains within approximately 1.2%. The AO-ABPF(Di) also exhibits lower sensitivity to W fingap . As W fingap increases from 0.100 mm to 0.250 mm, the AO-ABPF(Di) exhibits nearly unchanged T max , base and R th , remaining at approximately 320 K and 0.34 K/W, respectively. By contrast, the corresponding T max , base and R th of the AO-ABPF(Si) increase by 5.9 K and 26.5%, respectively. At q hs ′ ′ = 900 W/cm 2 , the AO-ABPF(Di) further reduces T max , base , R th , δ T , hotspot , and δ T , base by 18.9 K, 32.4%, 36.8%, and 34.5%, respectively, relative to the AO-ABPF(Si).

The authors' abstract, as published at the source. Applied Thermal Engineering, 2026 · DOI ↗