Abstract The continuing increase in CPU and GPU performances can be attributed mostly to the rise in frequency, scaling of chip area (bigger dies), advancements in thermal management, and improvements in thermal design power, among other factors. Over the past two decades, the size of a typical GPU die has increased from 100 mm2 to ~800 mm2 in the year 2020. Silicon-based single-phase embedded microchannels with a large area of 30 mm×30 mm, featuring three-dimensional (3D) manifold (MF) μ-coolers, can potentially remove heat and minimize pressure drop. However, previous computational fluid dynamics (CFD) simulation findings indicated considerable temperature nonuniformity resulting in increased thermal resistance and maximum temperature. The primary cause of temperature non-uniformity is the abrupt flow acceleration at the entrance and sudden deceleration at the end section of the 3D-manifold inlet channels. This leads to a significant temperature rise in the middle region of the microcooler. In this study, we introduce innovative microcooler designs and conduct extensive computational fluid dynamics (CFD) simulations to achieve low thermal resistance, low-pressure drop, and crucially, uniform temperature distribution across the entire surface area of the microprocessor. To deal with this issue, our initial approach involved integrating converging inlet and diverging outlet channels into the 3D manifold. Although this method effectively dealt with non-uniformity throughout the u-cooler's area, it still resulted in a large pressure drop. Consequently, we implemented a narrow opening at the end of the inlet channels in the 3D manifold. This allows a portion of the coolant (50-80%) to bypass the microchannels in the cold plate to the exit plenum. As a result, the pressure was reduced by ~66% compared to the conventional 3D manifold microchannel cooler.
Kong et al. (Mon,) studied this question.