A Learning-Based Thermal Simulation Framework for Emerging Two-Phase Cooling Technologies

Abstract

Heat transfer coefficient is a parameter that determines the rate of heat transfer per unit temperature difference.2 We will open source our modeling tool along with paper publication.

Neural network-based cooling design for high-performance processors

“…The existing cooling optimization methods (CMA-ES and MSA) have two main issues: (i) need to run a great number of thermal simulations which results in large simulation time and (ii) there is no guarantee that the selected cooling method and its cooling parameters are optimal. The accuracy of the optimization result selected by CMA-ES and MSA is determined by the sample size and the number of iterations ( Yuan et al., 2019b , 2020 ). Using the DL model, specifically, the CNN regression model, to predict the optimal cooling solution and its cooling parameters could be the solution to these two issues.…”

“…In addition to cooling design choice possibilities, the optimization flow needs to also account for the chip design and power profile changes. In this case, using a simple grid search to find the optimal cooling design for a small-sized chip floorplan and its typical power profile could take up to days ( Yuan et al., 2020 ). Previous work has investigated using machine learning or black-box optimization methods to optimize or model the system with emerging cooling technologies such as liquid cooling via microchannels and TECs ( Beneventi et al., 2012 ; Fan et al., 2018 ; Blackburn et al., 2020 ; Zhou et al., 2020 ; Tang et al., 2020 ).…”

“…A hybrid wick evaporator of nanoporous membrane and microchannels, as shown in Figure 1 D, has the potential to simultaneously improve both HTC and dry-out heat flux. Similar to two-phase VCs with micropillar wick evaporators, two-phase VCs with hybrid wick evaporators only replace the micropillar evaporator wicking structure with hybrid wicking structure of nanoporous membrane and microchannels ( Lu et al., 2019 ; Yuan et al., 2020 ). The hybrid wick evaporator geometry list in Table 1 determines the HTC and dry-out heat flux.…”

“…Researchers have developed various CTMs to model the cooling behaviors of emerging cooling solutions such as liquid cooling via microchannels ( Sridhar et al., 2013 ; Kaplan et al., 2017 ), TECs ( Kaplan et al., 2017 ; Yazawa et al., 2012 ), two-phase VCs with micropillar wick evaporators ( Yuan et al., 2019a ; 2019b ), two-phase VCs with hybrid wick evaporators ( Yuan et al., 2020 ), and hybrid cooling (of liquid cooling via microchannels and TECs) ( Kaplan et al., 2017 ; Yazawa et al., 2012 ).…”

“…The on-chip maximum temperature limit (temperature constraint) is set to 65 . The detailed experimental setup can be found in the previous work ( Yuan et al., 2019b , 2020 ). To select the optimal cooling solutions and cooling parameters, we use the covariance matrix adaptation evolution strategy (CMA-ES) and multi-start simulated annealing (MSA) to select the optimal cooling parameters for the aforementioned emerging cooling technologies ( Yuan et al., 2019b , 2020 ).…”