Abstract
The thermal management of central processing units (CPUs) has been critical as a result of the ever-increasing power density and heat generation of modern electronic devices. Physics-Informed Neural Networks (PINNs) are utilized in current work to evaluate the thermal performance of CPUs equipped with a thermoelectric cooler (TEC) module. Unlike the traditional numerical methods, the proposed method embeds the governing physical laws into the learning framework to enable accurate and stable prediction of the thermoelectric cooling behavior. Three bismuth-antimony-telluride based semiconductor materials, ( Bi 0.5 Sb 0.5 2 Te 3 , Bi 0.24 Sb 0.76 2 Te 3 , and Bi 0.24 Sb 0.76 2 Te 3 ) are considered to be thermoelectric leg materials. Simulations are performed for five different Thermal Design Power (TDP) levels corresponding to different CPU categories, so that the TEC performance can be evaluated under realistic heat generation conditions. The effect of electrical current on the cooling behavior and the efficiency of the system are investigated in a systematic way. A 1-D steady-state mathematical model is developed where the thermoelectric material properties are explicitly accounted for as functions of temperature. This overcomes a major limitation of many previous studies that assume constant properties. First, the governing equations are rewritten in non-dimensional form, for an efficient PINN implementation, and then solved using modeling in Python. Comparison with the published reference results demonstrates close consistency, thereby supporting the predictive capability of the developed model. The main criterion used to compare different operating conditions and material choices is the coefficient of performance (COP). The results demonstrate that ( Bi 0.24 Sb 0.76 2 Te 3 )) performs better than the other materials tested under the conditions investigated. Further, it is found that as the TDP increases, the COP of the system raises significantly by 2.035 times at an electrical current of 1A. Furthermore, lowering the applied current markedly improves the COP, which increases by up to approximately 16.5 times under the prescribed thermal-load conditions. The maximum COP achieved in this work is about 3.185 under the maximum TDP level with ( Bi 0.24 Sb 0.76 2 Te 3 )) and the electric current of 1 A. Overall, this work highlights the strong potential of PINNs as an advanced and reliable tool for the modeling of thermoelectric cooling systems and provides new insights related to the material selection and operating conditions for efficient CPU thermal management.