Seung-jin Seo

Methodology on thermal prediction of high-power memory modules considering air velocity, chip size and power consumption

High-speed and high-density RDIMMs (Registered Dual Inline Memory Modules) which are employed in server platforms consume significant amount of power, operating at close to 100 °C. Thus it is very important to predict operating temperature of high-power memory modules at the early stage of development procedure to determine which type of cooling solution to employ. In this paper, we propose an approach for thermal prediction of high-power memory modules in server platforms. Through comprehensive thermal simulations using commercial CFD tool, FloTHERM, matrices of thermal data were obtained for various memory thermal conditions. Using the database, we built characteristic curves via curve-fitting. Characteristic curves are such that we can predict operating temperature of memory modules for various chip size and power dissipation under various memory thermal conditions. Predicted thermal results are in good agreement with measurement data with deviation of less than 3 °C (~5%).