Study of Response Surface Methodology in Thermal Optimization Design of Multichip Modules

Abstract

A 3-D model of multichip module (MCM) is built with ANSYS and the temperature field distribution is studied. A regression equation describing the relationship of structure parameters and material properties with the maximum chip junction temperature of MCM is made, which integrates the response surface methodology and ANSYS. Quantitative analysis of the effect of four design parameters on the maximum chip junction temperature of MCM is studied. The four design parameters are the thickness of the substrate, thermal conductivity of the substrate, thermal conductivity of the thermal grease, and convection heat transfer coefficient, respectively. The accuracy and validity of the regression equation are validated by simulation with ANSYS. In addition, the maximum error between the calculation value of the regression equation and the simulation value with ANSYS is 0.541°C. With the regression equation, the thermal optimization design results of the four parameters are K t g = 5 W/m°C, δ = 2.5 mm, K s = 290 W/m°C, and h = 55 W/m 2°C , which lead to the maximum chip junction temperature T j max = 89.172°C as the minimum value. Index Terms-ANSYS, multichip module (MCM), optimization design, response surface methodology (RSM).

Polynomial-based surrogate modeling of microwave structures in frequency domain exploiting the multinomial theorem

“…The size of the training region X s is defined by a vector τ ∈ ℜ n containing the maximum relative deviation for each design variable with respect to x (0) . To train the surrogate model, we use L learning base points within X s , denoted as x (1) , x (2) , …, x (L) . To measure the generalization error of the surrogate model, we use T testing base points.…”

“…For the third-order surrogate model, we want to match the fine model response and the corresponding surrogate model response at the j-th learning base point and k-th frequency, R sk (0) (x (j) ) + w k (1)T q (1) (Δx (j) ) + w k (2)T q (2) (Δx (j) ) + w k (3)T q (3) (Δx (j) ) = R fk (x (j) ) (9) Applying (9) for j = 1, …, L, the surrogate model weighting factors can be calculated by solving for W (3) the following system of linear equations…”

“…RSM represents the relationship between the input parameters and the output responses by a second-order polynomial. The minimum number of function evaluations is equal to the number of unknowns in the polynomial equation [2,3]. ANNs can approximate arbitrary nonlinear input-output deterministic relationships by using the proper amount of training and testing data as well as a suitable complexity; their training is typically based on some optimization algorithm [4].…”

Applying Model Order Reduction to the Reliability Prediction of Power Electronic Module Wirebond Structure

“…In the surrogate modelling (SM) approach, the response surface technique is combined with design of experiment (DoE) sampling to generate the surrogate model representing the physical process being modelled. SM can be categorised into two distinct groups namely interpolation SM such as the Kriging model [1] and regression SM such as the quadratic polynomial model [2].…”

Thermal performance evaluation of cascode Paralleled-GaN-HEMTs packaging for high power switching applications

“…Several new techniques have yielded promising results toward establishing peak temperature for GaN devices in combination with detailed modeling and infrared (IR) imaging (Green et al, 2008;Raj & Bindra, August 2013;Salem, Ibitayo, & Geil, 2007;Sommet, Mouginot, Quere, Ouarch, & Camiade, 2012;J. Zhang & Zhang, 2013).…”