A data-driven framework to predict the morphology of interfacial Cu6Sn5 IMC in SAC/Cu system during laser soldering

Abstract

A data-driven approach combining together the experimental laser soldering, finite element analysis and machine learning, has been utilized to predict the morphology of interfacial intermetallic compound (IMC) in Sn-xAg-yCu/Cu (SAC/Cu) system. Six types of SAC solders with varying weight proportion of Ag and Cu, have been processed with fiber laser at different magnitudes of power (30-50 W) and scan speed (10 -240 mm/min), and the resultant IMC morphologies characterized through scanning electron microscope are categorized as prismatic and scalloped ones. For the different alloy composition and laser parameters, finite element method (FEM) is employed to compute the transient distribution of temperature at the interface of solder and substrates. The FEM-generated datasets are supplied to a neural network that predicts the IMC morphology through the quantified values of temperature dependent Jackson parameter (αJ). The numerical value of αJ predicted from neural network is validated with experimental IMC morphologies. The critical scan speed for the morphology transition between prismatic and scalloped IMC is estimated for each solder composition at a given power. Sn-0.7Cu having the largest critical scan speed at 30 W and Sn-3.5Ag alloy having the largest critical scan speed at input power 2 values of 40 W and 50 W, thus possessing the greatest likelihood of forming prismatic interfacial IMC during laser soldering, can be inferred as most suitable SAC solders in applications exposed to shear loads.

Combining multi-phase field simulation with neural network analysis to unravel thermomigration accelerated growth behavior of Cu6Sn5 IMC at cold side Cu–Sn interface

“…13, 15 and 18 , supported by material properties, initial conditions and boundary conditions, are solved using finite element method. Preconditioned Jacobian Free Newton Krylov (PJFNK) method is implemented in Multiphysics Object Oriented Simulation Environment (MOOSE) Framework [39,40,41,42,43,44,45] to perform the finite element analysis. For the binary Cu-Sn system involving multiple phases (3 phases), Kim-Kim-Suzuki ( KKS ) model [46] has been utilized and in accordance to this model , (a) the criteria of equality of chemical potential between two adjacent phases, and (b) relationship between global and phase compositions of Sn ( = ∑ ℎ ), have been established.…”

“…The growth rate constant (k ) of Cu 6 Sn 5 IMC at the cold side interface is set as the output feature of the NN. These input and output features are obtained from 1D phase field simulation performed in accordance to the methodology outlined in Section 3, and thus the dataset can be defined as the FEMgenerated datasets [45,49]. The total length of the 1D computational domain is 4 m. 60 phase field simulations were performed for different sets of values for the 7 input features -heats of transport and ΔT.…”

“…The system shown in Fig. 4 is a regression based feed forward neural network, and such network is useful when dealing with small datasets [45]. In all the n hidden layers, Rectified Linear Unit (ReLU) activation function is utilized, and linear activation function is used for the output layer.…”

Fabrication of cerium myristate coating for a mechanochemically robust modifier-free superwettability system to enhance the corrosion resistance on 316L steel by one-step electrodeposition

“…11(a) and for values of βs varied between 0 and -1.0, Eq. 1 is solved using finite element method (FEM) in Multiphysics Object Oriented Simulation Environment (MOOSE) Framework software [37,62].…”

Integration of machine learning with phase field method to model the electromigration induced Cu6Sn5 IMC growth at anode side Cu/Sn interface

“…Moreover, multiphysical models (finite element analysis), when coupled with experimental observation, can be utilized to construct the dataset for machine learning model. It has been demonstrated in [22] that finite element method generated (FEM-generated ) data for given laser soldering experiments can be used to train the neural network , for prediction of the morphology of interfacial intermetallic compound (IMC).…”