Prediction of Fan-out Panel Level Warpage using Neural Network Model with Edge Detection Enhancement

In recent years' fan-out panel level package (FOPLP) is playing a most important role in electronics packaging industry due to lower cost, thinner profile, better electrical and thermal performance. However, thermally induced warpage is the critical issue for each and every researcher's problem. Using larger wafer and panel size, and mismatch of coefficient of thermal expansion (CTE) among the constituent material, the thermally induced warpage is occurred which needs to be controlled effectively for fan-out package industry. In this work, finite element method (FEM) is used to build the FOPLP by considering equivalent CTE of molding material and then thermally induced warpage issue in the molding process is investigated. To reduce the experimental time and cost, FEM is the best method to simulate and analysis of warpage in FOPLP. Simulation can build a set of warpage result database of different geometric structures of panel level package (PLP). In addition to that, this work has been introduced Artificial Intelligence (AI) approach and the main purpose of this AI technique is to build a best model for warpage analysis by considering the design parameters like chip size, chip thickness, gap between the chips, thin film thick and so on. The best part of AI model is that, if numbers of different geometric parameters are used in this model then AI can predict the better warpage value without the help of FEM simulation. The implementation of machine learning concept which is an AI technique is to train the model from the different geometric model related to warpage datasets which are generated by FEM simulation. Convolution Neural Network (CNN) algorithm is applied to learn the relationship between geometry and amount of warpage occurred from the generated database. To enhance the training process, it has also considered Edge Detection technique. Eventually, this AI approach is successfully used to remove data points with so many similar warpage values and it could reduce training data as well as training time. Moreover, this paper successfully digitized warpage model obtained by the FEM and also successfully used the Artificial Neural Network (ANN) to train the database to learn and estimate the warpage for FOPLP.