Optimal design of thermo-compression bonder via ANN-based surrogate modeling under uncertainty

The miniaturisation of electronic devices demands advanced microelectronics packaging systems dominated by interconnection reliability. Thermo-compression bonding (TCB) is a promising approach to improve bonding quality by applying heat and pressure onto the chip. However, TCB has prompted severe mechanical warpage and electrical shortage when the diminutive interconnection is randomly shaped, which yields non-uniform stress and thermal distribution in bonding. Moreover, the mechanism is interpreted by excessive multi-physics simulations or experiments, which impedes computational cost-efficiency. Therefore, this research develops an optimised TCB design via a machine learning-based surrogate model to predict complex TCB behaviour, by certifying its best performance. First, a framework is proposed to extract and propagate the multivariate geometric uncertainties of the interconnection. The framework realistically represents the existing packaging system and predicts TCB behaviour through ANN, which enriches the accuracy and efficiency of the prediction model. Second, an optimal TCB is designed to minimise the warpage and electrical short. With the surrogate model, the optimisation attempts to find the best location and magnitude of a structural pressure, which derives uniform stress concentration or restricted interconnection deformation. The efficacy of the proposed framework is demonstrated with a practical example of an advanced packaging system utilised in actual commercial products.