Brand, S.S.BrandKoegel, M.M.KoegelAltmann, F.F.AltmannBach, Hoang LinhHoang LinhBach2022-05-062022-05-062021https://publica.fraunhofer.de/handle/publica/41745410.1109/ECTC32696.2021.00147In power electronics reliable die attach technologies with high electrical and high thermal performance as provided by Ag sintering are of major relevance. The current study continues a previously proposed approach for non-destructively assessing the relative porosity in Ag-sinter joints using ultrasonic signals obtained by acoustic microscopy. In this paper the approach is extended by sophisticated signal analysis employing methods of deep learning. For quantitative porosity estimation a 1D-convolutional neural network in combination with aregressor was trained and evaluated on two separate sample sets. In both cases a high prediction accuracy corresponding to a root-mean-square error (RMSE) of 0.5% was achieved.enacoustic microscopyAg-jointsdeep learningdie-attach for power electronicsmachine learningsignal analysissilver sintering670620530conference paper