An innovative training programme has been implemented to equip students with expertise in materials physics, numerical simulation, and machine learning – an essential but largely unprecedented combination for the microelectronics industry. The programme involves leading European industrial and academic experts in the field of ME, aiming to enhance the reliability and durability of microelectronics as industry demands grow. The key highlights of the training programme so far include individual and network-wide training sessions, featuring close supervision by at least two supervisors with academic and industrial expertise. All Doctoral Candidates (DCs) are enrolled in a local PhD programme. The training activities also include two on-site training weeks, a machine learning workshop, and quarterly webinars with expert talks.
The scientific achievements of the ongoing programme, MIRELAI, can be clustered into three categories which also represent the three scientific work packages (WPs).
Within WP4 – Physics of degradation – the individual DCs developed a thorough understanding of the problem-specific relevant failure modes and degradation behaviour. The degradation behaviour and potential failure of microelectronic systems is studied by the DCs on different levels. While some theses focus on defined features of Printed Circuit Board Assemblies (PCBAs), namely microvias, solder joints and Plated Through Holes (PTH), selected theses focus on system-level effects. On the feature level, a result example is the structured overview of potential failure modes for microvias based on literature data, as well as on in-house experience from the industry partner. Another example is a new analysis option for subsurface defects based on Lock-In Thermography (LIT). On the system level for example the effect of material, geometry and conducting path design variations on the warpage behaviour has been studied and summarised. Also, a new accelerated physics of degradation testing method on the system level is developed. The power cycle test, acting as a reference has been installed, and the resulting loading conditions have been analysed in detail. They will now be used to define and establish an accelerated mechanical substitute test in the ongoing project.
Within WP5 – Multi-scale modelling – the DCs generated virtual representations of their problem statements. Digital twins have been set up on local and global levels. At a local level, for example, a finite element model was developed at the microstructure level for the detailed investigation of the degradation of solder joints. It models the grain-scale behaviour of solder joints which can be used to detect how effects such as crystal orientation or Euler angles influence the mechanical response of solder joints in component assemblies. On a global level again a finite element-based thermal analysis of a complete PCBA using Reduced Order Modelling (ROM) was introduced. It allowed us to effectively use virtual sensing to correct the model parameter uncertainties based on a Kalman filter approach. The effectiveness of the approach was showcased in a study of different die attachment solutions. The model results of the individual problem statements are used to gain a further understanding of the failure modes and to generate training data considering uncertainties and parameter variations for potential metamodels.
WP6 – AI-based reliability – built upon the results of WP4 and WP5. On the one hand, specific identified failure modes and their physics are analysed testing the potential of Physics-Informed Neural Networks (PINN). One example is the non-linear buckling of a needle which has been examined using PINN. On the other hand, generated training data is used to generate metamodels to screen a wide parameter space. For example, an artificial neural network could be trained to predict the warpage of substrates for a wide range of material and geometry parameter variations. In a simplified application, the approach has already demonstrated its ability to optimise substrate design with the aim of minimising warpage.
