Automated FEM model generation for the simulation of microheaters

Abstract

The rapid growth of the semiconductor industry has seen an increase in demand for reliability tests which are essential for ensuring that the devices reach their expected lifetimes, and the overall well functioning of the devices is achieved. Experimental tests give plenty of insight into topics of interest, but nowadays with access to high performance computing facilities, it is possible to perform multiphysics simulations which are more cost efficient compared to their physical testing counterparts. However, there is a significant effort placed into developing and then refining the simulation models, so that they mimic reality up to a desired accuracy. In multiphysics simulations there are typically three well-known routines involved: preprocessing, solving, and post-processing. The aim of this work is to provide an automated workflow which goes through the pre-processing routine which, in most engineering/physics applications, is the most tedious and time consuming component. The pre-processing consists of the following tasks: Defining the geometry, meshing, defining the material properties, and specifying the initial and the boundary conditions. In most cases the latter two are derived from experiments or literature values, but the first two tasks are always problem-specific and can be of considerable difficulty. There are general good practices that one should follow, but in real engineering problems it is up to the simulation engineer to make compromises between modelling assumptions, meshing and the desired simulation accuracy compared to the experimental data. In this work, the finite element method is applied to analyse the temperature response in non-commercial microelectronic devices, which typically require many repetitive manual tasks to be executed during pre-processing. The design of these devices is first drawn in the graphic design system (GDS) format, which is an industry standard for microelectronic devices. In the scope of this thesis, a Python framework is developed to automate all the pre-processing parts of these designs and minor automation is also introduced to solving and the post-processing. For demonstration purposes the developed methods are applied to generate and validate models of test chips against experimental data. The model generation is fully automated, and is up to 50 times faster than the previous manual routine. Adjustments in the geometry pre-processing enhance the solving runtime, by making it up to 35 times faster than the previous solving methodology. The Python framework is scaled to work on a family of different device models and architectures which share certain similarities.