We show a way to gain a precise understanding how a MEMS hotplate as a mid-infrared emitter works. We used experimental methods to measure its main thermo-electric properties but also FEM simulation for gaining insight into not-so-easily accessible physical quantities. Those are for instance the current density field on a microscopic level or the amount of heat dissipation by convection. Since at the start of modelling the electrical and thermal conductivities of some materials were unknown at elevated temperatures, temperature characteristics were adjusted to fit the measured data. We discuss different ways to find matching combinations of characteristics, such as a) direct search in the calculated parameter set, b) a self-normalizing artificial neural network trained for regression, and c) physical intuition. When the influences of the packaging and of the surrounding air were properly included into the model, it reproduced the measured data quite well in vacuum and normal ambient conditions. We conclude, this modelling technique leads to models which are quantitatively verified in a large part of the applicable parameter space and cover all relevant physical effects.
|