Fraunhofer Institute for Electron BeamThe DigiMatUS project advances the digitalization of coating processes. A basis for machine learning is established by correlating process parameters and thin film properties for the synthesis of piezoelectric AlN/AlScN thin films for ultrasound microscopy. This enables a more effective development of new coatings or tailored coating properties as well as a more efficient transfer of technology between different machine platforms.
Piezoelectric thin films play a significant role in medical technology, microelectronics, and sensors, e. g., as active layers in lenses for ultrasound microscopes. These microscopes enable the examination of small structures, such as semiconductor components or biological tissue. To achieve the superior imaging quality, highly oriented and homogeneous piezoelectrically active layers are mandatory. The requirements for the quality of theses layers increase as the dimensions of the features to be examined become smaller. The properties of the thin films and ultrasound transducers result from the complex interplay of numerous process parameters and factors.
In the DigiMatUS project ("Digitalization of materials research on thin-film materials using the example of high-resolution piezoelectric ultrasonic sensors"), the efficient development of coating technologies is being advanced through a digital representation of the deposition processes, the development of a process and film ontology (a formal representation of knowledge, entities, their properties, and relationships), and machine learning methods. For this purpose, the synthesis of piezoelectric aluminum nitride (AlN) and aluminum scandium nitride (AlScN) thin films is carried out at Fraunhofer FEP across a broad range of parameters. This includes, for example, the variation of sputter power, process pressure, or gas flows, as well as the consideration of other influencing factors, like the underlying coatings of the electrode structures. Through enhanced optical in-situ process monitoring and layer diagnostics, as well as extensive characterization of the layers, the project aims to establish the process-parameter / thin-film-property relationships in more detail.
The combination of an ontological description of the processes and the incorporation of existing expert knowledge forms the basis for meaningful predictive models for machine learning utilizing a comparatively small data set.
The development and training of the predictive models is carried out by the project partners at Otto-von-Guericke University Magdeburg. As a result, an initial predictive model has already been created that predicts plasma spectra based on process parameters. Current and future work at Fraunhofer FEP will focus on further advancing the thin film ontology, continuing experiments for additional training, and validating the developed models.
In addition to developments for ultrasound sensors, the results of the project open up new opportunities for other applications based on thin film technologies. The digital representation of material and process data through the thin film ontology and AI models will be adapted for other relevant coating processes and materials. This will significantly accelerate research and development of new materials in the future.
