Deep Learning for real-time surveillance of single crystal growth

All discovery in material science and solid-state physics begins with synthesis. Notably, the growth of novel materials and or more high-quality materials was at the start of unearthing unconventional superconductors, candidates for quantum spin liquids and spin ice materials, topological insulators and semimetals and skyrmions to name just a few. This demonstrate that discovery of new materials with functional properties to fill the needs of modern society will clearly benefit from increasing the throughput of high-quality material synthesis methods.

However, the growth of single crystals of novel materials remains a complex process, which requires deep domain knowledge and constant monitoring and parameter optimisation. As the crystal growth process typically lasts between several hours to several days, the tasks of monitoring and optimising become increasingly cumbersome and require more hours of human supervision. 

The Solid State Chemistry Group is joining forces with the Laboratory for Neutron and Muon Instrumentation, in order to develop an automated crystal growth monitor system by harnessing the power of machine learning. This monitor system will have a dual functionality: a) detect or foresee failures in the growth process and alert the operator in time, and b) suggest parameter adjustments in order to avert such failures. 

At the starting phase of the project, we will use methods of supervised and/or reinforcement Machine Learning (in particular Deep Neural Networks), in order to build a model which can process in real time a video feed of the progressing crystal growth, and alert the human operator of ongoing or even imminent failures. At a later phase of the project, our ML model will be trained to perform parameter optimisation, in order to provide guidance to the human operator on adjusting the experimental parameters.


  • Participation in crystal growth experiments
  • Handling and preprocessing of video data
  • Some code development of DL models in Keras/Tensorflow or Keras/Theano or Pytorch
  • Training and optimising of DL models

Desired skills:

  • Knowledge and some experience on Neural Network architecture
  • Practical knowledge on one of the Deep Learning frameworks mentioned above

Required skills:

  • Excellent programing skills (Python preferred)

NUM laboratories: Laboratory for Neutron and Muon Instrumentation LIN and Laboratory for Multiscale Materials Experiments LMX

Contact person: Dr. Emmanouela Rantsiou , email, phone: +41 56 310 46 31

Back to overview