/Parallel evolutionary swarm optimization for the analysis of Rutherford backscattering

Parallel evolutionary swarm optimization for the analysis of Rutherford backscattering

Leuven | More than two weeks ago

Gain real-life in-depth experience in data science, computational physics modelling, computational optimization, and parallel processing

The Materials and Components Analysis expertise center of imec develops advanced characterization tools to investigate and monitor new sub-nanometer devices. Ion beam analysis, employing a 2 million Volt tandem particle accelerator, is one of the approaches. The experimental results (spectra) are compared to physical models to quantify the properties of the fabricated devices.

Traditionally, the optimization of the model towards the experimental result is achieved by least squares fitting (fitting) using grid-search or Levenberg-Marquardt algorithm. In this scheme, due to the computational demand, only a few parameters can be fitted (at once), and the analyst needs to supervise and steer the analysis constantly – a lengthy and labor-intensive activity.

Various advanced optimization algorithms are being explored to analyze Rutherford backscattering spectra. Noteworthy are published implementations of simulated annealing and of artificial neural networks. However, it proves difficult to construct an artificial neural network that accounts for all the possible experimental conditions and sample characteristics. Besides, the final stage of the optimization is reported to be computationally inefficient and is complemented with a local search routine.

Exploratory studies indicated that, in the case of Rutherford backscattering spectrometry, the differential evolution algorithm (Storn and Price, 1995) has a high potential to aid the analyst in finding the sample parameters. However, our exploratory studies were done with different codes from various contributing scientists and using various programming languages. Our goal is to develop a new and coherent software program for the analysis of the experimental results. The parallelization will be considered from the start of the software design. Fortunately, as it is a population-based optimization algorithm, the differential evolution algorithm is known to lend itself well to parallelization.

The main tasks will be:

  • To implement the simulator for Rutherford backscattering spectrometry in C++ or python, based on existing code that was developed in Java and Fortran.
  • To investigate the performance of the new simulator    
  • To prepare the new program to allow its execution on multiple cores.  

If time allows, the student may also compare the performance of Differential evolution with other nature-inspired optimization algorithms, like for example Moth-Flame optimization or Ant-Lion optimization.

This subject is an opportunity for those who wish to get in-depth experience in data science, computational optimization, and parallel processing.

Type of project: Combination of internship and thesis

Duration: 6 months+

Required degree: Master of Science, Master of Engineering Science

Required background: Computer Science, Physics

Supervising scientist(s): For further information or for application, please contact: Johan Meersschaut (Johan.Meersschaut@imec.be)

Imec allowance will be provided for students studying at a non-Belgian university.

Who we are
Accept marketing-cookies to view this content.
Cookie settings
imec's cleanroom
Accept marketing-cookies to view this content.
Cookie settings

Send this job to your email