Artificial Intelligence in Materials Science

The  amount  of  data  produced  by  detectors  has  increased  explosively.  Automated  tools  to  process  and analyze the data are necessary. Artificial intelligence (AI), in particular machine learning (ML), has  been  exploited  by  an  increasing  number  of  disciplines  to  automate  complex  problem-solving  tasks. Progress in Ml has led to decision rules that can in some cases be automatically derived by specific algorithms. One of the successful applications of ML in materials science lies in developing fully automated algorithms for analyzing high-throughput experiments.

Field  ion  microscopy  (FIM)  is  a  high  electric field technique, which enables the  imaging  of  surfaces  with  atomic  resolution. When exposing the tip not only  to  the  minimum  field  strength required for ionizing the imaging gas but also for evaporating the tip atoms themselves continually, the method is rendered  depth  sensitive  (3D-FIM). This means that the specimen can be investigated   tomographically   along   the tip’s longitudinal axis. 3D-FIM is capable  of  producing  large  and  accurate tomographic datasets containing  information  on  sequential  atomic  positions,  but  these  large  datasets  lead  to  a  new  tremendous  challenge  of how to manage the data.

Presently, there is a lack of efficient data  handling  algorithms  to  extract pertinent information from these data-sets in an (a) automated; (b) fast; (c) user-independent; and (d) error quantified  manner.  For  instance,  characterization of a volume of 0.001 μm3 (a typical 3D-FIM sample size) produces in the range of 2 × 105 images. Hence, there is a great need for efficient al-gorithms and data mining routines. To this end, we recently proposed a new method  to  extract  atomic  positions  from 3D-FIM datasets, using various image processing and machine learning algorithms [1, 2].

Fig. 1: Machine learning (Isomap) on the 3D-FIM dataset. Each point represents one picture in the reduced dimensional manifold, and its colour indicates the number of atoms in the first terrace of the corresponding picture. The numbers on the outer circle represent the averages over the number of atoms in the first terrace of pictures.

Applying machine learning to the 3D-FIM  images  (>  21000  dimensions)  and  projecting  the  data  to  a  low  dimensional   subspace   helped   us   to   understand the latent structure in the data.  The  exemplary  result  in  Fig.  1  reveals  a  cyclic  behaviour  in  the  dataset,  which  is  a  consequence  of  the field evaporation behaviour. Field evaporation typically occurs layer-by-layer and proceeds from atoms sitting at  the  ledge  of  a  terrace  towards  the  centre. When a terrace field evaporates  the  atoms  sitting  on  the  ledge  disappear,  decreasing  the  number  of  atoms on the terrace as the evaporation  proceeds.  By  evaporating  atoms  from the surface of the sample during FIM, the first terrace area decreases until  it  evaporates  completely.  The  number of cycles is a measure of the number of atomic terraces evaporated during the 3D-FIM.

With  the  use  of  such  advanced  algorithms for data extraction, we hope not only  to  improve  the  accuracy  of  the  data extraction from 3D-FIM but also to  identify  and  characterize  various material  defects.  Currently,  the  physics of image formation in FIM are still not  completely  understood,  but  it  is  our firm belief that ML can be a powerful tool in this direction.

References:

1.
S. Katnagallu, A. Nematollahi, M. Dagan,  M.  Moody,  B.  Grabowski,  B.  Gault, D. Raabe, J. Neugebauer
High fidelity reconstruction of experimental field  ion  microscopy  data  by  atomic relaxation simulations
Microsc Microanal, 23, 642 (2017)
2.
S. Katnagallu, B. Gault, B. Grabowski, J. Neugebauer, D. Raabe, A. Nematollahi
Advanced data min-ing in field ion microscopy
arX-iv:1712.10245

Authors: Gh. A. Nematollahi, B. Grabowski

Go to Editor View