Investigate the composition effect of Cu-Nb-B-Fe soft magnetic metallic glasses using deep learning method

Fe-based metallic glasses have attracted numerous research attention because of their outstanding soft magnetic properties. The major challenge is their large enthalpy to crystallize. To succeed in vitrification, metalloid elements and small amount of transition metals are added but at a cost of reduction of magnetic induction. There have been many experiment reports on individual compositions and a few deep learning studies. For atomic simulations, the origin of difficulty is that micro-structures are dependent on quenching speed. In this paper, we introduced a universal method to predict if a composition can succeed in amorphous glass without sample preparation. This method was based on the concept of amorphousness degree derived from bond orientational analysis. Compositions were lined up from high amorphous degree to low in simulation. When the quenching slowed down, the threshold of glass formation moved toward the higher amorphous end of the sequence. The conclusions were drawn from the results of molecular dynamics simulations with inter-atomic potentials by deep learning. We found that both decreasing Fe and increasing Cu/Nb contents could help succeed in vitrification. Only a small increase of Cu/Nb achieved the same effect as many more decreases of Fe atoms. When either Cu or Nb existed solely, the vitrification had a strong tendency to form amorphous and noncrystalline phase, but their coexistence greatly reduced this tendency. The effect might be caused by the strong interaction of Cu/Nb. At the end, we extracted a few glass formation temperatures and confirmed the validity of the calculations.