Fig. 3: Discovery of Ge4Sb6Te7 (GST467).
From: On-the-fly closed-loop materials discovery via Bayesian active learning
a Optimization of the Ge–Sb–Te system, with the objective of identifying the material with the largest ΔEg, bandgap difference between amorphous and crystalline states. Performance is shown for: the CAMEO live run (black) with GST467 discovered on iteration 19 (black star); for the mean and 95% confidence interval of 100 CAMEO runs computed post data collection (blue), 100 runs of Gaussian process-upper confidence bounds (GP-UBC, magenta), and random sampling (red). For the 100 runs, CAMEO gets within 1% of the optimal in the first 20 runs 31% of the time compared to GP-UBC’s 10%. b Left: structural phase map for the crystalline Ge–Sb–Te composition spread. Complete phase map constructed after all the diffraction measurements (beyond the live CAMEO run described in a)) is shown. Structural phase regions are color-coded as single-phase FCC–Ge–Sb–Te (GST) structure region (red, GST), single-phase Sb–Te region (blue), and region where GST and Sb–Te phases co-exist (orange). The materials measured during the live CAMEO run are indicated in black. Right: Complete mapping of ΔEg following analysis of entire ellipsometry data (beyond the live CAMEO run). The discovered GST467 and GST225 are indicated in both maps.
