Fig. 2: Film inflation method (FIM).

a Sample preparation: (1) An Al₂O₃ etch-mask is applied onto a silicon substrate and defines 241 square chips, each with a 2 mm circular area of exposed silicon. (2) A compositionally graded Mg–Cu–Y film library is co-sputtered onto the opposite substrate side. Elemental concentrations vary by 15–25 at.%, neighboring chips differ by ~1 at.%. (3) High-throughput film characterization for each chip: EDX measures chemical composition (RGB representation), XRD identifies amorphous or partially crystalline film (colored vs. gray representation), profilometry determines film thickness (contour line representation). (4) Through deep reactive ion etching (DRIE), exposed silicon is vertically etched down to the film. Individual square chips are cleaved along the gridlines. Each chip features a 2 mm circular hole, across which the glassy film is freely suspended. b Viscosity measurement: Stress is applied to the free-standing film through gas pressure p. Governed by its viscosity η at temperature T, the film deforms into a bubble of spherical geometry (see photograph and micrograph). The bubble’s height h is recorded over time to calculate η(T). In our automated setup, the silicon chip is pressed against a brass heater. p and T are recorded using a pressure gauge and thermocouple. The height h is recorded using a laser-micrometer, emitting a thin curtain of light, which the expanding bubble partially blocks. Operating in temperature-scanning mode (~25 K/min), the film reveals the viscosity–temperature dependence. c Example data: (1) Inflation data for Mg₆₄Cu₂₈Y₈ over time. (2) The resulting viscosity–temperature curve calculated using Eq. (2), along with the uncertainty estimate graph (green shaded graph). (3) Angell-plot: Two regimes are identified, the glass state and the metastable liquid state (compare Fig. 1b, and DSC trace in c.2). T_g is determined by linear extrapolation from the liquid regime to 10¹² Pa·s. T_x is determined where the viscosity curve bends upwards. The fragility m is determined as the slope extrapolated to T_g. (4) Using this approach, the fragility is measured at various compositions. While the example compositions here vary by only 4 at.%, the measured fragility exhibits a remarkably large variation from 25 to 41.