Dark-field electron correlation microscopy

The wires were heated into the supercooled liquid state between the glass transition temperature T_g = 507 K up to 523 K, near the crystallization temperature³⁷ using a DENSsolutions SH30 single-tilt heating holder, which can provide temperature stability of 0.1 K and temperature accuracy of 2%. Nanowires start to crystallize from the free surface when temperature increased to 527 K. The sample was equilibrated at temperature before data collection. At temperatures of 515, 511, and 507 K, the sample was heated to the target temperature at a rate of 20 K min⁻¹, and then held for 30–60 min before image time series acquisition. At 523 and 519 K, in order to avoid crystallization, the sample was first heated to 508 K at rate of 20 K min⁻¹ and held for 30–60 min, then heated to target temperature at the same heating rate and held isothermally for 2 min before data collection. At all temperatures, the equilibration time before data acquisition was at least five times the measured structural relaxation time shown in Fig. 2.

ECM measurements were carried out using tilted dark-field TEM imaging. Experiments used the University of Wisconsin-Madison FEI Titan with probe aberration corrector at 200 kV, operated in TEM mode. An objective aperture of 10 μm in diameter or 2.83 mrad of half angle was inserted, giving rise to speckles in the image ~0.7 nm in diameter, calculated from the Rayleigh criterion and confirmed by imaging. The speckle size sets the spatial resolution of the ECM experiment. An Orius 2.6 x 4 k fast CCD with 1 ms readout time was used to record the time series of images. In all, 256 by 256 pixel images were acquired with the magnification adjusted to yield a typical pixel size of 0.25 nm, so each speckle covers ~3 pixels.

The interval between frames in the image series was set as a function of temperature to ~0.005τ_med, where τ_med was acquired with a time interval short enough not to influence the results³⁰. The acquisition time per frame was set to 0.1, 0.25, 0.5, 1, and 2 s for temperatures 523, 519, 515, 511, and 507 K respectively. Every image series consists of ~4000 frames. The total time for the time series was set to ~20τ_med, to provide a balance between a time series that is too long, which may average together short and long relaxation process over temporally fluctuating dynamics, yielding artificial spatial homogeneity in the relaxation time⁶², and a time series that is too short, which yields autocorrelation functions that are not well converged^10,63. In all, 20τ_med can be thousands of seconds near T_g, so rigid image alignment was used to correct sample drift. Supplementary Movie ¹ is a typical image series with 4000 frames acquired at 523 K. The bright spot in the image arises from a crystallized chunk of nanowire, the intensity of which does not change over the whole image series. In addition, because the background scattering intensity from SiN_x is quite small compared with intensity from sample, the edge of nanowire can be well defined. Therefore, with the bright spot and the outline of nanowire as references, drift correction can be realized through rigid image alignment with single pixel precision using the convolution-based alignment in DigitalMicrograph software. Supplementary Movie ² shows the series in Supplementary Movie ¹ after alignment. The intensity of the pixels fluctuates, but the outline of the nanowire is stationary. Drift correction enables acquisition and analysis of long data series at low temperatures.