**Appendix A. Selection of Ramping Duration**

One aspect of these simulations which we found to be challenging was that the hydrostatic stress had to be applied carefully to the atomistic system. Otherwise, artifacts in the form of unwanted and uncontrolled defects would be introduced into the system prior to the holding phase. After trying out several strategies for applying stress to the atomistic system, we found that a simple linear ramp via an *NPT* ensemble was sufficient as long as the loading duration was sufficiently long. If the ramping duration were too short, we observed that void nucleation would occur prematurely via nucleation of dislocations. Figure A1 shows the time at which a void nucleated relative to the start of the holding phase (indicated by a 10% increase in the simulation cell volume relative to the start of the holding phase). Each data point is an average over five replicas. To ensure that the load ramping does not introduce artifacts, we need to confirm that the nucleation time is insensitive to the ramping duration. Figure A1 shows, within the scatter of the data, that 12 ps is a sufficiently long ramping duration. When ramped over durations less than 12 ps, the nucleation time decreases. We believe that this behavior results from stress waves and spikes introduced into the cell by the *NPT* barostat during rapid loading, which nucleate defects during ramping.

**Figure A1.** Void nucleation time, measured as time required for the simulation cell volume to increase by 10% during the holding phase, as a function of ramping duration.
