SPLTRAK Abstract Submission
From Femtoseconds to Gigaseconds: The SolDeg Project to Analyze Si Heterojunction Cell Degradation with Machine Learning 
Gergely Zimanyi1, Davis Unruh1, Reza Vatan2, Zitong Zhao1, Andrew Diggs1, Stephen Goodnick2
1University of California, Davis, Davis, CA, United States
/2Arizona State University, Tempe, AZ, United States

We are reporting the results of the SolDeg project for analyzing performance degradation in Si heterojunction solar cells. First, femtosecond molecular dynamics (MD) simulations were performed to create a-Si/c-Si stacks, using a Machine-Learning-based Si-Si Gaussian Approximation Potential GAP. The silicon- and hydrogen-related defects were determined next by combining MD and DFT methods. The defect generation energies were determined by the Nudged Elastic Band method. Finally, an accelerated Monte Carlo method was developed to simulate the thermally activated time dependent defect generation across the barriers, out to gigaseconds. We have shown that a stretched exponential analytical form can successfully describe the defect generation N(t) over at least ten orders of magnitude in time. We also developed the Time Correspondence Curve to calibrate and validate the accelerated testing of solar cells. We found a compellingly simple scaling relationship between accelerated and normal testing times: t(normal) ~ t(accel)^(T(accel)/T(normal). – Second, we used Machine Learning to develop our own Si-H GAP to reach unparalelled, DFT-level precision with computation times 10-100 times faster than DFT. We showed that in typical c-Si/a-Si:H HJ cells the hydrogen atoms experienced a potential gradient that sloped away from the interface, making the hydrogen atoms drift away from the interface and thus generating defect states at the interface. This degradation of the passivation is quite likely a key driver of the cell performance degradation. Finally, we discovered that the hydrogen potential gradient was caused by the crystallinity gradient of a-Si:H. Thus, the hydrogen potential gradient can be reversed to slope toward the interface by reversing the a-Si crystallinity gradient. In such reversed-gradient stacks, the hydrogen does not drift away from the interface. This is a key message of the Soldeg project: HJ cell degradation can be stopped by deposition protocols that have a crystallinity minimum at the HJ interface.