Metamodeling of numerical device simulations to rapidly create efficiency optimization roadmaps of monocrystalline silicon PERC cells

In this contribution, we present an approach to simulate the energy conversion efficiencies of monocrystalline p-type silicon passivated emitter and rear cells (PERC) using a state-of-the-art design of experiment (DoE) and metamodeling approach. We preserve the accuracy of numerical device simulations whilst reducing the time for a simulation of cell efficiency from potentially hours to milliseconds. We show that only 1000 numerical simulations arranged in a space-filling DoE and metamodeling by Gaussian process regression are sufficient to cover a 13-dimensional input space, with a small mean absolute error of only 0.056%abs determined in a 10-fold cross validation. In the second part of this work, we apply the metamodel to iteratively scan this input space for the technological improvements that allow the largest efficiency gains. Simultaneously, we consider physical and technological constraints on the input parameters and optimize technologically freely changeable parameters after each step for a fair comparison. We present a roadmap that enables the production of more than 23% efficient monocrystalline p-type silicon PERC cells. Steps on this roadmap are the permanent regeneration of the boron-oxygen defect and the introduction of a selective emitter, followed by an improved front side metallization, an improved local back surface field and an improved rear passivation.