# Performance optimization and machine learning-guided parameter sensitivity analysis of lead-free KGeCl3 perovskite solar cells

## TL;DR

This study explores lead-free perovskite solar cells using KGeCl3 and different electron transport layers, optimizing performance with simulation and machine learning.

## Contribution

The novel use of machine learning to guide parameter sensitivity analysis in lead-free perovskite solar cells is introduced.

## Key findings

- The FTO/CFTS/KGeCl3/WS2/Au configuration achieved the highest power conversion efficiency of 21.39%.
- CatBoost machine learning algorithm outperformed others with an R2 score of 0.984 and 99.344% accuracy.
- WS2 as an electron transport layer showed the best performance in terms of band alignment and recombination.

## Abstract

This study gives a realistic insight into the effectiveness of lead-free Ge-based perovskite solar cells (PSCs) using KGeCl3 as the absorber layer in combination with four different electron transport layers (ETLs), including WS2, ZnSe, PC60BM, and SnS2, with copper iron tin sulfide (CFTS) serving as the hole transport layer (HTL). Initially, key material parameters such as layer thickness, donor density (ND), acceptor density (NA), defect density (Nt), interface defect densities (IL1 & IL2), series resistance (Rs), shunt resistance (Rsh), operating temperature (K), and back contact work function (eV) are varied using a SCAPS-1D simulator to optimize device performance. Between the four cell configurations, the FTO/CFTS/KGeCl3/WS2/Au structure has achieved the highest performance with a power conversion efficiency (PCE) of 21.39%, short-circuit current density (JSC) of 39.526 mA cm−2, fill factor (FF) of 76.56%, and open-circuit voltage (VOC) of 0.706 V at a simulated temperature of 300 K. Other configurations using ZnSe, PC60BM, and SnS2 as ETLs showed PCE values of 21.38%, 21.05%, and 20.43%, respectively. Furthermore, an integrated machine learning framework with four supervised learning methods, i.e., Random Forest, XGBoost, CatBoost, and Decision Tree, has been utilized to effectively evaluate the importance of material features. Out of the algorithms, CatBoost has the highest performance with R2 and accuracy values of 0.984 and 99.344%, respectively.

Lead-free KGeCl3 perovskite solar cells are optimized using SCAPS-1D and machine learning with WS2, ZnSe, PC60BM, and SnS2 ETLs. The ETL WS2 provides the best band alignment, lowest recombination, and highest simulated efficiency.