Sheet resistance prediction of laser induced graphitic carbon with transformer encoder-enabled contrastive learning

Accurately predicting the sheet resistance of laser-induced graphitic carbon (LIGC) is crucial for optimizing process conditions and designing high-performance LIGC-based devices. However, identifying the most significant process parameter for predicting the sheet resistance of LIGC is challenging. In addition, training an accurate model with a small dataset remains a challenge. To address the first issue, a novel transformer encoder with a self-attention mechanism is introduced to determine the most and least significant process parameters affecting the sheet resistance of LIGC. To address the second issue, a contrastive learning method is developed to augment a small training dataset. Unlike conventional deep learning approaches that establish a direct relationship between process parameters and sheet resistance, the proposed method can learn the relationship between the difference in features extracted from process parameter settings and the difference in the corresponding sheet resistances. Experimental results have demonstrated that the proposed transformer encoder-enabled contrastive learning method accurately predicted the sheet resistance of LIGC and outperformed other machine learning methods.