Graph Neural Network Guided Selection of Functional Polymers for Charge Transfer Doping of 2D Materials

Fabia Farlin Athena, Tara Peña, Kalee Francis Rozylowicz, Anh Tuan Hoang, Andrew J. Mannix, H.-S. Philip Wong, Eric Pop, Alberto Salleo

Keywords: GNN, Polymer Informatics, High-Throughput Screening, Materials Discovery, 2D Materials, Charge Transfer Doping

TL;DR: We used a GNN to successfully screen for and experimentally validate new polymers that can effectively dope 2D semiconductors.

Abstract: Two-dimensional (2D) semiconductors are competing materials for post-Si transistors because of their desirable carrier mobilities and band gaps with atomic thinness. However, high contact resistances plague the performance of 2D-based devices. Thus, non-destructive doping strategies are needed in order to overcome this challenge for complementary (n-type and p-type) 2D logic. Here, we present the use of functional polymers to non-destructively introduce charge transfer n- and p-type doping in 2D materials. First, we utilize a multitask graph neural network (GNN), trained on density functional theory-calculated properties, to predict properties like ionization energy and electron affinity of candidate polymers. These predictions are then used to select polymers capable of doping n- and p-type monolayer MoS$_2$ and WSe$_2$ based on band-level alignment criteria. We validate our screening strategy with Raman spectroscopy, confirming successful p-type doping of WSe$_2$ with PEDOT:PSS and Nafion, and n-type doping of MoS$_2$ with PEI and WSe$_2$ with p(gPyDPP-MeOT2). The results demonstrate a practical model-to-lab pipeline that effectively narrows the search space for polymer dopants, bridging high-throughput computational materials screening with experimental validation.

Submission Track: Findings, Tools & Open Challenges

Submission Category: AI-Guided Design

Institution Location: California, USA

AI4Mat RLSF: Yes

Submission Number: 142