Organic field-effect transistors (OFETs) offer significant potential for flexible electronics owing to their low-cost processing and mechanical adaptability. This study systematically enhanced electrical performance in N2200-based top-gate bottom-contact OFETs through parametric optimization, demonstrating that shorter channel lengths (150 µm) boosted current density (with leakage control), while optimized N2200/poly(methyl methacrylate) (PMMA) concentration ratios (7/100 mg/mL) and annealing (80 °C, 3 h) improved crystallinity and interfacial properties, achieving stable electrical performance. Crucially, fabrication-parameter-driven performance prediction was established using 719 experimental data (superior to prevalent TCAD-generated datasets) via convolutional neural network (CNN), back propagation neural network (BPNN), and random forest (RF) models—further refined by our innovatively developed CNN-particle swarm optimization (PSO)-BP based hybrid architecture. These architectures autonomously extracted physical characteristics without predefined formulas, with CNN achieving R2>0.9 for all metrics (notably <InlineEquation ID="IEq1"> <EquationSource Format="TEX">${R^2}_{{V_{{\rm{th}}}}} = 0.95$</EquationSource> <EquationSource Format="MATHML"><math display="block"> <msub> <mrow> <msup> <mi>R</mi> <mn>2</mn> </msup> </mrow> <mrow> <mrow> <msub> <mi>V</mi> <mrow> <mrow> <mrow> <mi mathvariant="normal">t</mi> <mi mathvariant="normal">h</mi> </mrow> </mrow> </mrow> </msub> </mrow> </mrow> </msub> <mo>=</mo> <mn>0.95</mn> </math></EquationSource> </InlineEquation>, <InlineEquation ID="IEq2"> <EquationSource Format="TEX">${R^2}_{SS} = 0.96$</EquationSource> <EquationSource Format="MATHML"><math display="block"> <msub> <mrow> <msup> <mi>R</mi> <mn>2</mn> </msup> </mrow> <mrow> <mi>S</mi> <mi>S</mi> </mrow> </msub> <mo>=</mo> <mn>0.96</mn> </math></EquationSource> </InlineEquation>), and with CNN-PSO-BP based hybrid architecture delivering significant error reductions: 15.7% mean absolute error (MAE)/14.9% root mean square error (RMSE) for Vth, 10% MAE/9% RMSE for lg(Ion/Ioff), and 13.5% MAE/9.5% RMSE for SS, and enhanced μsat stability via outlier fitting. Leveraging PSO demonstrates superior navigation of OFET performance prediction, establishing machine learning-driven frameworks as a critical value for intelligent performance forecasting and accelerated high-throughput device performance tuning.

Deep learning-driven intelligent prediction for tailoring electrical properties of N2200-based donor-acceptor conjugated copolymer OFETs

期刊论文
发表日期：

2025年10月31日

查看全文

Jie Wu,
Xuqi Yang,
Jing Chen,
Yufan Mao,
Walid Boukhili,
Guangyao Chen,
Yuxuan Bao,
Lei Wang

摘要

Organic field-effect transistors (OFETs) offer significant potential for flexible electronics owing to their low-cost processing and mechanical adaptability. This study systematically enhanced electrical performance in N2200-based top-gate bottom-contact OFETs through parametric optimization, demonstrating that shorter channel lengths (150 µm) boosted current density (with leakage control), while optimized N2200/poly(methyl methacrylate) (PMMA) concentration ratios (7/100 mg/mL) and annealing (80 °C, 3 h) improved crystallinity and interfacial properties, achieving stable electrical performance. Crucially, fabrication-parameter-driven performance prediction was established using 719 experimental data (superior to prevalent TCAD-generated datasets) via convolutional neural network (CNN), back propagation neural network (BPNN), and random forest (RF) models—further refined by our innovatively developed CNN-particle swarm optimization (PSO)-BP based hybrid architecture. These architectures autonomously extracted physical characteristics without predefined formulas, with CNN achieving R²>0.9 for all metrics (notably ${R^2}_{{V_{{\rm{th}}}}} = 0.95$ ${R^{2}}_{V_{t h}} = 0.95$ , ${R^2}_{SS} = 0.96$ ${R^{2}}_{S S} = 0.96$ ), and with CNN-PSO-BP based hybrid architecture delivering significant error reductions: 15.7% mean absolute error (MAE)/14.9% root mean square error (RMSE) for V_th, 10% MAE/9% RMSE for lg(I_on/I_off), and 13.5% MAE/9.5% RMSE for SS, and enhanced μ_sat stability via outlier fitting. Leveraging PSO demonstrates superior navigation of OFET performance prediction, establishing machine learning-driven frameworks as a critical value for intelligent performance forecasting and accelerated high-throughput device performance tuning.