Ensemble machine learning digital twin for bulk and porous g-C<sub>3</sub>N<sub>4</sub> photodetectors
OPTICAL MATERIALS, cilt.178, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 178
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.optmat.2026.118280
- Dergi Adı: OPTICAL MATERIALS
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Chimica, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Recep Tayyip Erdoğan Üniversitesi Adresli: Evet
Özet
Graphitic carbon nitride (g-C3N4, GCN) has emerged as a promising metal-free semiconductor for photodetection applications owing to its visible-light-active bandgap (similar to 2.7 eV), facile synthesis, and chemically tunable electronic structure. In this work, two morphologically distinct GCN variants are synthesised via thermal condensation: bulk GCN (GCN-M) from melamine, and porous GCN (GCN-U + M) from a urea-melamine binary precursor mixture. Planar photodetector devices with the architecture Glass/ITO/g-C3N4/Al are fabricated and comprehensively characterised. X-ray diffraction (XRD) reveals characteristic (002) and (100) reflections at 2 theta approximate to 27.2 degrees and 13.1 degrees, with GCN-U + M showing a measurable (002) peak shift indicative of expanded interlayer d-spacing resulting from the porous network. Fourier-transform infrared (FTIR) spectroscopy confirms the heptazine/triazine-based polymeric network, characteristic s-triazine ring breathing at 810 cm(-1), C-N/CN stretching in the 1200-1650 cm(-1) region, and terminal N-H stretching above 3000 cm(-1). Photoluminescence (PL) measurements reveal excitation-dependent blue-green emission, with peak wavelengths of similar to 497 nm for GCN-M at lambda exc = 280 nm and similar to 478 nm for GCN-U + M at lambda exc = 340 nm. These values are reported under their respective excitation conditions without direct inter-sample comparison. Current-voltage (I-V) measurements under dark and AM1.5 illumination conditions reveal a photocurrent of 620 nA for GCN-M and 190 nA for GCN-U + M at +3 V bias, yielding responsivities of 15.5 mA/W and 4.7 mA/W and specific detectivities of 1.20 & times; 10(9) and 6.35 & times; 10(8) Jones, respectively. To transcend the limitations of physics only modelling, a novel ensemble machine-learning framework is developed, which achieves an exceptional R-2 = 0.9997 in predicting device I-V behaviour from engineered voltage domain features and a material type descriptor. Feature importance analysis identifies applied voltage (V), its cubic term (V-3), and the material morphology indicator as the dominant predictors, consistent with the non-linear space-charge-limited transport mechanism. This work provides the first demonstration of an ensemble ML-based digital twin for metal-free g-C3N4 photodetectors, establishing a pathway for accelerated design of next-generation visible-light sensing devices.