Scalable synthesis, characterization, and DFT-machine learning modelling of Ti₃C₂ MXene quantum dots

Abstract

Two-dimensional Ti₃C₂T_x MXenes exhibit exceptional electrical conductivity and tunable surface chemistry; however, reduction to zero-dimensional quantum dots (MQDs) affords quantum confinement and enriched functionalization, modulating the optoelectronic and electrochemical properties. In this research study, we developed a protocol for laboratory-level scale-up synthesis using TMAOH, yielding high-quantum-dot production with enhanced surface functionalization. XRD analysis confirmed quantum confinement, with the crystallite sizes sharply reducing from 8.68 nm (nanosheets) to ∼2 nm (MQDs). FTIR spectroscopy and XPS analysis revealed increased oxygen- and hydroxyl-functional groups replacing fluorine terminations on the MQD surfaces. Raman and photoluminescence spectroscopy reflected the combined influence of quantum confinement and surface termination chemistry, manifesting as electronic bandgap widening (1.77 to 2.32 eV), modified vibrational dynamics, and excitation-dependent PL emission. DLS and BET analyses documented size reduction, enhanced dispersion, and an increase in surface area from 31.8 to 140.6 m² g⁻¹ in MQDs. Electron microscopy validated preserved crystallinity and nanoscale morphology. Electrochemical studies showcased the superior charge transfer, stability, and catalytic activity of MQDs over MXene nanosheets, which was also supported by AFM surface roughness and Bode plot EIS analyses. A detailed DFT investigation quantified the metallic behaviour of 2D MXene with a near-zero bandgap versus the semiconducting behaviour of MQDs with a discrete 0.25 eV direct bandgap due to quantum confinement and oxygen termination. Machine learning models, trained on multimodal descriptors combining experimental and synthetic data, were used as an exploratory screening tool to identify candidate sensor-efficiency predictors, highlighting crystallite size, surface roughness, oxygen content, and particle size as the dominant influencers. MQDs consistently outperformed MXene nanosheets in sensing performance, attributed mechanistically to their nanoscale morphology and surface chemistry. Our work primarily focuses on the first scalable MQD synthesis correlating to the technoeconomic model, which supports a preliminary DFT-ML framework linking dimensionality to next-generation sensing applications.

From 2D Ti₃C₂T_x MXenes to 0D MXene quantum dots (MQDs): scalable synthesis, spectroscopic and microscopic characterization, density functional theory and machine learning.

1. Introduction

“Nanotechnology is not about making things small; it's about controlling things at the nanoscale. And this is where MXenes excel”, is a quote by Prof. Yury Gogotsi.¹ MXenes are a class of two-dimensional (2D) nanomaterials of transition-metal carbides, nitrides and/or carbonitrides that are being considered by the International Union of Pure and Applied Chemistry (IUPAC) as one of the top ten globally emerging technologies that can contribute to the United Nations Sustainable Development Goals (UNSDG) for a brighter future and a more sustainable circular economy.² MXenes provide vital energy and unprecedented storage capability for the modern era of nanotechnology, and they are employed as electrode materials in sensors and biomedical applications, like therapeutics, theranostics, medical imaging, smart textiles and wearables.^3–5 MXenes are extensively explored in energy storage devices, including batteries and supercapacitors and for environmental remediation.^6,7 In recent times, the field has been extended to both space technology and aviation industries, including electromagnetic shielding, lightweight components, and thermal management at extreme temperatures, making them ideal for diverse applications.^8–10

If one wonders how the new-age material group of MXenes competes with graphene, which is regarded as the industrial gold standard, the answer lies in their tunable surface chemistry, expressed by their general formula M_n+1X_nT_x, where M is an early transition metal (such as Ti, V, Nb, or Mo) from the d-block of the periodic table, X represents C and/or N atoms, and T_x denotes surface termination groups, like –O, –OH, and –F.¹¹ Unlike graphene, which consists solely of carbon arranged in a hexagonal lattice, MXenes possess layered structures with multiple metal atomic layers interleaved with carbon or nitrogen, endowing them with metallic conductivity combined with highly tunable surface chemistry.¹² This allows MXenes to exhibit customised electronic and electrochemical properties through controlled surface terminations and compositional variations, offering versatility in intercalation, catalysis, energy storage, and sensing applications. The presence of transition metals from groups 4 to 6 in the periodic table in MXenes contributes to their strong metallic bonding and high electron density at the Fermi level, differentiating them chemically and functionally from graphene's inert carbon network, thus explaining their rapidly growing prominence as advanced functional materials.¹³

Now imagine the same versatile properties of 2D MXene compressed into nanoscale 0D quantum dot (MQD): Wouldn't it be excellent if size-induced quantum confinement effects and abundant functional sites amplify their intrinsic properties, unlocking unprecedented ability for diverse applications¹⁴? Their ultra-small size leads to a high surface-to-volume ratio, resulting in superior reactivity, conductivity and tunable opto-electronic performance.¹⁵ It is evident that MQDs demonstrate better semiconductor properties due to their strong confinement effects and zero dimensionality, which induce a tunable bandgap, than the 2D MXenes, which typically exhibit metallic conductivity with a zero or negligible bandgap. This unique tunability and nanoscale control over properties is precisely why Aleksey Ekimov, Louis Brus, and Moungi Bawendi were awarded the 2023 Nobel Prize in Chemistry for the discovery and synthesis of 0D quantum dots, recognizing their profound impact on nanotechnology.¹⁶ Additionally, MQDs have proved to exhibit better dispersibility and biocompatibility than bulk 2D MXene sheets, positioning them as powerful nanoscale materials for next-generation semiconductor ecosystems.

While the basic cutting of MXenes using tetramethylammonium hydroxide (TMAOH) has been demonstrated in milligram-scale quantities, routine production remains limited by low product recovery and batch-to-batch structural variation. Here, we present a scalable, high-yield laboratory protocol that steps up production to gram-scale volumes of highly uniform MQDs. Furthermore, instead of relying purely on post-synthesis characterization, we introduce an integrated framework combining density functional theory (DFT) data with machine learning (ML) models to predict and map the electronic density of states based on surface termination chemistry. This combination of scalable physical production and predictive computational design establishes a highly optimized material matrix with superior charge-retention properties.

2. Experimental methods

2.1. Synthesis of Ti₃C₂T_x MXene from MAX phase

Titanium aluminium carbide (Ti₃AlC₂) MAX phase (powder, 99.8% trace metals basis, CAS Number: 196506-01-1), hydrofluoric (HF) acid (38–40%, EMPLURA, CAS Number: 7664-39-3) and dimethyl sulfoxide (DMSO) (ACS reagent, ≥99.9%, CAS Number: 67-68-5) were procured from Sigma-Aldrich, USA. The Ti₃AlC₂ MAX phase was etched using HF under magnetic stirring for 2.5 h in a fume hood.¹⁷ When treated with HF under magnetic stirring, the acid selectively removed the aluminium (Al) layers from the MAX phase through a topochemical reaction, as described by the equation:

1

This reaction effectively stripped the Al atoms, leaving behind the 2D multi-layered MXene. Subsequent reactions further functionalized the MXene surface by attaching termination groups such as –O, –OH, and –F, originating from HF and water interactions:

Ti₃C₂ + 2HF → Ti₃C₂F₂ + H₂ 2

Ti₃C₂ + 2H₂O → Ti₃C₂(OH)₂ + H₂ 3

Repeated washing was performed using an ultracentrifuge at 5500 rpm with DI water until the pH became neutral to ensure removal of residual acids, fluoride ions, metal salts, and other impurities generated during the synthesis that could possibly cover the MXene surface with corrosive or reactive species and affect the material quality and subsequent reactions. Post-washing, the precipitate, i.e., MXene, was dried using ethanol to prevent agglomeration and restacking by facilitating rapid evaporation and reducing surface tension while maintaining the material's high surface area and dispersibility.¹⁸ The material was then delaminated using DMSO under magnetic stirring for 24 h. The intercalation of DMSO into MXene was primarily a physical process in which DMSO molecules are inserted between the stacked MXene layers, expanding the interlayer spacing and weakening the van der Waals forces that hold the layers together.¹⁹ This expansion facilitated the delamination of bulk multilayer MXene stacks into few-layer or single-layer nanosheets. DMSO molecules, having a polar sulfoxide (S O) group and methyl groups, interact via dipole interactions and hydrogen bonding with surface-terminated groups (such as –OH) on the MXene layers, stabilizing the expanded structure as stated below:

Ti₃C₂T_x + DMSO → Ti₃C₂T_x⋅DMSO 4

After DMSO treatment, the delaminated MXene was gently washed with DI water at 3500 rpm to minimize agglomeration and restacking of the exfoliated nanosheets. This slow washing process prevented damage to the delicate layers and maintained good dispersion stability.²⁰ The material was again further washed with water to remove residual DMSO and impurities, followed by filtration to efficiently separate the MXene solids from the liquid without introducing contamination. Finally, the filtered MXene was dried using ethanol to yield the delaminated MXene nanosheets for subsequent applications.

2.2. Laboratory-based scale-up synthesis of MXene quantum dots by the reflux method

After confirmation of the formation of the desired 2D Ti₃C₂T_x MXene nanosheets from the MAX phase, it was used as a precursor for the synthesis of Ti₃C₂ MXene quantum dots (MQDs). Tetramethylammonium hydroxide (TMOAH) solution (1.0M ± 0.02 M in H₂O, solution, ACS reagent grade, CAS Number: 75-59-2) was procured from Sigma-Aldrich, USA. The scale-up synthesis of MQDs was initiated by dispersing the synthesized MXene powders of varying amounts (100 mg, 200 mg, 300 mg, 400 mg, and 500 mg) into corresponding volumes (10 mL, 20 mL, 30 mL, 40 mL, and 50 mL) of 3% TMAOH aqueous solution (Fig. 1). TMAOH acted as a mild etching and intercalating agent that further exfoliated the MXene layers into nanoscale quantum dots under heat. The dispersions were subjected to a borosilicate glass-reflux setup comprising a round-bottom flask, a condenser with continuous cool water supply, and an oil bath maintained at 140 °C with stirring at 290–300 rpm for 72 hours, during which the OH⁻ and (CH₃)₄N⁺ ions penetrated the MXene layers, facilitating layer separation and cleavage and forming the 0D MQDs through a combination of chemical bonding dissociation and physical delamination. The high temperature and prolonged reflux provide the energy required for partial breaking of Ti–C bonds at defect sites or edges, producing nanoscale fragments with abundant surface terminations (e.g., –OH, –O, and –F). The products were then centrifuged multiple times for 25 minutes at 10 000 rpm to separate the MQDs in the supernatant from the unreacted bulk 2D MXene. After filtration, the obtained MQDs were kept for drying in ethanol overnight at 65 °C to remove any remaining TMAOH. The MQD yields were measured using an Ohaus® PR Precision Balance (Model PR223/E). Furthermore, a comprehensive techno-economic analysis²¹ was performed for the scaled-up synthesis of MQDs by considering reagent usage at input scales spanning 100 mg to 500 mg of Ti₃AlC₂ precursor. Following the cost analysis, it was further decided to compare the optoelectronic properties of the MQDs to those of 2D MXene.

Fig. 1. Photographs of the laboratory-level scale-up synthesis of MQDs. The top and middle rows display the reflux setup and corresponding precursor dispersions for batches ranging from 100 mg to 500 mg of MXene. The sequence below demonstrates the key post-synthesis processing steps, including filtration, centrifugation, and drying. The final image shows the collected MQDs as a stable, dark colloidal product, highlighting the efficient and reproducible MQD production at increasing precursor scales.

2.3. Methodology for the cost analysis of the scaled-up synthesis of MQDs

A comprehensive techno-economic analysis²¹ was performed for the scaled-up synthesis of the MQDs by considering reagent usage at input scales spanning 100 mg to 500 mg of the Ti₃AlC₂ precursor. The catalogue prices of all the necessary chemicals, including Ti₃AlC₂, HF 40%, DMSO, TMAOH, ethanol, and deionized water, were obtained from Sigma-Aldrich, India. An 18% goods and services tax (GST) was applied to these prices to accurately reflect the total procurement cost. For each batch scale, the stoichiometric volume of reagents used was calculated proportionally to the MXene precursor input, and their costs were derived by multiplying required quantities by the respective unit prices inclusive of GST. Total production cost per batch was computed by summing all individual reagent costs and consumables. The yield of MQDs (%) obtained at each scale was recorded and incorporated into the analysis, allowing adjustment of cost estimates for actual product output efficiencies. Capital expenditure (CapEx) related to equipment and infrastructure was also considered consistently across batches to provide a comprehensive economic perspective on laboratory scale-up feasibility. This methodology allowed for the generation of a detailed cost table capturing reagent-wise and total expenses alongside the corresponding MQD yields, providing a comprehensive economic assessment that supported scale-up decisions and techno-economic feasibility evaluations of MQD synthesis. The synthesis of the MQDs was investigated across five distinct operational scales, governed by the initial precursor mass (MXene input). In this experimental design, the quantities of all seven cost-contributing reagents, MAX phase, HF 48%, DMSO, TMAOH, ethanol, and DI Water, were scaled in perfect proportionality to the MXene Input. Consequently, although these cost factors were conceptually independent variables, they exhibited extreme statistical collinearity. The primary response variable utilized for economic optimization was the cost efficiency (INR per mg), calculated as the ratio of total cost to MQD yield. To stabilize the model in the presence of severe multicollinearity and to quantify the contribution of all cost factors, principal component regression (PCR) was employed.²² A model was constructed where the eight independent features (seven cost factors plus MXene input) and the non-linear driver (MQD yield) were used to predict the cost efficiency of the scaled process. The seven cost factors were first subjected to principal component analysis (PCA) to transform the correlated variables into a set of uncorrelated principal components (PCs). A linear regression was then performed on the first resultant PC, effectively treating all cost factors as a single, combined investment factor. For visualization, the aggregated factor influence (total cost) and the scale factor (MXene input) were plotted against the cost efficiency response (INR per mg). A continuous predictive surface was generated using the radial basis function (RBF) interpolation to model the non-linear efficiency profile.²³ Optimization was completed by identifying the MXene input corresponding to the absolute minimum cost efficiency and the maximum MQD yield.

2.4. Material characterization by spectroscopy and microscopy

The synthesized nanoscale materials were characterized through powder XRD using a Panalytical X'Pert 3 diffractometer (manufactured in the Netherlands) equipped with 2.2 kW Cu-anode ceramic tubes as X-ray sources. The scans were conducted at a scan rate of 0.1° per minute to ensure high resolution and accuracy in the diffraction patterns. The crystallite size was calculated using the Scherrer equation. HighScore Plus software (Version: 3.0.5) was also used to analyse the measurements.²⁴ FTIR analyses were performed to observe the functional groups on the MXene and MQD powder samples using portable Agilent Cary 630 FTIR Benchtop Systems in the range 600 cm⁻¹ to 4000 cm⁻¹. A Cary 5000 UV-vis-NIR spectrophotometer (Agilent Technologies, USA) was used to compare the absorbance of the MXene and MQDs (1 mg dissolved in 2 mL DI water). The optical band gaps for each material were calculated from their corresponding Tauc plots. The hydrodynamic particle size and size distribution were determined using a Litesizer 500 instrument (Anton Paar GmbH, Austria) based on the dynamic light scattering (DLS) method. A laser Raman microscope (LabRAM HR Evolution, Horiba) was employed to identify the phases, phase transitions, bonding, structure, and defects over the spectral range 100 cm⁻¹ to 2000 cm⁻¹. Photoluminescence (PL) emission scans of MXene and MQDs in the spectral range 200 nm to 450 nm were recorded with a FLS1000-SS-s (Edinburgh Instruments, UK). The multi-point Brunauer–Emmett–Teller (BET) technique, employing 99.999% nitrogen gas (N₂) and an AutosorbiQ (Quantachrome, USA), was used to measure the nanomaterials' specific surface area. To determine the elemental composition and chemical states of the elements present on the surface of the materials, elemental scans having both survey and high-resolution spectra were recorded using an XPS ULVAC-PHI VersaProbe 4 (Japan). To observe the morphology of the nanomaterials, FE-SEM analysis was carried out with a FEI QUANTA 250 FEG (Thermo Fisher) operated at 20.00 kV, from 10 k to 300 k magnification at different resolutions. Further, to gain a better understanding of the 0D MQDs, HR-TEM and SAED analyses were carried out using a JEM-2100 Plus instrument (JEOL, Japan). The images were further analysed using a Digital Micrograph © GATAN (Version 3.69.4983.0) to calculate the interplanar d-spacing of the lattices. Furthermore, the electrode's surface topography was examined pre- and post-electrochemical analyses using an atomic force microscope (AFM) (Nanosurf AFM, Nanosurf, Switzerland). AFM analysis was conducted using the dynamic force operating mode with a Dyn190Al cantilever instrument. The image size was 2 µm × 2 µm, with 256 points and 256 lines scanned, and the data were interpreted using Gwyddion software.²⁵

2.5. Electrode preparation and electrochemical characterization

Commercial SPCEs (TE100) were purchased from Zensor R&D, Taiwan. The ready-to-use 3-electrode system did not require pre-treatment and had a polypropylene base (12.6 mm × 50 mm), with a 3 mm diameter carbon working electrode (WE), a Ag pseudo-reference electrode (RE), and a carbon counter electrode (CE), with working areas of 7.1 mm², 0.1 mm², and 10.6 mm², respectively. The SPCEs were fabricated with MXene and MQDs by the drop-casting method. The electrodes were air-dried at room temperature (25 °C) overnight. An electrode cable ECC100, Zensor R&D, Taiwan was used as a connector cable between the SPCEs and a CHI 760 F bulk potentiostat for recording electrochemical measurements via cyclic voltammetry (CV), differential pulse voltammetry (DPV), square wave voltammetry (SWV) and electrochemical impedance spectroscopy (EIS) for the [Fe (CN)₆]^3−/4− redox probe (2.5 mM K₃Fe(CN)₆ + 2.5 mM K₄Fe(CN)₆ in 0.1 M KCl). The catalytic performances of the materials compared to the bare material were characterized using CV (potential range −0.6 V to +1.2 V, 0.1 V s⁻¹ scan rate), DPV and SWV (potential range −0.4 V to +0.8 V). EIS was employed to study the electrode–electrolyte interface. EC-Lab® software V10.32 was used to obtain equivalent circuit models. For basic plots, OriginPro (Version 2025b, OriginLab Corporation, Northampton, MA, USA) was employed.

2.6. Density functional theory (DFT) calculations of MXene and MQDs

All DFT calculations to investigate the structural and electronic properties of Ti₃C₂ MXene were conducted using the Quantum Espresso v.7.4.1 package.²⁶ The Perdew–Burke–Ernzerhof (PBE) formulation of the generalized gradient approximation was employed as the exchange–correlation functional.²⁷ Ion-electron interactions were modelled using projector augmented-wave (PAW) pseudopotentials specifically designed for titanium and carbon atoms.²⁸ Initially, the structural optimization was performed through variable-cell relaxation (vc-relax) calculations to achieve simultaneous relaxation of atomic coordinates and lattice parameters. The plane-wave basis set was truncated at cutoff energies of 50 Ry for the wavefunctions and 400 Ry for the charge density to ensure a balanced accuracy and computational cost.²⁹ Ionic and cell relaxations proceeded until convergence thresholds of 10⁻⁵ Ry for the total energy and 10⁻⁴ Ry Bohr⁻¹ for forces were reached using the Broyden–Fletcher–Goldfarb–Shanno (BFGS) algorithm.³⁰ Following the successful structural optimization, a self-consistent field (SCF) calculation was employed with identical cutoff parameters to precisely determine the ground-state electronic charge density.³¹ Partial occupancies were described using Gaussian smearing with a smearing width (degauss) of 0.02 Ry to treat metallic-like states effectively.³² To obtain detailed electronic band energy information beyond the SCF charge density, non-self-consistent field (NSCF) calculations were performed.³³ Unlike a uniform Monkhorst–Pack grid, this step utilized a dense set of 100 k-points explicitly defined along key high-symmetry directions in the hexagonal Brillouin zone.³⁴ This path-based sampling facilitates accurate band structure plotting and captures intricate dispersion features. The NSCF calculations included 18 electronic bands, consistent with the valence and low-lying conduction states of interest, and adopted Gaussian smearing with a 0.02 Ry width to maintain consistency with the SCF step. Electronic density of states (DOS) computations were carried out using the DOS utility with an extended energy window from −50 eV to +20 eV relative to the Fermi level.³⁵ The fine energy resolution of 0.01 eV ensured a high-definition depiction of states deep within the valence bands as well as conduction band features. Projected density of states (PDOS) analyses were conducted using the projwfc.x post-processing tool, resolving contributions from atomic orbitals on titanium and carbon atoms.^36,37 This projection spanned the same energy window (−50 eV to +20 eV) as the DOS, with a consistent energy resolution of 0.01 eV. The projection did not employ any additional Gaussian broadening beyond the intrinsic smearing used in the NSCF calculation (with ngauss = 0 and degauss = 0.02 Ry), thus preserving a faithful representation of orbital-resolved states. Charge density distributions were extracted from the SCF results and visualized on a three-dimensional real-space grid with dimensions suitable for high-resolution representation (e.g., 200 × 200 × 100 points). The data files generated were compatible with visualization software VESTA (Version 30.9.5a), facilitating qualitative analysis of electron localization, chemical bonding, and spatial charge rearrangements upon relaxation.³⁸ Throughout all the calculations, the atomic coordinates were specified in crystal fractional units, while the lattice vectors were expressed in angstroms. Electronic convergence criteria for self-consistent field calculations were stringent, set at 10⁻⁸ Ry, with a mixing parameter of 0.7 to ensure robust charge density convergence. Highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energies were determined by parsing the eigenvalue spectrum relative to the Fermi level using custom Python scripts. These scripts identified the frontier orbitals and facilitated visualization of the HOMO–LUMO gap via graphical band diagrams. Finally, band structures and DOS/PDOS data were post-processed using Python-based plotting libraries, creating comprehensive visual representations, including band dispersions along selected high-symmetry k-points, elemental-resolved DOS plots, and clear demarcations of the Fermi level.

All DFT calculations on the Ti₃C₂O₂ MQDs were conducted using the Quantum Espresso v.7.4.1 software suite to investigate its structural and electronic properties. The Perdew–Burke–Ernzerhof revised for solids (PBEsol) formulation of the generalized gradient approximation was employed as the exchange–correlation functional to provide an accurate yet computationally efficient description of electron interactions in the system. Ion-electron interactions were modelled using PAW pseudopotentials explicitly generated for titanium, carbon, and oxygen atoms. Initial structural optimization was carried out through fixed-cell relaxation (relax calculation) with variable atomic positions, embedded within a large cubic vacuum supercell of approximately 20 Å in each spatial direction to avoid spurious interactions between periodic images of the isolated quantum dot. Plane-wave basis sets were truncated at cutoff energies of 50 Ry for wavefunctions and 400 Ry for charge density, balancing accuracy and computational expense. Ionic relaxations proceeded using the BFGS algorithm until convergence thresholds of 1.0×10⁻⁵ Ry for total energy and 1.0×10⁻⁴ Ry Bohr⁻¹ for forces were reached. Methfessel-Paxton smearing (order 1), with a smearing width of 0.02 Ry, was employed to treat partial electronic occupancies consistent with the quasi-metallic or semi-metallic character of the system. Subsequent SCF calculations utilized the optimized geometry from the relaxation stage to precisely determine the ground-state electronic charge density. The electronic wavefunctions were converged tightly to an energy threshold of 1.0 × 10⁻⁸ Ry, with a mixing parameter of 0.6, under the Davidson diagonalization scheme. As expected for an isolated zero-dimensional cluster, only the Γ-point was sampled in the Brillouin zone. NSCF calculations followed, employing the converged SCF charge density to calculate accurate Kohn–Sham eigenvalues and eigenstates essential for detailed electronic structure analysis. An extended number of 120 electronic bands were included to encompass unoccupied states well above the Fermi level. The same computational parameters and Γ-point sampling strategy were maintained for consistency. Electronic DOS and PDOS were evaluated using Quantum Espresso's dos.x and projwfc.x post-processing tools over an energy window from −50 eV to +20 eV relative to the Fermi level, with a fine energy resolution of 0.01 eV. Orbital-resolved projections focused particularly on the titanium 3d, carbon 2p, and oxygen 2p states. No additional Gaussian broadening beyond the intrinsic smearing was applied during PDOS generation, preserving an authentic depiction of orbital contributions. Charge density distributions were extracted from the SCF results and visualized on a high-resolution three-dimensional real-space grid (200 × 200 × 200 points) using the VESTA software (v.3.9.5a) to provide qualitative insights into electron localization and chemical bonding within the MQDs. Highest occupied molecular orbital (HOMO) and lowest unoccupied molecular orbital (LUMO) energies were determined by analyzing the energy eigenvalue spectrum relative to the Fermi level with custom Python scripts, facilitating frontier orbital identification and HOMO–LUMO gap estimation. Comprehensive electronic band diagrams, DOS, and PDOS plots were generated using Python-based plotting libraries, presenting clear insights into the quantum-confined electronic behaviour of the Ti₃C₂O₂ MQDs. Throughout all calculations, atomic positions were specified in Angstrom units, while vacuum-embedded lattice vectors ensured the molecular nature of the system. This rigorous and systematic computational workflow provided a detailed understanding of the structural and electronic properties critical to the potential applications of 2D Ti₃C₂ MXene and 0D Ti₃C₂O₂ MQDs.

2.7. Machine learning approaches for sensor performance evaluation of MXene vs. MQDs

A machine learning (ML) framework was implemented to explore the relationship between physicochemical, optical, and electrochemical descriptors and the sensor efficiency of 2D MXene nanosheets and 0D MXene quantum dots (MQDs). Because only a limited number of experimentally characterized samples were available for each material class, a synthetic training dataset was generated to provide the sample size required for conventional regression-based ML models. Two synthetic datasets, each comprising 50 samples, were generated to represent the 2D MXene and 0D MQD material classes, respectively, yielding a combined training set of 100 samples. For each of the 17 descriptors considered (crystallite size, FTIR surface group intensity, optical bandgap, UV-vis absorption peak position, particle size, BET surface area, XPS Ti/C/O content, electrochemical current, DFT HOMO–LUMO gap, AFM roughness before and after electrochemical testing, cyclic voltammetry peak anodic current, electroactive area, EIS charge-transfer resistance (R²), and PL intensity), sample values were drawn from independent Gaussian distributions, N(µ, σ²), in which µ was set to the corresponding experimentally measured or literature-reported central value for the given material class, and σ was set to reflect plausible measurement and batch-to-batch variability (typically 3–10% of µ, or an instrument-precision-based estimate where available). Sampled values were subsequently clipped to physically realistic upper and lower bounds for each descriptor to prevent the generation of implausible feature combinations. A fixed random seed was used throughout to ensure reproducibility. Sensor efficiency was treated as the target variable and was simulated using the same procedure, centered on the experimentally measured efficiency for each material class.

The above generation strategy was adopted specifically to enable model training in the absence of a large, independently replicated experimental dataset and was not intended to substitute for independent experimental replicates. Because the synthetic samples were generated by augmentation around a single experimental measurement per material class, the resulting models were understood to learn the local structure and covariance imposed by the sampling scheme, rather than necessarily capturing novel physical relationships beyond those already reflected in the seed measurements. Of the 17 descriptors used, the central values for nine were obtained directly from experimentally reported data elsewhere in this manuscript: crystallite size (XRD/Scherrer analysis), UV-vis absorption peak position and optical bandgap (Tauc plot analysis), BET surface area, particle size (dynamic light scattering), DFT HOMO–LUMO gap, AFM surface roughness before and after electrochemical testing, cyclic voltammetry peak anodic current, and EIS charge-transfer resistance. For the remaining eight descriptors (FTIR surface group intensity, XPS Ti/C/O content, generic electrochemical current, electroactive area, and PL intensity), no directly reported scalar value was available in the current characterization dataset; central values for these descriptors were therefore set to illustrative, literature-informed estimates rather than to values measured on the specific samples used in this study. The categorical variable Material_Type was numerically encoded (0 for 2D MXene, 1 for 0D MQDs) and included as a feature. Four regression algorithms were trained on the 100-sample synthetic dataset: random forest (200 estimators), support vector regression (radial basis function kernel), gradient boosting (200 estimators), and ridge regression (α = 1.0). All models were implemented using scikit-learn within a Python (Google Colab) environment, with data handling performed using Pandas and NumPy and visualization performed using Matplotlib and Seaborn. To assess internal generalization performance and to evaluate the risk of overfitting associated with the high descriptor-to-sample ratio (17 descriptors over 100 samples), 5-fold cross-validation (CV) was performed on the training set for each model, using both the coefficient of determination (R²) and root-mean-squared error (RMSE) as scoring metrics. Mean and standard deviation values were computed across the five folds. In addition, each model was subsequently retrained on the full training set, and the resulting in-sample (training) R² was compared against the mean cross-validated R² to quantify the gap between training and validation performance, an established diagnostic for overfitting. A gap exceeding 0.10 in R² was taken as an indication of a meaningful overfitting risk. Two samples, corresponding to the experimentally characterized 2D MXene and 0D MQD electrodes, respectively, were used as an additional point of comparison for model predictions. Because these two samples corresponded to the same central values used to parameterize the synthetic training distributions, this comparison was treated as an illustrative consistency check rather than as an independent, statistically meaningful validation set, and the resulting performance metrics were interpreted accordingly. Feature importance for the random forest model was quantified from the mean decrease in impurity across all decision trees. Pearson correlation coefficients between all descriptors and sensor efficiency were computed separately for the 2D MXene subset, the 0D MQD subset, and the combined dataset.

3. Results and discussion

3.1. Yield analysis of the scale-up synthesis of MQDs

The scale-up synthesis of MQDs was conducted primarily to produce a larger yield % for extended research and potential practical applications. Small-scale synthesis, while effective for fundamental studies, often yields insufficient material quantities for device fabrication or commercial use. Hence, scaling up allows material scarcity to be overcome and enables comprehensive performance testing at application-relevant scales. During scale-up, increasing amounts of MXene precursor (100 mg to 500 mg) were dispersed into appropriately scaled volumes of 3% TMOAH solution, which facilitated consistent MQD production (Fig. 1a). Yield analysis revealed that the mass of produced MQDs significantly exceeded the starting MXene amount, with yields ranging from approximately 150% to 170%. This larger yield weight percentage was initially attributed to surface functional groups (–OH, –O, –NH) introduced onto the MQD surface during synthesis; however, a quantitative bound on this contribution showed that surface functionalization alone accounts for only a minor fraction of the observed mass gain, with the majority instead attributed to retained bulk species from the TMAOH-assisted synthesis (Section 3.10). These surface terminations nonetheless enhanced the chemical stability and dispersibility of the MQDs.^39,40

3.2. Technoeconomic expenses for the lab-scaled MQD synthesis

The cost analysis of lab-level scaled-up synthesis of MQDs revealed clear insights into the economic implications at varying batch sizes.²¹ As the Ti₃C₂ MXene precursor input increased from 100 mg to 500 mg, the total reagent cost scaled proportionally, evidencing a consistent linear relationship between raw material consumption and cost, from approximately INR 4510 at the smallest scale to over INR 22 550 at the largest. This linear scaling underscored the predictability of material costs, a critical factor for budgeting industrial-scale synthesis. Notably, the primary contributors to expense were HF 40%, DMSO, TMAOH, and the MXene precursor itself, whose costs were comparable and significantly outweighed the minor expenses of ethanol and deionized water, emphasizing the need to optimize the use and sourcing of these chemicals to reduce total costs. Despite the increasing input scale, a decrease in MQD yield percentage was observed, dropping from 167.63% at 100 mg to 149.89% at 500 mg. This trend likely reflects scale-dependent effects such as reduced reaction efficiency, partial substrate loss, or limitations in uniform mixing and energy distribution at larger batch volumes. Although the yield percentage decreased, the absolute amount of MQDs obtained increased, showing that lab-production upscaling is beneficial but requires process optimization to sustain or improve yields. The fixed capital expenditure of INR 2 69 320 across scales highlighted the high initial investment in equipment, reinforcing the importance of maximizing throughput to distribute CapEx over greater output, ultimately decreasing the cost-per-unit of MQDs (Fig. 2b).

Fig. 2. Techno-economic yield and cost analysis for the scale-up synthesis of MQDs. (a) Table summarizing the initial MXene mass, resulting MQD output, and calculated yield percentage of each batch, reflecting the contributions of the retained bulk species (adsorbed TMAOH residues, solvent, and moisture) and, to a lesser extent, the surface functional groups, during quantum dot synthesis and a bar graph illustrating the proportional rise in the absolute MQD yield with increased MXene precursor from 100 to 500 mg, alongside a slight decline in the yield percentage at higher mass due to process-scale limitations and mass transfer inefficiencies. (b) Stacked bar chart and detailed table report of the reagent-wise expenditure (Ti₃AlC₂, HF 48%, DMSO, TMAOH, ethanol, and DI water) and corresponding MQD yield at scaled MXene inputs, demonstrating linearly escalating material and total costs and a decreasing yield trend at larger batches. (c) 3D visualization depicting the cost-yield surface, highlighting the optimal cost efficiency at 100 mg input (INR 26.91/mg) and quality-maximized output at 200 mg input, with the capital expenditure constant across batch sizes, emphasizing the impact of the process throughput on cost-effective production.

The statistical validity of the predictive model was confirmed by the principal component regression results (Fig. 2c). The model was found to exhibit exceptional fit, with an R-squared value of 0.9992 and a low mean squared error (MSE) of 1.6011. Furthermore, the initial PCA step provided crucial evidence regarding the factor influence. 97.49% of the total variance among the eight input features was captured by the first principal component. This result definitively validated the hypothesis that, within this experimental domain, all individual cost factors functioned collectively as a single, combined investment magnitude rather than as separable, independent economic drivers. The analysis established that the MQD yield (mg) was the critical parameter governing the non-linearity and, thus, the optimal point of the cost efficiency response surface. While the total cost was observed to increase linearly with MXene input, the MQD yield exhibited a maximum of 172.55 mg at the 200 mg input level, followed by a subsequent decline at higher scales. This non-proportional decrease in yield for greater investment was the physical mechanism responsible for the sharp increase in the INR per mg response at scale. Two critical optimal conditions were determined from the efficiency surface. The lowest cost efficiency was predicted at the smallest experimental scale of 100 mg MXene input, achieving a minimum cost of INR 26.91/mg. This point represented the highest economic return relative to capital outlay. For process scale-up, where maximizing product quality (MQD yield) is prioritized, the 200 mg MXene input was identified as the optimal condition. This scale achieved the maximum observed yield of 172.55 mg. It was concluded that the 200 mg input represented the most viable quality-maximized scaled optimum. Although the absolute INR per mg value was slightly higher than the 100 mg condition, the 200 mg scale successfully delivered the highest process conversion efficiency, thereby providing the best technical compromise between maintaining high product quality and optimizing the aggregated cost factor. Further scaling was confirmed to be sub-optimal, as it was associated with an unfavourable increase in resource utilization inefficiency. The analysis further revealed that improvements in yield and material efficiency could significantly enhance cost-effectiveness. Potential strategies include refining reaction parameters, adopting continuous or semi-continuous reactor designs to achieve better mass- and heat-transfer, and implementing solvent and reagent recycling,⁴¹ which are especially viable given MQD sensitivity to surface chemistry modifications. Moreover, the synthesis approach must balance maintaining crystallinity and particle size control while avoiding excessive oxidation or degradation during scale-up, which directly impacts functional yield. Overall, this study presented a robust techno-economic framework for scalable MQD synthesis, emphasizing that while material costs dominate expenditure, strategic process engineering and yield enhancement are key drivers to cost reduction.

3.3. XRD analysis

The XRD analysis revealed the structural transformation from the Ti₃AlC₂ MAX phase to 2D Ti₃C₂T_x MXene and ultimately to 0D MQDs, with comprehensive insights into the phase evolution, crystallite sizes, and the effects of TMAOH-driven quantum dot synthesis (Fig. 3a). Initially, the Ti₃AlC₂ MAX phase was confirmed by the appearance of characteristic sharp diffraction peaks at specific 2θ values, including (002), (004), (008), (100), (105), (106), (110), (107), and (113). The sharpness and intensity of these peaks indicated a high degree of crystallinity and orderly layered structure in the MAX phase. Upon etching and delamination of the MAX phase, the layered structure of Ti₃C₂T_x MXene emerged, as evident by the shift and broadening of the 8.9° to 5.8° (002) peak and the appearance of 14.3° (004) and 24.7° (006) peaks. The disappearance of the (100), (105), and (008) reflections was indicative of the removal of the Al layer, thus confirming successful etching and transformation to the MXene phase.⁴² The broad and less intense nature of the MXene peaks suggested disruption of the long-range order and the formation of few-layered or exfoliated MXene sheets with abundant surface functional groups, such as –OH, –O, and –F.⁴³

Fig. 3. (a) XRD patterns of the MQDs synthesized with varying precursor masses (100–500 mg), 2D MXene nanosheets, and Ti₃AlC₂ MAX phase. The transformation from MAX phase to MXene is indicated by the disappearance of the characteristic MAX reflections and the appearance of the broadened MXene peaks, while further peak broadening and intensity reduction confirm successful scission into MQDs with nanoscale domain sizes and enriched surface termination. (b) FTIR spectra of MQDs (100–500 mg) and MXene nanosheets, demonstrating the consistent functional group signatures of hydroxyl, carbonyl, alkyl, amide, nitrile, fluoride, Ti–C, and Ti–O across all MQD samples. Intensification and diversification of the surface terminations in MQDs compared with MXene are observed, reflecting quantum confinement and chemical modification during synthesis.

The synthesis of MQDs using TMAOH was reflected in the further broadening and decrease in intensity of the XRD peaks for all MQD samples prepared across different precursor MXene loadings (100–500 mg). The predominant reflections, assigned to 13.3° (004), 24.9° (006), 36.7° (101), 37.6° (105), and 54.6° (109), remained consistent across MQD samples.⁴⁴ The broadening of these peaks provided evidence for the significant reduction in particle size and loss of long-range order, typical for quantum dots and nanocrystalline domains. The absence of the (002) and other stacking-related peaks further corroborated the transformation from a 2D layered morphology to a 0D quantum dot structure.^45,46 The calculation of particle sizes was derived using the Scherrer equation,⁴⁷, where D represents the crystallite size in nanometres, K is the shape factor (typically 0.89 for spherical particles), λ is the X-ray wavelength for Cu Kα radiation (0.15406 nm), β is the full width at half maximum (FWHM) of the diffraction peak in radians, and θ is the Bragg angle, which is half of the observed 2θ value. Larger β values and broader peaks imply smaller crystallite sizes, as listed in Table 1. It was observed that the MQDs exhibit crystallite sizes averaged to 2.03 nm for the various precursor amounts; i.e., MQD-100 mg to MQD-500 mg. In contrast, the 2D MXene nanosheets displayed a much larger crystallite size of 8.68 nm. These results confirmed that severe peak broadening in MQDs was directly correlated to their quantum-confined particle size, while MXene sheets retained a much larger, sheet-like domain structure. The strict consistency of this ∼2.03 nm domain length across all processing scales confirms excellent batch-to-batch structural reproducibility during the scale-expansion protocol.

Table 1. Crystallite size calculation using the Scherrer equation from the XRD patterns.

Material	Crystallite size (nm)
MQD-100 mg	1.91
MQD-200 mg	2.31
MQD-300 mg	2.02
MQD-400 mg	2.00
MQD-500 mg	1.93
MXene	8.68

3.4. Functional group characterization using FTIR analysis

FTIR analysis was performed to investigate the surface chemistry and functional groups present in MQDs and their precursor MXene nanosheets (Fig. 3b). Broad absorption bands in the range of 3300–3600 cm⁻¹ were attributed to O–H stretching vibrations, and these were observed in both MXene and MQD samples.⁴⁴ Their presence reflected the incorporation of hydroxyl groups, which likely originated from water and TMOAH, enhancing the hydrophilic character and dispersion stability of the materials. Stronger signals for C O stretching associated with amide or carbonyl groups in the 1600–1700 cm⁻¹ region were found in all samples, but most prominently in MQDs. This enhancement indicated that quantum dot formation, with increased edge exposure and defect density, led to additional oxidation and amide functionalization. Distinct peaks near 2900 cm⁻¹ were assigned to C–H stretching, which appeared more prevalently in MQDs and suggested that surface alkylation occurred during TMAOH-assisted synthesis, with methyl groups passivating the quantum dots. The appearance of N–H bending around 1500 cm⁻¹ in MQDs further supported the incorporation of amine species from TMAOH. The absorption at 1060 cm⁻¹ was attributed to C–N stretching. This feature was prominent in MQDs, indicating that trace surface passivation involving nitrogen-containing groups had taken place, possibly improving the dispersibility and stability of the quantum dots. In the 1200–1350 cm⁻¹ range, C–F stretching vibrations were detected and were more intense in MXene nanosheets, confirming a higher degree of F-termination resulting from the initial HF etching process. In MQDs, competition from other functional groups during subsequent treatments likely reduced this feature. Finally, low-frequency bands at 600–800 cm⁻¹ and 500–700 cm⁻¹ were observed for Ti–C and Ti–O vibrations, respectively.⁴⁸ These were present in both MXene and MQD samples and confirmed the retention of the titanium carbide structure, along with partial oxidation at the edges. Across all scaled MQD preparations, the key FTIR bands remained consistent, demonstrating that scale-up was achieved with reproducible surface functionalization. In MQDs, competition from other functional groups during subsequent treatments likely reduced this feature. Finally, low-frequency bands at 600–800 cm⁻¹ and 500–700 cm⁻¹ were observed for Ti–C and Ti–O vibrations, respectively. These were present in both MXene and MQD samples and confirmed the retention of the titanium carbide structure, along with partial oxidation at the edges. Across all scaled MQD preparations, the key FTIR bands remained identical, demonstrating that scale-up was achieved with highly reproducible surface functionalization chemical profiles.

3.5. Absorbance and optical bandgap calculations using UV-vis spectroscopy

UV-vis absorption spectroscopic and Tauc plot analyses were conducted to gain a detailed understanding of the optical and electronic properties of MQDs compared to their parent MXene nanosheets (Fig. 4a). In the absorption spectra, the MQDs exhibited a distinct absorption peak at 315 nm, while the MXene nanosheets showed an absorption maximum at 234 nm. The enhanced and red-shifted absorption peak in MQDs corresponded to strong quantum confinement effects resulting from the significantly reduced particle size (∼2 nm). This confinement led to a discretization of electronic energy levels, which altered the light absorption characteristics compared to the bulk-like MXene. Additionally, surface passivation and functionalization were considered to contribute to the observed spectral shifts and enhanced absorbance. The optical bandgap energies (E_g) were quantitatively determined by constructing Tauc plots using the relationship: (αhν)ⁿ = A(hν − E_g), where α is the absorption coefficient, hν is the photon energy (calculated as 1240/λ eV nm⁻¹), A is a proportionality constant, E_g is the bandgap energy, and n varies with the type of electronic transition. For MQDs, an exponent n = 2 was used, representing a direct allowed transition, whereas for MXene n = 0.5 corresponded to an indirect allowed transition. From the Tauc plots (Fig. 4b), the bandgap of MQDs was found to be 2.32 eV, significantly larger than that of MXene nanosheets, which was 1.77 eV.^49,50 This substantial widening of the bandgap upon quantum dot formation was attributed jointly to the quantum size effect, where decreasing particle size increases bandgap energy through heightened electron–hole confinement, and to the concurrent change in surface termination chemistry (loss of –F, gain of –O/–OH), confirmed by XPS and FTIR, which is independently known to shift the electronic structure of Ti₃C₂T_x-derived materials. The relative contributions of these two effects could not be fully deconvoluted with the present data. The shift from an indirect bandgap in MXene to a direct bandgap in MQDs was further attributed to changes in electronic structure and symmetry, influenced by nanoscale effects and surface chemistry alterations.^51,52 These findings indicated that the electronic and optical properties of the material were fundamentally changed by the conversion from 2D MXene sheets to 0D MQDs, making the produced quantum dots more suitable for optoelectronic and photonic applications that benefit from direct bandgap materials with tunable energy gaps.

Fig. 4. (a) UV-vis absorbance spectra of MQDs and MXene nanosheets, highlighting a prominent absorption peak at 315 nm for MQDs and 234 nm for MXene. (b) Tauc plots for the determination of optical bandgaps. MQDs exhibit a direct bandgap of 2.32 eV (n = 2), and MXene shows an indirect bandgap of 1.77 eV (n = 0.5). The bandgap widening and transition-type alteration in MQDs are attributed to quantum confinement effects and surface functionalization induced by nanoscale size reduction and synthesis conditions. (c) Raman spectra comparison, showing the well-resolved phonon modes at 150.68, 405.77, and 611 cm⁻¹ in MXene, indicating high crystallinity and preserved layered structure and MQDs revealing broadened and intense features near 154.36 cm⁻¹, reflecting strong phonon confinement and increased disorder from quantum dot formation. Optical microscopy images collected via the integrated Raman system, displaying the morphology of the MXene flakes (d) and the granular, less uniform structure of MQDs (e), visually supporting the spectral evolution from 2D sheets to 0D quantum dots.

3.6. Raman spectroscopy

The Raman spectra of Ti₃C₂ MXene and the MQDs were analysed within the framework of quantum phonon confinement theory to reveal size-dependent vibrational dynamics (Fig. 4c). For MXene, Gaussian deconvolution of the experimental spectrum exhibited sharp, Lorentzian Raman peaks at about 150.68 cm⁻¹, 405.77 cm⁻¹, and 611.21 cm⁻¹, with additional lower-intensity features, matched closely by the Gaussian fitting parameters centred at 145.46, 404.36, and 611.69 cm⁻¹; full width at half maximum (FWHM) of 50.60, 71.38, and 126.33 cm⁻¹; and max amplitudes of 6216, 6079, and 4578 counts, respectively, which were assigned to zone-centre optical phonons (E_g and A_1g modes) characteristic of highly crystalline and extended Ti–C lattice vibrations and 2D lattice symmetry.⁵³ The sharpness and relatively narrow FWHM values signified long phonon coherence lengths and minimal scattering, consistent with the bulk-like phonon dispersion relation ω (q) being probed only near q = 0 (zone centre), as required by Raman selection rules.⁵⁴ The presence of peaks in the 200–400 cm⁻¹ region (such as 200.88 cm⁻¹ and 389.26 cm⁻¹ in MQDs) also suggested vibrations related to termination groups (Ti–O, Ti–OH), typical in the T_x region of Ti₃C₂T_x MXenes.⁵⁵ These peaks were detectable and broadened in MQDs due to enhanced surface area and disorder effects. While the dominant peaks in the spectrum were below 700 cm⁻¹, no graphitic or C–C related D or G bands (1000–1600 cm⁻¹) were prominent, indicating low or negligible carbon impurities or adsorbed aromatic carbonaceous species.^56,57

According to phonon confinement theory, the observed FWHM (Γ) grows as the quantum dot size decreases (L), mathematically expressed as Γ ∝ 1/L.⁵⁸ This parameter was consistent with the onset of phonon confinement, though increased lattice disorder and termination-group vibrations (Ti–O, Ti–OH, noted above at 200–400 cm⁻¹) were also expected to broaden the FWHM, so confinement and surface-chemistry effects could not be isolated from the FWHM trend alone. In contrast, MQDs demonstrated pronounced spectral changes. The dominant Raman peak near 154.36 cm⁻¹ displayed a very high intensity (max amplitude 42 766 counts), but a broader FWHM (31.06 cm⁻¹), indicating much shorter phonon coherence lengths and intense phonon-defect or surface scattering. Additional, often weaker, broadened peaks were observed at 200.88, 389.26, and 621.52 cm⁻¹ with FWHMs of 34.49, 53.20, and 65.57 cm⁻¹, respectively. This spectral broadening, together with the asymmetry observed in peak profiles, is a hallmark of spatial phonon confinement, where the Raman selection rules are relaxed, and phonons with q ≠ 0 make significant contributions to the observed response.^54,59–61 This led to convolution between the Gaussian confinement function and the phonon density of states, resulting in the signature broadened and asymmetric peaks typical of zero-dimensional quantum dots. Furthermore, a redshift in the peak position was noted for MQDs compared to MXene, which is interpreted as a softening of vibrational modes due to size reduction and surface effects. This redshift, consistent with the theoretical relation Δω ∝ 1/L², arises from the perturbation of phonon energies when vibrational modes are confined spatially, which is consistent with phonon confinement theory, although surface-termination-induced lattice softening could not be excluded as a contributing factor.⁶² The observed redshift in the position of the most intense MQD peak (153.06 cm⁻¹ compared to 150.68 cm⁻¹ for MXene) was consistent with a softening of the vibrational modes arising from a combination of confinement and surface-termination-induced disorder.⁶³ The phonon confinement model (PCM) states that spatial confinement in zero-dimensional quantum dots modifies the Raman intensity distribution by allowing phonons with q ≠ 0 to contribute, as modelled by the integral:

5

where C(q) encodes the phonon confinement profile dependent on MQD size, ω(q) represents the phonon dispersion, and Γ denotes the phonon linewidth associated with lifetime broadening. The combined effects of phonon confinement, size reduction, and enhanced surface-termination density in MQDs were considered responsible for the observed modifications in the vibrational dynamics relative to 2D MXene.⁶⁴ Specifically, the MQDs displayed a dominant, broadened Raman peak with increased intensity and FWHM relative to MXene's sharp and multiple well-defined peaks. This broadening, peak asymmetry, and spectral redshift point to increased phonon coherence disruption and relaxation of momentum conservation in the zero-dimensional quantum dots. These vibrational changes imply stronger electron-phonon coupling and an altered electronic band structure within MQDs, which is typically associated with greater excitonic confinement and discrete energy levels. The broader phonon density of states and defect-enhanced scattering evidenced by Raman provide indirect evidence that MQDs could facilitate enhanced light absorption, emission, and nonlinear optical interactions, translating to superior optical performance.

3.7. Photoluminescence analysis

To study the electronic band modulations, sub-bandgap transitions, and radiative recombination pathways of the synthesized materials, steady-state photoluminescence (PL) mapping was conducted. For the 2D MXene nanosheets, PL spectra acquired across an excitation wavelength range of λ_ex = 200 nm to 350 nm (deep UV to near-visible) revealed broad, structured emission bands extending from approximately 320 nm (3.88 eV) to over 650 nm (1.91 eV), accompanied by distinct satellite and shoulder features (Fig. 5a–c).⁶⁵ The optical transition energies were quantitatively estimated via:

6

Fig. 5. Representative photoluminescence (PL) emission spectra and time-resolved photoluminescence (TRPL) decay curves of 2D MXene and MQDs under various excitation wavelengths. Steady-state PL emission mapping profiles of (a–c) 2D MXene and (e–g) MQDs, highlighting their broad and structured spectral features, bathochromic shifts, and well-defined shoulder peaks across excitation ranges from deep-UV (200 nm) to visible (350 nm). Corresponding TRPL decay curves of (d) 2D MXene and (h) MXenes, where MQDs exhibit more prolonged and smoother microsecond-scale luminescence compared with 2D MXene, reflecting enhanced radiative efficiency, carrier retention, and quantum confinement effects.

The resolvable electronic transitions for the 2D sheets spanned from 3.88 eV (320 nm) at the highest-energy excitation down to 2.21 eV (560 nm) along the lower-energy emission tail. While pristine bulk MXene is metallic or narrow-semi-metallic with heavily overlapping bands (E_gap < 0.1 eV), the emergence of these wide, tunable optical windows indicates the formation of discrete sub-bandgap states induced by ultrathin chemical exfoliation. Under deep UV excitation (200–290 nm), the high-energy PL emission is governed by quantum-confined interband transitions arising from Ti 3d-C 2p orbital hybridization within the isolated 2D sheets. However, aggressive hydrofluoric (HF) acid etching and subsequent DMSO delamination introduce a highly unpassivated surface termination matrix dominated by mixed functional groups (–F, –O, –OH) and structural edge defects. These missing or broken surface bonds act exactly like the surface defects seen in standard quantum dots, creating a highly scattered network of electronic trap states inside the newly formed bandgap. When electrons drop through these intermediate defect levels, the light emission spreads out heavily (FWHM > 100 nm), leaves long spectral tails, and creates the visible shoulder peaks.

We ran the exact same PL mapping for the 0D MQDs using the same excitation range λ_ex = 200 nm to 350 nm (Fig. 5e–g). The MQDs showed broad, much brighter emission bands stretching from 300 nm (4.13 eV) all the way to 650 nm (1.91 eV). Just like the 2D sheets, the emission peaks red-shifted as the excitation wavelength increased, while dropping in height. Using eqn (1), we found the primary transitions for the MQDs were located between 3.8 eV (330 nm) and 3.1 eV (400 nm), with lower-energy visible emission extending to 1.91 eV (650 nm). This widening of the optical bandgap in MQDs compared to bulk and 2D MXene was consistent with a quantum size effect, though the TRPL results (Table 2) indicated that surface/defect-trap states, rather than confinement alone, dominated the emission pathway: the fractional trap contribution increased from 65.41% in MXene to 82.50% in MQDs, indicating that the majority of the observed PL response originated from surface-termination-derived trap states rather than intrinsic excitonic confinement. This happens because the physical size of the Ti₃C₂ crystal lattice is reduced below its exciton Bohr radius. On top of this size confinement, using TMAOH to break down the sheets cuts the material into isolated 0D dots while accumulating new edges with oxygen groups (–O, –OH). This changes the available electronic states, creating a massive number of surface trap sites. The shifting emission peaks (or PL tailing) happen because the core excitons within the inner Ti–C structure are competing directly with energy cascading down into these surface defect states. The long, low-energy emission tails are highly noticeable in the MQDs because these dense surface oxygen groups introduce extra intermediate steps at the edges or within small oxide sections formed during synthesis.

Table 2. Biexponential TRPL fitting parameters for the 2D MXene nanosheets and 0D MQDs at λ_ex = 340 nm.

Material	y 0	A 1	τ 1 (ns)	A 2	τ 2 (ns)	τ avg (ns)	Fractional trap contribution (%)	R 2
MXene	−7.92	5.79	1546.16	2.46	6872.59	5029.02	65.41	0.94
MQDs	−6.87	4.22	1004.24	3.07	6500.00	5541.20	82.50	0.96

To track how long the excited electrons stay alive before dropping back down, we measured time-resolved photoluminescence (TRPL) decay at an excitation wavelength of λ_ex = 340 nm. The decay curves for both the 2D nanosheets and 0D MQDs showed multi-channel decay (Fig. 5d and h). To model the decay, we linearized the data using natural logarithms (ln(I) vs. time) and fitted the lines using a biexponential decay model:

7

Here, y₀ is the background baseline, A₁ and A₂ are the scaling factors for each decay path, and τ₁ and τ₂ are the fast and slow electron lifetime components, respectively. To find the true overall lifetime of the electrons, we calculated the amplitude-weighted average lifetime (τ_avg):

8

We also calculated exactly what percentage of the total light emission came from the slower, defect-trapped pathway using this ratio:

9

For 2D MXene, the fast decay component was τ₁ = 1546.16 ns and the slow component was τ₂ = 6872.59 ns, giving an average lifetime (τ_avg) of 5029.02 ns. The fast component τ₁ represents electrons dropping quickly from the conduction band into basic internal electron trap states (e-TS). The wide error margin in the slow lifetime of the 2D sheets reflects a messy electronic environment (R² = 0.9364). This happens because the acid etching and layout separation steps create a wide mix of shallow and deep defect depths. Across this wide 2D sheet plane, moving charges run into unpassivated surface traps randomly, with these defect paths making up 65.41% of the total emission timeline.

When changing to the 0D MQD structure, the decay behaviour stabilizes. The fit becomes much tighter (R² = 0.9523) with small error bars (τ₁ = 1004.24 ± 65.27 ns and τ₂ = 6500 ± 855.58 ns). We can understand this change through basic electron wavefunctions. In large 2D layers, electron wavefunctions spread out widely across the surface, hitting a highly unstable mix of defects. Degrading the 2D sheets into 0D particles, MQDs confine the electrons, cutting off long-range loss paths and directing the excited states straight into specialized surface groups (–O, –OH) at the edges.

Crucially, the share of these long-lived trap states jumps to 82.50% in the MQDs, raising the overall average lifetime to a maximum of 5541.20 ns. This shift depends on how the chemistry controls the trap depths. Because the inner Ti–C core handles electrons differently than the outer –O and –OH groups, shrinking the material down to 0D forces the core bandgap (E_gap) to widen due to confinement. Since the energy levels of the surface defects stay fixed by their chemical identity, widening the main gap automatically makes the traps deeper. Because the traps are deeper, dropping into them is highly efficient, while bouncing back out via heat energy becomes difficult. This effectively locks the excited states inside these long-lasting edge states, keeping the material active for a longer time.⁶⁶

Recently, according to the study on trap states of quantum dots reported by Vishnu et al.,⁶⁶ we can hypothesize that when light hits the material, electrons jump from the valence band to the conduction band. In the 2D nanosheet, a large portion of these charges gets lost or scattered randomly across the open flat surface. In contrast, the ultra-fine 0D MQD structure splits the excited states into two very stable paths, i.e., fast trapping of electrons (τ₁ = 1004.24 ns) from the core into the dense layer of shallow active surface sites and secondly, deep trapping of the charges by surface –O, –OH groups. Now, since the charges are physically isolated in QDs, they are safe from immediate loss, resulting in a slow radiative drop (τ₂ = 6500 ns) or a slow thermal escape to the band edge. This long-lasting charge storage might help detect target analytes or biomolecules at the electrode surface. It can be modelled using a modified Perrin equation. When the analyte concentration is low, the connection between the MQD surface traps and the target molecule works like a fixed, tight lock (static binding complex). Here, the raw surface groups (–OH, –O) act as sticky chemical anchors, grabbing the target molecule using static charges or hydrogen bonds. Because the engineered 0D MQDs hold and save excited electrons and holes in these surface defect channels for microseconds, they keep the charges from disappearing early. This gives the dot plenty of time to pass the electron forward straight to the attached target molecule before the charge can slip backward or turn into wasted heat. When the target molecule concentration rises, more molecules crowd into the reaction zone as moving targets, hitting the open surface traps. Because of this, switching from 2D sheets to a 0D charge reservoir gives us the exact physical foundation needed to get strong signals and spot tiny amounts of target compounds in sensor development.^66,67

3.8. DLS particle size analysis

Particle size analysis was conducted to comprehensively characterize the distribution and hydrodynamic behaviour of Ti₃C₂ MXene nanosheets and the corresponding MQDs in dispersion (Fig. 6a). For the MXene sample, a single, sharp peak at 2.015 µm, representing 100% of the particle area, was detected, while the overall hydrodynamic diameter was measured at 32.621 µm. The polydispersity index of 26.1% and the relatively high standard deviation (0.434 µm) indicated that the MXene population exhibited considerable heterogeneity and was susceptible to aggregation, possibly due to van der Waals restacking and incomplete delamination, which are typical for high aspect ratio 2D materials. Absolute mean intensities and baseline values confirmed good measurement quality, with minimal systematic error.^68,69

Fig. 6. (a) Particle size distribution curves and tabulated quantitative results of 2D MXene and MQDs. MXene samples exhibited a single broad peak, larger hydrodynamic diameter, and higher polydispersity, indicating significant aggregation and heterogeneity. In contrast, MQDs displayed a bimodal distribution with reduced diameter and polydispersity, confirming efficient fragmentation and enhanced dispersion stability following chemical processing. Key parameters, i.e., peak sizes, distribution areas, diffusion coefficients, and standard deviations, quantitatively highlight the transition from aggregated 2D MXene layers to well-dispersed nanoscale MQDs. (b) BET surface area analysis comparing the nitrogen adsorption isotherms and summary parameters of 2D MXene and MQDs. MQDs exhibited a markedly higher adsorption volume and a calculated BET surface area of 140.600 m² g⁻¹, indicating a substantial increase in accessible surface compared with MXene (31.786 m² g⁻¹). The enhanced surface area and improved C constant for MQDs demonstrate the effect of quantum dot synthesis in increasing porosity and exposure of active sites, as supported by the corresponding BET model parameters.

In contrast, the MQD sample displayed a distinctly bimodal distribution: one peak at 2.62 µm (46.34% area) and a more prominent peak at 0.3187 µm (53.66% area), corresponding to the presence of smaller quantum dot populations and some aggregates. The hydrodynamic diameter was markedly reduced to 1.8909 µm, and the polydispersity index decreased to 17.2%. Standard deviation values for both observed peaks (1.3407 µm and 0.2430 µm) reflected improved distribution uniformity compared to MXene. The increase in the diffusion coefficient for MQDs (0.259 µm² s⁻¹versus 0.015 µm² s⁻¹ for MXene) provided further evidence of enhanced colloidal stability and mobility, which was attributed to the effectiveness of TMAOH-assisted etching and passivation during synthesis. This process was found to prevent restacking, promote scission, and yield well-dispersed MQDs with richer and more reactive surface chemistry. Overall, the conversion from MXene nanosheets to MQDs resulted in a pronounced shift toward smaller, more uniformly distributed, and less aggregated particles, thus enhancing the physical characteristics necessary for subsequent quantum dot applications in fields requiring stable colloidal dispersions and nanoscale features. The bimodal nature of the MQD sample further suggested the coexistence of primary quantum dot populations and minor aggregates: a consequence of both efficient top-down fragmentation and edge termination during chemical processing. To properly interpret these findings, a clear distinction must be made between the crystallite size, physical TEM size, and hydrodynamic diameter of the MQDs. The XRD crystallite size (∼2.03 nm) and the HR-TEM particle size (averaged at 1.776 nm) both probe the physical, highly crystalline Ti–C metallic core of individual isolated dots. Conversely, the much larger bimodal hydrodynamic values obtained via DLS (centred at 318.7 nm and 2.62 µm) describe the dynamic behaviour of the material in an aqueous matrix. This elevation in DLS size is due to two factors: first, the intense surface functionalization (–OH, –O, and –NH groups) and bound TMAOH counter-ions attract a thick, swelling hydration shell; second, the light-scattering intensity in DLS scales with the sixth power of the particle radius, meaning that even minor, trace populations of loosely aggregated edge-terminated clusters strongly dominate the hydrodynamic signal compared to the highly mobile primary quantum dots.

3.9. Multi-point BET analysis

BET surface area analysis was carried out to compare the specific surface characteristics of MQDs with the parent MXene nanosheets (Fig. 6b). It was observed that the MQDs exhibited markedly higher nitrogen adsorption volumes at all the investigated relative pressures (P/P₀), as shown by the steep increase in the low-pressure range of the isotherm.

The calculated BET surface area for MQDs was determined to be 140.600 m² g⁻¹, which far exceeded the value of 31.786 m² g⁻¹ obtained for MXene nanosheets.⁷⁰ The enhancement in the surface area of the MQDs was attributed to the top-down synthesis route, which fragmented the larger MXene sheets into nanoscale quantum dots. This process created a dramatic increase in accessible surface, not only from dimensional reduction but also from exposure of additional edge and defect sites. It was noted that the elevated C constant for the MQDs (233.420) reflected improved adsorptive capacity and heterogeneous surface energy, likely resulting from abundant functional group terminations and increased porosity introduced during the quantum dot formation. The correlation coefficients for both MQD and MXene measurements were very close to unity (0.999597 for MQDs and 0.998969 for MXene), indicating that the BET model accurately described the adsorption behaviour. Slope and intercept values substantiated the quantitative differences in adsorption affinity and surface accessibility between the two forms, while the intercept for MXene was found to be negative, signifying possible limitations in accessible pore volume due to aggregation or stacking effects. In summary, the conversion of MXene to quantum dots was shown to substantially increase the specific surface area, a finding supported by both experimental adsorption data and model fitting. This enhancement in surface properties is considered highly advantageous for applications requiring high surface reactivity, adsorption capacity, or catalytically active sites, revealing the transformative physical and chemical impact of quantum dot synthesis on 2D MXene materials.

3.10. FE-SEM and HR-TEM imaging

FE-SEM images were captured at magnifications ranging from approximately 100 00× to 1 000 00× to provide detailed surface morphology characterization of the Ti₃AlC₂ MAX phase and Ti₃C₂ MXene nanosheets. These magnifications enabled clear visualization of the large, stacked layered structure of the MAX phase, revealing smooth, micrometre-scale multi-layered plates (Fig. 7a). In the MXene nanosheets, the high-magnification images highlighted exfoliated, stacked nanosheets with reduced thickness and more exposed edges, indicative of successful selective etching and delamination that expanded interlayer spacing to approximately 1 and 2 nanometres (Fig. 7b). The few-layer MXene sheets were also observed to possess a characteristic accordion-like layered morphology.¹¹

Fig. 7. Electron microscopy analysis of the Ti₃AlC₂ MAX phase, Ti₃C₂ MXene, and MQDs. (a) FESEM image of MAX phase showing the stacked layered platelets. (b) FESEM image of MXene displaying the exfoliated, wrinkled nanosheets. (c and d) HR-TEM images under higher magnification resolving the individual MQDs with crystalline lattice fringes. (e) High-resolution TEM image of a single MQD revealing the atomic lattice spacing and confirming its crystalline structure at the nanoscale. (f) SAED pattern of MQDs showing the sharp polycrystalline diffraction rings and supporting the preservation of MXene crystallinity within quantum dots. (g–j) IFFT image processing to calculate the d-spacing of lattice fringes present in MQDs.

HR-TEM was conducted over a broad range of magnifications from around 3 000 00× in low-magnification mode to assess particle dispersion and aggregation at ultrahigh magnifications up to 80 000 00× for resolving atomic lattice fringes within the quantum dots. At lower magnifications (Fig. 7c and d), the MQDs appeared as dispersed clusters of small nanoparticles in the size range of a few nanometres, confirming effective fragmentation of larger MXene sheets. At ultrahigh magnification (Fig. 7e), lattice fringes corresponding to specific crystallographic planes could be directly observed, and the interplanar d-spacing was measured to be about 0.236 nm to 0.351 nm, consistent with the Ti₃C₂ MXene crystal planes, such as (004).⁷¹ Particle sizes were predominantly in the range of 1–2 nm, averaged at 1.776 nm, supported by HR-TEM imagery and correlated with XRD and BET results. The IFFT images (Fig. 7g–j) also suggested that the d-spacing was 0.326 nm.

The SAED pattern was acquired to further elucidate the crystallographic nature of the MQDs. Sharp, well-defined and symmetrically arranged diffraction spots forming hexagonal rings were observed, confirming the high crystallinity and hexagonal lattice structure characteristic of the MQD sample (Fig. 7f). The ring patterns corresponded to distinct lattice planes of the Ti₃C₂ MXene structure, demonstrating that despite the quantum dot's diminished size, the long-range hexagonally ordered atomic planes of the parent material were retained.⁷² The absence of diffuse amorphous halos indicated minimal structural degradation during synthesis and confirmed that the crystalline quality is retained in the MQDs. The SAED thus confirmed the preservation of crystallinity, which was also supported by the presence of clear lattice fringes in HR-TEM images, suggesting that the MQDs maintained a coherent Ti–C lattice core even after substantial size reduction. The radii (r) were measured from the SAED pattern and found to be 4.359, 8.529, and 11.945 nm⁻¹. The corresponding d-spacing values were calculated by taking the reciprocal of the measured radii using the formula d = 1/r. The d-spacing values were determined to be approximately 0.229 nm, 0.117 nm, and 0.084 nm, respectively, which corresponded closely to the (100), (110), and higher-index hexagonal planes of 2D T_i3C₂ MXene.⁷² Consequently, the collective physical, optical, and computational landscape provides unambiguous verification of true zero-dimensional quantum confinement within the synthesized MQDs rather than simple structural degradation or unconfined fragmentation. The transition from large micrometer-scale MAX phase blocks to expanded, accordion-like 2D sheets, and ultimately to isolated 0D particles, is geometrically pinned by a tightly controlled physical HR-TEM core size averaging 1.776 nm. The preservation of sharp hexagonal SAED spots and clean IFFT lattice fringes (0.326 nm) directly confirms that each nanosphere maintains a highly ordered, crystalline Ti–C core framework. When this crystalline boundary is reduced below its exciton Bohr radius, quantum size effects drive the discretization of continuous energy states, widening the direct allowed bandgap from 1.77 eV to 2.32 eV. This fundamental core-state restructuring shifts the relative position of chemically fixed surface edge terminations (–O, –OH), making surface trap states deeper and more energetically favourable to capture excited carriers, which directly accounts for the extended microsecond-scale TRPL lifetime (5541.20 ns) and bright excitation-dependent photoluminescence. Verified in tandem by first-principles DFT orbital calculations, these converging lines of evidence provide rigorous validation that the chemical scissoring protocol yields highly stable, true zero-dimensional quantum-confined structures.

3.11. XPS analysis

XPS elemental scanning was conducted to analyse the surface chemical composition and bonding environments of Ti₃C₂ MXene nanosheets and the derived MQDs (Fig. 8). The high-resolution X-ray photoelectron spectroscopy (XPS) analyses of 2D MXene nanosheets and MQDs were performed by deconvoluting the core-level spectra using Gaussian line shapes, following the mathematical model:

10

where y₀ is the baseline offset, A is the area under the peak, w is the peak width (FWHM), and x_c denotes the central binding energy of each component.

Fig. 8. XPS elemental scan and high-resolution spectra comparing the Ti₃C₂ MXene nanosheets and MQDs. Survey scans show the presence of the Ti, C, O, and F core levels, with the MXene spectra exhibiting prominent fluorine peaks and MQDs displaying reduced fluorine and intensified oxygen signals. The high-resolution C 1s, O 1s, and Ti 2p spectra illustrate the enhanced surface oxidation and hydroxylation in MQDs, while MXene retains substantial Ti–C and F-related bonds. Deconvolution of the core levels highlights the transition from F-terminations in MXene to dominant –OH and –O surface functionalities in MQDs, confirming the effective chemical transformation during quantum dot synthesis.

In the MXene nanosheets, titanium was found in mixed oxidation states, predominantly existing as Ti–C bonds within the carbide lattice, as well as partially oxidized Ti(iv) species due to surface oxidation.⁷³ The Ti 2p region was resolved into four peaks, confirming a distribution of titanium oxidation states and chemical environments. The prominent peak at x_c = 453.67 ± 0.09 eV was attributed to titanium bonded to carbon within the carbide lattice (Ti–C), evidenced by an area A = 2513.02 ± 168.09 and FWHM w = 3.45 ± 0.20 eV. Peaks at x_c = 457.29 ± 0.004 eV and x_c = 458.19 ± 0.07 eV, with areas A = 7653.70 ± 151.96 and 671.98 ± 270.88, respectively, corresponded to partially oxidized Ti(iv) species, indicative of Ti–O bonding formed by surface oxidation or intrinsic terminations. The Ti 2p_1/2 spin–orbit component was observed at x_c = 463.02 ± 0.01 eV, with A = 6909.02 ± 82.92 and w = 2.72 ± 0.03 eV, matching the expected theoretical separation from the Ti 2p_3/2 peak. The fitting was statistically robust, with an adjusted R-square of 0.99689, ensuring reliable peak assignment. The C 1s region exhibited three deconvoluted peaks. The strongest component at x_c = 283.17 ± 0.01 eV, area A = 3295.52 ± 98.70, and width w = 1.72 ± 0.03 eV was assigned to carbide carbon (C–Ti), revealing the preserved MXene lattice structure. The graphitic or adventitious carbon signal was distinguished at x_c = 284.85 ± 0.14 eV with a sizable area A = 2074.12 ± 148.95 and w = 4.87 ± 0.21 eV, and a minor component at x_c = 279.69 ± 0.08 eV (area A = 226.79 ± 34.04, width w = 1.44 ± 0.19 eV), which likely indicated defect-bound carbon or sub-stoichiometric sites. The fit achieved a high R-square of 0.99229, confirming the reliability of the deconvolution. Oxygen chemistry was elucidated through two primary peaks in the O 1s spectrum. The first, at x_c = 528.51 ± 0.01 eV, area A = 7303.17 ± 270.69, and width w = 1.32 ± 0.02 eV, was attributed to lattice oxygen (Ti–O–Ti), directly reflecting oxide-type terminations arising from exposure or residual etching. A second component at x_c = 529.89 ± 0.05 eV, area A = 8135.16 ± 330.43, and width w = 3.00 ± 0.07 eV corresponded to surface hydroxyl (–OH) functionalities, which are common in MXene materials after etching with HF or during ambient handling. Both species supported strong and broad O terminations at the MXene surface with an R-square of 0.99575. Fluorine was resolved into two primary peaks. Ti–F surface terminations were validated by a peak at x_c = 683.87 ± 0.59 eV, area A = 1426.54 ± 835.39, and width w = 2.53 ± 0.49 eV. A secondary signal at x_c = 682.77 ± 0.03 eV, area A = 2469.98 ± 812.81, and w = 1.69 ± 0.12 eV may be indicative of residual or loosely bound fluorine species remaining after the etching process. The overall fit showed high accuracy, as evidenced by the adjusted R-square of 0.98839.

Upon conversion to MQDs, significant chemical changes were observed.⁷⁴ The fluorine peak intensity decreased markedly, indicating loss or replacement of F-terminations, consistent with the introduction of hydroxyl (–OH) and oxide (–O) surface groups during TMAOH-assisted synthesis. The C 1s spectral envelope was deconvoluted into four Gaussian components, reflecting the diversity of carbon environments in the MQDs. The most intense feature at x_c = 283.11 ± 0.02 eV, area A = 2542.37 ± 72.02, and width w = 1.60 ± 0.02 eV was attributed to carbide carbon (C–Ti), though its relative intensity had decreased compared to that of bulk MXene, indicating significant surface modification during quantum dot formation. A major peak at x_c = 284.66 ± 0.03 eV, with area A = 1275.01 ± 99.45 and w = 1.62 ± 0.08 eV, corresponded to graphitic/adventitious carbon (C–C), commonly found at surfaces exposed to ambient conditions. Significantly, features at higher binding energies x_c = 286.83 ± 0.05 eV, A = 780.41 ± 57.66 and w = 2.22 ± 0.15 eV, and x_c = 289.88 ± 0.33 eV, A = 87.05 ± 29.83, w = 2.47 ± 0.77 eV were assigned to carbon-oxygen functionalities (C–O and O–C O, respectively). The observed increase in oxygenated carbon content compared to precursor MXene was a direct transformation of surface oxidation and functionalization, which occurred during the TMAOH-assisted quantum dot synthesis. The exceptional fit (adjusted R-square 0.99755) confirmed quantitative confidence in these assignments. The O 1s region displayed two distinct and intense peaks, further corroborating extensive surface modification. The first peak, at x_c = 528.13 ± 0.01 eV with A = 7629.32 ± 149.48 and w = 1.40 ± 0.01 eV, was characteristic of lattice oxygen (Ti–O–Ti), while the dominant broad feature at x_c = 530.16 ± 0.02 eV, A = 15571.06 ± 183.53, and w = 3.03 ± 0.03 eV was assigned to Ti–OH and possibly other oxygen-rich surface groups. This significant enhancement in hydroxyl and oxide-related oxygen, paired with high integrated intensities, supported the fact that TMAOH treatment removed F-terminations and facilitated the enrichment of O- and OH- groups at the quantum dot surface. The O 1s fit displayed a high degree of statistical certainty, with an adjusted R-square of 0.99812. The Ti 2p envelope was resolved into three major peaks, reflecting the altered titanium environment post-quantum-dot formation. The peak at x_c = 456.81 ± 0.004 eV, area A = 6477.93 ± 116.92, width w = 1.27 ± 0.01 eV was ascribed to Ti–O or Ti–OH groups, which had markedly higher relative intensity than in bulk MXene, signifying greater oxidation and functionalization. The peak at x_c = 457.45 ± 0.03 eV, A = 4571.55 ± 132.82, w = 3.31 ± 0.07 eV was attributed to partially oxidized titanium species at the nanosheet/quantum dot surface. Finally, the x_c = 462.55 ± 0.01 eV peak (area A = 5477.96 ± 61.40, width w = 2.63 ± 0.03 eV) represents Ti 2p_1/2 spin–orbit splitting. The increased Ti–O and Ti–OH content, paired with diminished Ti–C signal, was consistent with dominant oxide/hydroxyl surface terminations and reduced carbide connectivity. No significant nitrogen or other heteroatom peaks were detected, suggesting minimal doping or incorporation beyond oxygen and fluorine species. The overall spectral changes confirmed that MQDs possessed surface chemistry dominated by hydroxyl and oxide functional groups, accompanied by reduced fluorine content relative to MXene nanosheets, which is crucial for their enhanced dispersibility and potential application in sensing, catalysis, and photonics. To provide a quantitative bound on the contribution of surface functionalization to the mass yield reported in Section 3.1 (150–172%, Fig. 2), the O 1s/Ti 2p integrated area ratio was compared between MXene and MQDs, increasing from 0.650 to 1.404, a 2.16-fold enrichment in surface oxygen functionality.⁷⁵ Assuming complete monolayer hydroxyl coverage across the BET-measured surface area of the MQDs (140.600 m² g⁻¹, Fig. 6b) at a representative oxide-surface hydroxyl density of 2–5 groups nm⁻², the corresponding mass contribution was calculated as 0.8–2.0% of the total MQD mass, which is an upper bound, since it assumes saturation coverage across the entire measured surface. FTIR analysis (Fig. 3b) independently corroborated that the retained surface species were not limited to simple hydroxyl termination: prominent C O, C–H, N–H, and C–N bands in the MQD spectra, absent or markedly weaker in the precursor MXene, indicated retention of TMAOH-derived alkylammonium and amide-type residues in addition to the –OH/–O terminations quantified by XPS. As even an upper-bound monolayer functionalization estimate accounted for only ∼1–2% of the total mass, while FTIR confirmed additional nitrogen- and carbon-rich organic residues beyond a surface monolayer, the majority of the 50–70% mass increase reported in Section 3.1 was attributed to physiosorbed TMAOH-derived counter-ions, entrained solvent (DMSO), and residual moisture or reagent salts, rather than to surface functional groups alone.

3.12. AFM analysis of modified electrodes

The surface morphology and roughness characteristics of the bare SPCE, MXene/SPCE, and MQD/SPCE electrodes were quantitatively analysed using AFM before and after electrochemical characterization over a constant scanned area of 4.031 µm² (Table 3).⁷⁶ Before electrochemical cycling, the bare SPCE exhibited an average roughness (S_a) of 39.231 nm and root mean square roughness (S_q) of 48.665 nm, indicative of a relatively smooth and homogeneous surface. In contrast, the MXene-modified SPCE showed a substantial increase in roughness, with S_a and S_q values of 120.36 nm and 158.79 nm, respectively. This high roughness was attributed to the layered and stacked nature of the MXene flakes deposited on the electrode surface. The MQD-modified SPCE presented intermediate roughness values, with S_a and S_q measured at 36.585 nm and 52.088 nm, respectively, reflecting a more dispersed quantum dot distribution.

Table 3. Quantitative AFM surface roughness parameters of the bare SPCE, MXene/SPCE, and MQD/SPCE electrodes before and after electrochemical characterization.

Parameter	Before electrochemical characterization			After electrochemical characterization
	Bare SPE	MXene SPE	MQD SPE	Bare SPE	MXene SPE	MQD SPE
S a (nm)	39.231	120.36	36.585	21.574	80.844	26.222
S q (nm)	48.665	158.79	52.088	29.099	110.06	32.081
S y (nm)	365.4	2122.4	491.06	376.26	771.45	203.78
S p (nm)	118.66	1.411	244.7	136.66	210.27	120.31
S v (nm)	−246.74	−711.37	−246.36	−239.6	−56.19	−83.478
S m (pm)	−10.028	−10.015	−10.028	−10.021	−10.028	−10.028

The maximum height (S_y) and peak-to-valley metrics further corroborated these findings. MXene/SPCE displayed the most pronounced topography, with an initial S_y of 2122.4 nm and pronounced peaks and valleys characterized by S_p and S_v values of 1.411 nm and −711.37 nm, respectively. MQD/SPCE featured moderate topographical features (S_y = 491.06 nm, S_p = 244.7 nm, S_v = −246.36 nm), while the bare SPCE showed relatively minor vertical variations (S_y = 365.4 nm, S_p = 118.66 nm, S_v = −246.74 nm).

Following electrochemical characterization, all the electrodes exhibited significant morphological changes (Fig. 9). The bare SPCE's roughness decreased, with S_a and S_q reducing to 21.574 nm and 29.099 nm, respectively, indicating surface smoothing, likely caused by adsorption–desorption processes and electrochemical cycling. Notably, MXene/SPCE showed marked surface stabilization; S_a and S_q decreased to 80.844 nm and 110.06 nm, respectively, while S_y reduced significantly to 771.45 nm. The S_p and S_v values also became less extreme (S_p = 210.27 nm, S_v = −56.19 nm), suggesting partial densification or loss of loosely bound MXene flakes due to electrochemical cycling. MQD/SPCE demonstrated similar post-cycling trends, with S_a and S_q falling to 26.222 nm and 32.081 nm, respectively, and S_y decreasing to 203.78 nm. The reduction in peak and valley amplitudes (S_p = 120.31 nm, S_v = −83.48 nm) reflected improved uniformity and redistribution of quantum dots on the electrode surface. The profile mean spacing (S_m) remained nearly constant across all samples and conditions (∼−10 pm), implying that lateral feature distribution was maintained and morphological changes were predominantly vertical. These results demonstrated that while MXene modification initially enhanced roughness through the introduction of large, flaky nanostructures, electrochemical cycling induced significant topographic smoothing and densification, potentially compromising active surface area and electron transfer efficiency. In contrast, MQD electrodes preserved a more stable and uniform nanostructure post-cycling, with quantum dots maintaining a high surface area-to-volume ratio and facilitating efficient charge-transfer pathways. The observed morphological stabilization and superior roughness retention implied that MQD-modified electrodes would exhibit greater reproducibility, sensitivity, and long-term operational stability in electrochemical applications. Hence, it was concluded that MQD modification served as a better electrode strategy than 2D MXene coatings. The MQD/SPCE electrodes were more resilient to morphological degradation under electrochemical stress, preserving essential topographical features that support durable and efficient sensing performance.

Fig. 9. AFM three-dimensional surface topography images of bare SPCE, MXene/SPCE, and MQD/SPCE recorded before (top row) and after (bottom row) electrochemical characterization. The left inset shows a photograph of the three different electrodes before and after electrochemical measurements, illustrating both the macroscopic and nanoscale morphological changes resulting from electrode modification and cycling.

3.13. Electrochemical performance evaluation of modified electrodes

The electrochemical behavior of the bare and nanomaterial-modified SPCEs was probed using the outer-sphere, reversible one-electron redox couple [Fe(CN)₆]^3−/4− (2.5 mM K₃Fe(CN)₆ + 2.5 mM K₄Fe(CN)₆ in 0.1 M KCl). This couple is a standard diagnostic probe for electroactive surface area and interfacial electron-transfer kinetics, and it was not expected to report on any surface-bound faradaic chemistry of the electrode material itself.

The CVs recorded at 0.05 V s⁻¹ revealed a well-defined [Fe(CN)₆]^3−/4− couple at all four electrodes (Fig. 10a), although peak currents and peak-to-peak separations (ΔE_p) differed considerably among them. The bare SPCE produced a broadened, low-current signal with a comparatively large ΔE_p, consistent with sluggish heterogeneous electron transfer (HET) and the modest electroactive area typical of unmodified screen-printed carbon. The MAX Phase/SPCE showed only a marginal increase in peak current, which was attributed to the bulk, non-delaminated layered structure of the MAX phase: residual metallic Al interlayers were assumed to restrict electrolyte access to the inter-laminar Ti-containing planes, so that the area presented to the diffusing [Fe(CN)₆]^3−/4− probe was only slightly greater than that of bare carbon.

Fig. 10. Electrochemical performance and impedance characterization of the modified screen-printed carbon electrodes (SPCE). (a) CV curves of bare SPCE, MAX Phase/SPCE, MXene/SPCE, and MQD/SPCE in 5 mM K₃Fe(CN)₆/0.1 M KCl. (b) Scan-rate-dependent CV plots of MQD/SPCE from 0.01 to 0.20 V s⁻¹. (c) Linear calibration curves of the anodic and cathodic peak currents versus scan rate (v) for MQD/SPCE. Nyquist plots without (d) and with (e) equivalent-circuit simulation overlays, with the insets showing the high-frequency expansions for MXene/SPCE and MQD/SPCE and the unified Randles circuit model R₁ + C₂/(R₂ + W₂) used for data fitting. (f) Bode modulus plots (log |Z| vs. log frequency) and (g) Bode phase plots (−phase angle vs. log frequency) for all four electrode configurations.

The lowest peak current of the series was recorded for the MXene/SPCE, despite the intrinsically metallic character of Ti₃C₂T_x. This result was attributed to restacking of the exfoliated 2D nanosheets during drop-casting, whereby face-to-face aggregation of the flakes collapses the interlayer galleries and reduces the effective electrochemically active area available to the diffusing ferri/ferrocyanide ions, even though the individual sheets remain excellent conductors. This restacking behaviour is well documented for solution-processed 2D MXene films and is consistent with previous reports describing restacked Ti₃C₂T_x as behaving more like a compact, low-porosity film than an open 2D architecture.

By contrast, the highest and sharpest redox couple among the four electrodes (roughly 2.5 to 3 times that of the bare SPCE) was obtained at the MQD/SPCE, together with the smallest ΔE_p. Down-sizing of the MXene sheets into quantum dots was expected to eliminate restacking, generate a high density of exposed edge sites and Ti–O/–OH surface terminations, and yield a more open, porous surface film, all of which would be expected to increase the effective electroactive area and accelerate HET for the diffusing probe. This is consistent with recent reports demonstrating that electron-rich Ti–O terminal activation and Ti–O-enriched MXene surfaces substantially enhance charge-storage performance, further supporting the mechanistic link between oxygen-terminated surface chemistry and improved charge-transfer behaviour.⁷⁷ This response is consistent with the behaviour expected of a high-edge-density, quantum-confined 2D-derived nanomaterial acting as an electron-transfer mediator rather than as the redox-active species itself.

Having identified the MQD/SPCE as the best-performing platform, its kinetic behaviour was examined in more detail through a scan-rate study (Fig. 10b, 0.01–0.2 V s⁻¹). Both anodic and cathodic peak currents increased systematically with increasing scan rate (ν), and E_pa and E_pc diverged progressively at higher scan rates, a pattern typical of quasi-reversible electron transfer for the [Fe(CN)₆]^3−/4− couple at the MQD-modified interface. When I_p was plotted against ν directly, rather than ν^1/2, the best linear fit was obtained (R² = 0.975, anodic; R² = 0.964, cathodic; Fig. 10c). For a purely semi-infinite linear-diffusion process, as described by the Randles–Ševčík relationship, I_p would be expected to scale with ν^1/2 rather than ν. Because ν^1/2 and ν are difficult to distinguish statistically over such a narrow scan-rate window, spanning only a twenty-fold range, and because the MQD film is a porous, thin nanostructured coating rather than a bulk electrode, this near-linear dependence was interpreted as evidence of a thin-layer, restricted-diffusion regime superimposed on normal semi-infinite diffusion. In this regime, the diffusion-layer thickness at these scan rates becomes comparable to the thickness of the porous MQD coating, so that the I_p–ν dependence approaches linearity rather than following the ideal Randles–Ševčík ν^1/2 law. This behaviour has been reported previously for mesoporous and nanostructured film-modified electrodes and further suggests that the MQD layer does not simply sit on top of the SPCE but actively confines and channels the probe's access to the underlying carbon surface, consistent with the high-edge-density, quantum-confined morphology invoked earlier to account for the superior peak currents and lowest R_ct of this electrode. The near-unity ratio and similar magnitude of the anodic and cathodic slopes (2.91 × 10⁻⁴versus 2.41 × 10⁻⁴ A s V⁻¹) confirmed that the [Fe(CN)₆]^3−/4− couple remained chemically reversible and was only kinetically quasi-reversible at the MQD/SPCE surface. To quantitatively assess the possible contribution of non-faradaic (double-layer) charging to the observed scan-rate dependence, the interfacial capacitance obtained independently from EIS fitting (C₂ = 0.328 µF, Table 4) was used to estimate the expected capacitive charging current, i_c = C_dl × ν. Comparison of this predicted capacitive slope with the experimentally observed I_p–ν slopes (2.91 × 10⁻⁴ A s V⁻¹ anodic, 2.41 × 10⁻⁴ A s V⁻¹ cathodic; Fig. 10c) showed that capacitive charging could account for only approximately 0.11–0.14% of the total peak current at the MQD/SPCE across the scan-rate range tested. This confirmed that the near-linear I_p–ν dependence discussed above reflects a thin-layer-confined faradaic process rather than double-layer capacitive charging, consistent with the diffusion (Warburg) element required to fit the impedance response of this electrode.

Table 4. Fitted equivalent circuit diagrams and EIS parameters of the bare SPCE, MXene/SPCE, and MQD/SPCE electrodes, demonstrating the solution resistance, charge-transfer resistance, double-layer capacitance, and diffusion characteristics for comparative electrochemical impedance analysis.

Electrode	R 1 (Ω)	C 2 (µF)	R ct = R2 (Ω)	Warburg σ (Ω s−1/2)
Bare SPCE	107	0.308	8274	11.964
MAX phase/SPCE	120.1	0.206	9678	11.273
MXene/SPCE	139.5	0.198	886.2	1149
MQD/SPCE	118.8	0.328	311.3	1045

The Nyquist plots (Fig. 10d and e) were fitted to a modified Randles equivalent circuit, R₁ + C₂/(R₂ + W₂), where R₁ is the solution/ohmic resistance, C₂ the interfacial (double-layer) capacitance, R₂ the charge-transfer resistance (R_ct) governing [Fe(CN)₆]^3−/4− electron exchange, and W₂ is the Warburg element accounting for low-frequency semi-infinite diffusion of the redox probe to and from the electrode surface.

R _ct proved to be the most diagnostic parameter, as it reports directly on the kinetic barrier to interfacial electron transfer between the electrode and the [Fe(CN)₆]^3−/4− couple. The trend observed, in which the MAX Phase (9678 Ω) was roughly equal to or greater than Bare (8274 Ω), with both being well above MXene (886 Ω), which, in turn, exceeded MQD (311 Ω), indicated that R_ct tracked the degree of Ti₃C₂T_x delamination and downsizing rather than CV peak current. The roughly thirty-fold drop in R_ct from the MAX phase to the MQDs was taken as evidence that the intrinsically metallic conductivity of the Ti₃C₂T_x lattice was progressively unlocked as the material was exfoliated and then fragmented into quantum dots, providing a low-resistance electronic pathway that accelerated electron exchange with the probe once it reached the electrode surface. This interpretation was supported by the correspondingly low Warburg coefficients obtained for the MXene and MQD electrodes (1149 and 1045 Ω s^−1/2, respectively, versus 11 000 to 12 000 Ω s^−1/2 for the Bare and MAX phase electrodes), which pointed to reduced mass-transport impedance for the diffusing probe at these two more open, nanostructured surfaces. R_ct is governed mainly by the intrinsic conductivity and electron-transfer kinetics of the modifying film, whereas I_p is governed by the effective electroactive area accessible to the diffusing probe. The restacked MXene sheets appeared to form a highly conductive but low-porosity, low-accessible-area film, in which electron transfer was fast once the probe reached the interface, but relatively few pathways were available for it to do so. Both limitations were resolved simultaneously by the MQDs, which combine a highly conductive, high-edge-density nanostructure with a porous, non-restacked morphology that maximizes probe accessibility, accounting for the lowest R_ct and the highest I_p observed for this electrode.

The Bode |Z| versus log f plots (Fig. 10f) were consistent with this picture. An extended high-impedance plateau (log |Z| approximately 4.1 to 4.2) was observed at low-to-mid frequency for the Bare and MAX Phase/SPCE, characteristic of charge-transfer-limited behavior, whereas a substantially depressed, flatter |Z| profile (log |Z| approximately 3.0 to 3.2) was obtained for the MXene/SPCE and MQD/SPCE, consistent with their much lower R_ct values. The phase-angle plots (Fig. 10g) reinforced this trend: a broad, high phase-angle maximum (approximately 70° to 72°) at intermediate frequency was recorded for the Bare and MAX Phase/SPCE, approaching ideal capacitive/charge-transfer behavior, whereas markedly damped maxima (approximately 30° to 45°) were obtained for the MXene/SPCE and MQD/SPCE, indicating a more resistive, near-ohmic response dominated by the highly conductive Ti₃C₂T_x-derived network rather than by interfacial capacitive charging.

Based on these observations, it is hypothesized that exfoliation of the MAX phase into 2D MXene and its further top-down fragmentation into MQDs progressively unlocks the intrinsic metallic conductivity of the Ti₃C₂T_x lattice, reflected in the steep drop in R_ct. This benefit was found to translate into an improved faradaic response (I_p) only when the conductive network was also structurally open and non-restacked, a condition met by the MQDs but not by the 2D nanosheet form. Accordingly, the MQD/SPCE was identified as the electrode of choice, combining the lowest charge-transfer and Warburg impedance with the highest and sharpest [Fe(CN)₆]^3−/4− response; that is, the fastest electron-transfer kinetics together with the largest electroactive area among the electrodes tested. This combination is expected to translate directly into improved sensitivity and lower detection limits in downstream sensing applications, since both R_ct and the effective electroactive area are primary determinants of amperometric and voltammetric sensor response.

3.14. DFT analysis of structural and electronic properties of Ti₃C₂ MXene and Ti₃C₂O₂ MQDs

A comprehensive DFT investigation of Ti₃C₂ MXene was undertaken, and a series of numerical results were systematically obtained for the structural and electronic properties. During the structural optimization phase, the total energy was reduced to −385.8447 Ry, while the maximum residual force on any atom was constrained to 0.0289 Ry Bohr⁻¹, and an equilibrium cell stress of −58.77 kbar was established. The optimized cell parameters were determined to be a₁ = (3.5354, 0.0000, 0.0000) Å, a₂ = (−1.7677, 3.1049, −0.0124) Å, and a₃ = (0.0000, 0.0592, 4.2569) Å, thereby yielding a unit cell volume of 46.73 ų and a mass density of 3.83 g cm⁻³. Fractional coordinates of Ti1 (0.3186, 0.6372, 0.7475), Ti2 (0.0105, 0.0211, 0.2422), and C (0.3375, 0.6750, 0.2683) reflected robust structural convergence. Electronic ground-state properties were quantified through the SCF calculations. The convergence threshold was attained within 17 cycles, securing a total electronic energy of −385.8528 Ry. The SCF Fermi energy was determined to be 10.9121 eV, with one-electron, Hartree, exchange–correlation, Ewald, PAW, and smearing energy components found as −97.75 Ry, 58.78 Ry, −41.58 Ry, −168.48 Ry, −136.82 Ry, and −0.0091 Ry, respectively. The NSCF calculations were performed along a set of 100 k-points distributed explicitly along high-symmetry lines in the Brillouin zone, each sampling 18 electronic bands (Fig. 11c). The Fermi level was recalculated and extracted as 10.9028 eV, as evaluated directly from the distribution of occupations and Kohn–Sham band energies. The resultant band energies at the Γ-point spanned from −46.17 eV (deep valence region) to +12.64 eV (low conduction edge), and occupation statistics showcased full (1.0000) filling for all but the bands traversing the Fermi energy, which exhibited fractional occupation values, thereby substantiating a metallic ground state. A band gap (HOMO–LUMO difference) of 0.004 eV was calculated from across all sampled k-points, utilizing a maximum HOMO of 10.899 eV and a minimal LUMO of 10.903 eV, confirming that Ti₃C₂ MXene exists decisively in a gapless, metallic state (Fig. 11b). Numerical eigenvalue slopes (∂E/∂k) derived from the band structure showed that multiple bands crossed the Fermi energy, further supporting the interpretation of metallicity. The SCF charge density was rendered over a 200 × 200 × 100 real-space mesh, corresponding to a spatial resolution of approximately 0.02 Å. Charge maxima of 0.30 e Å⁻³ were established around Ti atoms, while C atoms localized charge densities up to 0.18 e Å⁻³. Interstitial regions displayed significant charge density (0.05–0.10 e Å⁻³), demonstrative of extended electronic delocalization and robust Ti–C hybridization. These charge density profiles were visualized as isosurfaces at 0.05 eÅ⁻³ in VESTA, revealing lobed anisotropy along the a–b plane and highlighting the two-dimensional, layered nature of the material (Fig. 11a). The total DOS, calculated from −50 to +20 eV with 0.01 eV intervals, produced a peak value of more than 2 states per eV at the Fermi energy (Fig. 11d). Spin-polarized calculations revealed that both alpha and beta channels retained finite states at EF, with any asymmetry between spin channels being minimal at EF. The PDOS calculations, carried out using projwfc.x and referencing the same energy window and Fermi energy, showed that the Ti 3d states reached a maximum intensity of ∼15 states per eV near EF, with sharp features primarily in the conduction region, while C 2p states were found mostly below EF, punctuated by deep peaks near −32 eV and a hybridization shoulder approaching EF. Ti 1s/1p and C 1s contributions were restricted to the deep valence region (∼−46 eV and below), and little contribution from these orbitals was detected near the conduction band minimum (Fig. 11e).

Fig. 11. (a) Visualization of the optimized Ti₃C₂ MXene atomic structure and electron charge density obtained from DFT calculations, with the isosurfaces and projected density contours highlighting the distribution of the Ti (blue) and C (red) electron clouds. (b) HOMO–LUMO energy level diagram showing the position of the highest occupied and lowest unoccupied states relative to the Fermi level (10.9028 eV), indicating a metallic character with an effectively vanishing gap. (c) Electronic band structure along the high-symmetry paths in the hexagonal Brillouin zone, displaying multiple bands intersecting the Fermi energy. (d) Spin-resolved total DOS plots with the Fermi level marked at 0 eV, demonstrating a finite DOS at the Fermi level and confirming metallicity. (e) PDOS plot spanning a wide energy window (−50 eV to +20 eV), quantitatively decomposing the contributions from Ti (3d, 1s, and 1p) and C (2p and 1s) orbitals and emphasizing the dominant Ti 3d states near the Fermi level, consistent with robust conduction and d-orbital metallic behaviour.

A thorough DFT examination of Ti₃C₂O₂ MQDs was also conducted, combining geometric optimization, ground-state electronic structure characterization, and quantum confinement analysis. During the variable-cell relaxation phase, the total energy was minimized to −2636.0306 Ry, with the maximum residual force on any atom curtailed to 0.0001 Ry Bohr⁻¹, reflecting robust atomic relaxation. Equilibrium stress in the supercell was minimized, as evidenced by a final average pressure less than 0.20 kbar. The optimized lattice vectors were determined as a₁ = (10.76, 0.07, −0.09) Å, a₂ = (0.07, 12.19, −0.20) Å, and a₃ = (−0.13, −0.24, 9.13) Å, yielding a cell volume of 1197.42 ų and a simulated mass density of 1.15 g cm⁻³. Atomic positions after relaxation shifted in the range of 0.1–0.8 Å for surface titanium and oxygen atoms, indicative of prominent lattice response to quantum size and functionalization effects. Electronic ground-state properties were quantified through SCF calculations, achieving convergence within 34 cycles and securing a total electronic energy of −2637.5697 Ry. The SCF Fermi energy was determined to be 12.9227 eV. Energy component breakdowns established the one-electron term as −590.32 Ry, Hartree as 414.39 Ry, exchange–correlation as −324.18 Ry, Ewald ionic as −1242.59 Ry, and one-center PAW contributions as −894.86 Ry, with a smearing component of −0.0135 Ry, reflecting only minor fractional occupations. The NSCF computations extended to 120 Kohn–Sham bands at the Γ-point, resolving eigenvalues from −45.40 eV (deeply bound valence) to +13.05 eV (unoccupied conduction). Full occupancy (1.0000) was observed below the Fermi level, while a discrete transition occurred at the frontier orbitals: the HOMO (12.8009 eV) and the LUMO (13.0536 eV) established a direct bandgap of 0.2527 eV for the oxygen-terminated Ti₃C₂O₂ cluster (Fig. 12b). This gap is an order of magnitude larger than the metallic 2D MXene gap (<0.01 eV). Because the DFT model incorporated both finite-size (0D) geometry and full oxygen termination simultaneously, the calculated gap reflects the combined influence of quantum confinement and termination chemistry; the present model does not isolate the two contributions, and a bare (unterminated) finite cluster of equivalent size would be required to deconvolve them. Charge density analysis, rendered on a 200 × 200 × 200 real-space mesh, revealed maximum electron density localizations of approximately 0.25 e Å⁻³ around Ti sites and 0.17 e Å⁻³ near O atoms, with peripheral regions displaying reduced densities (<0.10 e Å⁻³), consistent with quantum-dot-scale electron confinement (Fig. 12a). The SCF charge distribution, visualized via VESTA, showed multi-lobed anisotropy and charge transfer between Ti–O and Ti–C networks, essential for stabilizing the MQD electronic structure. Band structure analysis produced purely discrete energy levels, distinguishing MQDs from conventional periodic nanostructures (Fig. 12c). The TDOS calculations exhibited multiple sharp, quantized peaks spread from −45 eV to +13 eV, with the highest intensity between −10 and 0 eV and a distinct vanishing of DOS at the Fermi level, confirming semiconducting character (Fig. 12d). The PDOS attributes most of the states near the gap edge to Ti 3d orbitals (∼13 states per eV at the band edge), with C 2p and O 2p orbitals forming pronounced sub-peaks between −20 and −8 eV (Fig. 12e). These results numerically substantiate the transition from metallicity in 2D Ti₃C₂ MXene, where the Fermi level is intersected by multiple dispersive bands, and the calculated gap is 0.004 eV, to pronounced semiconducting quantum confinement for Ti₃C₂O₂ MQDs. The MQDs demonstrate a robust, direct bandgap, localized charge density, sharp DOS features, and strong orbital hybridization across Ti–C–O networks, confirming their suitability for next-generation optoelectronics, molecular sensing, and quantum nanodevice platforms requiring tunable electronic band structure and enhanced carrier isolation.

Fig. 12. Comprehensive electronic structure analysis of Ti₃C₂O₂ MQDs via DFT. (a) 3D charge density distribution highlighting the pronounced electron localization on the Ti, C and O atoms, visualized in the vacuum-embedded MQD supercell. (b) HOMO–LUMO diagram indicating the highest occupied molecular orbital (12.8009 eV), lowest unoccupied molecular orbital (13.0536 eV), and a quantum-confinement-induced bandgap of 0.2527 eV, with the Fermi energy marked at 12.9273 eV. (c) Electronic eigenvalue spectrum (band structure) of MQDs at the Γ-point, displaying fully discrete energy levels and demarcating the Fermi energy (dotted line). (d) Total and projected density of states (TDOS, PDOS) plots spanning −25 eV to +5 eV, resolved for the Ti (red), C (blue), and O (green) contributions. (e) Elemental-resolved TDOS and PDOS plots demonstrating that the Ti 3d orbitals dominate near the conduction states, while C 2p and O 2p primarily contribute to the deeper valence bands.

3.15. Machine learning evaluation of sensor performance

In this study, a machine learning (ML) framework was employed as a proof-of-concept computational tool to explore structure–property relationships governing the sensor efficiency of 2D MXene nanosheets and 0D MQDs, using a set of 17 multi-modal physicochemical, optical, and electrochemical descriptors. As described in the methodology, a synthetic training dataset of 100 samples was generated by Gaussian augmentation around experimentally measured or literature-reported central values, because a large, independently replicated experimental dataset was not available. This approach was adopted to enable model training and was not intended to substitute for independent experimental replicates; the implications of this choice for the interpretation of the results are discussed throughout this section. Correlation heatmaps were generated separately for the 2D MXene subset, the 0D MQD subset, and the combined dataset (Fig. 13a–c) to examine structure–property trends within the constructed dataset. In the combined dataset, sensor efficiency was found to be strongly correlated with AFM roughness after electrochemical testing (r = −0.89), UV-vis absorption peak position (r = 0.89), BET surface area (r = 0.89), AFM roughness before electrochemical testing (r = −0.88), CV peak anodic current (r = 0.88), and DFT HOMO–LUMO gap (r = 0.88), among other descriptors. These correlations were interpreted as reflecting the structure of the synthetic dataset as generated, rather than as independently confirmed physical relationships, given that the descriptors were not sampled independently of one another. Feature importance was quantified using the random forest model (Fig. 13d). Following correction of the AFM roughness and DFT HOMO–LUMO gap descriptors to match the experimentally reported values (Table 3; Section 3.13), AFM roughness before electrochemical testing was found to dominate the importance ranking (relative importance ≈ 0.266), reflecting the large, well-separated difference between the MXene (120.4 nm) and MQDs (36.6 nm) values relative to the assumed measurement variability. This was followed by XPS oxygen content (≈0.067), crystallite size (≈0.065), optical bandgap (≈0.060), and DFT HOMO–LUMO gap (≈0.055). PL intensity again contributed least to model predictions (≈0.007). Because AFM roughness before testing was also one of the more tightly constrained descriptors, its outsized importance was interpreted with caution: it was understood to reflect the descriptor's strong class-separating power within the synthetic dataset as constructed, rather than necessarily its true physical dominance among the mechanisms governing sensor efficiency.

Fig. 13. Machine learning-based feature sensitivity screening and endpoint consistency analysis for 2D MXene and 0D MXene quantum dots (MQDs). (a) Pearson correlation heatmap for the 2D MXene synthetically augmented sub-population. (b) Correlation heatmap for the 0D MQD sub-population. (c) Correlation heatmap for the combined dataset, illustrating the structural covariance boundaries and dimensionality-dependent descriptor relationships. (d) Random forest feature importance ranking based on the mean decrease in impurity, identifying the initial surface morphology (AFM roughness before testing), XPS oxygen content, and crystallite size as the primary drivers of model variance within the synthetic envelope. (e) Comparison of the experimental endpoints against the machine learning model projections (random forest, support vector regressor, gradient boosting, and ridge regression) for illustrative consistency tracking, featuring color-coded, non-overlapping annotations of each model-material coordinate pair. All computational profiling and visualizations were implemented within a physics-informed sensitivity workflow, serving as an exploratory tool to map multi-descriptor interaction pathways and guide baseline structure–property interpretations.

To address the risk of overfitting associated with training multi-descriptor models on a limited synthetic sample size, 5-fold cross-validation (CV) was performed on the 100-sample training set, and the resulting scores were compared against in-sample training performance (Table 5). Mean CV R² values of 0.709 (±0.079), 0.693 (±0.104), and 0.663 (±0.070) were obtained for random forest, ridge regression, and gradient boosting, respectively, while a mean CV R² of −0.102 (±0.098) was obtained for support vector regression. These cross-validated values were found to be substantially lower than the corresponding in-sample training R² values of 0.966, 0.809, and 1.000 for random forest, ridge regression, and gradient boosting, respectively, yielding train-CV gaps of 0.257, 0.117, and 0.337, respectively. Gaps exceeding 0.10 in R² were interpreted as indicating a meaningful overfitting risk, and this risk was found to be most pronounced for the tree-based ensemble methods (random forest, gradient boosting), whereas the smaller gap observed for ridge regression was attributed to the regularizing effect of its L2 penalty. The uniformly poor performance of support vector regression across both training and cross-validation (rather than a large gap between the two) was attributed to the unscaled, multi-order-of-magnitude ranges spanned by the descriptor set (for example, EIS R² values on the order of hundreds of ohms versus DFT HOMO–LUMO gap values on the order of 10⁻³–10⁰ eV), rather than to overfitting perse.

Table 5. Five-fold cross-validation and in-sample training performance for all four models.

Model	Mean CV R2	Mean CV RMSE	Train R2 (in-sample)
Random forest	0.709 (±0.079)	0.0205 (±0.0013)	0.966
Gradient boosting	0.663 (±0.070)	0.0223 (±0.0017)	1.000
Ridge regression	0.693 (±0.104)	0.0209 (±0.0015)	0.809
Support vector regressor	−0.102 (±0.098)	0.0407 (±0.0038)	−0.003

Model predictions were additionally compared against two manually curated, experimentally derived samples, one representing 2D MXene and one representing 0D MQDs (Fig. 13e). Predicted sensor efficiency values were found to lie close to the corresponding experimental values across all four models, with the largest deviation observed for support vector regression. However, because these two samples corresponded to the central values used to parameterize the synthetic training distributions, this comparison was treated as an illustrative consistency check rather than as statistically meaningful external validation, and the associated RMSE and R² values were not interpreted as evidence of model generalization. The five-fold cross-validation results described above, together with their associated standard deviations, were instead treated as the primary estimate of model generalization performance and its uncertainty.

Taken together, these results were interpreted as demonstrating the feasibility of a multi-descriptor ML workflow for exploring structure–property trends in MXene- and MQD-based sensor materials, rather than as constituting a fully validated predictive model. The reliance on synthetically augmented training data, the modest and variable cross-validated performance, the limited size and non-independence of the experimental test set, and the fact that eight of the seventeen descriptors were parameterized from illustrative rather than directly measured central values were identified as constraints on the strength of the conclusions that could be drawn from this analysis. Accordingly, the ML-derived feature rankings and predicted trends were presented as preliminary and hypothesis-generating rather than definitive, and external validation using an experimentally measured dataset of adequate sample size, together with direct measurement of the currently illustrative descriptors, was identified as a necessary direction for future work before the predictive framework could be considered validated for practical sensor performance forecasting.

4. Conclusion

This work fundamentally advances our understanding of dimensionality-driven properties in Ti₃C₂T_x MXenes by integrating a wide spectrum of experimental, computational, and data-driven approaches. The successful scale-up synthesis of MQDs with enriched surface functionalization demonstrated superior chemical stability and dispersibility compared to their 2D MXene nanosheet counterparts. Structural and surface characterizations via XRD, FTIR, XPS, Raman, and electron microscopy expansively confirmed the transition from metallic, bulk-like 2D structures to quantum-confined, semiconducting 0D entities with enhanced oxygen-functionalization and reduced fluorine terminations. Optical and vibrational analyses revealed bandgap widening, discrete energy states, and modified phonon dynamics arising from the combined influence of quantum confinement and surface-termination chemistry, vital for optoelectronic functionality. Electrochemical investigations revealed that MQDs provide enhanced charge transport, active surface area, and morphological stability over MXene nanosheets, backed by AFM surface roughness, Bode impedance spectra, and CV with Randles–Sevcik ECSA analyses. DFT simulations faithfully reproduced the experimental electronic structure evolution, quantifying the emergence of a robust direct bandgap in MQDs versus metallicity in 2D MXenes. The machine learning framework, used as an exploratory sensitivity-screening tool across multiple algorithms, identified crystallite size, surface roughness, and oxygen content as candidate primary drivers, furnishing a preliminary roadmap to guide future experimentally validated material performance optimization. Together, these multidisciplinary insights present a compelling case for MQDs as next-generation nanoarchitectures with tunable electronic, optical, and catalytic properties. This work not only extends our fundamental understanding from atomic-scale theory to device-level applications but also provides a lab-scalable, cost-effective synthesis and characterization paradigm. Such advances are pivotal for future innovations in nanoscale sensors, flexible optoelectronics, and quantum nanodevices, where control over dimensionality and surface chemistry enables unprecedented performance and reliability. The presented platform establishes a new benchmark for the rational design of MXene-based low-dimensional materials, heralding transformative impacts across materials science and nanotechnology domains.

Conflicts of interest

There are no conflicts to declare.

Data availability

The source codes used for computational studies are available in the Github. Any further data will be provided by the authors upon request.