Machine Learning Aided Optimization of P1 Laser Scribing Process on Indium Tin Oxide Substrates

Present study employes a picosecond laser (532 nm) for selective P1 laser scribing on the indium tin oxide (ITO) layer and subsequent fine‐tuning of P1 scribing conditions with machine learning (ML) techniques. Initially, the scribing is performed by varying different laser parameters and further evaluate them via an optical microscope and two probe resistivity measurements. The corresponding scribing width and sheet resistance data are used as input databases for ML analysis. The classification and regression tree (CART)‐based ML analysis revealed that median pulse energy <5.7 μJ insufficient to separate the adjacent scribing regions. While pulse energy >5.7 μJ, APL > 35%, LSO > 46%, and processing speed ≥1250 mm s−1 gives ≥16 μm of scribing width. Further, the decision tree (DT) analysis showed that pulse energy of ≥8.1 μJ, and LSO ≥ 37% are required for electrically isolated lines. The feature importance score suggests that laser fluence and pulse energy determined the scribing width, whereas electrical isolation strongly depends on LSO and processing speed. Finally, the ML achieved conditions experimentally validated and reassessed via scanning electron microscope, and atomic force microscopy aligns well with optical microscope measurements.


Introduction
Solar energy is one of the promising renewable energy source, that can meet the rising energy demands. [1]In recent times, there has been active research on various photovoltaic (PV) materials underway that can transform laboratory-scale device technology (<1 cm 2 area) to larger-area modules exceeding 25 cm 2 with the purpose of commercialization. [2,3]Making large-area solar cells or modules provides a higher surface area that facilitates enhanced photon absorption. [4]It subsequently enhances electricity generation and power output, lowering manufacturing costs.In developing lowcost, highly efficient, and large-area solar modules a laser scribing technique plays a pivotal role. [5]The laser scribing technique offers several advantages such as micrometer-scale precision, high reproducibility, uniformity, and selectivity. [6]Moreover, it also provides a significantly higher processing speed than the conventional mechanical scribing approach.Therefore, a laser scribing technique is widely adopted by industries and laboratories to prepare large-area solar cells and modules.In the modules, efficient interconnection with minimum dead area can be achieved with the laser scribing method. [7]Generally, the solar module is fabricated in three stages with laser scribing techniques, i.e., P1, P2, and P3. [8,9]The typical P1 scribing process includes scribing of the bottom electrode that electrically isolates strips and defines the individual cell area.These P1 boundaries exist with very precise equal spacing, and it is considered a reference for the P2 and P3 scribing. [10]In the case of thin film solar cells, diverse substrates such as indium tin oxide (ITO), fluorine-doped tin oxide (FTO), molybdenum (Mo), and Mo-coated polyimide (PI) are actively used as bottom electrodes. [11]Among them, ITO possesses excellent light transmittance and conductivity, so it is widely employed in different semitransparent, building-integrated photovoltaics (BIPV) and tandem solar cells. [12]ecently, ITO-based substrates extensively used in the fabrication of large-area perovskite and Cu(In,Ga)Se 2 (CIGS)-based semitransparent solar cells as well as modules. [13]However, during the fabrication of perovskite or CIGS-based modules over ITO substrates, it faces several issues due to imperfect or unoptimized P1 scribing conditions. [14,15]During the P1 process on the ITO/Glass substrate, the formation of debris, uplifted boundary edges, and unevenness/cracked scribing edges are usual. [16]t produces a less mechanically stable contact layer which can potentially delaminate the entire cell.The excess pulse energy causes thermal damage near the scribed region and also to the glass substrate. [17,18]It also modifies the electrical properties like sheet resistance of the ITO layer. [19]The imperfect ITO scribing disrupts the electrical isolation within two sub-cells and hampers the electric current flow, resulting in a lower efficiency.More importantly, the high transparency of the layer makes it challenging for the laser to scribe with different laser wavelengths. [18]Therefore, a systematic study should be carried out to achieve fine P1 scribing conditions.But then again performing extensive experiments eventually increases the production cost and human efforts.
Recently the ML-oriented fabrication of solar cells fostered the rapid and efficient development of PV technology. [20]The optimization of optoelectronic properties, exploring new PV materials, stability analysis, and controlling methodologies used different ML algorithms. [21]The integration of the ML approach with experimental methodologies effectively suppresses the experimental time, production cost, and human efforts.The random forest (RF), classification and regression tree (CART), decision tree (DT), gradient boost (GB), and artificial neuron network (ANN) are some of the commonly applied for these tasks. [20]ollowingly similar approaches have also been widely adopted in laser scribing technology.Processes such as image-based characterization of laser scribing quality, [22] optimization of laserscribed microgrooves, [23] laser drilling, and many others use these ML algorithms. [24,25]The integration of high throughput ML models with experimental data allows the efficient optimization of laser scribing conditions that effectively reduce additional experimental efforts by making smart decisions. [26]The CART and DT-based models provide the hidden rules and heuristics required for the target property, thus they have been extensively used in this work to optimize the laser scribing process.
In the present study, the P1 process is performed over an ITO/glass substrate with a picosecond laser, and fine boundary edges besides the electrical isolation within adjacent scribed lines were achieved via ML-assisted fine-tuning of laser conditions.The initial optimization study involves the effect of laser pumping power (LPP), attenuator power level (APL), pulse repetition frequency (PRF), processing speed, and laser spot overlap (LSO) parameters on ITO scribing.Later, the scribed lines were evaluated via an optical microscope and two probe resistivity measurements.The corresponding output data were used as input databases for ML.A detailed investigation of each laser parameter over the ITO/glass substrate and the application of different ML algorithms and their output are discussed below.

Effect of LPP
The LPP parameter controls the voltage applied to the Q-switch that adjusts the laser power.The change in LPP changes the pulse energy, laser fluence, and average power.Accordingly, to study the effect of laser power, the LPP was altered from 30% (minimum) to 100% (maximum).While investigating the effect of the LPP, the laser parameters such as LSO (70%), PRF (300 kHz), spot diameter (20 μm), and APL (100%) were kept constant.Before the laser scribing process, the laser output power was measured using a laser power meter embedded in the equipment.Figure 1a and Table 1 show the impact of different LPP values on laser output parameters.It reveals that the laser pulse energy and fluence increased from 1.31 to 40.38 μJ and 0.41 to 12.85 J cm À2 , respectively when the LPP increased from 30% to 100%.After the pulse energy measurement, the laser scribing was performed with different LPP values over ITO/glass substrates.The corresponding scribing width and quality of the edges of scribed lines were analyzed via an optical microscope shown in Figure 1b.At the lower LPP values, in particular, below 30%, there was no evidence of scribing lines or impressions observed due to low laser fluence. [27]Further, a slight increase in LPP value from 30% to 40% shows a laser impression over the ITO surface (yellow-colored circles).Generally, a laser pulse impression or laser ablation mark is observed, when the laser pulse interacts with the ITO surface and the regions where the material has been removed or altered by the laser.In this process, irradiation can change the texture or appearance of the material's surface and may lead to alterations in color due to laser-induced chemical or structural modifications.At LPP of 50% and 60% clear removal of the ITO layer was observed confirmed by the appearance of dark brown color.It was also later verified with complimentary scanning electron microscope (SEM), and atomic force microscopy (AFM) analysis techniques.The further increase in LPP values up to 100% shows a nearly similar scribing pattern with fine edges.The reduction in yellowcolored boundary edges with increasing LPP values discloses that at higher laser power more fine scribing edges can be achieved.The scribing width measured for scribed lines shows an increase in the width with an increase in LPP. [27]It increased from 14 to 20 μm when the LPP values were maintained from 50% to 100%, respectively, in contrast, lower than 50% no scribing conditions were observed.A nearly uniform scribing width of ≈20 μm was achieved with over 70% LPP values.
Further, to confirm the electrical isolation among adjacent scribed lines the resistance measurements were conducted using a two-probe multi-tester for all laser-scribed lines.It shows that, due to surface damage or laser heat affected zone (HAZ) at the low LPP values of 30%-40%, the sheet resistance values slightly increased from 9 to 18 Ω.When there was successful electrical isolation among the scribed lines was observed, the resistance value was found to be infinite.Above the 50% LPP, the wellelectrically isolated scribed lines can be achieved.In terms of pulse energy and fluence, the threshold energy of 8.78 μJ and fluence of 2.79 J cm À2 are required for ITO scribing.Below these energy levels, the laser pulse is unable to remove the ITO layer.From these observations, it can be said that, as the LPP values drop below 50% the energy for scribing the ITO layer is found to be insufficient, thus it results in the failure of electrical isolation.At the same time, a higher than 50% LPP value shows electrical isolation among the scribed lines.

Attenuation Power Level (APL)
Generally, the APL physically attenuates the power of the laser.In this context, during APL variation from 10% to 100%, the laser parameters such as LSO (70%), PRF (300 kHz), spot diameter (20 μm), and LPP (100%) were kept constant.Setting 100% APL signifies that the laser operates at its full power, whereas a decrease in APL signifies a decrease in laser power.The laser power measurements performed with different APL levels are shown in Figure 2a and Table 2.The alteration in the APL causes a change in laser average power, pulse energy, and fluence.As APL increases from 10% to 100%, the pulse energy and fluence increase linearly from 3.76 to 39.8 μJ and 1.19 to 12.66 J cm À2 , respectively.Thus, it can be inferred that the process energy for scribing can be easily adjusted using APL.The optical images acquired for the corresponding samples are shown in Figure 2b.It shows that at the APL values of 10%, only a laser pulse impression or laser ablation mark was observed as explained earlier.Despite this fact, a slight increase in APL value to 20% shows the successful removal of the ITO layer (green color).Due to the change in the color filter in the optical microscope, optical images exhibit a slight variation in scribed region color.The further increase in APL from 30% to 70% shows a well-scribed ITO layer.The scribing width measured for the lines revealed that as the APL values increased from 10% to 70% the scribing width also increased gradually from 0 μm to ≈17 μm (Table 2).At the high APL values from 80% to 100%, the scribing width remained constant at 18-20 μm.
The two probe measurements were performed for the scribed lines.At low attenuation levels of 10% and 20%, even though laser scribing shows some scribed regions, it was unable to  electrically isolate ITO.The adjacent lines also showed resistance around 16 and 120 Ω, which confirms that the electrical isolation failed.On the contrary, when APL increased from 30% to 100%, the well-scribed part can be easily distinguished via the extended green color.The corresponding resistance measurement showed effective electrical isolation.In the case of pulse energy and fluence, the threshold energy of 7.60 μJ and fluence of 2.41 J cm À2 are required for ITO scribing.Contrary to LPP, the APL provides better control over pulse energy and thereby on the scribing.Therefore, if fine power adjustment is required, the emitted laser energy can be adjusted through the APL.

PRF Variation
In this section, PRF values varied from 300 to 1000 kHz, and during PRF variation parameters such as processing speed (1800 mm s À1 ), APL (100%), spot diameter (20 μm), and LPP (100%) were kept constant.The laser output measurements performed at different PRFs show a change in pulse energy, laser fluence, average power, and LSO (Table 3). [28]An increase in PRF from 300 to 1000 kHz causes a change in LSO value, while a decrease in pulse energy and fluence is shown in Figure 3b.
A formula in Figure 3a proves increases in PRF cause an increase in LSO.As the LSO values increased from 66.66% to 90%, the pulse energy and fluence decreased from 40.45 to 7.34 μJ and 12.87 to 2.33 J cm À2 , respectively.Since the laser pulse emission occurs more frequently before energy is accumulated in the laser resonator, it results in a decrease in pulse energy.The optical images obtained for the different PRFs are shown in Figure 3c.All the samples show successful removal of the ITO layer irrespective of PRF applied.From PRF 300 to 1000 kHz the scribing width slightly decreased from 20 to 16 μm (Table 3).It can be observed that at higher PRF, smoother scribing edges are obtained, while at low PRF, the edges appear relatively broad.In other words, PRF significantly affects the spatial resolution of the laser system.A higher PRF allows for more closely spaced laser pulses, potentially resulting in finer and narrower scribing.While, low LSO situations result in broader edges, indicating that a change in LSO can impact the scribing width.Further, upon careful observation, uneven scribing edges along the scribed lines can be noticed.This may be attributed to a slight issue with beam alignment or the beam shaper.Such irregularities can impact the measured scribing width (inner scribed region), particularly for P2 and P3 processes, with P1 serving as a reference line.Consequently, the dead zone between two adjacent cells may be affected, potentially impacting the device's performance.Therefore, it is necessary to ensure an even scribing profile, especially during P1 scribing.The use of high LSO which provides fine edges and smaller variation in scribing width should be strictly considered; on that note, an average scribing width can stand as a target parameter.The electrical isolation performed via two probe tests also showed all the scribed lines were electrically isolated.It can be explained by, even if at high PRF the pulse energy is low (7.37 μJ) due to high LSO (>90%) the ITO layer gets effectively separated giving electrical isolation.However, it should be noted that the scribed line edges become sharper with increasing LSO and PRF (yellow color).At low PRF 300 kHz, laser scribing edges appear to be wide compared to other conditions.The change in scribing width with respect to applied PRFs was also reported by Cheng et al. [28] From the present experiment, it can be said that the PRF variation affects the LSO and pulse energy.The pulse energies get significantly reduced at high PRF, while it gets compensated with high LSO resulting in good electrical isolation and scribing.Even if better scribing results are obtained at higher PRF, due to the low energy output at high PRF, the lower PRF conditions are preferred for scribing.At high PRF the resultant pulse energy will be too low, so it becomes difficult to increase the pulse energy.On the contrary, the low PRF values enable further fine-tuning and better adjustment of laser power thereby scribing width.

Processing Speed Variation
Laser scribing speed is one of the important parameters, as it controls LSO and the delivery of pulse energy to the surface.In addition, the LSO controls the scribing edge properties, thus it is important to control the processing speed while laser scribing. [29]To explore the effect of processing speed, it was varied from 500 to 4500 mm s À1 and parameters such as PRF (300 kHz), APL (100%), spot diameter (20 μm), and LPP (80%) were kept constant.As shown in Figure 4 and Table 4 the change in the laser processing speed results in a change in LSO only from 91.66% to 25%.The other parameters such as pulse energy and fluence do not change and remain constant ≈36.2 μJ and 11.5 J cm À2 , respectively (Figure 4a).The optical images acquired with an increase in processing speed show a dramatic impact on the laser scribing output.When the processing speed was maintained as low as 500 mm s À1 due to the high pulse energy delivery the ITO layer was removed immediately, at the same time it also caused significant damage to the underlying glass substrate due to a high LSO of 91.66%.It also produces lots of melted ITO debris showing poor scribing properties.It was proposed that at low processing speed due to high thermal diffusion length, it extends the thermal damage up to the glass substrate. [30,31]An increase in processing speed from 1000 to 3500 mm s À1 demonstrated the effective removal of the ITO layer.Since the LSO decreased from 83.33% to 41.66% scribing width decreased from 24 to 14 μm.The analogous behavior of change in scribing width with LSO was reported by Farid et al. [31] In addition to this, Nakajima et al. [32] observed amount of debris decreased with an increase in processing speed.The decrease in the scan speed or processing speed which changes LSO causes a reduction in the delivery of pulse energy at the same spot, thus, the edge damage is eventually suppressed with an increase in processing speed.
The electrical isolation test performed for laser-scribed lines with a processing speed of 500 to 3500 mm s À1 showed good isolation among the lines.At a high processing speed of 4000 to 4500 mm s À1 , LSO decreases from 33.33% to 25%, it shows electrical resistance of 30 and 40 Ω and fails to make isolation.From this experiment, it can be concluded that process speed does not affect other laser output parameters except LSO.Therefore, compared to PRF the processing speed provides more acute control over the LSO while scribing, which effectively determines the scribing edge quality.

Machine Learning Assisted Fining Tuning of Scribing Conditions
The CART and DT are the supervised algorithms that can be effectively used to determine the target properties based on input data.In the case of CART, it is also called an unsupervised algorithm, since it does not require the classification of data and the model itself learns from existing data and provides output. [33]In contrast, DT is a supervised algorithm where we provide preclassified or grouped data, and the model provides output based on the classification/group.In the present work, scribing width and electrical resistance values are considered as target properties for the ML analysis, as scribing width determines the dead zone/effective cell area in the module, and resistance value gives an idea about the interconnection among the cells.The scribing width values analyzed in this work have continuous values, whereas due to abnormality in resistance values, sheet resistance has discreet data.The scribing width mostly varies in the range of 0-20 μm, whereas the resistance values change from 18 to infinity.This makes it difficult to employ similar algorithms for the analysis of scribing width and resistance input data.To overcome this difficulty and improve the ML efficacy we employed the CART algorithm for scribing width as the target property and the DT algorithm for resistance value as the target property.It should be noted that we have used all possible algorithms for both scribing width and resistance value.As the present work mainly focuses on laser processing and not on detailed ML, we emphasized only key algorithm results.In addition, determining the scribing width using an optical microscope provides only initial insights and faces certain limitations.However, a more comprehensive analysis performed through complementary techniques such as SEM, AFM, and others, can yield additional information, including scribing depth, width, and edge measurements.The corresponding data can also be utilized for ML analysis and the revalidation of optical image information, enhancing precision, and reliability.
The CART analysis was performed with scribing width as the target property and all laser parameters such as LPP, LSO, PRF, APL, processing speed, fluence, and pulse energy used as input.The corresponding results obtained from the CART are shown in Figure 5a.As shown in the figure, the first class of nodes has an average scribing width of 16 μm determined from 100% observation.So, the first rule of CART can be read as, if the laser pulse energy was maintained lower than 5.7 μJ (Yes), then the zeroscribing width (no scribing condition) was achieved projected with 9% observations; in contrast, if the laser pulse energy was maintained higher than 5.7 μJ (No) then the average scribing width of 17 μm achieved projected with the 91% of observations.The pulse energy of 5.7 μJ is not an experimentally observed value but it is a median value that comes out after CART excuation.Similarly, the second and third rule suggests, that, if an APL > 35% and LSO > 46% were maintained then a relatively higher scribing width of 18-19 μm can be achieved.Moreover, further CART rules suggest that >18 μm scribing width can be obtained by controlling the processing speed ≥1250 mm s À1 and pulse energy >28 μJ.As observed earlier use of low processing speed can provide a higher scribing width, yet, it can cause potential damage to the glass substrate due to higher fluence, predicted with 6% of observations.Thus, it can be realized that as per the major CART rules if the low pulse energy (<5.7 μJ), APL (<35%), and LSO (<46%) are used then it can not scribe effectively, and the maximum scribing width of 12 μm can be obtained, accordingly relatively higher LSO, APL, and pulse energy are required predicted with more than 74% of observations.The DT analysis was performed to obtain the well-electrically isolated scribed lines.Since the ITO sheet resistance values have discreet data, we broadly classified the resistance values into two groups or classes viz high and low values.The high values represent the laser-scribed lines exhibiting electrical isolation, while the low values represent the scribed lines that do not exhibit electrical isolation.The first DT rule suggests, that if the laser pulse energy was maintained >8.1 μJ (Yes), then the high resistance value (well electrical isolation) can be achieved as projected with 86% observations; in contrast, if the lower pulse energy was maintained <8.1 μJ (No) then low resistance values obtained projected with the 14% of observations (Figure 5b).The second sub-rule suggests along with high pulse energy it is necessary to maintain ≥37% LSO predicted with 80% observations, while the use of lower LSO at high pulse energy results in a low resistance value.Similarly, if the low pulse energy is maintained then it is necessary to use the high LSO values ≥80% predicted with 3% observations, otherwise, use of lower LSO can result in low resistance values.Overall DT rules suggest that the electrically isolated laser scribed line can be achieved in two scenarios: i) high pulse energy with low LSO; and ii) low pulse energy with high LSO.
The critical parameter that mainly influences the laser scribing width and electrical isolation was determined from the feature importance score based on the GB algorithm.It revealed that pulse energy and fluence are the most critical parameters that govern the laser scribing width (Figure 6a), while the LSO, processing speed, fluence, and pulse energy are the critical parameters for the electrical isolation or sheet resistance conditions (Figure 6b).As the CART analysis was performed for the scribing width due to the continuous data, for similar reasons the RF-based prediction model was used to predict the scribing width values.In the RF prediction model, 70% of the observations were used to train the model and 30% for the test.Figure 6c shows that the RF prediction model predicts the scribing width values with utmost accuracy and possesses an adjacent R 2 (Adj.R 2 ) value of 95.14%.Followingly the confusion matrix was obtained for the predictions of resistance values as shown in Figure 6d.The confusion matrix showed among a total of 11 observations tested the confusion matrix predicted all the high resistance values as high and two of the low values as low, thus it exhibits the accuracy = 1 and misclassification factor = 0 demonstrating excellent prediction accuracy.
Overall the CART and DT rules suggested the APL > 35%, LSO > 46% along with the processing speed ≥1250 mm s À1 , and pulse energy ≥28 μJ provide the better scribing quality in the studied case.Thus, for the validation of the ML conditions, we scribed six lines based on DT and CART conditions with a target scribing width of 20 μm and electrically isolated lines.The above-mentioned different laser conditions achieved maintaining the LPP 80% and 90%, LSO-50%, 60%, and 70% at APL-100%, and PRF-300 kHz, which deliver the corresponding pulse energy >28 μJ.The obtained scribing results analyzed via optical microscope and laser properties are shown in Figure 7 and Table 5, respectively.Figure 7 shows that at a low LSO of 50%, the fine edges were not obtained even though the pulse energy >28 μJ, at the same time increase in LSO from 60% and 70% delivered well-electrically isolated scribed lines with a scribing width of 19-20 μm.
We further analyzed the laser-scribed lines using SEM and AFM under optimal conditions of LPP-80%, APL-100%, PRF-300 kHz, and LSO at 50%, 60%, and 70%, with varying processing speed.To minimize the probability of substrate damage, we    scribing.It also verifies that the scribed lines have nearly uniform edges and almost negligible notches along the outer ITO region.The corresponding 3D surface topography is also shown in Figure S2, Supporting Information.At 50% LSO, there are heightened pillar-like structures close to the inner glass/ITOscribed region, representing some unremoved parts from the ITO layer.With an increase in LSO to 60% and 70%, the density of these pillars decreases.At higher LSO values of 60% and 70%, some white spots over the ITO surface are observed, they possibly occurred due to debris generated during the ITO scribing process.To obtain additional information, the line profile for similar scribed lines was also measured along horizontal and vertical directions (Figure 8g-l).The horizontal line profile for LSO 50% shows a two-step profile where the "U"-shaped curve along the edges, and then a nearly flat region at the bottom (Figure 8g).Since ITO has a thickness of ≈150 nm, the corresponding depth matches well with the AFM measurements, confirming the complete removal of the ITO layer.The scribing width measured from the horizontal line scan appears to be ≈17 μm.Further vertical line profile drawn along the scribed region shows a nearly linear profile with an average surface roughness of AE20 nm (Figure 8h).It also confirms that there was no deep damage along the glass substrate region observed.At higher LSO of 60% and 70%, horizontal line profiles show (Figure 8i,k) nearly halfsquared shapes with a flat bottom and a scribing width of ≈18 μm with a sharp depth region of 150-160 nm.The vertical line scan profile shows periodic trough-crest regions in both samples (Figure 8j,l).They have an average height difference of around AE35 to 40 nm and are particularly observed in the spot-overlapped region.It also shows partial damage caused to the substrate up to 40 nm.The trough-crest profile observed at higher LSO relies on the interaction between two successive laser pulses and the ridge formed by the previous pulse after solidification.In particular, the deeper regions are observed where the laser beam has a maximum incident intensity of the subsequent pulse and interacts with the ridge created by the previous pulse.We also tested similar scribing conditions with a relatively higher pulse energy of >36 μJ, and found more consistent scribing width and highly reproducible scribing results.But, apart from electrical resistance and scribing width analysis, factors such as scribing depth, debris, elevated edges, and substrate damage are also important and need to be considered while making a high-efficiency solar module.

Conclusions
The present investigation emphasizes the ML-assisted optimization of P1 laser processing parameters over ITO/glass substrate to ensure precise and effective scribing for module fabrication.The initial optimization studies were performed with different LPP, APL, PRF, and processing speeds.Further, the ML analysis was carried out for scribing width and electrical isolation as target properties.The CART analysis revealed that the lower pulse energy, APL, and LSO fail to separate the adjacent scribing regimes.The DT analysis also showed that electrical isolation can be achieved: i) high pulse energy with low LSO; and ii) low pulse energy with high LSO conditions.The feature importance score unveiled that the laser fluence and pulse energy principally control the scribing width, where the electrical isolation is strongly subject to LSO and processing speed.The optimal ML-based scribing conditions were further revalidated experimentally and assessed with AFM, SEM, and optical matches well with target experimental results.The present study outcomes provide key process parameters to be controlled while laser scribing in the fabrication of various solar cell technologies, such as BIPV, semitransparent, and tandem solar cells contributing to advancements in the field of PV applications.

Experimental Section
Laser System Assembly: Scheme 1 shows, a schematic view of the laser optical system setup used during the experiment, (μ-LAB KOS-NV12pG0FCS from KorTherm Science, S. Korea with their own developed control program -AOC, Ultrafast Laser Controller, and Laser Alignment Tool).It involves the typical optical system with reflecting mirrors that direct the laser beam to the Galvano head.The Galvano head controls the motion of the laser and processing providing effective laser scribing.In this work, a typical 532 nm based diode-pumped solid-state (DPSS) laser (AMT-532 from Advanced Optowave (Nd: YVO4)) was used as a laser source.Table 6 shows the specification of the DPSS laser used in this work.The top-hat profile for the laser beam used in this work is given in Figure S3, Supporting Information.We have used the beam profiler model DC-1545M-GL from the Thor labs to acquire the beam profile.
ITO Cleaning: To perform the laser scribing the commercially available ITO/Glass substrates sized (L Â B Â H) 50 Â 50 Â 1.1 mm, with an ITO layer thickness of about 150 nm purchased from i-TASCO (USA) were used.The ITO samples have an average sheet resistance of 9.0 Ω.Before laser scribing, the substrates were cleaned sequentially in the ultrasonication bath with deionized water, acetone, and ethanol for 15 min and then further dried in the oven.
Laser Processing: The ITO/glass substrates were scribed with a DPSS laser beam, at different PRF, LPP, LSO, and APL in a single-laser pass.To obtain uniform results in this study, a chiller (AOC-300 from Han's Scheme 1. Schematic representation of laser scribing system.laser and DI-water was used as a refrigerant) was used to maintain the temperature of the laser device at 22 °C during all processes.Before the scribing process, the stabilization process was performed for about 40 min to obtain stable laser pulse energy.During the scribing, the preoptimized Z-focus distance was kept, while the sample stage to Galvano head laser distance was maintained at about 25 cm.The corresponding system has a depth of focus (DOF) of around 300 μm.During the scribing process, the suction pump of NANOS (model-NS-1) was used to collect the generated debris and microparticles.After the scribing process, the thin films were separated from the stage and blown with an Ar gas gun that removes residual microparticles and debris.
Characterization Details: To evaluate the quality of laser-scribed lines over the ITO substrates an optical microscope (Olympus STM 6 from Dasan Trading with Mex program), with a magnification range of 50x to 100x, top and bottom LED illumination, and minimum measuring range of 1 μm was used.Generally, it was analyzed at 50x magnification, and when precision analysis is required, such as edge conditions, it was analyzed at 100x magnification.The surface morphology and EDS mapping analysis were performed with SEM (HITACHI: model-S-4800).Further, 2D and 3D surface topography of laser-scribed lines were acquired with AFM (INNOVA-LABRAM HR800) from HORIBA.Attenuator power level (%) 0≈100% Laser pumping power (LPP) (%) 10≈100% Beam diameter (μm) 20 μm Pulse width <12 ps

Figure 1 .
Figure 1.a) Effect of LPP variation on laser average power, fluence, and pulse energy, and b) optical images of laser scribed lines with different LPP from 30% to 100%.

Figure 2 .
Figure 2. a) Effect of attenuation power level (APL) on laser average power, fluence, and pulse energy, and b) optical images of laser scribed lines with different APL from 10% to 100%.

Figure 3 .
Figure 3. a) Laser spot overlap (LSO) schematic diagram, b) effect of pulse repetition frequency (PRF) on laser average power, fluence, and pulse energy, and c) optical images of laser scribed lines with different PRF from 300 to 1000 kHz.

Figure 4 .
Figure 4. a) Effect processing speed on laser average power, fluence, pulse energy, and LSO, and b) change in laser scribing properties with processing speed.

Figure 5 .
Figure 5. a) CART analysis of scribing width and b) DT sheet resistance of ITO.

Figure 6 .
Figure 6.a) Feature importance score for scribing width and b) feature importance score for sheet resistance of ITO, c) RF-based prediction for scribing width, and d) confusion matrix for the sheet resistance of ITO.

Table 1 .
Change in laser output and scribing properties with LPP variation.

Table 2 .
Change in laser output and scribing properties with APL variation.

Table 3 .
Change in laser output and scribing properties with different PRF.

Table 4 .
Change in laser scribing properties with processing speed at 80% LPP.

Table 5 .
Laser scribing properties based on ML conditions.