A flexible display device thin film encapsulation layer thickness online monitoring method

Abstract

The present application relates to the technical field of thin film thickness measurement in flexible display manufacturing process, and relates to a kind of flexible display device thin film encapsulation layer thickness online monitoring method, comprising: establishing multilayer film optical model and training depth neural network to obtain thickness fingerprint model;Pre-calibration flexible substrate warping topography and generate spectral compensation coefficient;Reflectance interferometric spectroscopy is collected on line scanning, after warping compensation, input neural network real-time inversion each layer thickness;Generate thickness distribution chart and close loop feedback to coating process controller.System integrates wide-spectrum reflectance interferometric probe, precision motion platform, industrial computer and OCT submodule.The present application migrates wafer-level precision thickness measurement algorithm to flexible display scene, overcomes substrate deformation and multilayer film signal coupling problem, realizes full-width, high-precision, online thickness monitoring, significantly improves the yield and stability of packaging process.

Description

Technical Field

[0001] This invention relates to the field of thin film thickness measurement technology in the manufacturing process of flexible displays, and more specifically, to a method for online monitoring of the thickness of the thin film encapsulation layer of a flexible display device. Background Technology

[0002] Flexible display devices, such as AMOLED flexible screens, are the core carriers of next-generation information displays. Their manufacturing process shares a high degree of technological similarity with semiconductor wafer processes: both require the construction of micron / nanoscale functional thin film structures on a substrate through processes such as thin film deposition (CVD / PVD / ALD), photolithography, and etching. Especially in the manufacturing stages of thin-film transistor (TFT) backplanes and the preparation stages of thin-film encapsulation layers (TFE), precise thickness control of multilayer dielectric films is required to ensure the electrical and barrier properties of the devices.

[0003] In wafer manufacturing, the thickness of thin films such as gate oxide layers, metal interconnect layers, and passivation layers are key parameters determining chip yield. The industry commonly uses equipment such as broadband reflectance interferometry, ellipsometers, and probe profilometers for high-precision measurement, and mature online / offline inspection standards have been established. However, in the field of flexible displays, especially in the online monitoring of thin-film encapsulation layer thickness, existing technologies have the following significant drawbacks: 1. Currently, production lines widely use offline sampling inspection: after coating is completed, the entire flexible substrate is removed from the vacuum chamber and sent to a dedicated thickness measurement station. This method not only introduces handling and positioning errors, but more importantly, the measurement results are severely delayed and cannot be fed back to the coating equipment in real time, resulting in batch defects when the thickness deviation exceeds the process window.

[0004] 2. Flexible encapsulation layers typically have a sandwich structure of "inorganic layer-organic layer-inorganic layer", with the thickness of each layer ranging from tens of nanometers to several micrometers. Traditional single-wavelength reflection methods or commercial spectroscopic ellipsometers have difficulty accurately decoupling the true thickness of each layer under multilayer film coupling signals, and the optical constants of the PI layer of the flexible substrate fluctuate with the process, further degrading the fitting results.

[0005] 3. Unlike rigid silicon wafers, large-size flexible substrates exhibit deformations such as warping, edge curling, and surface ripples. Conventional thickness measurement probes cannot adjust the focal plane in real time, leading to spot defocusing, spectral distortion of reflection, and decreased measurement repeatability. This makes it difficult to meet the stringent tolerance requirements of packaging processes for single-layer thickness deviations of <±5%.

[0006] 4. Currently, most online thickness measurement systems on production lines are single-point or single-line scanning systems. Completing full coverage of a G4.5 or larger carrier board takes several minutes, which seriously conflicts with the production line cycle time. Thickness uniformity in edge areas and between different panels cannot be effectively monitored.

[0007] In view of the above problems, there is an urgent need for an online monitoring method for the thickness of the thin film encapsulation layer of flexible display devices, in order to overcome the shortcomings of existing technologies such as offline detection lag, inaccurate decoupling of multilayer film thickness, interference from flexible substrate deformation, and low efficiency of full-area measurement. Summary of the Invention

[0008] To address the aforementioned technical problems, a first aspect of the present invention provides a method for online monitoring of the thickness of a thin-film encapsulation layer in a flexible display device, comprising the following specific steps: A multilayer optical model was established, and a dataset corresponding to the theoretical reflection interference spectrum and the thickness of each layer was generated. This dataset was used to train a deep neural network to obtain a thickness fingerprint model from the measured spectrum to the thickness of each layer. The warpage morphology of the flexible substrate is measured in advance, and a mapping database of position, defocus amount, and light intensity compensation coefficient is established. During or after the deposition of the thin film encapsulation layer, a motion platform carries a broadband reflection interferometry probe, which is dynamically focused according to the compensation coefficient to scan the entire area of ​​the flexible substrate and collect the reflection interferometry spectrum at each point. After preprocessing and warp compensation of the collected reflection interference spectrum, it is input into the thickness fingerprint model to realize the thickness of each film layer in real time. The thickness data at each point is mapped to coordinates to generate a thickness distribution map, and the thickness deviation is fed back to the coating process controller in real time to adjust the process parameters.

[0009] Preferably, the established multilayer film optical model includes at least a three-layer structure consisting of an upper inorganic layer, an organic layer, and a lower inorganic layer.

[0010] Preferably, the broadband reflection interferometry uses a composite light source in the 200nm-1700nm band with a spectral resolution ≤1nm.

[0011] Preferably, the deep neural network is a convolutional neural network, a residual network, or a Transformer network, and simulated flexible substrate warping noise and detector noise are added during training.

[0012] Preferably, the process of generating the warpage compensation coefficient includes the following steps: In offline mode, the surface morphology of the flexible substrate is measured using a white light interferometer or a laser profilometer to obtain the defocusing amount at each measurement location. and the reflection spectral intensity under the standard plane By comparison, the compensation coefficient is obtained. The obtained compensation coefficients are used to perform multiplicative compensation on the reflectance spectrum during online measurement; In the formula, This is the defocusing amount, which is the vertical offset between the actual position of the flexible substrate surface and the reference focal plane. The intensity of the reflected interference spectrum measured on the standard reference plane; To measure the defocusing amount of the same standard reflector during offline pre-calibration, The intensity of the reflected spectrum at that time.

[0013] Preferably, the scanning path of the motion platform is S-shaped, Z-shaped, or regionally spiral, and the display unit array is subjected to encrypted scanning of the region of interest, with the total scanning time not exceeding the production line cycle time.

[0014] Preferably, the coating process controller has a built-in PID model, which adjusts the deposition time, gas flow rate or plasma power according to the thickness deviation to form a closed-loop control.

[0015] A second aspect of the present invention provides an online monitoring system for the thickness of a thin film encapsulation layer of a flexible display device, which uses the above-described method for monitoring and includes: a broadband light source, a beam splitting optical path, an autofocus measurement probe, a high-resolution spectrometer, a precision motion platform, an industrial computer, and a process controller; the industrial computer stores a warp compensation database and a pre-trained thickness fingerprint deep neural network. The system also integrates an optical coherence tomography submodule for measuring the optical thickness and actual refractive index of the organic layer, so as to calibrate the optical constants of the organic layer in the thickness fingerprint model in real time.

[0016] Preferably, the autofocus measurement probe has a built-in microlens array and voice coil motor, with a response time of ≤10ms.

[0017] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0018] A fourth aspect of the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program: when the processor executes the computer program, it implements the steps of the above-described online monitoring method for the thickness of the thin film encapsulation layer of a flexible display device.

[0019] Compared with existing technologies: This invention achieves online high-precision measurement of the single-layer thickness of the thin-film encapsulation layer of flexible display devices with a resolution of 0.1 nm and repeatability better than ±0.5% by integrating the wafer-level wide-spectrum reflection interferometry measurement principle with a deep neural network thickness fingerprint model. It employs flexible substrate warpage pre-calibration and multiplicative spectral compensation technology to effectively eliminate spectral distortion caused by substrate deformation, ensuring system stability within a ±500 μm defocus range. Combined with S-shaped scanning of the motion platform and a region of interest densification strategy, it can complete full-width coverage of the G6 generation substrate within 90 seconds, meeting production line cycle time requirements. Thickness deviation... The process capability index Cpk was improved from 0.9 to 1.45 by adjusting the coating process parameters in real time through the PID controller, resulting in a relative increase in product yield of 11.8% and a reduction in the probability of encapsulation barrier failure by more than 70%. In addition, an optional optical coherence tomography submodule can be added to dynamically calibrate the refractive index of the organic layer, reducing the thickness measurement error from ±150nm to ±30nm. This fully realizes the technological leap from offline sampling inspection to online full inspection and from single-point measurement to closed-loop control of flexible display encapsulation layer thickness, significantly improving the consistency, reliability and mass production yield of the encapsulation process. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the drawings without creative effort. In the drawings: Figure 1 This is a flowchart of an online monitoring method for the thickness of a thin-film encapsulation layer in a flexible display device according to an embodiment of the present invention.

[0021] Figure 2 This is a system diagram for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to an embodiment of the present invention. Detailed Implementation

[0022] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0023] As mentioned in the background section, existing technologies face technical obstacles such as flexible substrate deformation, multilayer film signal decoupling, and rapid full-area scanning. This invention aims to provide an online monitoring method and system for the thickness of thin-film encapsulation layers in flexible display devices, overcoming the shortcomings of existing technologies such as offline detection lag, inaccurate decoupling of multilayer film thickness, interference from flexible substrate deformation, and low efficiency in full-area measurement. By integrating the broadband reflection interferometry principle from the field of wafer coating thickness measurement, the flexible substrate warpage pre-compensation algorithm, and a deep learning-based real-time "thickness fingerprint" inversion model, sub-nanometer-level online detection of single-layer encapsulation thickness is achieved, supporting closed-loop control of the coating process.

[0024] Example 1 Figure 1 A flowchart of an online monitoring method for the thickness of a thin-film encapsulation layer in a flexible display device according to an embodiment of the present invention is shown, including the following specific steps: This invention provides a method for online monitoring of the thickness of a thin-film encapsulation layer in a flexible display device, comprising the following specific steps: Step 1: Establish a multilayer film optical model and train a thickness fingerprint neural network; First, a multilayer optical model is established for the typical three-layer structure of the thin-film encapsulation layer of flexible display devices, including an upper inorganic layer, an organic layer, and a lower inorganic layer. Based on Fresnel's formula and the thin-film interference matrix, the interference spectrum reflected from the multilayer film system under perpendicular incidence conditions is calculated.

[0025] Define the characteristic matrix of the j-th membrane as:

[0026] in, For phase thickness, To correct the admittance, for perpendicular incidence , These are the refractive index, extinction coefficient, and physical thickness of the j-th layer, respectively. For wavelength, The angle of refraction is 0 when the incident light is perpendicular.

[0027] The reflectance coefficient of the entire film system is obtained by multiplying the characteristic matrices of all layers, and thus the reflectivity is obtained. Reflection interference spectral intensity ,in This represents the incident light intensity distribution. The thickness of each film layer is set. and optical constants The range of variation, for example: SiN x The layer thickness is 400-600 nm, the acrylate organic layer thickness is 6-10 μm, and the refractive index is typically 1.45-1.55. It is generated using the RCWA or FDTD method. 5A dataset of theoretical reflection interference spectra and corresponding thickness combinations. Each spectrum covers the 200nm-1700nm band, with approximately 1500 sampling points and a resolution ≤1nm.

[0028] In this embodiment, the deep neural network architecture uses a one-dimensional convolutional neural network as the main architecture of the thickness fingerprint model. The network structure is as follows: Input layer: Receives a one-dimensional vector with a length of 1500, corresponding to the preprocessed reflection interference spectrum intensity value, normalized to [0,1].

[0029] Convolutional layer: includes 3 convolutional blocks, each containing: Conv1D, kernel sizes of 11, 7, and 5 respectively, number of channels 64→128→256, batch normalization, ReLU activation, max pooling, pooling size 2, stride 2.

[0030] Flattening layer: Flattens the feature map output by convolution into a one-dimensional vector.

[0031] Fully connected layers: This includes two fully connected layers with 512 and 128 neurons respectively, both using ReLU activation and Dropout=0.3.

[0032] Output layer: The number of neurons equals the number of thickness variables to be measured. For a three-layer structure, the output layer has 3 nodes, each corresponding to the thickness of the inorganic layer. Organic layer thickness Thickness of the lower inorganic layer The activation function has a linear output.

[0033] The Adam optimizer was used during training; the learning rate was initially 0.001 and decreased to 0.9 times the original rate every 50 epochs; the training set / validation set / test set was split into 70% / 15% / 15%.

[0034] Data augmentation: During training, Gaussian noise (30dB signal-to-noise ratio), small baseline drift, and multiplicative light intensity perturbations simulating flexible substrate warping are randomly added to the input spectrum. The multiplicative factor ranges from 0.95 to 1.05 to enhance the model's robustness to real-world conditions. Training stops when the validation set loss does not decrease for 20 consecutive epochs, and the best model is saved. The model training is then considered complete, and subsequent steps are performed using the trained model.

[0035] Step 2: Pre-calibration of flexible substrate warpage morphology and generation of compensation coefficients; Before the coating process, a flexible substrate from the same batch is drawn from the production line and placed on the stage of a white light interferometer or laser profilometer. The entire substrate is scanned at a sampling interval of 1mm × 1mm to obtain the surface height distribution. The average height of the four corners of the substrate is defined as the ideal plane of the reference focal plane. Then each measurement location Defocus amount .

[0036] Subsequently, on the offline calibration device, using the same optical system as the online probe, a standard mirror with a surface flatness better than λ / 20 was measured: first, the probe was accurately focused. Recording spectra Then, a precision displacement stage is used to move the standard mirror along the optical axis to simulate different defocusing amounts. The range is -500μm to 500μm, with a step size of 10μm, and the spectrum is recorded at each defocus position. The compensation coefficient is defined as follows:

[0037] Because different wavelengths of light have different sensitivities to defocusing, It is a two-dimensional array that is related to both wavelength and defocus amount. In practical applications, it will... Discretize the data into integer micrometer steps and construct a lookup table.

[0038] During online measurement, the probe is based on the current coordinates of the motion platform. Obtain the point from the warp mapping database. Then, the corresponding compensation coefficient vector is read from the lookup table. For the raw spectra acquired in real time Perform multiplicative compensation:

[0039] Compensated spectrum It is approximately equal to the spectrum under ideal focusing conditions and can be directly input into the thickness fingerprint model.

[0040] Step 3: Online full-width reflection interferometry spectral acquisition; During the thin-film encapsulation layer deposition process, for example, after a certain layer of coating is completed, the substrate remains in a vacuum chamber or transmission chamber. The XYZ precision motion platform carries a broadband reflection interferometry probe and scans the entire area of ​​the flexible substrate according to a preset scanning path.

[0041] The motion platform adopts a gantry structure, with X-axis travel greater than or equal to the substrate width, Y-axis travel greater than or equal to the substrate length, and the Z-axis used for autofocus. The encoder resolution is 0.1μm. The probe incorporates a voice coil motor-driven autofocus mechanism, which adjusts the autofocus based on the defocus distance at the current coordinate. The lens group is moved in real time, keeping the measurement spot diameter below 50μm and ensuring clear imaging. The probe's sampling frequency can reach 200Hz, and the integration time for each measurement point is 5ms.

[0042] The scanning path is designed in an "S" shape: starting from one corner of the substrate, a line is quickly scanned along the X direction, then a step is taken in the Y direction, and the next line is scanned in reverse, repeating this process to cover the entire board. For the display unit array, the system automatically identifies the area range of each display unit and performs encrypted scanning of the region of interest within the unit: the standard grid step size is 10mm×10mm, while within the effective display area of ​​each display unit, the step size is encrypted to 2mm×2mm; in the cutting area between units, the step size remains at 10mm. This ensures the thickness sampling density of critical areas while keeping the total measurement time within the production line cycle time.

[0043] After acquiring the reflection interference spectrum at each measurement point, the probe automatically moves to the next point. All spectral data and position coordinates are then recorded. The data is packaged and uploaded to the industrial control computer in real time.

[0044] Step 4: Spectral preprocessing and real-time thickness fingerprint inversion; After receiving the raw spectral data, the industrial control computer performs the following preprocessing steps in sequence: (a) Dark current subtraction: Pre-recording the dark spectrum when the detector is completely dark. Subtract from the measured spectrum.

[0045] (b) Normalization: Divide the spectrum after deducting dark current by the reference spectrum to obtain the relative reflectance.

[0046] (c) Wavelength correction: The wavelength axis of the spectrometer is calibrated using known characteristic spectral lines.

[0047] (d) Warp compensation: Perform multiplicative compensation on the spectrum according to the method in step 2 to obtain... .

[0048] (e) Smoothing filtering: Use Savitzky-Golay filter to remove high-frequency noise.

[0049] Preprocessed spectrum The thickness fingerprint deep neural network model is input into the industrial control computer's memory. Since the network's forward computation takes only about 2ms, the system can output the thickness result for the current point before the next measurement point is completed. The output values ​​are three scalars: , , The unit is nm, thus achieving the goal of converting the organic layer into physical thickness through premultiplication of refractive index.

[0050] Step 5: Thickness distribution map generation and closed-loop feedback Each measurement point Coordinates are linked to their thickness values ​​to generate a full-area thickness distribution heatmap. The system automatically calculates the average, standard deviation, maximum, and minimum thickness of each layer, as well as the deviation from the target thickness. When the thickness deviation of a certain area exceeds the preset specification limit, that area is highlighted in red on the heatmap, and an alarm is triggered.

[0051] The processing unit communicates in real-time with the process controller of the coating equipment via Ethernet. The controller has a built-in PID model; its input is the thickness deviation distribution of the current substrate, and its output is the process parameter adjustment for the next substrate. The PID control law is as follows:

[0052] in For thickness deviation, Tuning based on experience to optimize the deposition of SiN by PECVD. x For example, the value can be set as Adjustable process parameters may include: deposition time (±2s), SiH4 / N2 gas flow ratio (±5 sccm), and RF plasma power (±10W). Through closed-loop control, the thickness of the encapsulation layer on the subsequent substrate is ensured to remain stable within the target window, with a process capability index Cpk ≥ 1.33.

[0053] In this embodiment, an optical coherence tomography (OCT) submodule can be used for further calibration. Specifically, before the formal scan, the probe is moved to the organic layer regions at the four corners and center of the substrate, emitting broadband low-coherence light with a center wavelength of 1310 nm and a bandwidth of 100 nm. By measuring the interference signal between the reference arm and the sample arm, the optical thickness of the organic layer is determined. At the same time, the actual refractive index is deduced by using the known physical thickness. This refractive index value is used to update the optical constants of the corresponding organic layer in the thickness fingerprint model, thereby further improving the inversion accuracy and effectively reducing errors.

[0054] The core advantage of this embodiment lies in integrating the mature wide-spectrum reflection interferometry principle in wafer manufacturing with a deep neural network thickness fingerprint model, achieving online high-precision measurement with a single-layer thickness resolution of 0.1nm and repeatability better than ±0.5%. By offline pre-calibrating the warp morphology of the flexible substrate and generating multiplicative spectral compensation coefficients, the defocusing of the light spot and spectral distortion caused by substrate deformation are effectively eliminated, enabling the system to maintain stable measurement within a defocusing range of ±500μm. At the same time, a one-dimensional convolutional neural network is used as the inversion model, and through data augmentation and early stopping strategy training, the thickness calculation only takes 2ms, meeting the requirements of high-speed online measurement. Combined with the S-shaped scanning of the motion platform and the region of interest encryption strategy, full coverage of the G6 generation substrate can be completed within 90 seconds. The built-in PID controller realizes real-time feedback of thickness deviation to adjust the coating process parameters, forming a complete closed-loop control link, which significantly improves the consistency of the packaging layer thickness and the process stability.

[0055] Example 2 To further verify the effectiveness of this solution, a specific case study is provided below for detailed explanation: On a 6th-generation flexible AMOLED mass production line, the glass substrate measures 1500mm × 1850mm, with a PI substrate attached and the low-temperature polycrystalline silicon TFT backplane fabrication completed. The lower inorganic layer SiN in the thin-film encapsulation layer needs further processing. x Target thickness 500nm, organic acrylate layer; target thickness 8μm, top inorganic SiN layer. x Online thickness monitoring was performed on a target thickness of 500nm.

[0056] Implement according to the method in Example 1: S1. Select a PI substrate from the same batch and scan its warpage morphology using a white light interferometer at 1mm steps to generate a defocusing database. Simultaneously, calibrate a compensation coefficient lookup table on a standard mirror. From -500μm to 500μm, in 10μm increments.

[0057] S2. Training the thickness fingerprint network: 200,000 sets of theoretical spectral-thickness data were generated using RCWA; the spectral range was 250nm-1500nm, with 1250 sampling points. A 1D-CNN was constructed and trained for 200 epochs on a GPU server, achieving a validation set MSE of 0.00012, with thickness measured in µm. 2 The trained model is exported as a TensorRT engine and deployed on the Jetson AGX Orin module of an industrial control computer.

[0058] S3. Online Measurement: After the inorganic layer deposition is completed, the motion platform drives an 8-channel parallel measurement probe array to scan the entire substrate in an "S" shaped path. The standard grid step size is 10mm, and within the effective area of ​​48 display units, it is refined to 2mm × 2mm, resulting in approximately 28,000 measurement points. The single-point sampling time is 5ms, the platform movement time is approximately 20ms, and each probe works independently and in parallel, with a total time of approximately 87.5 seconds, less than the production line cycle time of 120 seconds. The probes dynamically adjust focus based on real-time defocusing, and the data is input into the network after spectral compensation.

[0059] S4. Results Output and Feedback: The generated inorganic layer thickness distribution map showed that the thickness in the lower left corner was 485nm too thin, a deviation of -3%. The system automatically sent a warning to the PECVD process controller. The controller extended the deposition time in this area based on the PID output and achieved local compensation by adjusting the substrate scanning speed. The thickness in this area on subsequent substrates was restored to 498nm. The Cpk thickness of the entire batch of 200 wafers increased from 0.91 to 1.45, improving the yield by 11.8%.

[0060] This embodiment verifies the significant engineering effect of the method in Example 1 through practical application on a G6 generation flexible AMOLED mass production line. Specifically, a white light interferometer is used to generate a defocus database by scanning in 1mm step sizes, and a compensation coefficient lookup table is constructed to enable the probe to obtain a clear spectrum under dynamic focusing. RCWA is used to generate 200,000 sets of theoretical spectral-thickness data to train a 1D-CNN network, with a validation set MSE as low as 0.00012. After the model is deployed locally, high-speed inference is achieved. The actual scanning of 48 display units and approximately 28,000 measurement points takes only 85 seconds, less than the production line cycle time of 120 seconds. When the inorganic layer thickness deviation is -3%, the system automatically issues a warning and the PID controller extends the local deposition time to restore the thickness to the target value. The encapsulation layer thickness Cpk of the entire batch of 200 wafers increased from 0.91 to 1.45, and the product yield improved by 11.8%, fully demonstrating the reliability of this method in large-scale mass production and its practical value in improving yield.

[0061] Example 3 like Figure 2 As shown, this embodiment provides an online monitoring system for the thickness of the thin film encapsulation layer of a flexible display device, which uses the method in Embodiment 1 for monitoring. Specifically, it includes: a broadband light source, a beam splitting optical path, an autofocus measurement probe, a high-resolution spectrometer, a precision motion platform, an industrial computer, and a process controller; the industrial computer stores a warp compensation database and a pre-trained thickness fingerprint deep neural network. The system also integrates an optical coherence tomography submodule for measuring the optical thickness and actual refractive index of the organic layer to calibrate the optical constants of the organic layer in the thickness fingerprint model in real time; the autofocus measurement probe has a built-in microlens array and voice coil motor with a response time ≤10ms.

[0062] To further verify the effectiveness of this solution, a specific case is used for detailed explanation below: Similar to Example 2, this case is implemented on a 6th generation flexible AMOLED mass production line. The glass substrate size is 1500mm × 1850mm, with a PI substrate attached and the low-temperature polycrystalline silicon TFT backplane fabrication already completed. The lower inorganic layer SiN in the thin-film encapsulation layer needs to be addressed. x Target thickness 500nm, organic acrylate layer; target thickness 8μm, top inorganic SiN layer. x Online thickness monitoring was performed at a target thickness of 500 nm; the method described in Example 1 was followed. Unlike Example 2, for another batch of organic layer materials with significant refractive index fluctuations, the OCT submodule integrated into the system was activated. Before formal measurement, OCT scans were performed at five points: the four corners and the center of the substrate. The average optical thickness was measured to be 10.4 μm, and the physical thickness was initially estimated at 6.8 μm using reflection interferometry. The actual refractive index was then calculated. =1.529, a deviation of 1.9% from the original set value of 1.50. After updating this refractive index to the material parameter library of the thickness fingerprint model, the error between the organic layer thickness measurement results and the offline SEM cross-sectional verification was reduced from ±150nm to ±28nm, meeting the process requirement of ≤±5%.

[0063] This embodiment provides a complete online monitoring system for the thickness of thin-film encapsulation layers in flexible display devices, and integrates an optical coherence tomography (OCT) submodule based on Embodiment 2. This system not only includes core hardware such as a broadband light source, autofocus probe, high-resolution spectrometer, precision motion platform, and industrial control computer, but also uses the OCT submodule to measure the optical thickness and actual refractive index of the organic layer in real time, dynamically calibrating the material optical constants in the thickness fingerprint model. For organic layer materials with large refractive index fluctuations, OCT scans are performed on multiple points on the substrate before formal measurement to calculate the actual refractive index. After updating the model, the measurement error of the organic layer thickness is compressed from ±150nm to ±28nm, fully meeting the process tolerance requirement of ≤±5%. This solution effectively solves the problem of optical constant drift caused by batch differences in materials, further improving the accuracy and robustness of multilayer film thickness decoupling.

[0064] Example 4 For four- or five-layer thin-film encapsulation layer structures, the steps are similar to those in Example 1, as follows: The core difference between the four- or five-layer encapsulation structure and the aforementioned three-layer structure lies in the dimension of the multilayer film optical model, the number of output nodes of the deep neural network, the size of the training dataset, and the multi-input multi-output characteristics of the closed-loop controller. The remaining steps, such as warp topography pre-calibration and compensation coefficient generation, full-area scanning of the motion platform and encryption of the region of interest, and spectral preprocessing, are completely consistent with the three-layer structure. The same parts will not be elaborated here. Only the differences will be explained in detail below.

[0065] Step 1: Extending the multilayer film optical model and thickness fingerprint neural network For four-layer and five-layer structures, corresponding characteristic matrix multiplication models are established. The thickness ranges for each layer are defined: inorganic layer 400–600 nm, organic layer 6–10 μm; optical constants: refractive index of inorganic layer 1.8–2.0, organic layer 1.45–1.55. A dataset of theoretical reflection interference spectra combined with thicknesses is generated through rigorous coupled-wave analysis (RCWA). A 2×10⁻⁶ dataset is generated for the four-layer structure. 5 Group, five-layer structure generates 3×10 5 Each group has a spectrum covering 200nm~1700nm and 1500 sampling points.

[0066] Deep neural networks still use a one-dimensional convolutional neural network architecture, with the input layer and convolutional layer structure remaining unchanged. The difference lies in: The four-layer structure has 4 output layer nodes, and the five-layer structure has 5 nodes, corresponding to the physical thickness of each film layer. The original two fully connected layers are expanded to 1024 and 256, increasing the model capacity to fit a higher-dimensional parameter space. The dropout rate is increased from 0.3 to 0.4 to prevent overfitting.

[0067] The Adam optimizer is still used during training, with an initial learning rate of 0.001. The loss function is the mean squared error plus a thickness smoothing regularization term. The early stopping strategy stops the training if the learning rate does not decrease after 20 epochs. The data augmentation method remains unchanged.

[0068] The operations, including warp topography pre-calibration, compensation coefficient generation, S-shaped scanning and region of interest encryption of the motion platform, spectral preprocessing, and real-time thickness fingerprint inversion, are completely consistent with those in Examples 1-3, and will not be repeated here. The total scanning time is still controlled within 90 seconds, and the single-point thickness inversion time is ≤3ms.

[0069] Since four- or five-layer structures involve multiple film layers made of different materials, such as two organic layers or three inorganic layers, and each layer is controlled by different coating equipment or different deposition parameters, the controller needs to adjust multiple process parameters simultaneously. This embodiment upgrades the built-in single-input single-output PID model to a multi-input multi-output PID model.

[0070] This embodiment demonstrates that the method of the present invention can be seamlessly extended to four-layer, five-layer, or even more-layer thin-film encapsulation structures. Only the output dimension of the neural network needs to be adjusted, the training dataset expanded, and a MIMO-PID controller adopted. The core optical measurement hardware, warp compensation strategy, and scan path planning do not need to be changed, thus exhibiting good versatility and industrial promotion value.

[0071] Example 5 According to one aspect of the present invention, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the methods provided in the various alternative implementations described above.

[0072] In another aspect, the present invention also provides a computer-readable medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the online monitoring method for the thickness of the thin-film encapsulation layer of the flexible display device described in the above embodiments.

[0073] It should be noted that although several modules or units of the device for performing actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0074] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of the present invention.

[0075] Those skilled in the art will readily conceive of embodiments of the invention upon consideration of the specification and practice of the methods disclosed herein. The invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein.

[0076] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for online monitoring of the thickness of a thin-film encapsulation layer in a flexible display device, characterized in that, The specific steps include the following: A multilayer optical model was established, and a dataset corresponding to the theoretical reflection interference spectrum and the thickness of each layer was generated. This dataset was used to train a deep neural network to obtain a thickness fingerprint model from the measured spectrum to the thickness of each layer. The warpage morphology of the flexible substrate is measured in advance, and a mapping database of position, defocus amount, and light intensity compensation coefficient is established. During or after the deposition of the thin film encapsulation layer, a motion platform carries a broadband reflection interferometry probe, which is dynamically focused according to the compensation coefficient to scan the entire area of ​​the flexible substrate and collect the reflection interferometry spectrum at each point. After preprocessing and warp compensation of the collected reflection interference spectrum, it is input into the thickness fingerprint model to realize the thickness of each film layer in real time. The thickness data at each point is mapped to coordinates to generate a thickness distribution map, and the thickness deviation is fed back to the coating process controller in real time to adjust the process parameters.

2. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, The established multilayer optical model includes at least three layers: an upper inorganic layer, an organic layer, and a lower inorganic layer.

3. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, Broadband reflection interferometry uses a composite light source in the 200nm-1700nm band with a spectral resolution ≤1nm.

4. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, The deep neural network is a convolutional neural network, residual network, or Transformer network. Simulated flexible substrate warping noise and detector noise are added during training.

5. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, The process of generating the warpage compensation coefficient includes the following steps: In offline mode, the surface morphology of the flexible substrate is measured using a white light interferometer or a laser profilometer to obtain the defocusing amount at each measurement location. and the reflection spectral intensity under the standard plane By comparison, the compensation coefficient is obtained. The obtained compensation coefficients are used to perform multiplicative compensation on the reflectance spectrum during online measurement; In the formula, This is the defocusing amount, which is the vertical offset between the actual position of the flexible substrate surface and the reference focal plane. The intensity of the reflected interference spectrum measured on the standard reference plane; To measure the defocusing amount of the same standard reflector during offline pre-calibration, The intensity of the reflected spectrum at that time.

6. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, The scanning path of the motion platform is S-shaped, Z-shaped, or regionally spiral, and the display unit array is subjected to encrypted scanning of the region of interest. The total scanning time does not exceed the production line cycle time.

7. The method for online monitoring of the thickness of the thin-film encapsulation layer of a flexible display device according to claim 1, characterized in that, The coating process controller has a built-in PID model that adjusts the deposition time, gas flow rate, or plasma power according to the thickness deviation, forming a closed-loop control.

8. An online monitoring system for the thickness of a thin-film encapsulation layer in a flexible display device, wherein the monitoring is performed using the method described in any one of claims 1-7, characterized in that, include: Wide-spectrum light source, spectroscopic optical path, autofocus measurement probe, high-resolution spectrometer, precision motion platform, industrial computer and process controller; The industrial control computer stores a warp compensation database and a pre-trained thickness fingerprint deep neural network. The system also integrates an optical coherence tomography submodule for measuring the optical thickness and actual refractive index of the organic layer, so as to calibrate the optical constants of the organic layer in the thickness fingerprint model in real time.

9. The online monitoring system for the thickness of the thin film encapsulation layer of a flexible display device according to claim 8, characterized in that, The autofocus measurement probe incorporates a microlens array and a voice coil motor, with a response time of ≤10ms.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1-7.

Images