An automated ai-based repair process for microleds

Abstract

The present disclosure provides an automated repair of LED substrates system in which an automated repair inspection module is used to determine the quality of incoming processing material an automated repair process module is used to process the processing material, an automated repair post-process module is used to inspect the resultant process material for acceptability, and an automated repair historical module is used to analyze the post process results of the most recent process, against all historical data to find enhanced process changes in the automated repair process module, and an automated repair integration module to update the automated repair process module based upon the results of the automated repair historical module.

Description

[0001] AN AUTOMATED AI-BASED REPAIR PROCESS FOR

[0002] MICROLEDS

[0003] BACKGROUND AND FIELD OF THE INVENTION

[0004] [1] The present disclosure is generally related to an automated repair process of the microLED chips system.

[0005] [2] Currently, addressing the challenge of ensuring contaminant-free microLED arrays before inspection requires innovative solutions due to the intricate dimensions and high sensitivity of these arrays which poses a significant hurdle in the preparation stage. Also, detecting deviations in photoluminescence and electroluminescence ranges that fall outside specifications is a complex task, influenced by process variations or customer requirements, necessitating precise methodologies during the preparation and execution stages. Lastly, achieving robust image analysis for flaw detection in microLED arrays demands a deep understanding of machine vision and artificial intelligence which is crucial for both the preparation and execution stages to ensure effective repair decisions and maintain high-quality display standards. Thus, there is a need in the prior art for an automated repair process of microLED chips system.

[0006] SUMMARY

[0007] [3] The present invention relates to An automated repair method for microLED chips system, comprising, an automated repair inspection module that measures at least one parameter from a first group of parameters and calculates an output of at least one parameter from a second group of parameters to help control the automated repair processing module, an automated repair processing module which inputs the at least one parameter from said second group to control the repair using at least one parameter from a third group of parameters, an automated repair post-processing module which measures at least one parameter from a fourth group of parameters that determines how well the parameters of said first and second group were running, an automated repair historical module which calculates a plurality of group parameters based upon parameters of the first group, second group, third group, and fourth group and an automated repair integration module to update the automated repair processing module based upon the results of the automated repair historical module

[0008] BRIEF DESCRIPTIONS OF THE DRAWINGS

[0009] [4] FIG. 1 : Illustrates an automated Al-based repair process for microLEDs, according to an embodiment.

[0010] [5] FIG. 2: Illustrates a Base Module, according to an embodiment.

[0011] [6] FIG. 3 : Illustrates a Repair Inspection Module, according to an embodiment.

[0012] [7] FIG. 4: Illustrates a Repair Process Module, according to an embodiment.

[0013] [8] FIG. 5: Illustrates a Repair Post-Process Module, according to an embodiment.

[0014] [9] FIG. 6: Illustrates a Repair Historical Module, according to an embodiment.

[0015]

[0010] FIG. 7: Illustrates a Repair Integration Module, according to an embodiment.

[0016]

[0011] FIG. 8: Illustrates a Repair Inspection Database, according to an embodiment.

[0017]

[0012] FIG. 9: Illustrates a Repair Process Database, according to an embodiment.

[0018]

[0013] FIG. 10: Illustrates a Repair Post-Process Database, according to an embodiment.

[0019] DETAILED DESCRIPTION

[0020]

[0014] Embodiments of the present disclosure will be described more fully hereinafter with reference to the accompanying drawings in which like numerals represent like elements throughout the several figures, and in which example embodiments are shown. Embodiments of the claims may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. The examples set forth herein are non-limiting examples and are merely examples among other possible examples.

[0021]

[0015] FIG. 1 illustrates a system for an automated Al-based repair process for microLEDs. This system comprises of a 3rd party repair Al network 102 which integrates pre-quality inspection, processing, post-processing, historical analysis, and continuous improvement in a repair system. The repair inspection module 110 measures essential parameters and calculates outputs to guide the subsequent repair process. The repair process module 112 acts as the core controller, utilizing inputs from the repair inspection module 110 to govern the repair process to properly identify defects that also includes flatness, decide on replacement or repair defective LEDs, etc. The repair post-process module 114 conducts inspections to assess the execution of parameters and identify potential deviations. The repair historical module 116 analyzes data collected across various stages, allowing for continuous learning and pattern recognition. The repair integration module 118 then employs historical insights to update and optimize the process, enabling adaptive improvements over time.

[0022]

[0016] Further, embodiments may include a communication interface 104, which may be a hardware or software component that enables the communication between the 3rd party repair Al network 102 and the repair pre-process 128, repair process 134, and repair post-process 150. The communication interface 104 may include a set of protocols, rules, and standards that define how information is transmitted and received between the devices. The communication interface 104 may be a physical connector, wireless network, or software application and may include components such as drivers, software libraries, and firmware that may be used to control and manage the communication process. In some embodiments, the communication interface 104 may be compatible with USB, Bluetooth, or Wi-Fi. The communication interface 104 may communicate with a network. Examples of networks may include but are not limited to, the Internet, a cloud network, a Wireless Fidelity (Wi-Fi) network, a Wireless Local Area Network (WLAN), a Local Area Network (LAN), a telephone line (POTS), Long Term Evolution (LTE), and / or a Metropolitan Area Network (MAN).

[0023]

[0017] Further, embodiments may include a memory 106, which may include suitable logic, circuitry, and / or interfaces that may be configured to store a machine code and / or a computer program with at least one code section executable by a processor. Examples of implementation of the memory 106 may include, but are not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Hard Disk Drive (HDD), and / or a Secure Digital (SD) card.

[0024]

[0018] Further, embodiments may include a base module 108, which initiates the repair inspection module 110, the repair process module 112, the repair post-process module 114, the repair historical module 116, and the repair integration module 118.

[0025]

[0019] Further, embodiments may include a repair inspection module 110, which begins by being initiated by the base module 108. The repair inspection module 110 connects to the repair pre- process 130. The repair inspection module 110 collects the data from the repair pre-process 130. The repair inspection module 110 performs the inspection algorithm. The repair inspection module 110 sends the determined process data from the inspection algorithm to the repair process module 112. The repair inspection module 110 stores the data in the repair inspection database 120. The repair inspection module 110 returns to the base module 108.

[0026]

[0020] Further, embodiments may include a repair process module 112, which begins by being initiated by the base module 108. The repair process module 112 receives the process data, such as the data on the microLED processed wafer, from the repair inspection module 110. The repair process module 112 connects to the repair process 140. The repair process module 112 collects the data from the repair process 140. The repair process module 112 performs the control algorithm. The repair process module 112 adjusts the control parameters of the repair process 140. The repair process module 112 executes the repair process. The repair process module 112 stores the data in the repair process database 122. The repair process module 112 returns to the base module 108.

[0021] Further, embodiments may include a repair post-process module 114, which begins by being initiated by the base module 108. The repair post-process module 114 connects to the repair post-process 150. The repair post-process module 114 collects the post-process data. The repair post-process module 114 extracts the data from the repair inspection database 120 and the repair process database 122. The repair post-process module 114 performs the post-process algorithm. The repair post-process module 114 performs the post-process inspection. The repair post-process module 114 stores the data in the repair post-process database 124. The repair post-process module 114 returns to the base module 108.

[0027]

[0022] Further, embodiments may include a repair of historical module 116, which begins by being initiated by the base module 108. The repair historical module 116 connects to the repair pre- process 130, the repair process 140, and the repair post-process 150. The repair historical module 116 aggregates the data from the various types of processes and stores the data in the repair network database 126. The repair historical module 116 performs the historical machine learning algorithm on the historical data stored in the repair network database 126. The repair historical module 116 sends the process adjustments to the repair inspection module 110, the repair process module 112, and the repair post-process module 114. The repair historical module 116 returns to the base module 108.

[0028]

[0023] Further, embodiments may include a repair integration module 118, which begins by being initiated by the base module 108. The repair integration module 118 performs the integration machine learning algorithm. The repair integration module 118 connects to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. The repair integration module 118 sends the process adjustments to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. The repair integration module 118 returns to the base module 108.

[0029]

[0024] Further, embodiments may include a repair inspection database 120, which provides an example of the results of the inspection algorithm performed in the repair inspection module 110. The repair inspection database 120 displays an example of variations in photoluminescence and electroluminescence average lot measurements, along with corresponding temperature changes due to normal test process variations. The PL and EL average measurements are centered around 450 nm, a typical value for microLEDs, with variations of ±10%. Temperatures are set around room temperature, 25°C, with process variations observed during testing. The example demonstrates how PL and EL measurements may vary with temperature changes, reflecting the temperature sensitivity of microLED emission properties. This second example illustrates that predictive temperature control systems outputted by the inspection algorithm may achieve temperature regulation within ±1% of a desired set point ensuring optimal conditions for microLED testing and reducing the risk of temperature-induced variability in PL and EL measurements. The second example illustrates tighter control over temperature variations and PL and EL measurements for 10 substrates. The example includes a more controlled range of photoluminescence and electroluminescence measurements to reflect ±1% precision, alongside temperature changes, also maintained within a ±1% variation from 25°C. One substrate is significantly out of specification to illustrate a potential case for repair: The second example demonstrates a situation with no apparent up or down temperature trend from row 11 to row 20, providing an example of the variability across different substrates. Substrate_18 is out of specification for PL and EL measurements which indicates a need for repair or further investigation. In some embodiments, the repair inspection database 120 may contain parameters, optimized parameters, parameter adjustments from the inspection algorithm, etc.

[0030]

[0025] Further, embodiments may include a repair process database 122, which provides an example of the results of the control algorithm performed in the repair process module 112. The repair process database 122 may contain the substrate ID, the Delta E value, the position, the ambient temperature, and the repair decision outputted by the control algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair process 140. For example, a microLED display with minor inconsistencies in brightness across its surface but these variances fall within the acceptable quality specification range for uniformity. The decision would be not to repair as the quality already meets the set standards and intervention could introduce unnecessary risk without significant benefit. Another example may be a display with several microLEDs showing significant color shifts and dead pixels that are visible during standard operation and significantly impact the user experience. These issues fall outside the acceptable quality specification range making repair essential to meet the product's quality standards and ensure customer satisfaction. The control algorithm may decide on repairing or replacing based on the balance between the expected quality improvement and the resources required for repair. The control algorithm may use "Delta E" as a quality metric, which is a standard measure of color difference or accuracy in displays and the control algorithm may use a threshold for perceptible color difference is a Delta E value of 1. Delta E is a metric used to quantify the difference between two colors to understand how perceptibly different they are. Lower Delta E values may indicate less difference and a more uniform color appearance, which is essential for high-quality microLED display performance. This measurement helps identify when specific microLEDs deviate from expected color outputs to guide decisions on whether adjustments or repairs are necessary to meet stringent display quality standards. The control algorithm may use decision tree analysis which is a predictive modeling tool that splits data into branches to represent decision-making paths. It begins at a root node, the initial decision point, and splits into branches based on criteria, leading to leaf nodes representing outcomes. Evaluating how different features affect outcomes, helps in making informed decisions based on patterns found in historical data. For example, in the decision for substrate 18, the control algorithm may evaluate the Delta E value, which may exceed 1.5 and regardless of its position and ambient temperature, the substrate is directed towards the "Repair" decision due to the high Delta E value indicating significant color discrepancy. Another example may be if the control algorithm evaluates substrate 11 in which the Delta E value is above 1.0 but below the 1.5 thresholds, the control algorithm may then consider its position, such as edge, and ambient temperature, and given the lower risk associated with edge positions and the cooler temperature, which might mitigate some thermal effects on Delta E, the decision could lean towards "No Repair", assuming these conditions historically result in acceptable final quality without intervention. In some embodiments, the repair process database 122 may store other parameter data collected or calculated from the repair process module 112, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0031]

[0026] Further, embodiments may include a repair post-process database 124, which provides an example of the results of the post-process algorithm performed in the repair post-process module 114. The example data illustrates the substrate ID, the uniformity deviation, the pattern analysis outputted by the post-process algorithm, the decision to apply the demure algorithm from basic thresholds, and the decision to apply the demure algorithm from the pattern analysis outputted by the post-process algorithm, etc. The first example illustrates various substrates post-repair, where there is “Uniformity Deviation" as a post-repair test result, where higher values indicate worse uniformity across the display. In this example, a threshold for applying the Demura algorithm may be set around 4%, meaning any substrate with a uniformity deviation above this value would be considered for corrective measures. Examples of uniformity measurements for microLEDs postrepair may include luminance uniformity, color uniformity, and white point variation. These measurements may assess how evenly light is distributed across the display and how consistent the color output is from pixel to pixel. High deviations in these areas after repairs may indicate unresolved issues, necessitating further adjustments or corrections, such as applying a demura algorithm to improve display uniformity and overall visual quality. The second example illustrates how the post-process algorithm enhances the precision in identifying regions needing demura correction. Initially, optical scans may not fully capture subtle uniformity deviations, potentially overlooking areas that could benefit from demura. However, once the post-process algorithm is trained on a vast array of data, it may detect intricate patterns associated with visual discrepancies. The post-process algorithm may reveal that regions previously deemed acceptable require demura adjustment, or vice versa, even though a threshold is met.

[0032]

[0027] In some embodiments, the repair post-process database 124 may store other parameter data collected or calculated from the repair post-process module 114, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc.

[0033]

[0028] Further, embodiments may include a repair network database 126, which may contain historical data from the various processes performed by the repair pre-process 130, repair process 140, and repair post-process 150. The repair network database 126 may contain the data parameters collected during the inspection process, such as defect severity, defect type, the spatial distribution of defects, sample alignment, and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc., control parameters of the process, such as laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, feedback from sensors, monitoring systems, inspection equipment, start and stop times for process steps, delays or interruptions encountered, energy consumption measurements or power usage associated with the repair equipment and processes employed, etc., and data parameters collected during post-processing, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc.

[0034]

[0029] Further, embodiments may include cloud 128, which may be a network of remote servers that provide on-demand computing resources and services over the internet. Cloud 128 may consist of a collection of servers, storage devices, and networking equipment. Users may access cloud 128 through a variety of devices, such as computers, smartphones, and tablets, using internet connectivity.

[0035]

[0030] In some embodiments, the architecture of cloud 128 may be based on a distributed computing model, with multiple servers working together to provide services to users.

[0036]

[0031] Further, embodiments may include repair pre-process 130, which may include defect identification, defect classification, and decision-making procedures conducted before the repair processes and designed to ensure optimal performance during subsequent repair operations. The repair pre-process 130 may include collecting a plurality of parameter data including defect severity, defect type, spatial distribution of defects, etc.

[0037]

[0032] In some embodiments, the repair pre-process 130 may include pre-process equipment that may facilitate defect identification and handling procedures with precision and efficiency, such as a photoluminescence imaging system, electroluminescence testing, high-resolution cameras, machine vision systems, temperature control systems, data logging and documentation tools, etc.

[0038]

[0033] In some embodiments, the repair pre-process 130 may send and receive data from the 3rd party repair Al network 102, including parameter adjustments, control adjustments, material or workpiece data, decisions on material inspections, inputs to be used by the repair pre-process Al module 136, etc.

[0039]

[0034] Further, embodiments may include a communication interface 132, which may be a hardware or software component that enables communication between the repair pre-process 130 and the 3rd party repair Al network 102. In some embodiments, the repair pre-process 130 may communicate with the repair process 140, and the repair post-process 150. The communication interface 132 may include a set of protocols, rules, and standards that define how information is transmitted and received between devices. The communication interface 132 may be a physical connector, wireless network, or software application and may include components such as drivers, software libraries, and firmware that may be used to control and manage the communication process.

[0035] In some embodiments, the communication interface 132 may be compatible with USB, Bluetooth, or Wi-Fi. The communication interface 132 may communicate with a network. Examples of networks may include but are not limited to, the Internet, a cloud network, a Wireless Fidelity (Wi-Fi) network, a Wireless Local Area Network (WLAN), a Local Area Network (LAN), a telephone line (POTS), Long Term Evolution (LTE), and / or a Metropolitan Area Network (MAN).

[0040]

[0036] Further, embodiments may include a memory 134 may include suitable logic, circuitry, and / or interfaces that may be configured to store a machine code and / or a computer program with at least one code section executable by a processor. Examples of implementation of memory 134 may include, but are not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Hard Disk Drive (HDD), and / or a Secure Digital (SD) card.

[0041]

[0037] Further, embodiments may include a repair pre-process Al module 136 in which a predictive model may be performed to predict the final product of the repair process based upon the data collected from the repair pre-process 130, such as defect severity, defect type, spatial distribution of defects, etc. Predictive modeling involves the systematic analysis of historical data to identify patterns, correlations, and trends that can be used to build models capable of making accurate predictions about future outcomes.

[0042]

[0038] In some embodiments, the repair pre-process Al module 136 may receive a predictive model from the repair integration module 118. In some embodiments, the repair pre-process Al module 136 may receive and send data to the repair inspection module 110.

[0043]

[0039] In some embodiments, the data received from the repair inspection module 110 may be inputted into the predictive model to determine if any parameters of the pre-process should be adjusted, such as defect severity, defect type, spatial distribution of defects, etc.

[0044]

[0040] Further, embodiments may include a repair pre-process database 138 which may include data parameters collected from the repair pre-process 130, such as defect severity, defect type, spatial distribution of defects, etc. In some embodiments, the data stored in the repair pre-process database 138 may be sent to the repair inspection module 110 and / or the repair historical module 116.

[0045]

[0041] Further, embodiments may include repair process 140, which may be systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc. The repair process 140 may involve collecting parameter data, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed. The repair process 140 may include the repair equipment, such as cartridge-based solution and micro-solid printing technology. In some embodiments, the repair process 140 may send and receive data from the 3rd party repair Al network 102, including parameter adjustments, control adjustments, inputs to be used by the repair process Al module 146, etc.

[0046]

[0042] Further, embodiments may include a communication interface 142, which may be a hardware or software component that enables communication between the repair process 140 and the 3rd party repair Al network 102. In some embodiments, the repair process 140 may communicate with the repair pre-process 130, repair post-process 150. The communication interface 142 may include a set of protocols, rules, and standards that define how information is transmitted and received between the devices. The communication interface 142 may be a physical connector, wireless network, or software application and may include components such as drivers, software libraries, and firmware that may be used to control and manage the communication process.

[0047]

[0043] In some embodiments, the communication interface 142 may be compatible with USB, Bluetooth, or Wi-Fi. The communication interface 142 may communicate with a network. Examples of networks may include but are not limited to, the Internet, a cloud network, a Wireless Fidelity (Wi-Fi) network, a Wireless Local Area Network (WLAN), a Local Area Network (LAN), a telephone line (POTS), Long Term Evolution (LTE), and / or a Metropolitan Area Network (MAN).

[0048]

[0044] Further, embodiments may include a memory 144 may include suitable logic, circuitry, and / or interfaces that may be configured to store a machine code and / or a computer program with at least one code section executable by a processor. Examples of implementation of the memory 144 may include, but are not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Hard Disk Drive (HDD), and / or a Secure Digital (SD) card.

[0049]

[0045] Further, embodiments may include a repair process Al module 146 in which a predictive model may be performed to predict the final product of the repair process based upon the data collected from the repair process 140, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed. Predictive modeling involves the systematic analysis of historical data to identify patterns, correlations, and trends that can be used to build models capable of making accurate predictions about future outcomes.

[0050]

[0046] In some embodiments, the repair process Al module 146 may receive a predictive model from the repair integration module 118. In some embodiments, the repair process Al module 146 may receive and send data to the repair process module 112. In some embodiments, the data received from the repair process module 112 may be inputted into the predictive model to determine if any parameters of the repair process 140 should be adjusted, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0051]

[0047] Further, embodiments may include a repair process database 148 which may include data parameters collected from the repair process 140, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc. In some embodiments, the data stored in the repair process database 148 may be sent to the repair process module 112 and / or the repair historical module 116.

[0052]

[0048] Further, embodiments may include repair post-process 150, which may be the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The post-process may involve parameters such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. In some embodiments, the repair post- process 150 may send and receive data from the 3rd party repair Al network 102, including parameter adjustments, control adjustments, inputs to be used by the repair post-process Al module 156, etc.

[0053]

[0049] Further, embodiments may include a communication interface 152, which may be a hardware or software component that enables communication between the repair post-process 150 and the 3rd party repair Al network 102.

[0054]

[0050] In some embodiments, the repair post-process 150 may communicate with the repair pre- process 130, repair process 140. The communication interface 152 may include a set of protocols, rules, and standards that define how information is transmitted and received between the devices. The communication interface 152 may be a physical connector, wireless network, or software application and may include components such as drivers, software libraries, and firmware that may be used to control and manage the communication process.

[0055]

[0051] In some embodiments, the communication interface 152 may be compatible with USB, Bluetooth, or Wi-Fi. The communication interface 152 may communicate with a network. Examples of networks may include but are not limited to, the Internet, a cloud network, a Wireless Fidelity (Wi-Fi) network, a Wireless Local Area Network (WLAN), a Local Area Network (LAN), a telephone line (POTS), Long Term Evolution (LTE), and / or a Metropolitan Area Network (MAN).

[0056]

[0052] Further, embodiments may include a memory 154 may include suitable logic, circuitry, and / or interfaces that may be configured to store a machine code and / or a computer program with at least one code section executable by a processor. Examples of implementation of the memory 154 may include, but are not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Hard Disk Drive (HDD), and / or a Secure Digital (SD) card.

[0053] Further, embodiments may include a repair post-process Al module 156 in which a predictive model may be performed to optimize the future final products based upon the data collected from the repair post-process 150, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. Predictive modeling involves the systematic analysis of historical data to identify patterns, correlations, and trends that can be used to build models capable of making accurate predictions about future outcomes.

[0057]

[0054] In some embodiments, the repair post-process Al module 156 may receive a predictive model from the repair integration module 118. In some embodiments, the repair post-process Al module 156 may receive and send data to the repair post-process module 114. In some embodiments, the data received from the repair post-process module 114 may be inputted into the predictive model to determine if any parameters of the repair post-process 150 should be adjusted, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc.

[0058]

[0055] Further, embodiments may include a repair post-process database 158 which may include data parameters collected from the repair post-process 150, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. In some embodiments, the data stored in the repair post-process database 158 may be sent to the repair post-process module 114 and / or the repair historical module 116.

[0059]

[0056] FIG. 2 illustrates the base module 108. The process begins with the base module 108 initiating, at step 200, the repair inspection module 110. For example, the repair inspection module 110 begins by being initiated by the base module 108. The repair inspection module 110 connects to the repair pre-process 130. The repair inspection module 110 connects with the repair pre- process 130, such as defect identification, defect classification, and deci si on -making procedures conducted before the repair processes and designed to ensure optimal performance during subsequent repair operations.

[0060]

[0057] In some embodiments, the repair inspection module 110 may transmit and receive data from the repair pre-process 130. In some embodiments, the repair pre-process 130 may measure defect severity, defect type, spatial distribution of defects, etc. In some embodiments, the repair inspection module 110 may control or send inputs to control the repair pre-process 130. The repair inspection module 110 collects the data from the repair pre-process 130. The repair inspection module 110 collects the pre-processing data, such as defect severity, defect type, spatial distribution of defects, etc. For example, the repair pre-process 130 may include microLED array preparation to guarantee that the microLED array is sanitized and prepared for examination through photoluminescence imaging, electroluminescence testing, image analysis, classifying defects, process of decision making, documenting inspection results, etc. The repair inspection module 110 performs the inspection algorithm. For example, the inspection algorithm may be an artificial intelligence or machine learning algorithm in which the data collected from the repair pre-process 130 is used to determine if a control parameter of the repair process should be adjusted or altered. For example, The inspection algorithm may determine the photoluminescence, PL, and electroluminescence, EL, ranges that are out of specification. For example, fluctuations in the measured PL and EL signals may result from variations in the stability or intensity of the excitation source, such as electrical current for EL or lasers for PL. The accuracy of the measurements may be impacted by fluctuations in the sensitivity of the detectors employed to quantify PL and EL, as well as any disruptions in the calibration process, for example, sample alignment and positioning, ambient light interference, temperature variations, consistency of parameters, fluctuations in the excitation source, detectors, or optical components, unrelated results from algorithms for data analysis and processing, etc. The inspection algorithm may make predictive adjustments to achieve minimal variation in PL and EL measurements while testing microLEDs through the temperature control systems. For example, thermal chambers may be utilized to enclose the testing environment and maintain temperatures to within or better of ±0.1°C. In some embodiments, Peltier elements may be used, which are thermoelectric coolers that possess the ability to deliver accurate temperatures for heating and cooling purposes through the passage of an electric current and enable them to provide localized temperature regulation on the substrate or device holder. In some embodiments, liquid cooling systems may be used to eliminate heat and preserve a consistent temperature milieu. The inspection algorithm may analyze historical data to forecast temperature fluctuations and adjust cooling and heating systems in advance. The inspection algorithm may integrate with sensors to adjust temperature in real time to compensate for internal and external heat sources and maintain a constant range.

[0061]

[0058] In some embodiments, the inspection algorithm may receive input data from thermal probes and sensors, and automated environmental control systems. The inspection algorithm may be a predictive temperature control system to maintain the temperature of a thermal chamber within ±1% of the desired set point. The inspection algorithm may calibrate the thermal chamber and sensors to ensure accurate temperature readings. The inspection algorithm may establish a baseline by running the chamber without predictive control to record temperature fluctuations over a predefined period. The inspection algorithm may engage the predictive temperature control system, setting the desired temperature set point and continuously monitoring and recording the temperature inside the chamber with both the predictive control system active and inactive. The inspection algorithm may be a machine learning algorithm that is trained on historical temperature data to predict future fluctuations based on external and internal influences. The inspection algorithm may factor in thermal inertia, expected changes in external temperature, and the operational cycles of heating and cooling systems. For example, without predictive control, the temperature within the chamber may fluctuate by up to ±5% from the set point due to external temperature changes and the delayed response of the heating and cooling systems. The results of the inspection algorithm may limit temperature deviations to within ±1% of the set point. The inspection algorithm may be able to accurately forecast temperature trends adjust the chamber's heating and cooling in advance and maintain the set point with high precision. In some embodiments, the inspection algorithm may perform reinforcement learning to dynamically adapt to changes in the environment by continually learning and optimizing its policy, the inspection algorithm may find the most efficient way to maintain the target temperature. Over time, the inspection algorithm may develop a policy that precisely controls temperature with minimal deviation from the set point, even under fluctuating conditions.

[0062]

[0059] In some embodiments, the inspection algorithm may be a neural network that predicts future temperature trends based on historical data, current sensor readings, and known external influences and accurately models the non-linear relationships between the control actions, such as heating, cooling, ventilation, and the resulting temperature changes within the environment. The inspection algorithm may provide precise temperature regulation by learning the specific dynamics of the thermal system, including compensating for delayed effects, such as thermal inertia, predicting the impact of external temperature changes, and automatically adjusting control strategies for optimal performance. The repair inspection module 110 sends the determined process data from the inspection algorithm to the repair process module 112. For example, the repair inspection module 110 may send the temperature to the repair process module 112 to optimize the repair process.

[0060] In some embodiments, the repair inspection module 110 may send sample alignment and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc. The repair inspection module 110 stores the data in the repair inspection database 120. The repair inspection module 110 stores the data outputted from the inspection algorithm in the repair inspection database 120, such as the substrate ID, PL, and EL average measurements, temperature outputted by the inspection algorithm, etc. In some embodiments, the repair inspection database 120 may include other parameter data collected or calculated during the repair inspection module 110 process, such as defect severity, defect type, spatial distribution of defects, sample alignment and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc. The repair inspection module 110 returns to the base module 108. The base module 108 initiates, at step 202, the repair process module 112. For example, the repair process module 112 begins by being initiated by the base module 108. The repair process module 112 receives the pre-process data, such as the data on the microLED processed wafer, from the repair inspection module 110. The repair process module 112 receives the outputted data from the inspection algorithm performed in the repair inspection module 110, such as the temperature, allowing the repair process 140 to determine the optimal control parameters to prevent any additional defects in the repair process or decide to repair or replace components of the microLED wafer. In some embodiments, the repair process module 112 may receive defect severity, defect type, the spatial distribution of defects, sample alignment, and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc. The repair process module 112 connects to the repair process 140. The repair process module 112 connects with the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc. In some embodiments, the repair process module 112 may transmit and receive data from the repair process 140. In some embodiments, the repair process module 112 may control or send inputs to control the repair process 140, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0063]

[0061] The repair process module 112 collects the data from the repair process 140. The repair process module 112 collects the processing data, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc. For example, the process of fabricating microLEDs on a wafer, involving sacrificial layers, may include substrate preparation, deposition of sacrificial layers, epitaxial growth of microLEDs, wafer processing, laser lift-off or LLO, transfer to the temporary carrier or final substrate, repair, and finalization, etc. For example, any defective microLEDs identified during inspection can be repaired or replaced and the final substrate, now populated with microLEDs, undergoes additional processing to complete the display or device assembly, including connecting the microLEDs to the control electronics. Repairing faulty microLEDs may require the use of sophisticated laser-based techniques that guarantee high precision and efficiency, such as Laser-Induced Forward Transfer LIFT, Laser Lift-Off or LLO, etc. In some embodiments, the repair process 140 may include local repair via a focused ion beam or FIB, electrical maintenance methods, laser annealing, photolithography -based repair, etc. The repair process module 112 performs the control algorithm. For example, The control algorithm may decide to repair or replace faulty micro LEDs based on the impact on display quality, cost, and time.

[0064]

[0062] The control algorithm may identify defects, assess defect impact, review repair techniques, evaluate repair feasibility, and analyze quality improvement. For example, a microLED display with minor inconsistencies in brightness across its surface but these variances fall within the acceptable quality specification range for uniformity. The decision would be not to repair as the quality already meets the set standards and intervention could introduce unnecessary risk without significant benefit. Another example may be a display with several microLEDs showing significant color shifts and dead pixels that are visible during standard operation and significantly impact the user experience. These issues fall outside the acceptable quality specification range making repair essential to meet the product's quality standards and ensure customer satisfaction. The control algorithm may decide on repairing or replacing based on the balance between the expected quality improvement and the resources required for repair. The control algorithm may use "Delta E" as a quality metric, which is a standard measure of color difference or accuracy in displays and the control algorithm may use a threshold for perceptible color difference is a Delta E value of 1. Delta E is a metric used to quantify the difference between two colors to understand how perceptibly different they are. Lower Delta E values may indicate less difference and a more uniform color appearance, which is essential for high-quality microLED display performance. This measurement helps identify when specific microLEDs deviate from expected color outputs to guide decisions on whether adjustments or repairs are necessary to meet stringent display quality standards. The control algorithm may use decision tree analysis which is a predictive modeling tool that splits data into branches to represent decision-making paths. It begins at a root node, the initial decision point, and splits into branches based on criteria, leading to leaf nodes representing outcomes. Evaluating how different features affect outcomes, helps in making informed decisions based on patterns found in historical data.

[0065]

[0063] For example, the control algorithm may evaluate the Delta E value, which may exceed 1.5 and regardless of its position and ambient temperature, the substrate is directed towards the "Repair" decision due to the high Delta E value indicating significant color discrepancy. Another example may be if the control algorithm evaluates another substrate in which the Delta E value is above 1.0 but below the 1.5 threshold, the control algorithm may then consider its position, such as edge, and ambient temperature, and given the lower risk associated with edge positions and the cooler temperature, which might mitigate some thermal effects on Delta E, the decision could lean towards "No Repair", assuming these conditions historically result in acceptable final quality without intervention. In some embodiments, the control algorithm may include support vector machines, in which the substrates may be classified into repaired required or non-repair categories, and neural networks, in which subtle imperfections within the microLEDs may be detected through image and pattern recognition. The repair process module 112 adjusts the control parameters of the repair process 140. In some embodiments, the repair process module 112 may adjust repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0066]

[0064] The repair process module 112 executes the repair process. The repair process 140 may include identification of defects, replacement or repair of defective LEDs, repair of electrical connections, etc. The defects that may be identified may locate non-functioning or underperforming microLEDs on the display substrate. Once identified, defective LEDs may be repaired or replaced, which may involve adding new microLEDs to replace the defective ones or using laser repair techniques to reactivate non-functioning LEDs.

[0067]

[0065] In some embodiments, in which the microLEDs are functional, but the electrical connections are faulty, repair processes may focus on re-establishing these connections to ensure each microLED is correctly powered. The repair process module 112 stores the data in the repair process database 122. The repair process module 112 stores the data in the repair process database 122, such as the data collected from the repair process 140, the output of the inspection algorithm, and the output of the control algorithm. The repair process database 122 may contain the substrate ID, the Delta E value, the position, the ambient temperature, and the repair decision outputted by the control algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair process 140. The repair process module 112 returns to the base module 108. The base module 108 initiates, at step 204, the repair post-process module 114. For example, the repair post-process module 114 begins by being initiated by the base module 108. The repair post-process module 114 connects to the repair post-process 150. The repair postprocess module 114 may connect to the repair post-process 150, such the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair post-process module 114 collects the post-process data. The repair post-process module 114 may collect the parameter data, such as initial moisture.

[0066] The post-process may involve parameters such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. The repair post-process module 114 extracts the data from the repair inspection database 120 and the repair process database 122. The repair post-process module 114 extracts the data, such as material data, parameter data, etc. of the pre-process and the repair process. The repair post-process module 114 performs the post-process algorithm. The postprocess algorithm may detect and rectify through the inputs from high-resolution imaging devices, such as colorimeters and photometers, which are capable of capturing intricate images of the output of a display allowing an examination of brightness and color uniformity at the pixel level. The post-process algorithm may utilize demura techniques, which is a corrective measure utilized to address visual discrepancies in displays and to rectify issues of uniformity, demura modifies the color and luminance of specific pixels or sub-pixels. This is accomplished via software algorithms that detect inconsistencies and implement precise corrections to every impacted pixel to guarantee a consistent visual presentation across the entire display. For example, a threshold for applying the demura algorithm could be set around 4%, meaning any substrate with a uniformity deviation above this value would be considered for the corrective measure. Examples of uniformity measurements for microLEDs post-repair may include luminance uniformity, color uniformity, and white point variation. These measurements assess how evenly light is distributed across the display and how consistent the color output is from pixel to pixel. High deviations in these areas after repairs can indicate unresolved issues, necessitating further adjustments or corrections, such as applying a demura algorithm to improve display uniformity and overall visual quality. The postprocess algorithm may be a convolutional neural network that may be used to analyze high- resolution images to detect visual discrepancies in microLED displays and identify uniformity issues. The post-process algorithm may use reinforcement learning to suggest specifics of the demura algorithms needed and then optimize demura parameters by learning from previous corrections, effectively enhancing display quality based on specific detected issues.

[0068]

[0067] An example criterion for training the post-process algorithm may involve defining a 'bad' quality standard, such as "a sequence of three pixels in a row operating at very low luminance" being indicative of an area needing demura correction. By categorizing images with these patterns as 'bad', the post-process algorithm is trained to recognize similar patterns in new images and identify regions where demura algorithms should be applied to enhance display uniformity. The post-process algorithm may obtain visual quality standards that detail acceptable luminance and color uniformity levels, receive measurements from imaging photometers on luminance and color values across the microLED display, identify specific patterns indicating quality issues, such as sequences of dim pixels or color shifts, compare measured data against the standards to identify discrepancies, mark regions failing to meet standards, documenting the extent and nature of deviations, analyze the logged data to determine necessary adjustments for each identified region, and prepare software fixes to correct the discrepancies.

[0069]

[0068] In some embodiments, the post-process algorithm may enhance the precision in identifying regions needing demura correction. Initially, optical scans may not fully capture subtle uniformity deviations, potentially overlooking areas that could benefit from demura. However, a post-process algorithm trained on a vast array of data may detect intricate patterns associated with visual discrepancies, which may reveal that regions previously deemed acceptable require demura adjustment, or vice versa, even though a threshold is met to ensure optimal display quality. In some embodiments, the post-process algorithm may utilize support vector machines to classify regions of microLED displays as defective or non-defective by analyzing features derived from optical scans, such as pixel intensity, color variance, and pattern irregularities.

[0070]

[0069] In some embodiments, the post-process algorithm may utilize autoencoders that may learn to compress and decompress microLED display images, with the reconstruction error highlighting anomalies to identify subtle defects that may not significantly alter the overall pattern but still deviate from the norm and regions with high reconstruction errors may then be flagged for further inspection or direct demura correction. The repair post-process module 114 performs the postprocess inspection. For example, the repair post-process may include functional testing, uniformity testing, encapsulation, etc. The functional testing may include performing an exhaustive battery of tests to validate the electrical capabilities of every microLED, encompassing precise brightness modulation and on / off capabilities. The uniformity testing may include verifying the consistency of visual output throughout the entire display and fixing any anomalies that are identified. The encapsulation step may protect the microLEDs from environmental factors such as dust and moisture, which have the potential to degrade their performance over time, by applying protective layers over them. The repair post-process module 114 stores the data in the repair post-process database 124. The repair post-process module 114 stores the data in the repair post-process database 124, such as the data collected from the repair post-process 150, the output of the postprocess algorithm, etc.

[0071]

[0070] The repair post-process database 124 may contain the substrate ID, the uniformity deviation, the pattern analysis outputted by the post-process algorithm, the decision to apply the demura algorithm from basic thresholds, and the decision to apply the demura algorithm from the pattern analysis outputted by the post-process algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair post-process 150. The repair post-process module 114 returns to the base module 108. The base module 108 initiates, at step 206, the repair historical module 116. For example, the repair of historical module 116 begins by being initiated by the base module 108. The repair historical module 116 connects to the repair pre- process 130, the repair process 140, and the repair post-process 150. The historical module 116 connects to the repair pre-process 130, such as processes, such defect identification, defect classification, and decision-making procedures conducted prior to the repair processes and designed to ensure optimal performance during subsequent repair operations, the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc., and the repair post -process 150, such as the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair historical module 116 aggregates the data from the various types of processes and stores the data in the repair network database 126. The repair network database 126 may contain the historical data from the various processes performed by the repair pre-process 130, repair process 140, and repair post-process 150. The repair network database 126 may contain the data parameters collected during the inspection process, such as defect severity, defect type, the spatial distribution of defects, sample alignment, and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc., control parameters of the process, such as laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, feedback from sensors, monitoring systems, inspection equipment, start and stop times for process steps, delays or interruptions encountered, energy consumption measurements or power usage associated with the repair equipment and processes employed, etc., and data parameters collected during post-processing, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. The repair historical module 116 performs the historical machine learning algorithm on the historical data stored in the repair network database 126.

[0072]

[0071] The historical machine learning algorithm may improve each stage of the microLED repair process, such as repair pre-process 130, repair process 140, and repair post-process 150. For example, the historical machine learning algorithm may identify defects and assess the need for repair using initial photoluminescence and electroluminescence measurements and temperature data. The historical machine learning algorithm may utilize convolution neural networks to perform pattern recognition on photoluminescence and electroluminescence data to predict potential defects. The historical machine learning algorithm may be a machine learning regression model to correlate temperature variations with photoluminescence and electroluminescence measurements to identify anomalies that indicate repair needs. The historical machine learning algorithm may be trained on historical PL, EL, and temperature data, to learn patterns that lead to successful repairs and predict areas that need attention before actual repair to optimize planning and resource allocation by focusing on high-risk regions. The historical machine learning algorithm may provide real-time assistance and optimization during the repair process by guiding the process using Delta E values to ensure targeted and efficient repairs. The historical machine learning algorithm may perform reinforcement learning to adapt repair strategies based on Delta E measurements and initial repair decisions. The historical machine learning algorithm may utilize support vector machines to classify areas as needing repair based on Delta E thresholds and historical outcomes.

[0073]

[0072] The historical machine learning algorithm may dynamically adjust repair techniques, focusing on areas where Delta E values exceed thresholds that historically correlate with necessary repairs, utilize decision trees, or support vector machine outputs to confirm or adjust initial repair decisions to ensure precision in targeting defects. In some embodiments, a comprehensive history of Delta E metrics and the success rates of subsequent repairs may allow the historical machine learning algorithm to set more accurate thresholds for repair decisions. In some embodiments, the data may reveal which Delta E values are true indicators of a need for repair to reduce unnecessary interventions and focus efforts on genuinely impactful corrections. The historical machine learning algorithm may provide quality assessment for the repair post-process by validating repair quality and fine-tuning future repair actions based on uniformity deviation and convolutional neural network analysis. The historical machine learning algorithm may be a supervised learning model that evaluates post-repair uniformity against quality standards. The historical machine learning algorithm may utilize convolutional neural networks to identify subtle patterns in uniformity deviations. The historical machine learning algorithm may assess post-repair uniformity deviation and apply basic demura corrections where deviations exceed acceptable thresholds and further refine the analysis with convolutional neural networks to identify areas where demura based on basic thresholds might be insufficient or unnecessary. The historical machine learning algorithm may implement demura corrections informed by convolutional neural network analysis to target specific areas with precision to optimize the visual quality of the display. In some embodiments, the historical data on post-repair uniformity and the effectiveness of applied demura algorithms may allow the historical machine learning algorithm to assess repair quality and the necessity for further adjustments. In some embodiments, the historical machine learning algorithm may learn from past successes and failures and adjust demura parameters to achieve optimal results with greater efficiency. In some embodiments, the historical machine learning algorithm adjusts the parameters in real-time for the repair inspection module 110, repair process module 112, and repair post-process module 114 to correct any defects. The repair historical module 116 sends the process adjustments to the repair inspection module 110, the repair process module 112, and the repair post-process module 114. The historical machine learning algorithm may identify defects and assess the need for repair using initial PL and EL measurements and temperature data, guide the repair process using Delta E values to ensure targeted and efficient repairs, validate repair quality, and fine-tune future repair actions based on uniformity deviation and pattern analysis. The adjustments are sent to the repair inspection module 110 and repair process module 112 to create a more consistent final product which is then verified by the repair post-process module 114. The repair historical module 116 returns to the base module 108. The base module 108 initiates, at step 208, the repair integration module 118. For example, the repair integration module 118 begins by being initiated by the base module 108. The repair integration module 118 performs the integration machine learning algorithm. The integration machine learning algorithm may facilitate interactions between the repair pre-process 130, repair process 140, and repair post-process 150 stages of repairing microLEDs. For example, the integration machine learning algorithm may optimize each step, such as before repair, during repair, and after repair.

[0074]

[0073] The integration machine learning algorithm may use the initial analysis of photoluminescence and electroluminescence measurements along with temperature data to predict potential defects and the predictions and characterizations made before repairs, such as identifying regions at risk based on photoluminescence, electroluminescence, and temperature correlations may provide a baseline for evaluating the effectiveness of the repair process. This data may inform the after-repair assessment by setting expectations for the uniformity and quality improvements that repairs should achieve to help to calibrate the demura algorithms more accurately for specific types of discrepancies identified in the pre-repair phase. The integration machine learning algorithm may use insights from the during-repair phase, particularly in which Delta E thresholds led to successful repairs, which are fed back to refine the predictive models used before repair.

[0075]

[0074] This feedback loop may enhance the accuracy of defect predictions, allowing for more nuanced decisions about which regions truly require repair and potentially reducing unnecessary repair interventions. The integration machine learning algorithm may use the specific Delta E values and the corresponding decisions to repair as input for assessing the after-repair quality. In some embodiments, this may include evaluating how well the repairs addressed the identified issues and whether the chosen strategies led to the desired uniformity improvements. This data may be used to fine-tune the demura algorithms applied after repair to ensure they are appropriately calibrated to the nature and severity of the defects addressed during the repair process. The integration machine learning algorithm may use before-repair predictions to inform after-repair evaluations by establishing a baseline expectation for display quality improvements, the pre-repair analysis may influence how after-repair quality is assessed to ensure a targeted approach to applying demura corrections, where the integration machine learning algorithm adjusts its parameters based on the specific defects identified initially. The integration machine learning algorithm may be used during repair insights to refine before-repair predictions by learning from the outcomes of repair decisions to predict future repair needs more accurately allowing for a more effective allocation of resources and focus repair efforts where they are most needed and likely to have a significant impact. The integration machine learning algorithm may be used during repair decisions to enhance after-repair quality control by closely analyzing the outcomes of specific repair interventions to improve the after-repair assessment and demura application process and ensure that the demura algorithms are precisely tailored to the types of uniformity issues the repair process typically encounters which may lead to higher quality displays.

[0076]

[0075] In some embodiments, the integration machine learning algorithm may use the data stored in the repair network database 126 to perform a predictive modeling algorithm to make predictions or forecasts of the final product based on historical data and patterns from the previously created products. Predictive modeling involves the systematic analysis of historical data to identify patterns, correlations, and trends that can be used to build models capable of making accurate predictions about future outcomes. For example, the integration machine learning algorithm may be used for validating the repair quality and fine-tuning future repair actions across the repair pre- process 130, repair process 140, and repair post-process 150. The historical data in the repair network database 126 may include information on uniformity deviations and relevant operational parameters such as photoluminescence and electroluminescence average measurements, temperature, Delta E values, the decision to repair, the effectiveness of demura algorithms, etc. The data is cleaned and preprocessed, addressing any missing values and ensuring that all variables are in a suitable format for modeling. This may involve normalization or scaling of numerical features. The dataset is split into training and testing sets. The training set is used to train the predictive model, and the testing set assesses its performance on unseen data. The model used may be linear regression, decision trees, ensemble methods, neural networks, etc. Then key features are identified that influence the uniformity deviations and engineer new features if necessary. For example, repair data from the repair process 140 may have a significant impact on the uniformity deviations. The selected predictive model is trained using the training dataset. The model learns patterns and relationships between operational parameters and moisture removal. The model's performance is validated on a separate validation dataset and hyperparameters are fine-tuned to optimize its accuracy and generalization. The model is then evaluated and deployed by being sent to the repair pre-process 130, repair process 140, and repair post-process 150. The repair integration module 118 connects to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. The repair integration module 118 connects to the repair pre- process 130, such as processes, such defect identification, defect classification, and decisionmaking procedures conducted prior to the repair processes and designed to ensure optimal performance during subsequent repair operations, the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc., and the repair post-process 150, such as the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair integration module 118 sends the process adjustments to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. For example, the repair integration module 118 may predict potential defects based on the initial analysis of PL and EL measurements and temperature data. The integration machine learning algorithm may decide to repair based on Delta E values to finetune the demura algorithm that is applied after the repair.

[0077]

[0076] In some embodiments, the repair integration module 118 may send the integration machine learning algorithm to the processes, allowing the systems to further enhance the optimization. The repair integration module 118 returns to the base module 108.

[0078]

[0077] FIG. 3 illustrates the repair inspection module 110. The process begins with the repair inspection module 110 being initiated, at step 300, by the base module 108. The repair inspection module 110 connects, at step 302, to the repair pre-process 130. The repair inspection module 110 connects with the repair pre-process 130, such as defect identification, defect classification, and decision-making procedures conducted before the repair processes and designed to ensure optimal performance during subsequent repair operations.

[0079]

[0078] In some embodiments, the repair inspection module 110 may transmit and receive data from the repair pre-process 130. In some embodiments, the repair pre-process 130 may measure defect severity, defect type, spatial distribution of defects, etc.

[0080]

[0079] In some embodiments, the repair inspection module 110 may control or send inputs to control the repair pre-process 130. The repair inspection module 110 collects, at step 304, the data from the repair pre-process 130. The repair inspection module 110 collects the pre-processing data, such as defect severity, defect type, spatial distribution of defects, etc. For example, the repair pre- process 130 may include microLED array preparation to guarantee that the microLED array is sanitized and prepared for examination through photoluminescence imaging, electroluminescence testing, image analysis, classifying defects, process of decision making, documenting inspection results, etc.

[0081]

[0080] Edge Al and TinyML Example 1; In some embodiments, the repair inspection module 110 may deploy a TinyML-enabled Edge Al model directly on the embedded controller of the repair inspection equipment 130 to execute localized defect classification and wafer flagging in real time. For example, a technician may request, “Flag microcrack clusters exceeding 10 pm on die corners during live scan,” while the embedded system may autonomously trigger, “Tag wafers for reinspection if edge halo patterns deviate from reference symmetry by more than 5%.” In operation, the repair inspection module 110 may capture image frames and edge morphology signatures from the inspection equipment 130, and apply a compact convolutional model trained on defect examples archived in the repair inspection database 120. Internally, the model may use memoryefficient convolution layers and reduced precision inference to detect critical features without transferring data to external systems. Upon identifying at-risk regions, the device may flag the wafer ID, trigger a visual indicator for operator review, and queue a metadata packet for later upload to the repair network database 126. Outputs may include local inspection flags, defect overlays, and statistical summaries of inline classifications. By embedding TinyML at the inspection hardware level, the repair inspection module 110 may deliver immediate classification and filtering of critical defects while preserving throughput and minimizing data movement overhead. The repair inspection module 110 performs, at step 306, the inspection algorithm. For example, the inspection algorithm may be an artificial intelligence or machine learning algorithm in which the data collected from the repair pre-process 130 is used to determine if a control parameter of the repair process should be adjusted or altered. For example, The inspection algorithm may determine the photoluminescence, PL, and electroluminescence, EL, ranges that are out of specification. For example, fluctuations in the measured PL and EL signals may result from variations in the stability or intensity of the excitation source, such as electrical current for EL or lasers for PL.

[0082]

[0081] The accuracy of the measurements may be impacted by fluctuations in the sensitivity of the detectors employed to quantify PL and EL, as well as any disruptions in the calibration process, for example, sample alignment and positioning, ambient light interference, temperature variations, consistency of parameters, fluctuations in the excitation source, detectors, or optical components, unrelated results from algorithms for data analysis and processing, etc. The inspection algorithm may make predictive adjustments to achieve minimal variation in PL and EL measurements while testing microLEDs through the temperature control systems. For example, thermal chambers may be utilized to enclose the testing environment and maintain temperatures to within or better of ±0.1°C. In some embodiments, Peltier elements may be used, which are thermoelectric coolers that possess the ability to deliver accurate temperatures for heating and cooling purposes through the passage of an electric current and enable them to provide localized temperature regulation on the substrate or device holder. In some embodiments, liquid cooling systems may be used to eliminate heat and preserve a consistent temperature milieu. The inspection algorithm may analyze historical data to forecast temperature fluctuations and adjust cooling and heating systems in advance. The inspection algorithm may integrate with sensors to adjust temperature in real time to compensate for internal and external heat sources and maintain a constant range.

[0083]

[0082] In some embodiments, the inspection algorithm may receive input data from thermal probes and sensors, and automated environmental control systems. The inspection algorithm may be a predictive temperature control system to maintain the temperature of a thermal chamber within ±1% of the desired set point. The inspection algorithm may calibrate the thermal chamber and sensors to ensure accurate temperature readings. The inspection algorithm may establish a baseline by running the chamber without predictive control to record temperature fluctuations over a predefined period. The inspection algorithm may engage the predictive temperature control system, setting the desired temperature set point and continuously monitoring and recording the temperature inside the chamber with both the predictive control system active and inactive. The inspection algorithm may be a machine learning algorithm that is trained on historical temperature data to predict future fluctuations based on external and internal influences. The inspection algorithm may factor in thermal inertia, expected changes in external temperature, and the operational cycles of heating and cooling systems. For example, without predictive control, the temperature within the chamber may fluctuate by up to ±5% from the set point due to external temperature changes and the delayed response of the heating and cooling systems. The results of the inspection algorithm may limit temperature deviations to within ±1% of the set point. The inspection algorithm may be able to accurately forecast temperature trends adjust the chamber's heating and cooling in advance and maintain the set point with high precision. In some embodiments, the inspection algorithm may perform reinforcement learning to dynamically adapt to changes in the environment by continually learning and optimizing its policy, the inspection algorithm may find the most efficient way to maintain the target temperature. Over time, the inspection algorithm may develop a policy that precisely controls temperature with minimal deviation from the set point, even under fluctuating conditions.

[0084]

[0083] In some embodiments, the inspection algorithm may be a neural network that predicts future temperature trends based on historical data, current sensor readings, and known external influences and accurately models the non-linear relationships between the control actions, such as heating, cooling, ventilation, and the resulting temperature changes within the environment. The inspection algorithm may provide precise temperature regulation by learning the specific dynamics of the thermal system, including compensating for delayed effects, such as thermal inertia, predicting the impact of external temperature changes, and automatically adjusting control strategies for optimal performance.

[0085]

[0084] Al Agent Example 1; In some embodiments, the repair inspection module 110 may employ a localized Al agent configured to dynamically adjust inspection sampling rates, reinspection targeting, and defect scoring thresholds in response to evolving substrate conditions and defect clustering behavior. For example, a quality engineer may configure, “Escalate sampling to 100% if void cluster density exceeds 5 per mm2,” while the Al agent may autonomously trigger, “Reduce scan resolution and activate edge-focus pass if microcrack patterns consistently emerge at die corners.” In execution, the agent may monitor defect heatmaps and spatial patterning from the repair inspection database 120, cross-reference tool-specific motion and temperature logs from the repair process database 122, and assess rework trends stored in the repair network database 126. When predefined or learned patterns suggest emerging risk conditions, the agent may adjust scan density, sampling coverage, or image classification thresholds in real time, while also flagging suspect wafers for downstream verification in the repair post-process module 114. Internally, the agent may rely on probabilistic risk models and reinforcement-learned sampling policies tuned to historical defect escape events. Outputs may include an updated inspection plan, annotated trigger rationale, and priority flags passed to subsequent process stages. By embedding a responsive Al agent into the repair inspection module 110, the system may shift from fixed sampling logic to dynamic, risk-aware inspection coverage, increasing detection accuracy without sacrificing throughput.

[0086]

[0085] Multimodal Al Example 1 : In some embodiments, the repair inspection module 110 may integrate a multimodal Al model to improve defect classification and causality analysis by fusing visual inspection data, localized beam parameters, and technician-entered annotations. For example, a process engineer may request, “Determine if corner dark spots are correlated with beam trajectory fluctuations,” while a system-level diagnostic prompt may query, “Identify repeatable defect shapes linked to operator-flagged lift-off issues.” In response, the repair inspection module 110 may retrieve high-resolution inspection images from the repair inspection equipment 130, align them with beam dwell time, power, and alignment offset logs from the repair process database 122, and incorporate inspection flags and textual comments from the repair inspection database 120. The multimodal Al model may jointly embed and compare these inputs, applying cross-attention to identify causal features across modalities — such as repeated elliptical voids correlated with subthreshold trajectory misalignments during operator-noted lot events. Internally, the model may match these to known failure signatures stored in the repair network database 126 and assign classification confidence along with causal probability scores. Outputs may include a fused defect report, highlighting the co-occurrence of beam and image features with operator observations, and a supporting literature citation describing similar beam-instability-induced defect geometries. By embedding multimodal reasoning into the repair inspection module 110, the system may unify previously siloed defect indicators into actionable, pattern-driven insights that accelerate troubleshooting and increase first-pass yield.

[0087]

[0086] Generative Video Al Example 1 : In some embodiments, the repair inspection module 110 may employ a generative video Al model to simulate in-process defect dynamics — such as lift-off instability, beam-induced drift, or void propagation — using early-cycle inspection frames and partial beam control telemetry. For example, a process engineer may prompt, “Visualize how an initial alignment offset might evolve into die edge peeling under standard beam trajectory,” while a quality system may request, “Project void growth paths for incomplete bonding zones identified in frame 3 of inspection scan.” In response, the repair inspection module 110 may retrieve initial visual frames from the repair inspection equipment 130, align these with subcycle beam timing, path geometry, and actuation logs from the repair process database 122, and condition a generative video model trained on annotated defect evolution patterns stored in the repair network database 126. Internally, the model may use time-sequenced generative layers — such as recurrent GANs or physics-aware video transformers — to forecast likely morphological changes, predicting edge delamination, void elongation, or fracture initiation over the remainder of the process cycle. Outputs may include a projected defect evolution video, annotated with causal beam features, and an external reference linking similar trajectories to instability in precision laser repair environments. By embedding generative video Al into the repair inspection module 110, the system may support anticipatory process control and preemptive rework decisions based on dynamic defect risk simulation. The repair inspection module 110 sends, at step 308, the determined process data from the inspection algorithm to the repair process module 112. For example, the repair inspection module 110 may send the temperature to the repair process module 112 to optimize the repair process.

[0088]

[0087] In some embodiments, the repair inspection module 110 may send sample alignment and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc.

[0089]

[0088] Synthetic Data and AI-Generated Digital Twins Example 1; In some embodiments, the repair inspection module 110 may employ synthetic data generation to augment training of its defect classification model by producing labeled image sets representing rare or emerging defect types not yet captured in production runs. For example, a machine learning engineer may request, “Generate synthetic corner-peeling defects for die bonded with experimental substrates,” while a yield analyst may prompt, “Create annotated microvoid patterns under atypical beam power fluctuations for training set expansion.” In response, the repair inspection module 110 may analyze defect frequency distributions in the repair inspection database 120 and identify underrepresented defect modes based on substrate metadata, beam telemetry, and inspection classifications. A generative model trained on historical defect morphologies stored in the repair network database 126 may then produce realistic synthetic image samples with accurate spatial, textural, and process- aligned labels. Internally, the model may apply conditional generative adversarial networks (GANs) or diffusion models constrained by process condition parameters such as layer composition, beam dwell time, and local alignment offsets. Outputs may include a batch of labeled synthetic defect images, metadata tags aligned to training inputs, and a classifier retraining report demonstrating improved recall for previously low-coverage failure modes. By embedding synthetic data generation in the repair inspection module 110, the system may accelerate adaptation to new materials and tool configurations while improving generalization and robustness of defect detection models. The repair inspection module 110 stores, at step 310, the data in the repair inspection database 120. The repair inspection module 110 stores the data outputted from the inspection algorithm in the repair inspection database 120, such as the substrate ID, PL, and EL average measurements, temperature outputted by the inspection algorithm, etc. In some embodiments, the repair inspection database 120 may include other parameter data collected or calculated during the repair inspection module 110 process, such as defect severity, defect type, spatial distribution of defects, sample alignment and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc. The repair inspection module 110 returns, at step 312, to the base module 108.

[0090]

[0089] A 306 perform inspection module advances quality control or manufacturing procedures, including those in the production of microLED arrays. It highlights how control parameters can be changed based on gathered data to optimize the repair process through the use of artificial intelligence (Al) or machine learning techniques. In particular, Step 306 emphasizes how the inspection algorithm uses data to make predictive modifications, minimizing variation in important parameters such as photoluminescence (PL) and electroluminescence (EL) measurements, and using predictive temperature control systems to keep the temperature within precise bounds. A decision tree is an easy-to-use Al algorithm that can be applied in this situation since it is capable of handling intricate deci si on -making procedures whilst being simple to use. In order for decision trees to function, a dataset must be divided into smaller subsets. At the same time, an associated decision tree must be gradually created. A tree with decision nodes and leaf nodes is the end product. Two or more branches, each representing a value for the tested characteristic, make up a decision node. A judgment or classification is represented by a leaf node. Like the inspection algorithm above, this method might be used to decide whether a repair process control parameter has to be changed or adjusted. For example, in the inspection method mentioned, data on temperature changes, PL and EL measurements, and other variables could be analyzed by a decision tree to determine how to modify the repair procedure. This could entail analyzing current sensor readings and previous temperature data to determine whether the temperature within a thermal chamber needs to be altered in order to maintain temperature variations within a target range. Historical temperature changes, PL and EL measurements, sensor readings (from environmental sensors and thermal probes, for example), and information on the spatial distribution, nature, and severity of faults are likely among the datasets employed in such an algorithm. The repair inspection module's databases, which include thorough records of earlier inspections, repairs, and results, may be the source of this information. The inspection algorithm could forecast future situations and make wise decisions to guarantee the best possible performance of the repair procedure by training the decision tree on this extensive dataset. A perform inspection module advances quality control or manufacturing procedures, including those in the production of microLED arrays. It highlights how control parameters can be changed based on gathered data to optimize the repair process through the use of artificial intelligence (Al) or machine learning techniques. In particular, Step 306 emphasizes how the inspection algorithm uses data to make predictive modifications, minimizing variation in important parameters such as photoluminescence (PL) and electroluminescence (EL) measurements, and using predictive temperature control systems to keep the temperature within precise bounds. A decision tree is an easy-to-use Al algorithm that can be applied in this situation since it is capable of handling intricate decisionmaking procedures whilst being simple to use. In order for decision trees to function, a dataset must be divided into smaller subsets. At the same time, an associated decision tree must be gradually created. A tree with decision nodes and leaf nodes is the end product. Two or more branches, each representing a value for the tested characteristic, make up a decision node. A judgment or classification is represented by a leaf node. Like the inspection algorithm above, this method might be used to decide whether a repair process control parameter has to be changed or adjusted. For example, in the inspection method mentioned, data on temperature changes, PL and EL measurements, and other variables could be analyzed by a decision tree to determine how to modify the repair procedure. This could entail analyzing current sensor readings and previous temperature data to determine whether the temperature within a thermal chamber needs to be altered in order to maintain temperature variations within a target range. Historical temperature changes, PL and EL measurements, sensor readings (from environmental sensors and thermal probes, for example), and information on the spatial distribution, nature, and severity of faults are likely among the datasets employed in such an algorithm. The repair inspection module's databases, which include thorough records of earlier inspections, repairs, and results, may be the source of this information. The inspection algorithm could forecast future situations and make wise decisions to guarantee the best possible performance of the repair procedure by training the decision tree on this extensive dataset.

[0091]

[0090] FIG. 4 illustrates the repair process module 112. The process begins with the repair process module 112 being initiated, at step 400, by the base module 108. The repair process module 112 receives, at step 402, the pre-process data, such as the data on the microLED processed wafer, from the repair inspection module 110. The repair process module 112 receives the outputted data from the inspection algorithm performed in the repair inspection module 110, such as the temperature, allowing the repair process 140 to determine the optimal control parameters to prevent any additional defects in the repair process or decide to repair or replace components of the microLED wafer. In some embodiments, the repair process module 112 may receive defect severity, defect type, the spatial distribution of defects, sample alignment, and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc. The repair process module 112 connects, at step 404, to the repair process 140. The repair process module 112 connects with the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc.

[0092]

[0091] In some embodiments, the repair process module 112 may transmit and receive data from the repair process 140. In some embodiments, the repair process module 112 may control or send inputs to control the repair process 140, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc. The repair process module 112 collects, at step 406, the data from the repair process 140. The repair process module 112 collects the processing data, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., process monitoring data including feedback from sensors, monitoring systems, inspection equipment, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including measurements of energy consumption or power usage associated with the repair equipment and processes employed, etc. For example, the process of fabricating microLEDs on a wafer, involving sacrificial layers, may include substrate preparation, deposition of sacrificial layers, epitaxial growth of microLEDs, wafer processing, laser lift-off or LLO, transfer to the temporary carrier or final substrate, repair and finalization, etc. For example, any defective microLEDs identified during inspection can be repaired or replaced and the final substrate, now populated with microLEDs, undergoes additional processing to complete the display or device assembly, including connecting the microLEDs to the control electronics. Repairing faulty microLEDs may require the use of sophisticated laser-based techniques that guarantee high precision and efficiency, such as Laser- Induced Forward Transfer LIFT, Laser Lift-Off or LLO, etc. In some embodiments, the repair process 140 may include local repair via a focused ion beam or FIB, electrical maintenance methods, laser annealing, photolithography-based repair, etc. The repair process module 112 performs, at step 408, the control algorithm. For example, The control algorithm may decide to repair or replace faulty microLEDs based on the impact on display quality, cost, and time. The control algorithm may identify defects, assess defect impact, review repair techniques, evaluate repair feasibility, and analyze quality improvement. For example, a microLED display with minor inconsistencies in brightness across its surface but these variances fall within the acceptable quality specification range for uniformity. The decision would be not to repair as the quality already meets the set standards and intervention could introduce unnecessary risk without significant benefit. Another example may be a display with several microLEDs showing significant color shifts and dead pixels that are visible during standard operation and significantly impact the user experience. These issues fall outside the acceptable quality specification range making repair essential to meet the product's quality standards and ensure customer satisfaction. The control algorithm may decide on repairing or replacing based on the balance between the expected quality improvement and the resources required for repair. The control algorithm may use "Delta E" as a quality metric, which is a standard measure of color difference or accuracy in displays and the control algorithm may use a threshold for perceptible color difference is a Delta E value of 1. Delta E is a metric used to quantify the difference between two colors to understand how perceptibly different they are. Lower Delta E values may indicate less difference and a more uniform color appearance, which is essential for high-quality microLED display performance.

[0093]

[0092] This measurement helps identify when specific microLEDs deviate from expected color outputs to guide decisions on whether adjustments or repairs are necessary to meet stringent display quality standards. The control algorithm may use decision tree analysis which is a predictive modeling tool that splits data into branches to represent decision-making paths. It begins at a root node, the initial decision point, and splits into branches based on criteria, leading to leaf nodes representing outcomes. Evaluating how different features affect outcomes, helps in making informed decisions based on patterns found in historical data. For example, the control algorithm may evaluate the Delta E value, which may exceed 1.5 and regardless of its position and ambient temperature, the substrate is directed towards the "Repair" decision due to the high Delta E value indicating significant color discrepancy. Another example may be if the control algorithm evaluates another substrate in which the Delta E value is above 1.0 but below the 1.5 threshold, the control algorithm may then consider its position, such as edge, and ambient temperature, and given the lower risk associated with edge positions and the cooler temperature, which might mitigate some thermal effects on Delta E, the decision could lean towards "No Repair", assuming these conditions historically result in acceptable final quality without intervention.

[0094]

[0093] In some embodiments, the control algorithm may include support vector machines, in which the substrates may be classified into repaired required or non-repair categories, and neural networks, in which subtle imperfections within the microLEDs may be detected through image and pattern recognition.

[0095]

[0094] Synthetic Data and ALGenerated Digital Twins Example 2; In some embodiments, the repair process module 112 may incorporate a digital twin framework configured to simulate repair process outcomes across variable beam parameters, substrate stackups, and actuator conditions — enabling virtual experimentation prior to physical execution. For example, a development engineer may prompt, “Predict surface distortion patterns if beam dwell is reduced by 15% for multilayer sapphire substrates,” while a tool qualification agent may request, “Evaluate defect risk if actuator lag exceeds 0.4 ms during Z-axis alignment.” In response, the repair process module 112 may extract historical chamber performance data, material-dependent failure signatures, and control profiles from the repair process database 122 and repair network database 126. A digital twin model may then simulate physical responses — including beam-induced thermal fields, mechanical deformation, and trajectory variance — based on parameterized virtual tool configurations. Internally, the model may apply finite-element physics constraints and learnable correction layers aligned to prior real-world outcomes. Outputs may include defect risk maps, predicted failure geometries, and scenario-graded confidence reports with reference to known yield detractors under similar conditions. By embedding a digital twin model in the repair process module 112, the system may support low-risk recipe development and predictive tuning of process behavior without requiring material expenditure or line stoppage. The repair process module 112 adjusts, at step 410, the control parameters of the repair process 140. In some embodiments, the repair process module 112 may adjust repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0096]

[0095] Edge Al and TinyML Example 2: In some embodiments, the repair process module 112 may incorporate a TinyML-based Edge Al model deployed directly on beam alignment actuators and thermal controllers to perform localized, low-latency corrections during high-speed repair operations. For example, a process engineer may configure, “Adjust beam offset in real time if submicron drift exceeds 2o,” while an embedded trigger may initiate, “Suppress thermal ramp rate if temperature rise exceeds 0.5°C per second during dwell phase.” In operation, the repair process module 112 may continuously acquire beam trajectory, alignment drift, and temperature data from the repair process equipment 140 and execute a quantized neural model trained on stabilization patterns stored in the repair process database 122. Internally, the model may use edge-optimized filters and inference routines to detect anomalies within milliseconds and autonomously issue micro-adjustments to actuator voltage or thermal drive signals. Outputs may include a localized event log, in-situ correction traces, and metadata queued for synchronization with the repair network database 126. By embedding real-time correction logic into device-level controllers, the repair process module 112 may maintain beam fidelity and substrate safety in fast-switching environments without dependence on upstream optimization modules. The repair process module 112 executes, at step 412, the repair process. The repair process 140 may include identification of defects, replacement or repair of defective LEDs, repair of electrical connections, etc. The defects that may be identified may locate non-functioning or underperforming microLEDs on the display substrate. Once identified, defective LEDs may be repaired or replaced, which may involve adding new microLEDs to replace the defective ones or using laser repair techniques to reactivate nonfunctioning LEDs. In some embodiments, in which the microLEDs are functional, but the electrical connections are faulty, repair processes may focus on re-establishing these connections to ensure each microLED is correctly powered. The repair process module 112 stores, at step 414, the data in the repair process database 122. The repair process module 112 stores the data in the repair process database 122, such as the data collected from the repair process 140, the output of the inspection algorithm, and the output of the control algorithm. The repair process database 122 may contain the substrate ID, the Delta E value, the position, the ambient temperature, and the repair decision outputted by the control algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair process 140. The repair process module 112 returns, at step 416, to the base module 108.

[0097]

[0096] A key part of the disclosed repair process module 112 is the control algorithm that is run at step 408 and is responsible for determining whether to replace or repair defective microLEDs. This decision-making process, which balances variables including display quality, cost, and maintenance time, is essential to preserving the effectiveness and quality of microLED displays. Decision trees are a straightforward and powerful Al technique that can be used in this framework. The way this method operates is by building a model that uses a number of input variables to predict the value of a target variable. The tree's interior nodes each represent an input variable, its branches each stand for decision rules, and its leaf nodes each represent the result. Within the framework of the control algorithm, a decision tree could be utilized to evaluate if a microLED needs to be replaced or repaired by examining multiple data points, including defect kind, defect severity, and Delta E values, which indicate the variation in color accuracy. To determine whether repair actions are required, the decision tree could, for instance, examine the Delta E value. It may label the tiny LED as broken if this value rises above a set threshold and shows a discernible color difference. The algorithm may determine that no repair is necessary if the Delta E value is within acceptable bounds, taking into consideration additional variables that may affect the overall quality, such as the location of the microLED or the surrounding temperature.

[0097] A multitude of parameters would be included in the dataset required for the decision tree model and the control method in particular. Data from the repair inspection module 110, which offers information on the microLED processed wafer prior to repair, and the inspection algorithm's output, which may contain metrics like temperature and fault details, are included in this. It would also contain specifics about the repair procedure itself, including energy consumption data, feedback from sensors and monitoring systems, and characteristics about the repair techniques utilized (such as laser power duration and ion beam current). Furthermore, information on quality metrics — in particular, Delta E values — is essential since it directly affects choices about the degree and necessity of repairs. This extensive dataset is used to train the decision tree, which gives the control algorithm the ability to learn from past repair decisions and optimize the deci si on -making process for subsequent repairs. By avoiding needless repairs that do not result in appreciable improvements, this method guarantees the effective use of resources while upholding high standards of display quality. Repairs are only performed when they are anticipated to significantly improve the display's quality.

[0098]

[0098] FIG. 5 illustrates the repair post-process module 114. The process begins with the repair post-process module 114 being initiated, at step 500, by the base module 108. The repair postprocess module 114 connects, at step 502, to the repair post-process 150. The repair post-process module 114 may connect to the repair post-process 150, such the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair post-process module 114 collects, at step 504, the post-process data. The repair postprocess module 114 may collect the parameter data, such as initial moisture. The post-process may involve parameters such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. The repair post-process module 114 extracts, at step 506, the data from the repair inspection database 120 and repair process database 122. The repair post-process module 114 extracts the data, such as material data, parameter data, etc. of the pre-process and the repair process.

[0099]

[0099] Multimodal Al Example 2: In some embodiments, the repair post-process module 114 may incorporate a multimodal Al model to assess repair quality and residual substrate risk by fusing photoluminescence images, temperature excursion profiles, and reinspection commentary. For example, a reliability engineer may prompt, “Correlate low-luminance edge artifacts with thermal anomalies during beam realignment,” while a quality system trigger may request, “Identify defect patterns associated with annotated substrate warpage alerts.” In response, the repair post-process module 114 may retrieve high -resolution photoluminescence outputs and surface scan images from the repair post-processing equipment 150, align them with historical repair chamber thermal profiles and actuation sequences from the repair process database 122, and integrate technician reinspection tags and qualitative feedback stored in the repair post-processing database 124. The multimodal model may process these inputs through shared embedding layers, identifying spatial, temporal, and textual associations — such as fringe photoluminescence loss consistently linked to late-cycle beam overshoot events and operator-labeled substrate bowing. Internally, the model may compare these signals to known failure trajectories stored in the repair network database 126 and issue a risk- weighted post-repair report. Outputs may include a composite quality summary with cross-modal attribution, flags for tool-side recalibration, and a citation referencing literature on beam-induced thermal drift effects in microLED materials. By embedding multimodal Al into the repair post-process module 114, the system may deliver high-fidelity yield risk assessments grounded in correlated signals across image, process, and technician insight layers.

[0100]

[0100] Generative Video Al Example 2: In some embodiments, the repair post-process module 114 may incorporate a generative video Al model to simulate the time-evolved behavior of micro defects — such as edge delamination, substrate bowing, or luminance decay — under projected operating conditions following repair. For example, a reliability engineer may prompt, “Simulate photoluminescence degradation over 72 hours under elevated soak conditions,” while a process owner may request, “Visualize how marginal die lift may propagate into corner delamination during thermal cycling.” In response, the repair post-process module 114 may retrieve final surface and photometric measurements from the repair post-processing equipment 150, align those with repair chamber conditions and actuation logs from the repair process database 122, and input annotated outcome data from the repair post-processing database 124 into a generative model trained on historical defect evolution sequences stored in the repair network database 126. Internally, the model may use spatiotemporal prediction layers — such as recurrent attention modules or conditional video diffusion — to simulate frame-by-frame defect morphology under stress scenarios, including surface curling, subpixel failure, and light output decay. Output may include a projected video of defect development, a causal parameter summary highlighting the contributing repair or tool factors, and a literature reference linking thermal reflow conditions to stress-amplified failure in laminated microstructures. By embedding generative video Al into the repair post-process module 114, the system may support proactive reliability screening and virtual stress testing, reducing reliance on time-intensive physical aging studies. The repair post-process module 114 performs, at step 508, the post-process algorithm. The post-process algorithm may detect and rectify through the inputs from high-resolution imaging devices, such as colorimeters and photometers, which are capable of capturing intricate images of the output of a display allowing an examination of brightness and color uniformity at the pixel level. The post-process algorithm may utilize demura techniques, which is a corrective measure utilized to address visual discrepancies in displays and to rectify issues of uniformity, demura modifies the color and luminance of specific pixels or sub-pixels. This is accomplished via software algorithms that detect inconsistencies and implement precise corrections to every impacted pixel to guarantee a consistent visual presentation across the entire display. For example, a threshold for applying the demura algorithm could be set around 4%, meaning any substrate with a uniformity deviation above this value would be considered for corrective measures. Examples of uniformity measurements for microLEDs post-repair may include luminance uniformity, color uniformity, and white point variation. These measurements assess how evenly light is distributed across the display and how consistent the color output is from pixel to pixel. High deviations in these areas after repairs can indicate unresolved issues, necessitating further adjustments or corrections, such as applying a demura algorithm to improve display uniformity and overall visual quality.

[0101]

[0101] The post-process algorithm may be a convolutional neural network that can be used to analyze high-resolution images to detect visual discrepancies in microLED displays and identify uniformity issues. The post-process algorithm may use reinforcement learning to suggest specifics of the demura algorithms needed and then optimize demura parameters by learning from previous corrections, effectively enhancing display quality based on specific detected issues. An example criterion for training the post-process algorithm may involve defining a 'bad' quality standard, such as "a sequence of three pixels in a row operating at very low luminance" being indicative of an area needing demura correction. By categorizing images with these patterns as 'bad', the post-process algorithm is trained to recognize similar patterns in new images and identify regions where demura algorithms should be applied to enhance display uniformity. The post-process algorithm may obtain visual quality standards that detail acceptable luminance and color uniformity levels, receive measurements from imaging photometers on luminance and color values across the microLED display, identify specific patterns indicating quality issues, such as sequences of dim pixels or color shifts, compare measured data against the standards to identify discrepancies, mark regions failing to meet standards, documenting the extent and nature of deviations, analyze the logged data to determine necessary adjustments for each identified region, and prepare software fixes to correct the discrepancies. In some embodiments, the post-process algorithm may enhance the precision in identifying regions needing demura correction. Initially, optical scans may not fully capture subtle uniformity deviations, potentially overlooking areas that could benefit from demura. However, a post-process algorithm trained on a vast array of data may detect intricate patterns associated with visual discrepancies, which may reveal that regions previously deemed acceptable require demura adjustment, or vice versa, even though a threshold is met to ensure optimal display quality. In some embodiments, the post-process algorithm may utilize support vector machines to classify regions of microLED displays as defective or non-defective by analyzing features derived from optical scans, such as pixel intensity, color variance, and pattern irregularities.

[0102]

[0102] In some embodiments, the post-process algorithm may utilize autoencoders that may learn to compress and decompress microLED display images, with the reconstruction error highlighting anomalies to identify subtle defects that may not significantly alter the overall pattern but still deviate from the norm and regions with high reconstruction errors may then be flagged for further inspection or direct demura correction. The repair post-process module 114 performs, at step 510, the post-process inspection. For example, the repair post-process may include functional testing, uniformity testing, encapsulation, etc. The functional testing may include performing an exhaustive battery of tests to validate the electrical capabilities of every microLED, encompassing precise brightness modulation and on / off capabilities. The uniformity testing may include verifying the consistency of visual output throughout the entire display and fixing any anomalies that are identified. The encapsulation step may protect the microLEDs from environmental factors such as dust and moisture, which have the potential to degrade their performance over time, by applying protective layers over them. The repair post-process module 114 stores, at step 512, the data in the repair post-process database 124. The repair post-process module 114 stores the data in the repair post-process database 124, such as the data collected from the repair post-process 150, the output of the post-process algorithm, etc. The repair post-process database 124 may contain the substrate ID, the uniformity deviation, the pattern analysis outputted by the post-process algorithm, the decision to apply the demura algorithm from basic thresholds, and the decision to apply the demura algorithm from the pattern analysis outputted by the post-process algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair postprocess 150. The repair post-process module 114 returns, at step 514, to the base module 108.

[0103]

[0103] The technique focuses on the role that the repair post-process module 114 plays in the quality assurance stage following microLED repairs. The post-process algorithm, which is carried out at step 508, uses advanced methods such as demura to rectify visual disparities in microLED displays and guarantee consistency in color and brightness throughout the display. Convolutional Neural Networks (CNNs) are a straightforward yet powerful Al method appropriate for this stage. Given their specialization in visual data analysis, CNNs are perfect for analyzing high-resolution images of microLED displays in order to identify and address uniformity problems. A CNN could be taught, for example, to identify patterns that indicate non-uniformity, like a series of faint pixels or regions with notable color variations. In order to achieve a uniform visual presentation across the display, the CNN might direct the use of demura methods upon identifying such abnormalities, modifying the hue and luminance of particular pixels or sub-pixels. Because CNNs can handle complicated visual data, they are ideal for this kind of work because they can spot minor differences that other algorithms or even human inspectors might miss. A CNN or other post-process technique would require high-resolution photos of microLED displays taken by colorimeters and photometers as training datasets. To guarantee that the CNN can effectively learn to spot inconsistencies, these photos should reflect a wide range of uniformity deviations, from slight to substantial. The outcomes of functional testing, which evaluates each microLED's electrical capacity, and uniformity testing data, which looks at the display's visual output consistency, could be considered additional data. These datasets are used to train the CNN, which helps the post-process algorithm identify and categorize uniformity problems according to their severity. Furthermore, by including reinforcement learning techniques into the system, the parameters of demura adjustments could be improved based on the results of prior corrections, so generating a feedback loop that continuously enhances display quality. By precisely adjusting to the unique problems found in each display, this adaptive technique maximizes visual quality and prolongs the life of the tiny LEDs. In conclusion, the repair post-process module 114 uses cutting-edge Al methods, such as CNNs, to guarantee the best possible display quality after repairs. The module resolves any residual disparities by evaluating comprehensive visual data and implementing remedial techniques like demura, guaranteeing that every tiny LED display satisfies strict quality standards prior to being purchased by customers. This procedure adds to the technology's longevity and dependability while also improving the end-user's visual experience.

[0104]

[0104] FIG. 6 illustrates the repair of historical module 116. The process begins with the repair of historical module 116 being initiated, at step 600, by the base module 108. The repair historical module 116 connects, at step 602, to the repair pre-process 130, the repair process 140, and the repair post-process 150. The historical module 116 connects to the repair pre-process 130, such as processes, such defect identification, defect classification, and deci si on -making procedures conducted prior to the repair processes and designed to ensure optimal performance during subsequent repair operations, the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc., and the repair post-process 150, such as the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair historical module 116 aggregates, at step 604, the data from the various types of processes and stores the data in the repair network database 126. The repair network database 126 may contain the historical data from the various processes performed by the repair pre-process 130, repair process 140, and repair post-process 150.

[0105]

[0105] The repair network database 126 may contain the data parameters collected during the inspection process, such as defect severity, defect type, the spatial distribution of defects, sample alignment, and positioning, ambient light interference, consistency of parameters, fluctuations in the spectral response of the measurement system, results from algorithms for data analysis and processing, etc., control parameters of the process, such as laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, feedback from sensors, monitoring systems, inspection equipment, start and stop times for process steps, delays or interruptions encountered, energy consumption measurements or power usage associated with the repair equipment and processes employed, etc., and data parameters collected during postprocessing, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc. The repair historical module 116 performs, at steps 606, the historical machine learning algorithm on the historical data stored in the repair network database 126. The historical machine learning algorithm may improve each stage of the microLED repair process, such as repair pre- process 130, repair process 140, and repair post-process 150. For example, the historical machine learning algorithm may identify defects and assess the need for repair using initial photoluminescence and electroluminescence measurements and temperature data. The historical machine learning algorithm may utilize convolution neural networks to perform pattern recognition on photoluminescence and electroluminescence data to predict potential defects. The historical machine learning algorithm may be a machine learning regression model to correlate temperature variations with photoluminescence and electroluminescence measurements to identify anomalies that indicate repair needs. The historical machine learning algorithm may be trained on historical PL, EL, and temperature data, to learn patterns that lead to successful repairs and predict areas that need attention before actual repair to optimize planning and resource allocation by focusing on high-risk regions. The historical machine learning algorithm may provide real-time assistance and optimization during the repair process by guiding the process using Delta E values to ensure targeted and efficient repairs. The historical machine learning algorithm may perform reinforcement learning to adapt repair strategies based on Delta E measurements and initial repair decisions. The historical machine learning algorithm may utilize support vector machines to classify areas as needing repair based on Delta E thresholds and historical outcomes. The historical machine learning algorithm may dynamically adjust repair techniques, focusing on areas where Delta E values exceed thresholds that historically correlate with necessary repairs, utilize decision trees, or support vector machine outputs to confirm or adjust initial repair decisions to ensure precision in targeting defects. In some embodiments, a comprehensive history of Delta E metrics and the success rates of subsequent repairs may allow the historical machine learning algorithm to set more accurate thresholds for repair decisions.

[0106]

[0106] In some embodiments, the data may reveal which Delta E values are true indicators of a need for repair to reduce unnecessary interventions and focus efforts on genuinely impactful corrections. The historical machine learning algorithm may provide quality assessment for the repair post-process by validating repair quality and fine-tuning future repair actions based on uniformity deviation and convolutional neural network analysis. The historical machine learning algorithm may be a supervised learning model that evaluates post-repair uniformity against quality standards. The historical machine learning algorithm may utilize convolutional neural networks to identify subtle patterns in uniformity deviations. The historical machine learning algorithm may assess post-repair uniformity deviation and apply basic demura corrections where deviations exceed acceptable thresholds and further refine the analysis with convolutional neural networks to identify areas where demura based on basic thresholds might be insufficient or unnecessary. The historical machine learning algorithm may implement demura corrections informed by convolutional neural network analysis to target specific areas with precision to optimize the visual quality of the display. In some embodiments, the historical data on post-repair uniformity and the effectiveness of applied demura algorithms may allow the historical machine learning algorithm to assess repair quality and the necessity for further adjustments.

[0107]

[0107] In some embodiments, the historical machine learning algorithm may learn from past successes and failures and adjust demura parameters to achieve optimal results with greater efficiency.

[0108]

[0108] In some embodiments, the historical machine learning algorithm adjusts the parameters in real-time for the repair inspection module 110, repair process module 112, and repair post-process module 114 to correct any defects.

[0109]

[0109] LLM Example 1 : In some embodiments, the repair historical module 116 may incorporate a large language model (LLM) configured to generate causal explanations and actionable summaries from multi-run defect and repair records stored within the repair network database 126. For example, a process engineer may prompt, “Why did microcrack recurrence increase on repair line 4 over the past five shifts?” while a yield analyst may query, “Summarize the most frequent corrective actions linked to lift-off errors this quarter.” In response, the repair historical module 116 may retrieve classification logs from the repair inspection database 120, thermal profile and mechanical actuation data from the repair process database 122, and post-repair evaluation outcomes from the repair post-processing database 124. The LLM may semantically segment these records, correlate feature combinations with elevated failure recurrence, and compose a natural language explanation attributing the rise in microcracks to an increased incidence of mechanical overshoot during beam alignment. Internally, the model may rank potential causes by co- occurrence weight and cite similar trends from prior timeframes or literature sources on microLED repair-induced damage mechanisms. Outputs may include an annotated summary detailing causal linkages and operator-correctable parameters, along with supporting references aligned to internal corrective action frameworks. By embedding an LLM within the repair historical module 116, the system may convert dense historical data into semantically structured insights that inform both short-term troubleshooting and long-term yield strategy. The repair historical module 116 sends, at step 608, the process adjustments to the repair inspection module 110, the repair process module 112, and the repair post-process module 114. The historical machine learning algorithm may identify defects and assess the need for repair using initial PL and EL measurements and temperature data, guide the repair process using Delta E values to ensure targeted and efficient repairs, validate repair quality, and fine-tune future repair actions based on uniformity deviation and pattern analysis. The adjustments are sent to the repair inspection module 110 and repair process module 112 to create a more consistent final product which is then verified by the repair post-process module 114. The repair historical module 116 returns, at step 610, to the base module 108.

[0110] [HO] As described, the repair historical module 116 plays a fundamental function in utilizing historical data to improve and optimize the microLED repair procedures throughout their whole lifecycle, encompassing pre-processing, active repair phases, and post-processing evaluations. The originality of this module is mostly found in step 608's implementation of a historical machine learning algorithm, which evaluates gathered data to guide and enhance subsequent repair operations. This machine learning approach is multidimensional rather than solitary, using a variety of algorithms that are specific to the data's characteristics and the analytical goals at different points in the repair process. The decision tree method is a straightforward yet effective Al algorithm that can be very useful in this situation, especially for categorizing flaws and directing repair choices. Decision trees are useful for analyzing the intricate datasets used in the repair processes since they are simple to use and can handle both continuous and categorical data. They function by dividing a dataset into more manageable portions and concurrently creating a decision tree that goes along with it. The most informative elements in this tree — which are chosen using statistical metrics — provide the basis for decision-making. To anticipate possible flaws that might require repairs, for example, a decision tree could be used to evaluate preliminary electroluminescence (EL) and photoluminescence (PL) measurements along with temperature information. This forecast can aid in determining the order of importance for repairs, concentrating resources on areas most likely to fail, and eventually improving the effectiveness of the repair procedure. Decision trees may also be utilized in post-repair quality evaluations, where the success of the repairs and the necessity for additional changes are assessed by analyzing uniformity deviations and Delta E values. The datasets that are essential to the historical machine learning algorithm comprise a diverse range of parameters that are gathered throughout the restoration process. This comprises information from the pre-process stage regarding defect kind, severity, and spatial distribution; control parameters from the active repair phase, such as voltage, ion beam current, and laser power; and data from the post-repair assessments regarding electrical testing and uniformity testing. These datasets all offer insightful information that, when processed by machine learning algorithms, can greatly improve the accuracy and efficiency of the repair procedures. The decision tree algorithm is trained on this historical data to identify trends that result in successful repairs and pinpoint regions that require attention, ultimately optimizing resource allocation and planning. This learning process is not static; instead, the models of the algorithm are updated in response to fresh data, guaranteeing that the decision-making process is always evolving to take into account the most recent discoveries. Additionally, other advanced techniques like support vector machines for classification tasks and convolutional neural networks for pattern recognition are used by the historical machine learning algorithm. By offering sophisticated analytic skills for intricate patterns in data that the decision tree might oversimplify, these algorithms work in tandem with decision trees. These algorithms work together to create a thorough analytical framework that allows the historical module to offer useful insights and steer the repair process in the direction of more reliable and superior results.

[0111] [Hl] FIG. 7 illustrates the repair integration module 118. The process begins with the repair integration module 118 being initiated, at step 700, by the base module 108. The repair integration module 118 performs, at step 702, the integration machine learning algorithm. The integration machine learning algorithm may facilitate interactions between the repair pre-process 130, repair process 140, and repair post-process 150 stages of repairing microLEDs. For example, the integration machine learning algorithm may optimize each step, such as before repair, during repair, and after repair. The integration machine learning algorithm may use the initial analysis of photoluminescence and electroluminescence measurements along with temperature data to predict potential defects and the predictions and characterizations made before repairs, such as identifying regions at risk based on photoluminescence, electroluminescence, and temperature correlations may provide a baseline for evaluating the effectiveness of the repair process. This data may inform the after-repair assessment by setting expectations for the uniformity and quality improvements that repairs should achieve to help to calibrate the demura algorithms more accurately for specific types of discrepancies identified in the pre-repair phase. The integration machine learning algorithm may use insights from the during-repair phase, particularly in which Delta E thresholds led to successful repairs, which are fed back to refine the predictive models used before repair. This feedback loop may enhance the accuracy of defect predictions, allowing for more nuanced decisions about which regions truly require repair and potentially reducing unnecessary repair interventions. The integration machine learning algorithm may use the specific Delta E values and the corresponding decisions to repair as input for assessing the after-repair quality. In some embodiments, this may include evaluating how well the repairs addressed the identified issues and whether the chosen strategies led to the desired uniformity improvements. This data may be used to fine-tune the demura algorithms applied after repair to ensure they are appropriately calibrated to the nature and severity of the defects addressed during the repair process. The integration machine learning algorithm may use before-repair predictions to inform after-repair evaluations by establishing a baseline expectation for display quality improvements, the pre-repair analysis may influence how after-repair quality is assessed to ensure a targeted approach to applying demura corrections, where the integration machine learning algorithm adjusts its parameters based on the specific defects identified initially. The integration machine learning algorithm may be used during repair insights to refine before-repair predictions by learning from the outcomes of repair decisions to predict future repair needs more accurately allowing for a more effective allocation of resources and focus repair efforts where they are most needed and likely to have a significant impact. The integration machine learning algorithm may be used during repair decisions to enhance after-repair quality control by closely analyzing the outcomes of specific repair interventions to improve the after-repair assessment and demura application process and ensure that the demura algorithms are precisely tailored to the types of uniformity issues the repair process typically encounters which may lead to higher quality displays.

[0112]

[0112] In some embodiments, the integration machine learning algorithm may use the data stored in the repair network database 126 to perform a predictive modeling algorithm to make predictions or forecasts of the final product based on historical data and patterns from the previously created products. Predictive modeling involves the systematic analysis of historical data to identify patterns, correlations, and trends that can be used to build models capable of making accurate predictions about future outcomes. For example, the integration machine learning algorithm may be used for validating the repair quality and fine-tuning future repair actions across the repair pre- process 130, repair process 140, and repair post-process 150. The historical data in the repair network database 126 may include information on uniformity deviations and relevant operational parameters such as photoluminescence and electroluminescence average measurements, temperature, Delta E values, the decision to repair, the effectiveness of demura algorithms, etc. The data is cleaned and preprocessed, addressing any missing values and ensuring that all variables are in a suitable format for modeling. This may involve normalization or scaling of numerical features. The dataset is split into training and testing sets. The training set is used to train the predictive model, and the testing set assesses its performance on unseen data. The model used may be linear regression, decision trees, ensemble methods, neural networks, etc. Then key features are identified that influence the uniformity deviations and engineer new features if necessary. For example, repair data from the repair process 140 may have a significant impact on the uniformity deviations. The selected predictive model is trained using the training dataset. The model learns patterns and relationships between operational parameters and moisture removal. The model's performance is validated on a separate validation dataset and hyperparameters are fine-tuned to optimize its accuracy and generalization. The model is then evaluated and deployed by being sent to the repair pre-process 130, repair process 140, and repair post-process 150.

[0113]

[0113] Al Agent Example 2: In some embodiments, the repair integration module 118 may incorporate a dedicated Al agent configured to autonomously monitor failure trends, chamber state changes, and thermal event distributions across the repair inspection module 110, repair process module 112, and repair post-process module 114, and selectively trigger execution of the integration optimization algorithm at step 702 when real-time deviation thresholds are exceeded. For example, a staff engineer may specify, “Trigger optimization if subpixel edge deformation exceeds 1.5o in three consecutive lots,” while the agent may independently initiate, “Update beam trajectory compensation if chamber temperature variance exceeds 2°C within a 10-minute rolling window.” In operation, the agent may analyze defect classifications from the repair inspection database 120, process telemetry such as beam latency, power overshoot, and thermal drift from the repair process database 122, and post-repair uniformity scores from the repair post-processing database 124. When cross-module indicators meet or exceed pre-defined deviation logic, the agent may autonomously activate step 702 and dispatch revised control parameters through the repair integration module 118. Internally, the agent may use reinforcement learning policies trained on recovery efficacy and yield impact metrics stored in the repair network database 126. Outputs may include updated parameters, a rationale trace with associated thresholds and trigger sources, and a timestamped control event record for traceability and review. By embedding an autonomous Al agent in the repair integration module 118, the system may shift from passive optimization to active, data-driven coordination across modules, improving responsiveness to evolving repair conditions. The repair integration module 118 connects, at step 704, to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. The repair integration module 118 connects to the repair pre-process 130, such as processes, such defect identification, defect classification, and decision-making procedures conducted prior to the repair processes and designed to ensure optimal performance during subsequent repair operations, the repair process 140, such as systems or equipment for automated repair of microLEDs to identify and fix defects or errors that occur during the production of microLED displays, such as defects occurring during the growth of the LED material, the fabrication of the microLED chips, the transfer of microLEDs onto the display substrate, the assembly of the final display, etc., and the repair post-process 150, such as the processes and methods after the completion of the repair process 140, such as adhering the microLED substrate that has been repaired to the display backplane to guarantee alignment and interconnection of every microLED with its respective electronic driver through functional testing, uniformity testing, encapsulation, etc. The repair integration module 118 sends, at step 706, the process adjustments to the repair pre-process Al module 136, the repair process Al module 146, and the repair post-process Al module 156. For example, the repair integration module 118 may predict potential defects based on the initial analysis of PL and EL measurements and temperature data. The integration machine learning algorithm may decide to repair based on Delta E values to fine-tune the demura algorithm that is applied after the repair.

[0114]

[0114] In some embodiments, the repair integration module 118 may send the integration machine learning algorithm to the processes, allowing the systems to further enhance the optimization.

[0115]

[0115] LLM Example 2: In some embodiments, the repair integration module 118 may employ a large language model (LLM) to generate operator-facing explanations and rationale reports for parameter updates produced by the integration optimization logic at step 702. For example, a shift lead may request, “Why were laser pulse timing and cooling interval adjusted on Tool 17 for Lot R654?” while a training coordinator may prompt, “Summarize the last five parameter changes issued for lift-off stabilization and their associated root causes.” In response, the repair integration module 118 may retrieve the optimization output issued at step 702, correlate it with process signatures from the repair process database 122, recent inspection triggers from the repair inspection database 120, and surface outcome flags from the repair post-processing database 124. The LLM may then construct a natural language explanation linking the pulse timing shift to a detected rise in subpixel edge curling, citing temperature overshoot during recent beam realignments as the inferred cause. Internally, the model may reference similar historical adaptations in the repair network database 126 and apply attend on -weighted reasoning to rank contributing features. Outputs may include a decision trace with semantic justifications, technician- readable annotations, and citations to relevant tooling bulletins or defect evolution patterns. By embedding an LLM into the repair integration module 118, the system may improve explainability of Al-driven control adjustments, support technician retraining, and reduce intervention ambiguity during live operations. The repair integration module 118 returns, at step 708, to the base module 108.

[0116]

[0116] An integration machine learning algorithm is used in step 702 of the repair integration module 118, as described in the description. This algorithm plays a crucial role in coordinating the many phases of the microLED repair procedure. To guarantee a unified and clever strategy for microLED repair and optimization, this algorithm serves as a link between pre-repair analysis, real repair interventions, and post-repair quality evaluations. Predictive modeling is an Al method that is easy to use and very successful at performing this function. This kind of algorithm is very useful for anticipating possible flaws, maximizing repair choices, and guaranteeing quality control since it is skilled at evaluating both history and present data to predict future results. One example of this kind of application may be the use of linear regression, a simple yet effective predictive modeling method. In order to forecast regions of the microLED display that would exhibit possible flaws, linear regression could examine historical temperature data as well as photoluminescence (PL) and electroluminescence (EL) measurement data. By concentrating on high-risk regions during the repair process, this predictive capability improves the effectiveness and efficiency of repairs. Additionally, in order to help refine the thresholds for subsequent repairs, linear regression models might assess the correlation between the Delta E values acquired during the repair process and the success rate of these repairs in obtaining the necessary uniformity. The integrated machine learning method would use extensive datasets that included information from every stage of the healing process. This comprises the pre-repair phase's spatial distribution of defects, temperature readings, initial PL and EL measurements, defect severity and type, control parameters like laser power, ion beam current and voltage during the repair phase, and uniformity measures after the repair. It would also be crucial to collect more information on the efficacy of demura algorithms, and the quality gains made possible by repairs. The predictive modeling method can find patterns and connections that might not be immediately obvious by training on this large dataset. For example, it might identify particular Delta E value ranges that, according to past data, are associated with a high chance of successful repair, offering a more sophisticated foundation for repair choices. Likewise, by the examination of post-repair uniformity data, the system can adjust the demura adjustments to more precisely target certain disparities, resulting in displays of improved quality. In real life, feedback loops allow the integrated machine learning algorithm to continuously improve its models.

[0117]

[0117] The consequences of repair decisions provide insights that are sent back to enhance the calibration of demura algorithms and the precision of fault predictions. This iterative method helps to focus efforts where they are most needed and likely to provide major changes, improving repair precision while also contributing to a more efficient use of resources. The integrated machine learning algorithm, which constantly modifies its settings depending on an ongoing examination of past and present data, essentially serves as the repair process's central intellect. This makes sure that decisions about repairs are based on the most thorough and current information available, which promotes a more flexible, effective, and efficient repair procedure.

[0118]

[0118] FIG. 8 illustrates the repair inspection database 120. The repair inspection database 120 provides an example of the results of the inspection algorithm performed in the repair inspection module 110. The repair inspection database 120 displays an example of variations in photoluminescence and electroluminescence average lot measurements, along with corresponding temperature changes due to normal test process variations. The PL and EL average measurements are centered around 450 nm, a typical value for microLEDs, with variations of ±10%. Temperatures are set around room temperature, 25°C, with process variations observed during testing. The example demonstrates how PL and EL measurements may vary with temperature changes, reflecting the temperature sensitivity of microLED emission properties. This second example illustrates that predictive temperature control systems outputted by the inspection algorithm may achieve temperature regulation within ±1% of a desired set point ensuring optimal conditions for microLED testing and reducing the risk of temperature-induced variability in PL and EL measurements. The second example illustrates tighter control over temperature variations and PL and EL measurements for 10 substrates. The example includes a more controlled range of photoluminescence and electroluminescence measurements to reflect ±1% precision, alongside temperature changes, also maintained within a ±1% variation from 25°C. One substrate is significantly out of specification to illustrate a potential case for repair: The second example demonstrates a situation with no apparent up or down temperature trend from row 11 to row 20, providing an example of the variability across different substrates. Substrate_18 is out of specification for PL and EL measurements which indicates a need for repair or further investigation. In some embodiments, the repair inspection database 120 may contain parameters, optimized parameters, parameter adjustments from the inspection algorithm, etc.

[0119]

[0119] FIG. 9 illustrates the repair process database 122. The repair process database 122 provides an example of the results of the control algorithm performed in the repair process module 112. The repair process database 122 may contain the substrate ID, the Delta E value, the position, the ambient temperature, and the repair decision outputted by the control algorithm, etc. The repair process database 122 may contain the data for each substrate that is being processed by the repair process 140. For example, a microLED display with minor inconsistencies in brightness across its surface but these variances fall within the acceptable quality specification range for uniformity. The decision would be not to repair as the quality already meets the set standards and intervention could introduce unnecessary risk without significant benefit. Another example may be a display with several microLEDs showing significant color shifts and dead pixels that are visible during standard operation and significantly impact the user experience. These issues fall outside the acceptable quality specification range making repair essential to meet the product's quality standards and ensure customer satisfaction. The control algorithm may decide on repairing or replacing based on the balance between the expected quality improvement and the resources required for repair. The control algorithm may use "Delta E" as a quality metric, which is a standard measure of color difference or accuracy in displays and the control algorithm may use a threshold for perceptible color difference is a Delta E value of 1. Delta E is a metric used to quantify the difference between two colors to understand how perceptibly different they are. Lower Delta E values may indicate less difference and a more uniform color appearance, which is essential for high-quality microLED display performance. This measurement helps identify when specific microLEDs deviate from expected color outputs to guide decisions on whether adjustments or repairs are necessary to meet stringent display quality standards. The control algorithm may use decision tree analysis which is a predictive modeling tool that splits data into branches to represent decision-making paths. It begins at a root node, the initial decision point, and splits into branches based on criteria, leading to leaf nodes representing outcomes. Evaluating how different features affect outcomes, helps in making informed decisions based on patterns found in historical data. For example, in the decision for substrate 18, the control algorithm may evaluate the Delta E value, which may exceed 1.5 and regardless of its position and ambient temperature, the substrate is directed towards the "Repair" decision due to the high Delta E value indicating significant color discrepancy. Another example may be if the control algorithm evaluates substrate 11 in which the Delta E value is above 1.0 but below the 1.5 thresholds, the control algorithm may then consider its position, such as edge, and ambient temperature, and given the lower risk associated with edge positions and the cooler temperature, which might mitigate some thermal effects on Delta E, the decision could lean towards "No Repair", assuming these conditions historically result in acceptable final quality without intervention.

[0120]

[0120] In some embodiments, the repair process database 122 may store other parameter data collected or calculated from the repair process module 112, such as repair method parameters including laser power and duration, ion beam current and voltage, deposition rate, chemical treatment conditions, etc., time and duration data including start and stop times for process steps, delays or interruptions encountered, etc., energy and power consumption data including energy consumption or power usage associated with the repair equipment and processes employed, etc.

[0121]

[0121] FIG. 10 illustrates the repair post-process database 124. The repair post-process database 124 provides an example of the results of the post-process algorithm performed in the repair postprocess module 114. The example data illustrates the substrate ID, the uniformity deviation, the pattern analysis outputted by the post-process algorithm, the decision to apply the demura algorithm from basic thresholds, and the decision to apply the demura algorithm from the pattern analysis outputted by the post-process algorithm, etc. The first example illustrates various substrates post-repair, where there is “Uniformity Deviation" as a post-repair test result, where higher values indicate worse uniformity across the display. In this example, a threshold for applying the demura algorithm may be set around 4%, meaning any substrate with a uniformity deviation above this value would be considered for the corrective measure. Examples of uniformity measurements for microLEDs post-repair may include luminance uniformity, color uniformity, and white point variation. These measurements may assess how evenly light is distributed across the display and how consistent the color output is from pixel to pixel. High deviations in these areas after repairs may indicate unresolved issues, necessitating further adjustments or corrections, such as applying a demura algorithm to improve display uniformity and overall visual quality. The second example illustrates how the post-process algorithm enhances the precision in identifying regions needing demura correction. Initially, optical scans may not fully capture subtle uniformity deviations, potentially overlooking areas that could benefit from demura. However, once the postprocess algorithm is trained on a vast array of data it may detect intricate patterns associated with visual discrepancies. The post-process algorithm may reveal that regions previously deemed acceptable require demura adjustment, or vice versa, even though a threshold is met.

[0122]

[0122] In some embodiments, the repair post-process database 124 may store other parameter data collected or calculated from the repair post-process module 114, such as testing electrical capabilities, uniformity, determining anomalies or defects, etc.

[0123]

[0123] The functions performed in the processes and methods may be implemented in differing order. Furthermore, the outlined steps and operations are only provided as examples, and some of the steps and operations may be optional, combined into fewer steps and operations, or expanded into additional steps and operations without detracting from the essence of the disclosed embodiments.

Claims

CLAIMSWhat is claimed is:

1. An automated repair method (ARM) for microLED chips system, comprising: an automated repair inspection module that measures at least one parameter from a first group of parameters and calculates an output of at least one parameter from a second group of parameters to help control the automated repair processing module; an automated repair processing module which inputs the at least one parameter from said second group to control the repair using at least one parameter from a third group of parameters; an automated repair post-processing module which measures at least one parameter from a fourth group of parameters that determines how well the parameters of said first and second group were running; an automated repair historical module which calculates a plurality of group parameters based upon parameters of the first group, second group, third group, and fourth group; and an automated repair integration module to update the automated repair processing module based upon the results of the automated repair historical module.

2. The method of claim 1, wherein the ARM process module acts as the core controller, utilizing inputs from the ARM inspection module to govern the repair process to properly identify defects that also includes flatness, decide on replacement or repair defective LEDs.

3. The method of claim 1, wherein the ARM post-process module conducts inspections to assess the execution of parameters and identify potential deviations, and the ARM historical module analyzes data collected across various stages, allowing for continuous learning and pattern recognition and subsequently an ARM integration module then employs historical insights to update and optimize the process, enabling adaptive improvements over time.

4. The method of claim 1, wherein a communication interface, which is a hardware or a software component that enables communication between an ARM Al network and an ARM pre- process, the ARM process, and a ARM post-process.

5. The method of claim 4, wherein the communication interface comprises a set of protocols, rules, and standards that define how information is transmitted and received between devices is a physical connector, wireless network, or software application and include components such as drivers, software libraries, and firmware that are used to control and manage the communication process.

6. The method of claim 4, wherein the communication interface is compatible with a USB, aBluetooth, or a Wi-Fi and the communication interface communicates with a network.

7. The method of claim 1, wherein there is a memory that comprises suitable logic, circuitry, and / or interfaces that are configured to store a machine code and / or a computer program with at least one code section executable by a processor.

8. The method of claim 1, wherein there is a base module, which initiates the ARM inspection module, the ARM process module, the ARM post-process module, the ARM historical module, and the ARM integration module.

9. The method of claim 8, wherein the ARM inspection module begins by being initiated by the base module wherein the ARM inspection module connects to the ARM pre-process wherein the ARM inspection module collects data from the ARM pre-process.

10. The method of claim 8, wherein the ARM inspection module performs an inspection algorithm, and the ARM inspection module sends the determined process data from the inspection algorithm to the ARM process module wherein the ARM inspection module stores the data in the ARM inspection database and ARM inspection module returns to the base module.

11. The method of claim 8, wherein the ARM process module begins by being initiated by the base module wherein the ARM process module receives the process data, such as the data on the microLED processed wafer, from the ARM inspection module and the ARM process module connects to the ARM process.

12. The method of claim 11, wherein the ARM process module collects the data from the ARM process and the ARM process module performs a control algorithm and the ARM process module adjusts the control parameters of the ARM process and executes the ARM process wherein ARM process module stores the data in the ARM process database and the ARM process module returns to the base module.

13. The method of claim 8, wherein the DB post-process module begins by being initiated by the base module and the DB post-process module connects to the DB post-process wherein the DB post-process module collects the post-process data.

14. The method of claim 13, wherein the ARM post-process module extracts the data from the ARM inspection database and ARM process database and performs the post-process algorithm subsequently the ARM post-process module performs the post-process inspection wherein the ARM post-process module stores the data in the ARM post-process database and the ARM postprocess module returns to the base module.

15. The method of claim 8, wherein the ARM historical module begins by being initiated by the base module and the ARM historical module connects to the ARM pre-process, the ARM process, and the ARM post-process wherein the ARM historical module aggregates data from the various types of processes and stores the data in the ARM network database.

16. The method of claim 15, wherein the ARM historical module performs a historical machine learning algorithm on the historical data stored in the ARM network database and sends the process adjustments to the ARM inspection module, the ARM process module, and the ARM post-process module and subsequently the ARM historical module returns to the base module.

17. The method of claim 8, wherein the ARM integration module begins by being initiated by the base module and the ARM integration module performs the integration machine learning algorithm and connects to the ARM pre-process Al module, the ARM process Al module, and the ARM post-process Al module.

18. The method of claim 17, wherein the ARM integration module sends the process adjustments to the ARM pre-process Al module , the ARM process Al module , and the ARM post-process Al module and the ARM integration module returns to the base module.

19. The method of claim 10, wherein there is an ARM inspection database which provides results of the inspection algorithm performed in the ARM inspection module displaying data such as, but not limited to, variations in photoluminescence and electroluminescence average lot measurements, along with corresponding temperature changes due to normal test process variations.

20. The method of claim 19, wherein the ARM inspection database includes other parameter data collected or calculated during the ARM inspection modules process, such as die characteristics, substrate cleanliness, adhesive thickness and electrical resistance.

21. The method of claim 19, wherein PL and EL average measurements are centered around 450 nm, a typical value for microLEDs, with variations of ±10% and temperatures are set around room temperature, 25°C, with process variations observed during testing showing PL and EL measurements may vary with temperature changes, reflecting the temperature sensitivity of microLED emission properties.

22. The method of claim 21, wherein predictive temperature control systems output by the inspection algorithm achieve temperature regulation within ±1% of a desired set point ensuring optimal conditions for microLED testing and reducing the risk of temperature-induced variability in PL and EL measurements23. The method of claim 22, wherein there is tighter control over temperature variations and PL and EL measurements for substrates.

24. The method of claim 22, wherein there is a more controlled range of photoluminescence and electroluminescence measurements to reflect ±1% precision, alongside temperature changes, also maintained within a ±1% variation from 25°C.

25. The method of claim 22 wherein there is a situation with no apparent up or down temperature trend providing variability across different substrates wherein if a substrate is out of specification for PL and EL measurements it indicates a need for repair.

Images