An automated ai-based die bonder for microleds

Abstract

The present disclosure provides an automated die bonding of LED substrates system in which an automated die bonding inspection module is used to determine the quality of incoming processing material and an automated die bonding process module is used to process the processing material, an automated die bonding post-process module is used to inspect the resultant process material for acceptability, and an automated die bonding 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 die bonding process module and an automated die bonding integration module to update the automated die bonding process module based upon the results of the automated die bonding historical module.

Description

[0001] AN AUTOMATED AI-BASED DIE BONDER FOR MICROLEDS

[0002] BACKGROUND AND FIELD OF THE INVENTION

[0003] [1] The present disclosure is generally related to an automated die bonding process of microLED chips system.

[0004] [2] Currently, the pre-process stage of microLED manufacturing faces significant challenges in cleaning and surface preparation. Conventional cleaning techniques struggle to effectively remove particulate matter and residues from microLED surfaces, potentially compromising bond quality. Furthermore, surface preparation methods must address the unique dimensions and susceptibility of microLEDs to ensure optimal adhesion and bonding during subsequent processing steps. Also, the die bonding process in microLED manufacturing encounters challenges related to precision and uniformity. Achieving precise placement and bonding of microLED dies onto substrates is essential for ensuring device performance and reliability. However, factors such as material conductivity, curing control in epoxy attachment, temperature regulation, pressure application, and quality evaluation pose challenges in maintaining consistent and uniform bonds across large-scale production. Lastly, the post-process stage of microLED manufacturing presents challenges in quality control and inspection. Ensuring the accuracy and reliability of microLED devices requires sophisticated inspection and measurement techniques capable of addressing the minute dimensions and high pixel density of microLED arrays. Challenges include precise placement on substrates, variability in luminance and color, and the need for advanced inspection methods to maintain consistent performance across millions of pixels. Thus, there is a need in the prior art for an automated die bonding process of microLED chips system. SUMMARY

[0005] [3] The present invention relates to an automated die bonding method for microLED chips system, comprising: an automated die bonding 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 die bonding processing module; an automated die bonding processing module which inputs the at least one parameter from said second group to control the die bonding using at least one parameter from a third group of parameters; an automated die bonding 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 die bonding 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 die bonding integration module to update the automated die bonding processing module based upon the results of the automated die bonding historical module.

[0006] BRIEF DESCRIPTIONS OF THE DRAWINGS

[0007] [4] FIG. 1 : Illustrates an automated die bonder for microLEDs, according to an embodiment.

[0008] [5] FIG. 2: Illustrates a Base Module, according to an embodiment.

[0009] [6] FIG. 3 : Illustrates a DB Inspection Module, according to an embodiment.

[0010] [7] FIG. 4: Illustrates a DB Process Module, according to an embodiment.

[0011] [8] FIG. 5: Illustrates a DB Post-Process Module, according to an embodiment.

[0012] [9] FIG. 6: Illustrates a DB Historical Module, according to an embodiment.

[0013]

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

[0014]

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

[0015]

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

[0016]

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

[0017] DETAILED DESCRIPTION

[0018]

[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.

[0019]

[0015] FIG. 1 illustrates a system for an automated die bonder for microLEDs. This system comprises a 3rd party die bonding Al network 102 which integrates pre-quality inspection, processing, post-processing, historical analysis, and continuous improvement in a die bonding system. The DB inspection module 110 measures essential parameters and calculates outputs to guide the subsequent die bonding process. The DB process module 112 acts as the core controller, utilizing inputs from the DB inspection module 110 to govern the die bonding process to ensure optimal pressure, temperature, duration, etc. The DB post-process module 114 conducts inspections to assess the execution of parameters and identify potential deviations. The DB historical module 116 analyzes data collected across various stages, allowing for continuous learning and pattern recognition. The DB integration module 118 then employs historical insights to update and optimize the process, enabling adaptive improvements over time.

[0020]

[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 die bonding Al network 102 and the die bonding pre-process 128, die bonding process 134 and die bonding 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).

[0021]

[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.

[0022]

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

[0023]

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

[0024]

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

[0025]

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

[0026]

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

[0027]

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

[0028]

[0024] Further, embodiments may include a DB inspection database 120, which provides an example of the results of the inspection algorithm performed in the DB inspection module 110. The DB inspection database 120 displays an example of data for various groups of microLED wet etching that will be die bonded to a substrate, where each group has an average electrical resistance where the variation may be based on normal variations of other tool parameters, such as pressure, time, atmospheric conditions. This example represents example data of electrical resistance changes against the normal temperature variations of the tool, where it may be controlled at 250C but varies due to tool control which in this example may vary +-10% for example. An analysis of data of all the tool parameters may show that time at a particular temperature is also a factor that changes electrical resistance in a run, for example, by a regression analysis of time against electrical resistance. For example, a strong correlation may be found between time and electrical resistance. Given this correlation, the next example shows example data where the set point was set at 250 C, but the time of each run was changed to compensate for the actual temperature changes resulting in electrical resistance that is far less and varies less. In some embodiments, the DB inspection database 120 may include other parameter data collected or calculated during the DB inspection module 110 process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc.

[0029]

[0025] Further, embodiments may include a DB process database 122, which provides an example of the results of the control algorithm performed in the DB process module 112. For example, die bonding pressure may refer to the force applied during the bonding process to attach a semiconductor die, such as a microLED, to its substrate or package and ensures proper contact between the die and the substrate, allowing for efficient heat dissipation, electrical connectivity, and mechanical stability. The factors that influence die bonding pressure are material properties, die size and shape, adhesive type, process temperature, etc. In some embodiments, the bonding pressure for microLEDs may be between 50 g / mm2to 500 g / mm2. There may be pressure variations while maintaining pressure control in die bonding which may produce defects, such as voids. These variations may be caused by mechanical wear, calibration issues, temperature fluctuations, hydraulic system performance, material properties, process variability, etc. The first example displays the variations in pressure applied during the die bonding process and the voids produced during the process. The second example displays example data where the die bonding pressure changed within normal process variation, but the temperature was changed to compensate, where the temperature change was calculated by regression analysis from pressure versus temperature, that is as the pressure went up, the temperature would be reduced. The example data shows the voids are reduced and there is less variation than the example data above.

[0030]

[0026] In some embodiments, the control algorithm may control the process parameters, such as temperature, to adjust the pressure being exerted by the hydraulic system and components of the die bonder which can be altered by thermal expansion or contraction. In some embodiments, the DB process database 122 may store other parameter data collected or calculated from the DB process module 112, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc.

[0027] Further, embodiments may include a DB post-process database 124, which provides an example of the results of the post-process algorithm performed in the DB post-process module 114. The example data illustrates the average misalignment in the x-direction, which should be zero, to the measured final average luminescence of the group. For example, alignment accuracy may result in adverse effects on luminance output during the die bonding process. The typical alignment accuracy for calibrated alignment falls within the range of approximately 5 pm, or micrometers. The alignment accuracy may be influenced by roll stamp structure, X-Y theta stage performance, stage-roller orthogonality, roller roundness, etc. which can lead to mechanical stress, damage of the microLEDs, luminance variations, defects, etc. The alignment accuracy may be enhanced through machine calibration and precision, roll stamp design and geometry, roller roundness and orthogonality, alignment sensors, and feedback systems, material properties and wafer warpage, self-alignment techniques, active optical alignment, underfill material and void reduction, etc. The post-process algorithm may categorize the wafers based on their warpage, which is the degree of deformation or curvature in the wafer surface resulting from the etching process. By grouping wafers with similar warpage characteristics together, the post-process algorithm ensures that during the subsequent die bonding step, wafers with comparable warpage are bonded together to minimize overall warpage within each group, leading to reduced alignment challenges during bonding, as shown in the second example of data. For example, when wafers with similar warpage are bonded together, the overall warpage variation within each group is reduced. As a result, the alignment process during die bonding becomes more consistent and predictable. On average, the luminescence of microLEDs bonded using this optimized approach tends to be higher and exhibits a more tightly distributed pattern. This improvement is attributed to the reduced impact of warpage on alignment accuracy. With less variation in alignment, the bonding of microLEDs is more uniform, leading to improved luminous output and overall performance. In some embodiments, the DB post-process database 124 may store other parameter data collected or calculated from the DB post-process module 114, such as bond quality, defect detection, electrical performance, encapsulation quality, etc.

[0031]

[0028] Further, embodiments may include a DB network database 126, which may contain the historical data from the various processes performed by the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150. The DB network database 126 may contain the data parameters collected during the inspection process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc., control parameters of the process, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc., and data parameters collected during post-processing, such as bond quality, defect detection, electrical performance, encapsulation quality, etc.

[0032]

[0029] Further, embodiments may include a cloud 128, which may be a network of remote servers that provide on-demand computing resources and services over the internet. The cloud 128 may consist of a collection of servers, storage devices, and networking equipment. Users may access the cloud 128 through a variety of devices, such as computers, smartphones, and tablets, using internet connectivity. In some embodiments, the architecture of the cloud 128 may be based on a distributed computing model, with multiple servers working together to provide services to users.

[0033]

[0030] Further, embodiments may include die bonding pre-process 130, which may be the processes, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc. utilized before the die bonding process 140, designed to prepare the microLED processed wafers for optimal performance during the die bonding process. The pre-process may involve collecting a plurality of parameter data including die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. In some embodiments, the die bonding pre-process 130 may send and receive data from the 3rd party die bonding Al network 102, including parameter adjustments, control adjustments, material or workpiece data, decisions on material inspections, inputs to be used by the die bonding pre-process Al module 136, etc.

[0034]

[0031] Further, embodiments may include a communication interface 132, which may be a hardware or software component that enables communication between the die bonding pre-process 130 and the 3rd party die bonding Al network 102. In some embodiments, the die bonding pre- process 130 may communicate with the die bonding process 140, and die bonding 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 the 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. 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).

[0035]

[0032] 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 the 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.

[0036]

[0033] Further, embodiments may include a die bonding pre-process Al module 136 in which a predictive model may be performed to predict the final product of the die bonding process based upon the data collected from the die bonding pre-process 130, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, 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 In some embodiments, the die bonding pre-process Al module 136 may receive a predictive model from the DB integration module 118. In some embodiments, the die bonding pre-process Al module 136 may receive and send data to the DB inspection module 110. In some embodiments, the data received from the DB inspection module 110 may be inputted into the predictive model to determine if any parameters of the pre-process should be adjusted, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc.

[0037]

[0034] Further, embodiments may include a die bonding pre-process database 138 which may include data parameters collected from the die bonding pre-process 130, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. In some embodiments, the data stored in the die bonding pre-process database 138 may be sent to the DB inspection module 110 and / or the DB historical module 116.

[0038]

[0035] Further, embodiments may include die bonding process 140, which may be systems or equipment for automated die bonding of microLED chips that places the microLED dies onto a prepared substrate, secures the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength. The die bonding process 140 may involve collecting parameter data, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. The die bonding process 140 may include the bonding equipment, such as fully automatic die bonding machines, semi-automatic die bonders, flip chip bonders, thermocompression bonding machines, ultrasonic bonding systems, etc. In some embodiments, the die bonding process 140 may be performed by the FINEPLACER lambda 2, FINEPLACER femto 2, ASM ASMPT Xpeed 8, West-Bond 7476E semi-automatic die bonder, etc. In some embodiments, the die bonding process 140 may send and receive data from the 3rd party die bonding Al network 102, including parameter adjustments, control adjustments, inputs to be used by the die bonding process Al module 146, etc.

[0039]

[0036] Further, embodiments may include a communication interface 142, which may be a hardware or software component that enables the communication between the die bonding process 140 and the 3rd party die bonding Al network 102. In some embodiments, the die bonding process 140 may communicate with the die bonding pre-process 130, and die bonding 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. 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).

[0040]

[0037] 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.

[0041]

[0038] Further, embodiments may include a die bonding process Al module 146 in which a predictive model may be performed to predict the final product of the die bonding process based upon the data collected from the die bonding process 140, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, 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 In some embodiments, the die bonding process Al module 146 may receive a predictive model from the DB integration module 118. In some embodiments, the die bonding process Al module 146 may receive and send data to the DB process module 112. In some embodiments, the data received from the DB process module 112 may be inputted into the predictive model to determine if any parameters of the die bonding process 140 should be adjusted, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc.

[0042]

[0039] Further, embodiments may include a die bonding process database 148 which may include data parameters collected from the die bonding process 140, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. In some embodiments, the data stored in the die bonding process database 148 may be sent to the DB process module 112 and / or the DB historical module 116.

[0043]

[0040] Further, embodiments may include die bonding post-process 150, which may be the processes and methods after the completion of the die bonding process 140, such as a cleaning and inspection, electrical testing, encapsulation, etc. The post-process may involve parameters such as bond quality, defect detection, electrical performance, encapsulation quality, etc. The post-process may include using a visual inspection system, testing system, electrical testing system, etc. In some embodiments, the die bonding post- process 150 may send and receive data from the 3rd party die bonding Al network 102, including parameter adjustments, control adjustments, inputs to be used by the die bonding post-process Al module 156, etc.

[0044]

[0041] Further, embodiments may include a communication interface 152, which may be a hardware or software component that enables communication between the die bonding postprocess 150 and the 3rd party die bonding Al network 102. In some embodiments, the die bonding post-process 150 may communicate with the die bonding pre-process 130, and die bonding 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. 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).

[0045]

[0042] 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.

[0046]

[0043] Further, embodiments may include a die bonding 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 die bonding post-process 150, such as bond quality, defect detection, electrical performance, encapsulation quality, 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 In some embodiments, the die bonding postprocess Al module 156 may receive a predictive model from the DB integration module 118. In some embodiments, the die bonding post-process Al module 156 may receive and send data to the DB post-process module 114. In some embodiments, the data received from the DB post-process module 114 may be inputted into the predictive model to determine if any parameters of the die bonding post-process 150 should be adjusted, such as bond quality, defect detection, electrical performance, encapsulation quality, etc.

[0047]

[0044] Further, embodiments may include a die bonding post-process database 158 which may include data parameters collected from the die bonding post-process 150, such as bond quality, defect detection, electrical performance, encapsulation quality, etc. In some embodiments, the data stored in the die bonding post-process database 158 may be sent to the DB post-process module 114 and / or the DB historical module 116.

[0048]

[0045] FIG. 2 illustrates the base module 108. The process begins with the base module 108 initiating, at step 200, the DB inspection module 110. For example, the DB inspection module 110 begins by being initiated by the base module 108. The DB inspection module 110 connects to the die bonding pre-process 130. The DB inspection module 110 connects with the die bonding pre- process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc. In some embodiments, the DB inspection module 110 may transmit and receive data from the die bonding pre-process 130. In some embodiments, the die bonding pre-process 130 may measure die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. In some embodiments, the DB inspection module 110 may control or send inputs to control the die bonding pre-process 130. The DB inspection module 110 collects the data from the die bonding pre-process 130. The DB inspection module 110 collects the pre-processing data, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB 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 die bonding pre-process 130 is used to determine if a control parameter of the die bonding process should be adjusted or altered. The inspection algorithm may determine the die bonding temperature for the die bonding process based on the electrical resistance of the die determined in the die bonding pre-process 130. For example, the inspection algorithm may perform a process window experiment to optimize the temperature during die bonding process 140 based on the electrical testing data collected from the die bonding pre-process 130.

[0049]

[0046] The inspection may identify the key variables that may affect the electrical resistance of the die, such as temperature, pressure, time, material composition, etc. For each variable, the inspection algorithm may determine the range of values to be tested, including expected operational values as well as extreme values to understand the limits of the process. The inspection algorithm may design the experimental setup, including the number of runs and the sequence of experiments, such as testing all possible combinations of the variables within the specified range. The inspection algorithm then conducts the experiments according to the predefined design ensuring that each combination of variables is tested accurately and consistently. The inspection algorithm records the data for each experiment run, including measurements, observations, and any other relevant information. The inspection algorithm may analyze the data to understand how changes in variables affect the electrical resistance, such as performing statistical analysis to identify significant factors and their interactions. The inspection algorithm determines the optimal conditions based on the analysis.

[0047] In some embodiments, the inspection algorithm may validate the results through additional testing or verification, implement any necessary changes to the process based on the findings of the experiment, and monitor and perform adjustments to maintain optimal performance. In some embodiments, the inspection algorithm may perform linear regression to determine linear relationships between time and electrical resistance to forecast the bonding time.

[0050]

[0048] In some embodiments, the inspection algorithm may be a neural network to simulate nonlinear bonding time and resistance relationships and may be trained to forecast ideal bonding times given resistance values. In some embodiments, the inspection algorithm may perform a random forest technique which may generate predictions by employing multiple decision trees. In some embodiments, the inspection algorithm may be an artificial intelligence or machine learning algorithm in which the data collected from the die bonding pre-process 130 is used to determine if a control parameter of the die bonding process should be adjusted or altered. The inspection algorithm may determine the electrical resistance and use this measurement to determine the temperature during the die bonding process. For example, the inspection algorithm may identify the electrical resistance and modify die bonding parameters to maintain consistent electrical resistance during the die bonding process, such as temperature, bonding pressure, alignment, etc. The inspection algorithm utilizes the captured data, from the die bonding pre-process 130, to send data about the microLED processed wafer to the DB process module 112 to be used to adjust die bonding parameters used by the die bonding process 138.

[0051]

[0049] In some embodiments, the DB inspection module 110 may send the output data from the inspection algorithm to the DB process module 112 to optimize the parameters used in the process. In some embodiments, the DB inspection module 110 may send the output data from the inspection algorithm to the DB process module 112 to optimize the parameters used in the process. The DB inspection module 110 sends the determined process data from the inspection algorithm to the DB process module 112. For example, the DB inspection module 110 may send the calculated time at a predetermined temperature for the die bonding process to the DB process module 112 to optimize the die bonding process.

[0052]

[0050] In some embodiments, the DB inspection module 110 may send die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB inspection module 110 stores the data in the DB inspection database 120. The DB inspection module 110 stores the data outputted from the inspection algorithm in the DB inspection database 120, such as the wafer group number or ID, the average pre-bonding electrical resistance of the contacts, the calculated time in minutes at 250 degrees Celsius for die bonding, etc. In some embodiments, the DB inspection database 120 may include other parameter data collected or calculated during the DB inspection module 110 process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB inspection module 110 returns to the base module 108. The base module 108 initiates, at step 202, the DB process module 112. For example, the DB process module 112 begins by being initiated by the base module 108. The DB process module 112 receives the pre- process data, such as the data on the microLED processed wafer, from the DB inspection module 110. The DB process module 112 receives the outputted data from the inspection algorithm performed in the DB inspection module 110, such as the calculated time at a predetermined temperature for the die bonding process, allowing the die bonding process 140 to determine the optimal control parameters to prevent any defects in the die bonding process. In some embodiments, the DB process module 112 may receive die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB process module 112 connects to the die bonding process 140. The DB process module 112 connects with the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that place the microLED dies onto a prepared substrate, and secure the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength.

[0053]

[0051] In some embodiments, the DB process module 112 may transmit and receive data from the die bonding process 140. In some embodiments, the DB process module 112 may control or send inputs to control the die bonding process 140, such as controlling parameters of the die bonding process, for example, pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. The DB process module 112 collects the data from the die bonding process 140. The DB process module 112 collects the processing data, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. The DB process module 112 performs the control algorithm. The control algorithm may output the pressure, temperature, pressure, etc. for the die bonding process to achieve optimal thermal management, mechanical robustness, and electrical performance in microLED die bonding.

[0052] In some embodiments, the die bonding process may include ultrasonic scanning to detect various defects, including delamination, voids, cracks, incomplete bonding, etc. For example, die bonding pressure may refer to the force applied during the bonding process to attach a semiconductor die, such as a microLED, to its substrate or package and ensures proper contact between the die and the substrate, allowing for efficient heat dissipation, electrical connectivity, and mechanical stability. The factors that influence die bonding pressure are material properties, die size and shape, adhesive type, process temperature, etc.

[0054]

[0053] In some embodiments, the bonding pressure for microLEDs may be between 50 g / mm2to 500 g / mm2. There may be pressure variations while maintaining pressure control in die bonding which may produce defects, such as voids. These variations may be caused by mechanical wear, calibration issues, temperature fluctuations, hydraulic system performance, material properties, process variability, etc. The control algorithm may control the process parameters, such as temperature, to adjust the pressure being exerted by the hydraulic system and components of the die bonder which can be altered by thermal expansion or contraction.

[0055]

[0054] In some embodiments, the control algorithm may utilize proportional-integral-derivative controllers to adjust the heating or cooling elements based on the difference between the desired setpoint temperature and the actual temperature by calculating the proportional, integral, and derivative terms to fine-tune the control action.

[0056]

[0055] In some embodiments, the control algorithm may utilize model predictive control to optimize control inputs, such as heater power, over a finite time horizon to minimize deviations from the setpoint and adapt to changing conditions and disturbances.

[0057]

[0056] In some embodiments, the control algorithm may utilize fuzzy logic control by using linguistic variables, such as hot or cold, and rules to adjust control actions. In some embodiments, the control algorithm may use the output of the inspection algorithm performed in the DB inspection module 110 and the data collected from the die bonding process 140 as input. The control algorithm may clean and preprocess the collected data to handle missing values, and outliers, and ensure uniformity, and the dataset is divided into training and testing sets. The key features are identified that significantly impact void defects occurring during the die bonding process, such as the temperature of the system and components while applying pressure during the die bonding process. The control algorithm may utilize the training dataset to train a linear regression model, which establishes a linear relationship between the selected features and the target variable, such as void defects occurring during the die bonding process. The trained model is then evaluated using the testing dataset to ensure its predictive accuracy.

[0058]

[0057] In some embodiments, metrics such as Mean Squared Error (MSE) or R-squared may be used to assess the performance of the linear regression model. The new data from the die bonding process 140 and the output of the inspection algorithm is inputted into the trained linear regression model, such as the die bonding pressure and the void defects and the control algorithm outputs the predicted temperature for the die bonding process.

[0059]

[0058] In some embodiments, new data during the process may be continuously inputted into the control algorithm, allowing the algorithm to dynamically adapt to changing conditions and minimize void defects for optimal results. The DB process module 112 adjusts the control parameters of the die bonding process 140. For example, the DB process module 112 adjusts the die bonding pressure of the die bonding process.

[0060]

[0059] In some embodiments, the DB process module 112 may adjust the temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, curing parameters, etc. The DB process module 112 executes the die bonding process. For example, die bonding may include utilizing a die bonding machine to join the aligned substrate and die. The die is retrieved by a pickup tool and maintained under vacuum pressure while being transferred to the leadframe or header. In some embodiments, the die bonding process may mount the die to the die pad of the leadframe utilizing soft solder in the form of a wire roll in which the die is removed from the wafer and positioned onto the die attach area using a pick-and-place instrument. In some embodiments, the die bonding process may form a eutectic bond between the die and leadframe without the use of soft solder preforms, achieving a eutectic die attachment, in which silver is plated on the bonding area of the leadframe and gold on the reverse side of the die, followed by heating above the melting point, to produce a rigid joint.

[0061]

[0060] In some embodiments, the die bonding process may adhere the die to the lead frame utilizing epoxy adhesive by applying a drop of epoxy to the package and positioning the die on top, and the adhesive is cured at the designated temperature.

[0061] In some embodiments, the die bonding process may control the adhesive properties of acrylic adhesive using UV curing technology to ensure that the die attach film is perfectly adhered to via UV curing. In some embodiments, the die bonding process may perform thermocompression die bonding, which is a metallurgical process that enables the formation of a metallurgical bond by facilitating atomic contact between two metals through the application of thermal and mechanical pressure and heating.

[0062]

[0062] In some embodiments, the die bonding process may perform an inverted chip die attachment in which the die is inverted before attachment, and conductive polymer bumps, or solder are utilized to establish mechanical and electrical connections.

[0063]

[0063] In some embodiments, the die bonding process may include the control parameter settings outputted by the control algorithm being used by the controller of the die bonding process 140 equipment to set the die bonding pressure of the die bonding process. The DB process module 112 may monitor the die bonding process of the die bonding process 140 equipment and perform the control algorithm in real-time to continuously adjust the control parameters. The DB process module 112 stores the data in the DB process database 122. The DB process module 112 stores the data in the DB process database 122, such as the data collected from the die bonding process 140, the output of the inspection algorithm, and the output of the control algorithm. The DB process database 122 may contain the wafer group number or ID, the die bonding pressure with Al temperature control, the void defects per unit standard area, etc. The DB process database 122 may contain the data for each die that is being processed by the die bonding process 140. The DB process module 112 returns to the base module 108. The base module 108 initiates, at step 204, the DB post-process module 114. For example, the DB post-process module 114 begins by being initiated by the base module 108. The DB post-process module 114 connects to the die bonding post-process 150. The DB post-process module 114 may connect to the die bonding post-process 150, such as cleaning and inspection, electrical testing, encapsulation, etc. The DB post-process module 114 collects the post-process data. The DB post-process module 114 may collect the parameter data, such as testing, etc. The post-process may involve parameters such as bond quality, defect detection, electrical performance, encapsulation quality, etc. The DB post-process module 114 extracts the data from the DB inspection database 120 and DB process database 122. The DB postprocess module 114 extracts the data, such as material data, parameter data, etc. of the pre-process and the die bonding process. The DB post-process module 114 performs the post-process algorithm.

[0064]

[0064] The post-process algorithm may be used to maintain consistent color and luminance across the pixels. For example, luminance may be the intensity of the light that a microLED emits, each microLED manifests distinct variations in luminance and color, and to attain visual uniformity throughout the entire display, accurate measurement and correction are necessary. The luminance of the microLEDs may be determined by imaging colorimeters and other specialized instruments that quantify luminance and color uniformity, the information provided by these instruments is used to calibrate displays and to attain uniformity, it is necessary to individually measure and adjust each microLED. In some embodiments, color and luminance may be modified to conform to a specified standard. The luminance of microLEDs may be measured in candelas per square meter (cd / m2). In some embodiments, the luminance for indoor displays may be within the range of 40 to 300 cd / m2l, sRGB monitors may be around 100 cd / m2, consumer desktop monitors may be between 200 to 300 cd / m2, HDR displays may fall within 450 to above 1600 cd / m2. The luminance of the microLEDs may be changed or influenced by various parameters in the die bonding process, such as variations in the thickness or uniformity of the die attach material, inadequate temperature control, pressure control, alignment accuracy, etc. For example, alignment accuracy may result in adverse effects on luminance output during the die bonding process. The typical alignment accuracy for calibrated alignment falls within the range of approximately 5 pm, or micrometers. The alignment accuracy may be influenced by roll stamp structure, X-Y theta stage performance, stage-roller orthogonality, roller roundness, etc. which can lead to mechanical stress, damage of the microLEDs, luminance variations, defects, etc. The alignment accuracy may be enhanced through machine calibration and precision, roll stamp design and geometry, roller roundness and orthogonality, alignment sensors, and feedback systems, material properties and wafer warpage, self-alignment techniques, active optical alignment, underfill material and void reduction, etc. The post-process may categorize the wet etched wafers based on their warpage, which is the degree of deformation or curvature in the wafer surface resulting from the etching process. By grouping wet etched wafers with similar warpage characteristics together, the post-process algorithm ensures that during the subsequent die bonding step, wafers with comparable warpages are bonded together to minimize overall warpage within each group, leading to reduced alignment challenges during bonding. For example, when wet etched wafers with similar warpage are bonded together, the overall warpage variation within each group is reduced. As a result, the alignment process during die bonding becomes more consistent and predictable. On average, the luminescence of microLEDs bonded using this optimized approach tends to be higher and exhibits a more tightly distributed pattern. This improvement is attributed to the reduced impact of warpage on alignment accuracy. With less variation in alignment, the bonding of microLEDs is more uniform, leading to improved luminous output and overall performance.

[0065]

[0065] In some embodiments, the post-process algorithm may be a clustering algorithm, such as a K-Means clustering to clusters chips according to the degree of similarity in wafer warpage, hierarchical clustering which enables the grouping of chips at varying degrees of granularity through the formation of a hierarchy of clusters, density-based spatial clustering of applications with noise which is a technique utilized to detect dense regions of chips within the warpage space.

[0066]

[0066] In some embodiments, the post-process algorithm may perform a principal component analysis which may reduce the dimensionality of the warpage data while preserving its variance and chips can then be grouped based on the transformed features.

[0067]

[0067] In some embodiments, the post-process algorithm may perform self-organizing maps which is a neural network-based technique that maps high-dimensional data onto a lower-dimensional grid, and chips with similar warpage patterns are grouped.

[0068]

[0068] In some embodiments, the post-process algorithm may perform autoencoders which are neural networks that learn compact representations of the warpage data, and the encoded features can be used for clustering. In some embodiments, the post-process algorithm may use machine learning algorithms to analyze parameters related to luminescence. The post-process algorithm may focus on the process data outputted from the inspection algorithm performed in the DB inspection module 110 and the effectiveness of the parameters outputted by the control algorithm performed in the DB process module 112 during the die bonding process. The post-process algorithm may adjust control parameters of the current microLED processed dies in the process to result in an improved die bonding process.

[0069]

[0069] For example, the post-process algorithm may optimize alignment to group the wafers based on warpage to improve the luminescence. The DB post-process module 114 performs the postprocess inspection. For example, the post-process may include defect inspection to detect flaws or defects in chips or wafers to avert their integration into the final product, color and brightness variation checks to ensure color and brightness uniformity across pixels, performing inline metrology and automated optical inspection to identify and rate any areas of nonuniformity or defects, etc. The DB post-process module 114 stores the data in the DB post-process database 124. The DB post-process module 114 stores the data in the DB post-process database 124, such as the data collected from the die bonding post-process 150, the output of the post-process algorithm, etc. The DB post-process database 124 may contain the wafer group number or ID, the alignment measure based on binned together based on wafer warpage, the luminescence, etc. The DB process database 122 may contain the data for each die that is being processed by the die bonding postprocess 150. The DB post-process module 114 returns to the base module 108. The base module 108 initiates, at step 206, the DB historical module 116. For example, the DB historical module 116 begins by being initiated by the base module 108. The DB historical module 116 connects to the die bonding pre-process 130, the die bonding process 140, and the die bonding post-process 150. The historical module 116 connects to the die bonding pre-process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc., the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that places the microLED dies onto a prepared substrate, secures the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength, and the die bonding post-process 150, such processes and methods after the completion of the die bonding process 140, such as a cleaning and inspection, electrical testing, encapsulation, etc. The DB historical module 116 aggregates the data from the various types of processes and stores the data in the DB network database 126. The DB network database 126 may contain the historical data from the various processes performed by the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150. The DB network database 126 may contain the data parameters collected during the inspection process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc., control parameters of the process, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc., and data parameters collected during post-processing, such as bond quality, defect detection, electrical performance, encapsulation quality, etc. The DB historical module 116 performs the historical machine learning algorithm on the historical data stored in the DB network database 126. The historical machine learning algorithm may improve each stage of the microLED manufacturing process, such as die bonding pre-process, die bonding process, and die bonding post-process. For example, the historical machine learning algorithm may receive input data from the DB inspection module 110, such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, and analyze the historical data of electrical resistance and die bonding time to adjusts die bonding time predictions based on variations in electrical resistance which improves accuracy in predicting die bonding time for different electrical resistance values and leads to optimized die bonding pre-process.

[0070]

[0070] The historical machine learning algorithm may receive input data from the DP process module 112, such as die bonding pressure with calculated temperature control, and void defects per unit standard area, and may incorporate reinforcement learning techniques to adjust die bonding pressure and temperature control based on real-time feedback of void defects detected, to enhanced pressure and temperature control during die bonding which leads to a reduction in void defects and improved bonding quality. The historical machine learning algorithm may receive input data from the DP post-process module 114, such as alignment measurement warpage Al binned, Luminescence, etc., to implement deep learning models to analyze alignment measurements and luminescence data to learn patterns of alignment-related warpage and luminescence variations which provides insights into the relationship between alignment, warpage, and luminescence, facilitating adjustments in the post-die bonding process to improve microLED chip quality. In some embodiments, the historical machine learning algorithm may enable iterative improvements in each stage of the microLED manufacturing process by learning from historical data. In some embodiments, the historical machine learning algorithm may enhance process parameters and controls which leads to better-quality microLED chips and higher manufacturing efficiency.

[0071]

[0071] In some embodiments, the historical machine learning algorithm may perform data-driven decision making which allows for informed decision-making based on data analysis and leads to more accurate predictions and optimizations.

[0072]

[0072] In some embodiments, the historical machine learning algorithm may use the historical data stored in the DB network database 126 to determine trends to understand how varying parameters affect the final quality and identify optimal parameter settings to create the best outcomes. The historical machine learning algorithm adjusts the parameters in real time for the DB inspection module 110, DB process module 112, and DB post-process module 114 to correct any defects. For example, the data may show defects, such as voids, that occurred during the die bonding process. The historical machine learning algorithm may minimize the void defects in the wafer by adjusting the time at a predetermined temperature to improve the electrical resistance in the DB inspection module 110. The historical machine learning algorithm may also adjust the temperature occurring in the die bonding pressure process in the DB process module 112 to minimize the void defects. The DB historical module 116 sends the process adjustments to the DB inspection module 110, the DB process module 112, and the DB post-process module 114. The historical machine learning algorithm may optimize the DB inspection module 110 by improving the accuracy in predicting die bonding time for different electrical resistance values. The historical machine learning algorithm may also reduce void defects by enhancing pressure and temperature control during the DB process module 112. The historical machine learning algorithm may improve microLED chip quality by providing relationships and adjustments between alignment, warpage, and luminescence to the DB post-process module 114. The adjustments are sent to the DB inspection module 110 and DB process module 112 to create a more consistent final product which is then verified by the DB post-process module 114. The DB historical module 116 returns to the base module 108. The base module 108 initiates, at step 208, the DB integration module 118. For example, the DB integration module 118 begins by being initiated by the base module 108. The DB integration module 118 performs the integration machine learning algorithm. The integration machine learning algorithm may facilitate interactions between the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150 stages of microLED manufacturing. For example, the integration machine learning algorithm may receive input data, such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, alignment measurement, luminescence, etc., to analyze the relationship between die bonding pre-process 130 parameters and die bonding post-process 150 outcomes and may adjust die bonding pre-process 130 parameters to optimize die bonding post-process 150 results which improve alignment and luminescence of microLED chips by adjusting die bonding pre-process 130 parameters based on their impact on die bonding post-process 150 outcomes. The integration machine learning algorithm may receive input data, such as die bonding pressure with Al temperature control, void defects per unit standard area, alignment measurement, luminescence, etc., and implement reinforcement learning algorithms to analyze the relationship between the die bonding process 140 parameters and die bonding post-process 150 outcomes and adjusts die bonding process 140 parameters to optimize die bonding post-process 150 results in which a reduction of void defects and improvement in alignment and luminescence is achieved by adjusting die bonding process 140 parameters based on their impact on die bonding post-process 150 outcomes. The integration machine learning algorithm may receive input data, such as alignment measurement, luminescence, die bonding pressure with Al temperature control, void defects per unit standard area, etc., and utilizes deep learning models to analyze the relationship between die bonding post-process 150 outcomes and die bonding process 140 parameters by adjusts die bonding process 140 parameters to optimize die bonding post-process 150 results to improve alignment and luminescence by adjusting die bonding process 140 parameters based on their impact on die bonding post-process 150 outcomes.

[0073]

[0073] In some embodiments, the DB integration module 118 may facilitate the optimization of the entire microLED manufacturing process by considering interactions between different stages. In some embodiments, the DB integration module 118 may leverage Al to adjust parameters in one stage based on their impact on outcomes in subsequent stages to enhance the overall process efficiency and product quality.

[0074]

[0074] In some embodiments, the DB integration module 118 may adapt to changes in process conditions and requirements by continuously learning data and adjusting parameters accordingly.

[0075]

[0075] In some embodiments, the integration machine learning algorithm may use the data stored in the DB 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 improving luminescence across the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150. The historical data in the DB network database 126 may include information on luminescence and relevant operational parameters such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, die bonding pressure with Al temperature control, void defects per unit standard area, 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 luminescence and engineer new features if necessary. For example, the interaction between the electrical resistance of the contacts and the calculated time at 250°C temperature for die bonding may have a significant impact on the luminescence. The selected predictive model is trained using the training dataset. The model learns patterns and relationships between operational parameters and luminescence. 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 die bonding pre- process 130, die bonding process 140, and die bonding post-process 150. The DB integration module 118 connects to the die bonding pre-process Al module 136, the die bonding process Al module 146, and the die bonding post-process Al module 156. The DB integration module 118 connects to the die bonding pre-process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc., the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that places the microLED dies onto a prepared substrate, secures the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength, and the die bonding post-process 150, such processes and methods after the completion of the die bonding process 140, such as a cleaning and inspection, electrical testing, encapsulation, etc. The DB integration module 118 sends the process adjustments to the die bonding pre-process Al module 136, the die bonding process Al module 146, and the die bonding post-process Al module 156. For example, the DB integration module 118 may send the die bonding time, pressure, temperature, etc. adjustments to each of the processes to optimize the results of the die bonding process.

[0076]

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

[0077] FIG. 3 illustrates the DB inspection module 110. The process begins with the DB inspection module 110 being initiated, at step 300, by the base module 108. The DB inspection module 110 connects, at step 302, to the die bonding pre-process 130. The DB inspection module 110 connects with the die bonding pre-process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc. In some embodiments, the DB inspection module 110 may transmit and receive data from the die bonding pre-process 130.

[0077]

[0078] In some embodiments, the die bonding pre-process 130 may measure die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. In some embodiments, the DB inspection module 110 may control or send inputs to control the die bonding pre-process 130. The DB inspection module 110 collects, at step 304, the data from the die bonding pre-process 130. The DB inspection module 110 collects the pre-processing data, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc.

[0078]

[0079] Edge Al and TinyML Example 1 : In some embodiments, the DB inspection module 110 may deploy a TinyML-based Edge Al model directly on in-line camera or vision sensor hardware to perform immediate, energy -efficient defect detection during die placement without relying on centralized inspection servers. For example, a process technician may prompt, “Detect skew or tilt on contact and flag for reinspection locally,” while the embedded vision model may automatically trigger, “Tag wafers if fillet boundary symmetry is violated during high-speed placement.” In operation, the DB inspection module 110 may capture image frames from the DB inspection equipment 128 and execute a quantized convolutional neural network locally to classify bonding anomalies — such as partial contact, edge overflow, or centerline deviation — using visual feature embeddings previously trained on samples from the DB inspection database 120. Internally, the model may utilize optimized pooling layers and lightweight activation functions to meet edge compute constraints and deliver sub-second inference. Upon detecting a defect, the system may locally annotate the wafer ID and location, trigger an operator-facing visual alert, and queue metadata for asynchronous upload to the DB network database 126. Outputs may include immediate visual flags, localized defect logs, and summary statistics on edge-level classification performance. By embedding vision-based Al logic directly at the point of image capture, the DB inspection module 110 may reduce latency, increase inspection autonomy, and preserve yield in fast-moving packaging environments where centralized analysis may introduce unacceptable delay. The DB 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 die bonding pre-process 130 is used to determine if a control parameter of the die bonding process should be adjusted or altered. The inspection algorithm may determine the die bonding temperature for the die bonding process based on the electrical resistance of the die determined in the die bonding pre-process 130. For example, the inspection algorithm may perform a process window experiment to optimize the temperature during the die bonding process 140 based on the electrical testing data collected from the die bonding pre-process 130. The inspection may identify the key variables that may affect the electrical resistance of the die, such as temperature, pressure, time, material composition, etc. For each variable, the inspection algorithm may determine the range of values to be tested, including expected operational values as well as extreme values to understand the limits of the process. The inspection algorithm may design the experimental setup, including the number of runs and the sequence of experiments, such as testing all possible combinations of the variables within the specified range.

[0079]

[0080] The inspection algorithm then conducts the experiments according to the predefined design ensuring that each combination of variables is tested accurately and consistently. The inspection algorithm records the data for each experiment run, including measurements, observations, and any other relevant information. The inspection algorithm may analyze the data to understand how changes in variables affect the electrical resistance, such as performing statistical analysis to identify significant factors and their interactions. The inspection algorithm determines the optimal conditions based on the analysis.

[0080]

[0081] In some embodiments, the inspection algorithm may validate the results through additional testing or verification, implement any necessary changes to the process based on the findings of the experiment, and monitor and perform adjustments to maintain optimal performance.

[0081]

[0082] In some embodiments, the inspection algorithm may perform linear regression to determine linear relationships between time and electrical resistance to forecast the bonding time.

[0083] In some embodiments, the inspection algorithm may be a neural network to simulate nonlinear bonding time and resistance relationships and may be trained to forecast ideal bonding times given resistance values.

[0082]

[0084] In some embodiments, the inspection algorithm may perform a random forest technique which may generate predictions by employing multiple decision trees.

[0083]

[0085] In some embodiments, the inspection algorithm may be an artificial intelligence or machine learning algorithm in which the data collected from the die bonding pre-process 130 is used to determine if a control parameter of the die bonding process should be adjusted or altered. The inspection algorithm may determine the electrical resistance and use this measurement to determine the temperature during the die bonding process. For example, the inspection algorithm may identify the electrical resistance and modify die bonding parameters to maintain consistent electrical resistance during the die bonding process, such as temperature, bonding pressure, alignment, etc. The inspection algorithm utilizes the captured data, from the die bonding pre-process 130, to send data about the microLED processed wafer to the DB process module 112 to be used to adjust die bonding parameters used by the die bonding process 138. In some embodiments, the DB inspection module 110 may send the outputted data from the inspection algorithm to the DB process module 112 to optimize the parameters used in the process. In some embodiments, the DB inspection module 110 may send the outputted data from the inspection algorithm to the DB process module 112 to optimize the parameters used in the process.

[0084]

[0086] Al Agents Example 1 : In some embodiments, the DB inspection module 110 may incorporate a task-specific Al agent configured to dynamically adapt inspection sample rates and reinspection targeting based on detected defect clusters, tool state drift, and substrate -level metadata. For example, a quality engineer may request, “Increase inspection frequency for units with residual adhesive patterns,” while the agent may autonomously trigger, “Escalate reinspection rate if radial void anomalies increase on Tools 3 and 5 over two consecutive lots.” In execution, the agent may monitor real-time defect classification outputs from the DB inspection database 120, correlate spatial defect distribution patterns with tool runtime parameters from the DB process database 122, and evaluate deviations from nominal tool behavior profiles archived in the DB network database 126. If a risk pattern is detected — such as edge-aligned void propagation under rising nozzle pressure drift — the agent may increase inspection density for selected units, initiate reinspection using alternate metrology paths, and tag affected lots for downstream yield containment. Internally, the agent may leverage multi-condition policy rules trained on past falsenegative rates, balancing detection sensitivity with throughput constraints. Outputs may include a revised inspection schedule annotated with risk rationale, and an alert signal sent to the DB integration module 118 to inform process control updates. By embedding an adaptive Al agent within the DB inspection module 110, the system may respond to emerging defect risks in real time, maintaining inspection efficiency while reducing the probability of latent yield escape.

[0085]

[0087] Multimodal Al Example 1 : In some embodiments, the DB inspection module 110 may integrate a multimodal Al model to enhance defect root cause analysis by fusing visual inspection images, die bond process telemetry, and technician-entered annotations. For example, a process engineer may ask, “What combination of conditions caused inconsistent fillet shape on Lot M442?” while a process planner may query, “Identify spatial void patterns linked to pressure anomalies and adhesive type changes.” In response, the DB inspection module 110 may retrieve high-resolution bond images from the DB inspection equipment 128, align them with dispense pressure and temperature logs from the DB process database 122, and incorporate historical technician notes and defect tags from the DB inspection database 120. The multimodal Al model may fuse these image, numerical, and textual inputs using shared embedding spaces and attention-weighted cross- modal alignment, enabling identification of correlated failure mechanisms such as fillet asymmetry due to late-stage adhesive flow shift caused by nozzle wear. Internally, the model may reference known defect archetypes and condition-parameter mappings stored in the DB network database 126 to strengthen causal attribution. Outputs may include a fused insight summary that links specific fillet distortions to real-time dispense instability and an external reference noting similar adhesive flow behavior in industry literature for thermally-variable die attach systems. By embedding multimodal reasoning into the DB inspection module 110, the system may transform siloed inspection data into unified failure insights, accelerating diagnostics and supporting corrective action planning.

[0086]

[0088] Generative Video Al Example 1; In some embodiments, the DB inspection module 110 may employ a generative video Al model to simulate die attach flow dynamics and void formation behavior during the initial bonding sequence, based on early-cycle imaging, temperature profiles, and dispense telemetry. For example, a packaging engineer may prompt, “Simulate adhesive flow and fillet evolution for Lot Al 6 at current stage velocity,” while a tool setup agent may request, “Visualize void pattern emergence if dispense offset is increased by 200 microns.” In response, the DB inspection module 110 may retrieve early bond cycle images from the DB inspection equipment 128, correlate those with dispense rate, stage motion, and heater zone telemetry from the DB process database 122, and apply a generative video model trained on temporal bonding sequences stored in the DB network database 126. Internally, the model may use time-conditioned generative layers — such as diffusion-based frame prediction — to synthesize visual forecasts of fillet shape, adhesive coverage, and defect development under the modified conditions. Outputs may include a predictive video illustrating probable void nucleation or asymmetrical flow behavior, paired with an annotation referencing literature on the sensitivity of adhesive wet-out to motion delay and thermal boundary layer variation. By embedding forward-looking video simulation in the DB inspection module 110, the system may support tooling validation and pre-run optimization by visualizing bond integrity outcomes before full production engagement. The DB inspection module 110 sends, at step 308, the determined process data from the inspection algorithm to the DB process module 112. For example, the DB inspection module 110 may send the calculated time at a predetermined temperature for the die bonding process to the DB process module 112 to optimize the die bonding process.

[0087]

[0089] In some embodiments, the DB inspection module 110 may send die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc.

[0088]

[0090] Synthetic Data and AI-Generated Digital Twins Example 1: In some embodiments, the DB inspection module 110 may employ synthetic data generation to enhance model robustness by creating labeled defect images for underrepresented bonding conditions, such as rare die sizes, adhesive materials, or tool-specific geometries. For example, a machine learning specialist may request, “Generate synthetic void cases for large die bonded with high -viscosity adhesive,” while a yield analyst may prompt, “Augment the training set with tilt variation examples for new Tool 5 fixture configurations.” In response, the DB inspection module 110 may analyze defect frequency distributions within the DB inspection database 120 and identify sparse data clusters based on process metadata retrieved from the DB process equipment 128. A generative model trained on annotated bond image data stored in the DB network database 126 may then synthesize realistic defect images — including center voids, asymmetric fillet profiles, or marginal tilt artifacts — labeled with relevant bonding conditions such as die dimensions, dispense offset, and contact pressure. Internally, the model may leverage conditional GANs or diffusion-based architectures to ensure fidelity to known morphological patterns while filling gaps in the data distribution. Outputs may include a batch of synthetic training images with structured metadata tags and a retraining report demonstrating improved classifier performance on previously low-recall defect classes. By embedding synthetic image generation into the DB inspection module 110, the system may enable more balanced and generalizable model training, accelerating adaptation to new process configurations and tool variants without requiring costly physical rework. The DB inspection module 110 stores, at step 310, the data in the DB inspection database 120. The DB inspection module 110 stores the data outputted from the inspection algorithm in the DB inspection database 120, such as the wafer group number or ID, the average pre-bonding electrical resistance of the contacts, the calculated time in minutes at 250 degrees Celsius for die bonding, etc.

[0089]

[0091] In some embodiments, the DB inspection database 120 may include other parameter data collected or calculated during the DB inspection module 110 process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB inspection module 110 returns, at step 312, to the base module 108.

[0090]

[0092] A simple and efficient method for the inspection algorithm for a die bonding process would be to use a linear regression algorithm. The necessity to comprehend and forecast the link between factors including electrical resistance, temperature, pressure, and time — all of which are critical in figuring out the ideal circumstances for the die bonding process — motivates this decision. Because it can simulate the linear connection between a dependent variable (like die bonding time or temperature) and one or more independent variables (like electrical resistance, pressure), linear regression is especially well suited for this purpose. The use of linear regression is one example of the technique. A basic statistical and machine learning method called linear regression is used to forecast a quantitative result based on one or more predictor variables. The method involves determining which coefficients in a linear equation comprising one or more independent variables best predict the dependent variable's value. One practical example would be to use the electrical resistance of the die, which was determined during the pre-process, to calculate the ideal temperature for the die bonding procedure. Here, the temperature required for the ideal die bonding procedure is the dependent variable (Y), and electrical resistance is the independent variable (X).

[0093] The procedure carries out Data Collection, gathering information from the die bonding preliminary procedure, such as electrical resistance readings and associated temperatures that resulted in successful bonding results. In order to train a Linear Regression model, the method uses a dataset that consists of pairs of electrical resistance measurements and the successful bonding temperatures. By understanding the correlation between resistance and temperature, this model is able to forecast the ideal temperature in response to a fresh measurement of resistance. Adjustment and Prediction are used by the model. Here, the technique uses the trained model to forecast, using fresh electrical resistance measurements, the ideal die bonding temperature. To get the required results, modify the die bonding process parameters as necessary. Typically, experimental or operational data gathered throughout the die bonding pre-process and process stages would comprise the datasets for such an application. The following would be crucial elements of this dataset: (1) Electrical Resistance Measurements: Dies' resistance measured prior to bonding (2) Temperature: The values at which die bonding was accomplished successfully, (3) Additional Variables: To increase the precision and resilience of the model, other pertinent data, if available, such as pressure, time, and material composition, could be included. To guarantee that the model is well-trained and capable of producing precise predictions in a variety of scenarios, this dataset should ideally encompass a range of values for each variable, including both operational (normal) and severe conditions. The inspection method can efficiently assess the relationship between important variables and improve the die bonding process, improving productivity and product quality, by utilizing linear regression in this situation.

[0091]

[0094] FIG. 4 illustrates the DB process module 112. The process begins with the DB process module 112 being initiated, at step 400, by the base module 108. The DB process module 112 receives, at step 402, the pre-process data, such as the data on the microLED processed wafer, from the DB inspection module 110. The DB process module 112 receives the outputted data from the inspection algorithm performed in the DB inspection module 110, such as the calculated time at a predetermined temperature for the die bonding process, allowing the die bonding process 140 to determine the optimal control parameters to prevent any defects in the die bonding process. In some embodiments, the DB process module 112 may receive die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc. The DB process module 112 connects, at step 404, to the die bonding process 140. The DB process module 112 connects with the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that place the microLED dies onto a prepared substrate, and secure the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength. In some embodiments, the DB process module 112 may transmit and receive data from the die bonding process 140. In some embodiments, the DB process module 112 may control or send inputs to control the die bonding process 140, such as controlling parameters of the die bonding process, for example, pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. The DB process module 112 collects, at step 406, the data from the die bonding process 140. The DB process module 112 collects the processing data, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc. The DB process module 112 performs, at step 408, the control algorithm.

[0092]

[0095] The control algorithm may output the pressure, temperature, pressure, etc. for the die bonding process to achieve optimal thermal management, mechanical robustness, and electrical performance in microLED die bonding. In some embodiments, the die bonding process may include ultrasonic scanning to detect various defects, including delamination, voids, cracks, incomplete bonding, etc. For example, die bonding pressure may refer to the force applied during the bonding process to attach a semiconductor die, such as a microLED, to its substrate or package and ensures proper contact between the die and the substrate, allowing for efficient heat dissipation, electrical connectivity, and mechanical stability. The factors that influence die bonding pressure are material properties, die size and shape, adhesive type, process temperature, etc. In some embodiments, the bonding pressure for microLEDs may be between 50 g / mm2to 500 g / mm2. There may be pressure variations while maintaining pressure control in die bonding which may produce defects, such as voids. These variations may be caused by mechanical wear, calibration issues, temperature fluctuations, hydraulic system performance, material properties, process variability, etc. The control algorithm may control the process parameters, such as temperature, to adjust the pressure being exerted by the hydraulic system and components of the die bonder which can be altered by thermal expansion or contraction. In some embodiments, the control algorithm may utilize proportional-integral-derivative controllers to adjust the heating or cooling elements based on the difference between the desired setpoint temperature and the actual temperature by calculating the proportional, integral, and derivative terms to fine-tune the control action. In some embodiments, the control algorithm may utilize model predictive control to optimize control inputs, such as heater power, over a finite time horizon to minimize deviations from the setpoint and adapt to changing conditions and disturbances.

[0093]

[0096] In some embodiments, the control algorithm may utilize fuzzy logic control by using linguistic variables, such as hot or cold, and rules to adjust control actions. In some embodiments, the control algorithm may use the output of the inspection algorithm performed in the DB inspection module 110 and the data collected from the die bonding process 140 as input. The control algorithm may clean and preprocess the collected data to handle missing values, and outliers, and ensure uniformity, and the dataset is divided into training and testing sets. The key features are identified that significantly impact void defects occurring during the die bonding process, such as the temperature of the system and components while applying pressure during the die bonding process. The control algorithm may utilize the training dataset to train a linear regression model, which establishes a linear relationship between the selected features and the target variable, such as void defects occurring during the die bonding process. The trained model is then evaluated using the testing dataset to ensure its predictive accuracy.

[0094]

[0097] In some embodiments, metrics such as Mean Squared Error (MSE) or R-squared may be used to assess the performance of the linear regression model. The new data from the die bonding process 140 and the output of the inspection algorithm is inputted into the trained linear regression model, such as the die bonding pressure and the void defects and the control algorithm outputs the predicted temperature for the die bonding process.

[0095]

[0098] In some embodiments, new data during the process may be continuously inputted into the control algorithm, allowing the algorithm to dynamically adapt to changing conditions and minimize void defects for optimal results.

[0096]

[0099] Synthetic Data and AI-Generated Digital Twins Example 2 In some embodiments, the DB process module 112 may integrate an Al-generated digital twin framework to simulate bonding outcomes under varying dispense profiles, die geometries, and thermal settings — enabling predictive validation of process changes without requiring physical lots. For example, a process engineer may request, “Simulate adhesive spread and tilt risk for a new 6 mm die under reduced stage acceleration,” while a tooling group may prompt, “Predict fillet non -uniformity for Lot J82 with alternate dispense delay timing.” In response, the DB process module 112 may extract historical bonding cycle trajectories and defect-tagged outcomes from the DB process database 122, and generate synthetic die placement and adhesive behavior scenarios using a generative model trained on defect-process pairings stored in the DB network database 126. Internally, the model may use parametric conditioning on variables such as die size, contact velocity, and heater ramp profile to produce projected bonding surface states — including adhesive distribution, fillet shape, and tilt deformation — mapped across the wafer layout. Outputs may include a synthetic wafer summary with annotated risk zones, parameter-adjusted bond maps, and an external reference noting thermal-induced edge lifting documented in prior work on asymmetric die attach systems. By embedding digital twin simulation capabilities into the DB process module 112, the system may enable low-risk recipe experimentation and tool validation that preserves material and accelerates production qualification cycles. The DB process module 112 adjusts, at step 410, the control parameters of the die bonding process 140. For example, the DB process module 112 adjusts the die bonding pressure of the die bonding process.

[0097]

[0100] In some embodiments, the DB process module 112 may adjust the temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, curing parameters, etc.

[0098]

[0101] Edge Al and TinyML Example 2. In some embodiments, the DB process module 112 may incorporate a TinyML-powered Edge Al model deployed directly on clamp force controllers or stage motion actuators to enable ultra-low-latency correction of mechanical misalignments during the die bonding sequence. For example, a field engineer may prompt, “Detect and correct clamp overdrive if force exceeds spec in the first 200 ms,” while an autonomous logic block may initiate, “Reduce Y-axis stage velocity if sub-micron drift is detected during die placement.” In response, the DB process module 112 may capture high-frequency actuator current, force sensor, and encoder data from the DB process equipment 138, and execute a compact neural network locally to infer deviation patterns based on bonding trajectories archived in the DB process database 122. If anomalies such as overshoot, oscillation, or alignment drift are identified, the embedded model may issue local adjustments to controller parameters without requiring communication with centralized servers. Internally, the model may use quantized convolutional kernels and edge- optimized inference routines to maintain sub -millisecond response times and minimal memory usage. Outputs may include a control log with adjustment timestamps, local alert flags for technician access, and queued metadata packets for later sync with the DB network database 126. By embedding Edge Al models within bonding motion and clamp systems, the DB process module 112 may enhance bonding precision, prevent early-stage mechanical failures, and maintain yield quality in high-speed production environments. The DB process module 112 executes, at step 412, the die bonding process. For example, the die bonding may include utilizing a die bonding machine to join the aligned substrate and die. The die is retrieved by a pick-up tool and maintained under vacuum pressure while being transferred to the leadframe or header.

[0099]

[0102] In some embodiments, the die bonding process may mount the die to the die pad of the leadframe utilizing soft solder in the form of a wire roll in which the die is removed from the wafer and positioned onto the die attach area using a pick-and-place instrument. In some embodiments, the die bonding process may form a eutectic bond between the die and leadframe without the use of soft solder preforms, achieving a eutectic die attachment, in which silver is plated on the bonding area of the leadframe and gold on the reverse side of the die, followed by heating above the melting point, to produce a rigid joint.

[0100]

[0103] In some embodiments, the die bonding process may adhere the die to the lead frame utilizing epoxy adhesive by applying a drop of epoxy to the package and positioning the die on top, and the adhesive is cured at the designated temperature.

[0101]

[0104] In some embodiments, the die bonding process may control the adhesive properties of acrylic adhesive using UV curing technology to ensure that the die attach film is perfectly adhered via UV curing.

[0102]

[0105] In some embodiments, the die bonding process may perform thermocompression die bonding, which is a metallurgical process that enables the formation of a metallurgical bond by facilitating atomic contact between two metals through the application of thermal and mechanical pressure and heating.

[0103]

[0106] In some embodiments, the die bonding process may perform an inverted chip die attachment in which the die is inverted before attachment, and conductive polymer bumps, or solder are utilized to establish mechanical and electrical connections. In some embodiments, the die bonding process may include the control parameter settings outputted by the control algorithm being used by the controller of the die bonding process 140 equipment to set the die bonding pressure of the die bonding process. The DB process module 112 may monitor the die bonding process of the die bonding process 140 equipment and perform the control algorithm in real-time to continuously adjust the control parameters. The DB process module 112 stores, at step 414, the data in the DB process database 122. The DB process module 112 stores the data in the DB process database 122, such as the data collected from the die bonding process 140, the output of the inspection algorithm, and the output of the control algorithm. The DB process database 122 may contain the wafer group number or ID, the die bonding pressure with Al temperature control, the void defects per unit standard area, etc. The DB process database 122 may contain the data for each die that is being processed by the die bonding process 140. The DB process module 112 returns, at step 416, to the base module 108.

[0104]

[0107] FIG. 5 illustrates the DB post-process module 114. The process begins with the DB postprocess module 114 being initiated, at step 500, by the base module 108. The DB post-process module 114 connects, at step 502, to the die bonding post-process 150. The DB post-process module 114 may connect to the die bonding post-process 150, such as cleaning and inspection, electrical testing, encapsulation, etc. The DB post-process module 114 collects, at step 504, the post-process data. The DB post-process module 114 may collect the parameter data, such as testing, etc. The post-process may involve parameters such as bond quality, defect detection, electrical performance, encapsulation quality, etc. The DB post-process module 114 extracts, at step 506, the data from the DB inspection database 120 and DB process database 122. The DB post-process module 114 extracts the data, such as material data, parameter data, etc. of the pre-process and the die bonding process.

[0105]

[0108] Multimodal Al Example 2 In some embodiments, the DB post-process module 114 may incorporate a multimodal Al model to correlate post-bond reliability signatures — such as die tilt, delamination, or substrate warpage — with inspection images, bonding cycle telemetry, and postprocess technician observations. For example, a reliability engineer may prompt, “What upstream bonding signals contributed to tilt failures on Lots 721-724?” while a materials team may request, “Identify defect clusters tied to die attach material batch changes and substrate warp deviation.” In response, the DB post-process module 114 may retrieve geometric and visual data from the DB post-processing equipment 148, access bonding cycle variables such as platen pressure curves and stage velocity profiles from the DB process database 122, and include structured comments and defect descriptions from the DB post-processing database 124. The multimodal model may fuse these heterogeneous inputs using feature alignment and attention layers, comparing extracted patterns to previously recorded failure typologies from the DB network database 126. Internally, the model may use cross-modal embedding similarity and condition-specific weighting to explain how mismatched ramp profiles and batch-specific adhesive behavior correlate with observed die misalignment. Outputs may include a confidence-ranked root cause attribution report, highlighting cross-source signal relationships, and a literature citation linking ramp-induced stress vectors to tilt amplification in asymmetric laminate systems. By embedding multimodal Al into the DB postprocess module 114, the system may deliver high-resolution, cross-phase defect diagnostics that accelerate feedback loops between bonding, inspection, and reliability analysis teams.

[0106]

[0109] Generative Video Al Example 2: In some embodiments, the DB post-process module 114 may incorporate a generative video Al model to simulate the time-evolved progression of bonding defects such as die tilt, void propagation, or delamination under projected thermal and mechanical stress conditions. For example, a process engineer may prompt, “Show how tilt deformation may evolve over time if platen ramp-down is shortened by 20%, ” while a reliability planning tool may request, “Simulate void expansion patterns for hybrid laminate stacks under elevated soak conditions.” In response, the DB post-process module 114 may retrieve image-based outputs from the DB post-processing equipment 148, align them with temperature, pressure, and displacement telemetry from the DB post-processing database 124, and apply a generative model trained on historical defect progression sequences stored in the DB network database 126. Internally, the model may use spatiotemporal generation layers conditioned on stress evolution variables and substrate properties to generate frame-by-frame projections of defect morphology over time.

[0107] [HO] The resulting video simulation may visualize tilt amplification, void expansion paths, or delamination initiation based on hypothetical parameter shifts. Output may include the predictive video, a summary of the causal input conditions, and a literature citation linking asymmetric thermal soak rates to delamination risks in high-aspect-ratio die bonding. By embedding generative video simulation into the DB post-process module 114, the system may enable proactive reliability planning and process sensitivity exploration without requiring physical wafer sacrifice. The DB post-process module 114 performs, at step 508, the post-process algorithm. The post-process algorithm may be used to maintain consistent color and luminance across the pixels. For example, luminance may be the intensity of the light that a microLED emits, each microLED manifests distinct variations in luminance and color, and to attain visual uniformity throughout the entire display, accurate measurement and correction are necessary. The luminance of the microLEDs may be determined by imaging colorimeters and other specialized instruments that quantify luminance and color uniformity, the information provided by these instruments is used to calibrate displays and to attain uniformity, it is necessary to individually measure and adjust each microLED. In some embodiments, color and luminance may be modified to conform to a specified standard. The luminance of microLEDs may be measured in candelas per square meter (cd / m2). In some embodiments, the luminance for indoor displays may be within the range of 40 to 300 cd / m2l, sRGB monitors may be around 100 cd / m2, consumer desktop monitors may be between 200 to 300 cd / m2, HDR displays may fall within 450 to above 1600 cd / m2. The luminance of the microLEDs may be changed or influenced by various parameters in the die bonding process, such as variations in the thickness or uniformity of the die attach material, inadequate temperature control, pressure control, alignment accuracy, etc. For example, alignment accuracy may result in adverse effects on luminance output during the die bonding process.

[0108] [Hl] The typical alignment accuracy for calibrated alignment falls within the range of approximately 5 pm, or micrometers. The alignment accuracy may be influenced by roll stamp structure, X-Y theta stage performance, stage-roller orthogonality, roller roundness, etc. which can lead to mechanical stress, damage of the microLEDs, luminance variations, defects, etc. The alignment accuracy may be enhanced through machine calibration and precision, roll stamp design and geometry, roller roundness and orthogonality, alignment sensors, and feedback systems, material properties and wafer warpage, self-alignment techniques, active optical alignment, underfill material and void reduction, etc. The post-process may categorize the wet etched wafers based on their warpage, which is the degree of deformation or curvature in the wafer surface resulting from the etching process. By grouping wet etched wafers with similar warpage characteristics together, the post-process algorithm ensures that during the subsequent die bonding step, wafers with comparable warpages are bonded together to minimize overall warpages within each group, leading to reduced alignment challenges during bonding. For example, when wet etched wafers with similar warpage are bonded together, the overall warpage variation within each group is reduced.

[0112] As a result, the alignment process during die bonding becomes more consistent and predictable. On average, the luminescence of microLEDs bonded using this optimized approach tends to be higher and exhibits a more tightly distributed pattern. This improvement is attributed to the reduced impact of warpage on alignment accuracy. With less variation in alignment, the bonding of microLEDs is more uniform, leading to improved luminous output and overall performance. In some embodiments, the post-process algorithm may be a clustering algorithm, such as a K-Means clustering to clusters chips according to the degree of similarity in wafer warpage, hierarchical clustering which enables the grouping of chips at varying degrees of granularity through the formation of a hierarchy of clusters, density-based spatial clustering of applications with noise which is a technique utilized to detect dense regions of chips within the warpage space.

[0109]

[0113] In some embodiments, the post-process algorithm may perform a principal component analysis which may reduce the dimensionality of the warpage data while preserving its variance and chips can then be grouped based on the transformed features.

[0110]

[0114] In some embodiments, the post-process algorithm may perform self-organizing maps which is a neural network-based technique that maps high-dimensional data onto a lowerdimensional grid, and chips with similar warpage patterns are grouped. In some embodiments, the post-process algorithm may perform autoencoders which are neural networks that learn compact representations of the warpage data and the encoded features can be used for clustering.

[0111]

[0115] In some embodiments, the post-process algorithm may use machine learning algorithms to analyze parameters related to luminescence. The post-process algorithm may focus on the process data outputted from the inspection algorithm performed in the DB inspection module 110 and the effectiveness of the parameters outputted by the control algorithm performed in the DB process module 112 during the die bonding process. The post-process algorithm may adjust control parameters of the current microLED processed dies in the process to result in an improved die bonding process. For example, the post-process algorithm may optimize alignment to group the wafers based on warpage to improve the luminescence. The DB post-process module 114 performs, at step 510, the post-process inspection. For example, the post-process may include defect inspection to detect flaws or defects in chips or wafers to avert their integration into the final product, color and brightness variation checks to ensure color and brightness uniformity across pixels, performing inline metrology and automated optical inspection to identify and rate any areas of nonuniformity or defects, etc. The DB post-process module 114 stores, at step 512, the data in the DB post-process database 124. The DB post-process module 114 stores the data in the DB postprocess database 124, such as the data collected from the die bonding post-process 150, the output of the post-process algorithm, etc. The DB post-process database 124 may contain the wafer group number or ID, the alignment measure based on binned together based on wafer warpage, the luminescence, etc. The DB process database 122 may contain the data for each die that is being processed by the die bonding post-process 150. The DB post-process module 114 returns, at step 514, to the base module 108.

[0112]

[0116] In microLED manufacture, the K-Means Clustering algorithm is a particularly useful technique for optimizing the post-process with the goal of achieving consistent brightness and color across pixels. This decision is motivated by the algorithm's ability to cluster data based on similarity, which in this case enables the arrangement of microLEDs or dies based on shared properties such as wafer warpage. These properties have a major impact on the brightness and color consistency of the tiny LEDs, which is important for maintaining visual consistency amongst displays. One well-known unsupervised learning approach is K-Means Clustering, which divides a dataset into discrete subsets or clusters. The technique works by maximizing the variation between clusters and decreasing the variance within clusters. When wafers or dies with similar qualities need to be identified and grouped together during the post-processing stage of microLED manufacture, this feature comes in handy. This allows for the die bonding process to be adjusted to minimize deviations that can negatively impact the brightness and color uniformity of the finished product. Imagine a situation where improving the brightness uniformity throughout the display is the main objective of the production process for microLED displays. The strategy would take multiple steps, the first of which would be gathering extensive post-process data. This data comprises precise measurements of the warpage of each wafer, the correctness of the alignment of each die or wafer, and the measurements of each microLED's luminance. Such thorough data collection is essential because it establishes the framework for the actions that follow. The next stage is to choose features from this dataset that significantly affect brightness uniformity after the pertinent data has been gathered. Measurements of wafer warpage and alignment accuracy data are typically included in this. The dies or wafers are then grouped using the K-Means Clustering technique after the pertinent features have been found. The algorithm's job is to identify groups of dies or wafers that share alignment and warpage traits, which are a sign of possible profiles for brightness and color uniformity. Following the clustering, the outcomes are examined to identify particular die or wafer groups that may need modifications throughout the die bonding procedure.

[0113]

[0117] These modifications could involve modifying temperature, pressure, or alignment procedures in an effort to lessen the observed variances and improve the display's overall brightness uniformity. Comprehensive post-process data from the die bonding processes is included in the datasets used for this purpose. The wafer warpage measurements, which describe the amount and pattern of warpage for each wafer, the alignment accuracy data, which records departures from ideal conditions, and the luminance measurements for each microLED, which shed light on uniformity throughout the display, are important components of these datasets. In order to ensure that the algorithm can discover meaningful categories that appropriately reflect variances experienced in the manufacturing process, these datasets should ideally span a wide variety of operational and severe situations. By leveraging K-Means Clustering in this manner, the postprocess algorithm can effectively categorize dies or wafers with similar characteristics, enabling targeted process adjustments. This methodology not only optimizes the luminance and color uniformity across microLED displays but also enhances the overall quality and visual appeal of the final product.

[0114]

[0118] FIG. 6 illustrates the DB historical module 116. The process begins with the DB historical module 116 being initiated, at step 600, by the base module 108. The DB historical module 116 connects, at step 602, to the die bonding pre-process 130, the die bonding process 140, and the die bonding post-process 150. The historical module 116 connects to the die bonding pre-process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc., the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that places the microLED dies onto a prepared substrate, secures the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength, and the die bonding post-process 150, such processes and methods after the completion of the die bonding process 140, such as a cleaning and inspection, electrical testing, encapsulation, etc. The DB historical module 116 aggregates, at step 604, the data from the various types of processes and stores the data in the DB network database 126. The DB network database 126 may contain the historical data from the various processes performed by the die bonding pre-process 130, die bonding process 140, and die bonding postprocess 150. The DB network database 126 may contain the data parameters collected during the inspection process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc., control parameters of the process, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc., and data parameters collected during post-processing, such as bond quality, defect detection, electrical performance, encapsulation quality, etc. The DB historical module 116 performs, at step 606, the historical machine learning algorithm on the historical data stored in the DB network database 126. The historical machine learning algorithm may improve each stage of the microLED manufacturing process, such as die bonding pre-process, die bonding process, and die bonding post-process.

[0115]

[0119] For example, the historical machine learning algorithm may receive input data from the DB inspection module 110, such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, and analyze the historical data of electrical resistance and die bonding time to adjusts die bonding time predictions based on variations in electrical resistance which improves accuracy in predicting die bonding time for different electrical resistance values and leads to optimized die bonding pre-process. The historical machine learning algorithm may receive input data from the DP process module 112, such as die bonding pressure with calculated temperature control, and void defects per unit standard area, and may incorporate reinforcement learning techniques to adjust die bonding pressure and temperature control based on real-time feedback of void defects detected, to enhanced pressure and temperature control during die bonding which leads to a reduction in void defects and improved bonding quality. The historical machine learning algorithm may receive input data from the DP post-process module 114, such as alignment measurement warpage Al binned, Luminescence, etc., to implement deep learning models to analyze alignment measurements and luminescence data to learn patterns of alignment-related warpage and luminescence variations which provides insights into the relationship between alignment, warpage, and luminescence, facilitating adjustments in the post-die bonding process to improve microLED chip quality.

[0120] In some embodiments, the historical machine learning algorithm may enable iterative improvements in each stage of the microLED manufacturing process by learning from historical data. In some embodiments, the historical machine learning algorithm may enhance process parameters and controls which leads to better-quality microLED chips and higher manufacturing efficiency. In some embodiments, the historical machine learning algorithm may perform data- driven decision making which allows for informed decision-making based on data analysis and leads to more accurate predictions and optimizations.

[0116]

[0121] In some embodiments, the historical machine learning algorithm may use the historical data stored in the DB network database 126 to determine trends to understand how varying parameters affect the final quality and identify optimal parameter settings to create the best outcomes. The historical machine learning algorithm adjusts the parameters in real time for the DB inspection module 110, DB process module 112, and DB post-process module 114 to correct any defects. For example, the data may show defects, such as voids, that occurred during the die bonding process. The historical machine learning algorithm may minimize the void defects in the wafer by adjusting the time at a predetermined temperature to improve the electrical resistance in the DB inspection module 110. The historical machine learning algorithm may also adjust the temperature occurring in the die bonding pressure process in the DB process module 112 to minimize the void defects.

[0117]

[0122] LLM Example 1 : In some embodiments, the DB historical module 116 may utilize a large language model (LLM) to autonomously generate trend-based interpretations and corrective action narratives from high-dimensional data contained within the DB network database 126. For example, a user may input the query, “Why is group 54 exhibiting persistent luminescence variability?”, while a downstream request may involve an inferred inquiry such as, “What adjustments should be prioritized for alignment accuracy improvements next quarter?”. The LLM may analyze inspection values from DB inspection database 120, process parameter logs from DB process database 122, and outcome data from DB post-process database 124 to generate a response that identifies historical deviations in adhesive thickness and thermal gradient non-uniformity as contributing factors to the observed luminance variability. The system may further retrieve peer- reviewed literature on die attach uniformity and cite a published technique for predictive underfill placement to mitigate luminance anomalies. Internally, the LLM may perform a multi-layered reasoning chain involving data aggregation across temporal batches, clustering of correlated defect modes, semantic summarization of intervention records, and mapping of these insights into alignment with recognized best practices. Through such contextual interpretation and actionable synthesis, the use of LLMs within the DB historical module 116 enhances long-range fault analysis and embeds generalized domain knowledge within the continuous improvement cycle. The DB historical module 116 sends, at step 608, the process adjustments to the DB inspection module 110, the DB process module 112, and the DB post-process module 114. The historical machine learning algorithm may optimize the DB inspection module 110 by improving the accuracy in predicting die bonding time for different electrical resistance values. The historical machine learning algorithm may also reduce void defects by enhancing pressure and temperature control during the DB process module 112. The historical machine learning algorithm may improve microLED chip quality by providing relationships and adjustments between alignment, warpage, and luminescence to the DB post-process module 114. The adjustments are sent to the DB inspection module 110 and DB process module 112 to create a more consistent final product which is then verified by the DB post-process module 114. The DB historical module 116 returns, at step 610, to the base module 108.

[0118]

[0123] In the context of improving the microLED manufacturing process at step 606 using a historical machine learning algorithm, the Random Forest algorithm presents itself as a highly versatile and powerful choice. This algorithm stands out for its ability to handle a wide range of data types and complexities, making it particularly suitable for analyzing the diverse datasets involved in the microLED manufacturing process, from pre-processing and bonding to postprocessing stages. The Random Forest algorithm operates by constructing multiple decision trees during the training phase and outputting the mode of the classes (classification) or mean prediction (regression) of the individual trees. Random Forest is known for its high accuracy, ability to deal with unbalanced and missing data, and its feature of providing a clear indication of the importance of each feature on the prediction. This makes it ideal for dissecting the intricate relationships and dependencies between various process parameters and outcomes in microLED manufacturing. The goal is to enhance the overall quality and efficiency of the microLED manufacturing process. The DB historical module would begin by aggregating and storing extensive historical data from various stages of the manufacturing process, including die sorting, bonding techniques, and postprocess quality checks. This dataset would encompass a wide range of parameters such as die characteristics, adhesive thickness, bond strength, defect rates, and luminescence measurements, among others. The Random Forest algorithm would then be applied to this comprehensive dataset to perform both predictive and analytical tasks. For instance, it could predict optimal die bonding times based on variables like electrical resistance and adhesive characteristics, identify patterns and correlations between die placement accuracy and subsequent bond quality, or determine the most significant factors affecting luminescence and color uniformity in the final product. Through this analysis, the algorithm could recommend adjustments to process parameters in real-time, aiming to reduce defects, enhance bond quality, and ensure uniformity in luminescence across microLED displays. These adjustments might include modifying bonding pressure and temperature based on real-time feedback, optimizing adhesive application based on the characteristics of the dies and substrates, or adjusting alignment protocols to account for wafer warpage and minimize luminescence variation. The datasets used in this process are pivotal, encompassing historical data from the entire spectrum of the manufacturing process. This includes detailed records of process parameters (such as temperature, pressure, duration, and accuracy of die placement), material characteristics (like substrate cleanliness and adhesive thickness), and outcomes (including bond quality, defect detection, electrical performance, and encapsulation quality).

[0119]

[0124] By analyzing these datasets, the Random Forest algorithm can uncover trends and dependencies that would not be apparent otherwise, facilitating a deeper understanding of how different variables influence the manufacturing process and product quality. Employing the Random Forest algorithm in this manner enables a data-driven approach to continuous improvement in the microLED manufacturing process. It leverages historical data to make informed decisions, thereby enhancing process efficiency, reducing the incidence of defects, and ultimately leading to the production of higher quality microLED chips. This approach exemplifies how machine learning algorithms can transform manufacturing processes by enabling a deeper understanding of complex relationships between process parameters and outcomes. The Random Forest algorithm shows up as a very adaptable and potent option when it comes to utilizing a historical machine learning algorithm to improve the microLED production process at step 606. This method is particularly useful for analyzing the different datasets involved in the microLED production process, from pre-processing and bonding stages to post-processing stages. It stands out for its ability to handle a wide range of data kinds and complexities. During the training phase, the Random Forest algorithm builds numerous decision trees, and it outputs the mean prediction (regression) or the mode of the classes (classification) for each tree. High accuracy, flexibility in handling missing and imbalanced data, and the ability to clearly display the contribution of each feature to the prediction are all attributes of Random Forest. Because of this, it's perfect for breaking down the complex connections and dependencies between different process parameters and results in the production of tiny LEDs. Imagine a situation where improving the entire quality and productivity of the microLED production process is the aim. First, a large amount of historical data from different manufacturing phases, such as die sorting, bonding methods, and post-process quality checks, would be combined and stored in the DB historical module. This dataset would include several different metrics, including luminescence measurements, defect rates, adhesive thickness, bond strength, and die properties. Then, using this large dataset, the Random Forest algorithm would be used to carry out analytical and predictive tasks.

[0120]

[0125] It might, for example, identify patterns and correlations between die placement accuracy and subsequent bond quality, predict ideal die bonding times based on variables like electrical resistance and adhesive characteristics, or identify the key elements influencing luminescence and color uniformity in the finished product. Based on this research, the program might suggest realtime modifications to process settings with the goal of lowering faults, improving bond quality, and guaranteeing luminosity consistency across tiny LED displays. These modifications might be as simple as tweaking alignment techniques to limit luminescence variation and account for wafer warpage, or they could involve optimizing adhesive application depending on die and substrate properties, or changing bonding pressure and temperature based on real-time feedback. The historical data from every stage of the manufacturing process is included in the crucial datasets used in this process. This includes thorough documentation of material properties (like adhesive thickness and substrate cleanliness), process parameters (like temperature, pressure, duration, and die placement accuracy), and results (like bond quality, defect detection, electrical performance, and encapsulation quality). The Random Forest algorithm's analysis of these datasets allows it to identify patterns and dependencies that would not be visible otherwise, allowing for a better understanding of the ways in which various factors affect the production process and final product quality. This method of using the Random Forest algorithm allows for a data-driven approach to ongoing process optimization in the production of tiny LEDs. It makes judgments based on past data, which improves process efficiency, lowers defect rates, and eventually results in the manufacturing of microLED chips of greater quality. This method provides a deeper knowledge of the intricate links between process parameters and outcomes, illuminating how machine learning algorithms might revolutionize industrial processes.

[0121]

[0126] FIG. 7 illustrates the DB integration module 118. The process begins with the DB integration module 118 being initiated, at step 700, by the base module 108. The DB integration module 118 performs, at step 702, the integration machine learning algorithm. The integration machine learning algorithm may facilitate interactions between the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150 stages of microLED manufacturing. For example, the integration machine learning algorithm may receive input data, such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, alignment measurement, luminescence, etc., to analyze the relationship between die bonding pre-process 130 parameters and die bonding post-process 150 outcomes and may adjust die bonding pre-process 130 parameters to optimize die bonding post-process 150 results which improve alignment and luminescence of microLED chips by adjusting die bonding pre-process 130 parameters based on their impact on die bonding post-process 150 outcomes. The integration machine learning algorithm may receive input data, such as die bonding pressure with Al temperature control, void defects per unit standard area, alignment measurement, luminescence, etc., and implement reinforcement learning algorithms to analyze the relationship between die bonding process 140 parameters and die bonding post-process 150 outcomes and adjusts die bonding process 140 parameters to optimize die bonding post-process 150 results in which a reduction of void defects and improvement in alignment and luminescence is achieved by adjusting die bonding process 140 parameters based on their impact on die bonding post-process 150 outcomes. The integration machine learning algorithm may receive input data, such as alignment measurement, luminescence, die bonding pressure with Al temperature control, void defects per unit standard area, etc., and utilizes deep learning models to analyze the relationship between die bonding post-process 150 outcomes and die bonding process 140 parameters by adjusts die bonding process 140 parameters to optimize die bonding post-process 150 results to improve alignment and luminescence by adjusting die bonding process 140 parameters based on their impact on die bonding post-process 150 outcomes.

[0122]

[0127] In some embodiments, the DB integration module 118 may facilitate the optimization of the entire microLED manufacturing process by considering interactions between different stages. In some embodiments, the DB integration module 118 may leverage Al to adjust parameters in one stage based on their impact on outcomes in subsequent stages to enhance the overall process efficiency and product quality.

[0123]

[0128] In some embodiments, the DB integration module 118 may adapt to changes in process conditions and requirements by continuously learning data and adjusting parameters accordingly.

[0124]

[0129] In some embodiments, the integration machine learning algorithm may use the data stored in the DB 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 improving luminescence across the die bonding pre-process 130, die bonding process 140, and die bonding post-process 150. The historical data in the DB network database 126 may include information on luminescence and relevant operational parameters such as the average electrical resistance of the contacts, the calculated time at 250°C temperature for die bonding, die bonding pressure with Al temperature control, void defects per unit standard area, 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 luminescence and engineer new features if necessary. For example, the interaction between the electrical resistance of the contacts and the calculated time at 250°C temperature for die bonding may have a significant impact on the luminescence. The selected predictive model is trained using the training dataset. The model learns patterns and relationships between operational parameters and luminescence. 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 die bonding pre- process 130, die bonding process 140, and die bonding post-process 150.

[0130] Al Agents Example 2: In some embodiments, the DB integration module 118 may deploy a persistent Al agent configured to autonomously monitor defect trends and tool state signals across the DB inspection module 110, DB process module 112, and DB post-process module 114, and selectively invoke the integration machine learning algorithm at step 702 in response to real-time yield degradation triggers. For example, a process integration engineer may issue a standing instruction such as, “Adjust dispense ramp profiles automatically if edge void rate increases for die sizes over 5 mm,” while the Al agent may independently trigger, “Rerun optimization if die tilt exceeds 3o across three consecutive lots on Tool 7.” The agent may continuously analyze image- tagged defect outputs from the DB inspection database 120, pressure and temperature logs from the DB process database 122, and substrate stress patterns recorded in the DB post-processing database 124. Upon detecting a multi-lot deviation pattern associated with compound yield risk, the agent may autonomously invoke step 702, generate new control recommendations, and issue real-time updates through the DB integration module 118. Internally, the agent may use reinforcement-learned policy functions and defect risk trajectory modeling trained on data archived in the DB network database 126 to determine optimal trigger timing and action prioritization.

[0125]

[0131] Outputs may include updated parameter dispatches with rationale tags and a continuous decision log stored for traceability. By embedding an autonomous Al agent in the DB integration module 118, the system may elevate from passive batch optimization to proactive, event-driven control adaptation, improving real-time defect suppression and process agility. The DB integration module 118 connects, at step 704, to the die bonding pre-process Al module 136, the die bonding process Al module 146, and the die bonding post-process Al module 156. The DB integration module 118 connects to the die bonding pre-process 130, such as die sorting and inspection, cleaning processes, electrical testing, substrate preparation, alignment marking, adhesive dispensing, etc., the die bonding process 140, such as systems or equipment for automated die bonding of microLED chips that places the microLED dies onto a prepared substrate, secures the dies to the substrate through bonding techniques, such as pressure bonding, thermocompression bonding, ultrasonic bonding, etc., performs a curing process to solidify the adhesive and enhance the bond strength, and the die bonding post-process 150, such processes and methods after the completion of the die bonding process 140, such as a cleaning and inspection, electrical testing, encapsulation, etc. The DB integration module 118 sends, at step 706, the process adjustments to the die bonding pre-process Al module 136, the die bonding process Al module 146, and the die bonding post-process Al module 156. For example, the DB integration module 118 may send the die bonding time, pressure, temperature, etc. adjustments to each of the processes to optimize the results of the die bonding process.

[0126]

[0132] In some embodiments, the DB integration module 118 may send the integration machine learning algorithm to the processes, allowing the systems to further enhance the optimization.

[0127]

[0133] LLM Example 2: In some embodiments, the DB integration module 118 may incorporate a large language model (LLM) to generate human-readable rationales and training summaries based on control parameter decisions issued by the integration machine learning algorithm at step 702, thereby supporting interpretability and cross-team knowledge transfer. For example, a packaging engineer may prompt, “Why were dispense pressure and stage speed modified for Lot B 13?” while a training lead may request, “Summarize the last five integration updates with context for technician onboarding.” In response, the DB integration module 118 may extract historical parameter changes from the DB process database 122, correlate these changes with inspection results from the DB inspection module 110 and defect classification trends in the DB network database 126, and apply the LLM to generate plain-language explanations of the cause-effect relationships underlying each adjustment. Internally, the model may tokenize parameter deltas, link them to past intervention records, and semantically rank explanatory factors based on cooccurrence with known defect signatures such as die tilt or void presence. Output may include a rationale summary for each modified parameter with defect correlation traceability, along with an onboarding document citing literature-supported best practices for managing adhesive flow variability. By embedding LLM-generated explanation capabilities within the DB integration module 118, the system may improve the transparency of Al-driven control changes and accelerate alignment between automated decision systems and technician understanding. The DB integration module 118 returns, at step 708, to the base module 108.

[0128]

[0134] At step 702 of the microLED manufacturing process, the application of an integration machine learning algorithm, specifically a Reinforcement Learning (RL) algorithm, could offer substantial benefits. This algorithm is particularly adept at facilitating interactions and optimizations across various stages of the manufacturing process, including the die bonding pre- process, process, and post-process stages. Reinforcement Learning stands out for its ability to learn optimal actions through trial and error, directly interacting with the environment to achieve specific goals. This capability makes it uniquely suited for dynamically adjusting process parameters in real time to enhance outcomes such as alignment and luminescence in microLED chips. In the context of microLED manufacturing, the Reinforcement Learning algorithm can continuously receive and analyze input data from various stages of the process, such as electrical resistance measurements, die bonding times and temperatures, alignment accuracies, luminescence levels, and defect rates. Based on this data, the RL algorithm can identify the most effective adjustments to parameters at each stage of the manufacturing process to optimize overall product quality and process efficiency.

[0129]

[0135] For instance, it might adjust the die bonding pressure and temperature in response to detected variations in electrical resistance or void defects, with the goal of minimizing defects and improving bond quality. A simple example of this in action might involve the algorithm initially receiving data indicating a correlation between certain adhesive dispensing patterns in the pre- process stage and void defects observed in the post-process stage. By experimenting with different adhesive dispensing parameters and observing the resulting impact on void defects, the RL algorithm can learn to recommend adjustments that consistently reduce void occurrences. The datasets used for this application are comprehensive, spanning across all stages of the microLED manufacturing process. These datasets include detailed records of operational parameters such as die bonding pressure, temperature, adhesive thickness, electrical resistance of the contacts, and outcomes like defect rates, bond quality, alignment accuracy, and luminescence levels. This data is critical for the RL algorithm, as it relies on a feedback loop of actions and outcomes to learn and propose improvements. For preprocessing, the data must be cleaned and formatted correctly, with any missing values addressed and features normalized or scaled as necessary.

[0130]

[0136] The dataset would typically be divided into subsets for training and testing the algorithm, ensuring that the model can generalize well to new, unseen data. Throughout the training process, the algorithm identifies key features and patterns that impact the final product quality, such as the relationship between pre-process temperatures and luminescence uniformity. It learns to make decisions that optimize these key outcomes based on real-time data and feedback. By deploying a Reinforcement Learning algorithm within the DB integration module, the microLED manufacturing process benefits from an Al-driven approach that continuously adapts and optimizes process parameters across all stages. This not only enhances the efficiency and quality of the manufacturing process but also leads to improved final products with better alignment and luminescence characteristics. This dynamic and adaptive approach exemplifies the potential of integrating advanced machine learning techniques into complex manufacturing processes. The use of an integration machine learning algorithm, more especially a Reinforcement Learning (RL) algorithm, at step 702 of the microLED production process may provide significant advantages. Die bonding pre-process, process, and post-process stages are just a few of the manufacturing stages where this algorithm excels at enabling interactions and optimizations. Reinforcement learning is distinguished by its capacity to discover the best course of action via trial and error, engaging with the environment directly to accomplish predetermined objectives. Because of this feature, it is especially well-suited for dynamically modifying process settings in real time to improve microLED chip alignment and luminosity. The Reinforcement Learning algorithm is capable of continuously receiving and analyzing input data from different stages of the process in the context of microLED manufacturing. These stages include electrical resistance measurements, die bonding times and temperatures, alignment accuracies, luminescence levels, and defect rates. The RL algorithm can determine the best parameter adjustments at each stage of the manufacturing process based on this data, maximizing both process efficiency and overall product quality. In order to reduce defects and enhance bond quality, it could, for example, modify the die bonding pressure and temperature in response to fluctuations in electrical resistance or void defects that are identified.

[0131]

[0137] To illustrate this, the algorithm is first presented with data suggesting a relationship between specific adhesive dispensing patterns during the pre-process stage and void defects found during the post-process stage. The RL algorithm can learn to suggest modifications that reliably lower the frequency of void events by experimenting with various adhesive dispensing parameters and tracking the impact on void defects. The datasets utilized in this application are extensive, covering every phase of the production of tiny LEDs. These files contain comprehensive records of operational characteristics, including defect rates, bond quality, alignment accuracy, and luminescence levels, as well as operational factors including die bonding pressure, temperature, adhesive thickness, and electrical resistance of the contacts. The RL algorithm depends on a feedback loop of actions and results to learn and suggest changes, therefore this data is essential to its operation. The data needs to be properly cleaned, structured, and preprocessed in order to resolve any missing values and normalize or scale features as needed. For the purpose of training and testing the algorithm, the dataset would normally be partitioned into subsets to make sure the model can adapt well to new, untested data. The algorithm learns important characteristics and trends during training that affect the quality of the finished output, like the correlation between luminescence uniformity and pre-process temperatures. It gains the ability to make choices based on real-time data and feedback that optimize these important outcomes. The DB integration module's implementation of a Reinforcement Learning algorithm gives the microLED production process an Al-driven strategy that continuously adjusts and optimizes process parameters at every level. This improves the production process' effectiveness and quality while also producing superior end products with increased alignment and luminescence properties. This flexible and dynamic method shows how cutting-edge machine learning techniques may be successfully incorporated into intricate industrial processes.

[0132]

[0138] FIG. 8 illustrates the DB inspection database 120. The DB inspection database 120 provides an example of the results of the inspection algorithm performed in the DB inspection module 110. The DB inspection database 120 displays an example of data for various groups of microLED wet etching that will be die bonded to a substrate, where each group has an average electrical resistance where the variation may be based on normal variations of other tool parameters, such as pressure, time, atmospheric conditions. This example represents example data of electrical resistance changes against the normal temperature variations of the tool, where it may be controlled at 250C but varies due to tool control which in this example may vary +-10% for example. An analysis of data of all the tool parameters may show that time at a particular temperature is also a factor that changes electrical resistance in a run, for example, by a regression analysis of time against electrical resistance. For example, a strong correlation may be found between time and electrical resistance. Given this correlation, the next example shows example data where the set point was set at 250 C, but the time of each run was changed to compensate for the actual temperature changes resulting in electrical resistance that is far less and varies less. In some embodiments, the DB inspection database 120 may include other parameter data collected or calculated during the DB inspection module 110 process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance, etc.

[0133]

[0139] FIG. 9 illustrates the DB process database 122. The DB process database 122 provides an example of the results of the control algorithm performed in the DB process module 112. For example, die bonding pressure may refer to the force applied during the bonding process to attach a semiconductor die, such as a microLED, to its substrate or package and ensures proper contact between the die and the substrate, allowing for efficient heat dissipation, electrical connectivity, and mechanical stability. The factors that influence die bonding pressure are material properties, die size and shape, adhesive type, process temperature, etc. In some embodiments, the bonding pressure for microLEDs may be between 50 g / mm2to 500 g / mm2. There may be pressure variations while maintaining pressure control in die bonding which may produce defects, such as voids. These variations may be caused by mechanical wear, calibration issues, temperature fluctuations, hydraulic system performance, material properties, process variability, etc. The first example displays the variations in pressure applied during the die bonding process and the voids produced during the process. The second example displays example data where the die bonding pressure changed within normal process variation, but the temperature was changed to compensate, where the temperature change was calculated by regression analysis from pressure versus temperature, that is as the pressure went up, the temperature would be reduced. The example data shows the voids are reduced and there is less variation than the example data above.

[0134]

[0140] In some embodiments, the control algorithm may control the process parameters, such as temperature, to adjust the pressure being exerted by the hydraulic system and components of the die bonder which can be altered by thermal expansion or contraction. In some embodiments, the DB process database 122 may store other parameter data collected or calculated from the DB process module 112, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality, etc.

[0135]

[0141] FIG. 10 illustrates the DB post-process database 124. The DB post-process database 124 provides an example of the results of the post-process algorithm performed in the DB post-process module 114. The example data illustrates the average misalignment in the x-direction, which should be zero, to the measured final average luminescence of the group. For example, alignment accuracy may result in adverse effects on luminance output during the die bonding process. The typical alignment accuracy for calibrated alignment falls within the range of approximately 5 pm, or micrometers. The alignment accuracy may be influenced by roll stamp structure, X-Y theta stage performance, stage-roller orthogonality, roller roundness, etc. which can lead to mechanical stress, damage of the microLEDs, luminance variations, defects, etc. The alignment accuracy may be enhanced through machine calibration and precision, roll stamp design and geometry, roller roundness and orthogonality, alignment sensors, and feedback systems, material properties and wafer warpage, self-alignment techniques, active optical alignment, underfill material and void reduction, etc. The post-process algorithm may categorize the wafers based on their warpage, which is the degree of deformation or curvature in the wafer surface resulting from the etching process. By grouping wafers with similar warpage characteristics together, the post-process algorithm ensures that during the subsequent die bonding step, wafers with comparable warpage are bonded together to minimize overall warpage within each group, leading to reduced alignment challenges during bonding, as shown in the second example of data. For example, when wafers with similar warpage are bonded together, the overall warpage variation within each group is reduced. As a result, the alignment process during die bonding becomes more consistent and predictable. On average, the luminescence of microLEDs bonded using this optimized approach tends to be higher and exhibits a more tightly distributed pattern. This improvement is attributed to the reduced impact of warpage on alignment accuracy. With less variation in alignment, the bonding of microLEDs is more uniform, leading to improved luminous output and overall performance. In some embodiments, the DB post-process database 124 may store other parameter data collected or calculated from the DB post-process module 114, such as bond quality, defect detection, electrical performance, encapsulation quality, etc.

[0136]

[0142] 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 die bonding (DB) method for microLED chips system, comprising: an automated die bonding 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 die bonding processing module; an automated die bonding processing module which inputs the at least one parameter from said second group to control the die bonding using at least one parameter from a third group of parameters; an automated die bonding 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 die bonding 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 die bonding integration module to update the automated die bonding processing module based upon the results of the automated die bonding historical module.

2. The method of claim 1, wherein the DB process module acts as the core controller, utilizing inputs from the DB inspection module to govern the die bonding process to ensure optimal pressure, temperature and duration.

3. The method of claim 1, wherein the DB post-process module conducts inspections to assess the execution of parameters and identify potential deviations, and the DB historical module analyzes data collected across various stages, allowing for continuous learning and pattern recognition and subsequently a DB 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 a DB Al network and a DB pre-process, the DB process, and a DB 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 DB inspection module, the DB process module, the DB post-process module, the DB historical module, and the DB integration module.

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

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

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

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

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

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

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

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

19. The method of claim 10, wherein there is a DB inspection database which provides results of the inspection algorithm performed in the DB inspection module displaying data for various groups of microLED wet etching that will be die bonded to a substrate, where each group has an average electrical resistance where the variation may be based on normal variations of other tool parameters, such as pressure, time and atmospheric conditions.

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

21. The method of claim 12, wherein a DB process database provides results of the inspection algorithm performed in the DB process module.

22. The method of claim 21, wherein the control algorithm controls process parameters, such as temperature, to adjust a pressure being exerted by a hydraulic system and components of a die bonder which are alterable by a thermal expansion or contraction.

23. The method of claim 21, wherein the DB process database stores other parameter data collected or calculated from the DB process module, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment and cure quality.

24. The method of claim 14, wherein a DB post-process database provides results of the postprocess algorithm performed in the DB post-process module.

25. The method of claim 24, wherein, the DB post-process database stores other parameter data collected or calculated from the DB post-process module, such as bond quality, defect detection, electrical performance and encapsulation quality.

26. The method of claim 24, wherein data illustrates an average misalignment in the x-direction, which should be zero, to the measured final average luminescence of the group wherein a typical alignment accuracy for calibrated alignment falls within the range of approximately 5pm, or micrometers and the alignment accuracy is enhanced through machine calibration and precision, roll stamp design and geometry, roller roundness and orthogonality, alignment sensors, and feedback systems, material properties and wafer warpage, selfalignment techniques, active optical alignment, underfill material and void reduction wherein the post-process algorithm categorizes the wafers based on their warpage, which is the degree of deformation or curvature in the wafer surface resulting from the etching process.

25. The method of claim 16, wherein include the DB network database contains the historical data from the various processes performed by the DB pre-process, DB process, and DB post-process, the data parameters collected during the inspection process, such as die characteristics, substrate cleanliness, adhesive thickness, electrical resistance and control parameters of the process, such as pressure, temperature, duration, die placement accuracy, bond strength, bond line thickness, alignment, cure quality and data parameters collected during post-processing, such as bond quality, defect detection, electrical performance and encapsulation quality.

Images