Awafer array module bonding method

Abstract

The invention relates to a non-wafer array module bonding method, which comprises the following steps of: cutting a Micro-LED chip wafer and a driving substrate wafer into a plurality of square wafers which are arranged in an up-and-down matching manner according to a specified cutting channel, bonding a plurality of Micro-LED chips with a corresponding driving substrate at the same time, and after the bonding is finished, carrying out bonding on the driving substrate and the Micro-LED chips, so as to obtain the non-wafer array module. And the bonded multi-chip module is divided into independent single chip units through a scratching process for subsequent assembly and packaging. Batch bonding is carried out on multiple Micro-LED chips at the same time, the one-by-one operation step of single-chip bonding is greatly reduced, and the integration efficiency is remarkably improved; the influence of single-point defects on the overall yield can be reduced, meanwhile, the interface problem caused by stress or uneven machining in wafer bonding is avoided, and the integration reliability is improved; by combining the subsequent scribing process, the manufacturing process is optimized, the complex post-processing steps are reduced, the production cost is reduced, and economic and feasible technical support is provided for large-scale application of Micro-LEDs.

Description

Technical Field

[0001] The present invention relates to the technical field of micro light emitting diode manufacturing technology, and in particular to a non-wafer array module bonding method. Background Art

[0002] Micro-LED (micro light-emitting diode) is an emerging display technology that uses tiny LED chips as independent pixel units to achieve high brightness, wide color gamut, low power consumption, and long life. However, due to the extremely small size of Micro-LED chips and the high density they require, efficient integration with driver circuits has become a core technical challenge for industrialization. Current integration methods mainly include monolithic bonding and wafer bonding, but both technologies have significant shortcomings in efficiency and yield, limiting the large-scale production of Micro-LEDs.

[0003] Single-wafer bonding achieves the connection between the chip and the substrate through a one-by-one operation. Its advantage lies in high operational precision, but its extremely low efficiency makes it difficult to meet the needs of large-scale mass production. At the same time, the alignment and bonding stability during the bonding process may lead to a decrease in yield. While wafer bonding can achieve the connection of an entire wafer at a time, significantly improving batch efficiency, its strict requirements for matching the thermal expansion coefficients of materials, its high dependence on the overall quality of the wafer, the complexity of subsequent processing, and its lack of process flexibility make it also limited in the high-density application of Micro-LEDs. Summary of the Invention

[0004] In response to the existing problems, a waferless array module bonding method is proposed to improve production efficiency, reduce manufacturing costs, and improve product yield and adaptability.

[0005] The technical solution of the present invention is: a non-wafer array module bonding method, first cutting the Micro-LED chip wafer and the driver substrate wafer into multiple square wafers arranged in a matching number up and down according to the specified cutting lanes, each sliced wafer contains several single-chip chips, and then multiple Micro-LED chips are bonded to the corresponding driver substrate at the same time. After the bonding is completed, the bonded multi-chip module is divided into independent single chip units through a cleavage process for subsequent assembly and packaging.

[0006] Furthermore, the process parameters in the bonding are optimized using a multi-layer perceptron model.

[0007] Furthermore, the specific method for optimizing the multi-layer perceptron model is as follows: S1. Obtain process parameters related to bonding quality parameters and the value range of each process parameter, randomly combine process parameters within the corresponding value range, collect bonding quality results under different process parameter conditions through experiments, and generate a data set; S2. Construct a multi-layer perceptron model, learn the laws between the bonding process parameters and the bonding quality in the data set of step S1, and convert the relationship between the bonding process parameters and the bonding quality results into a mathematical model to provide data-driven guidance for optimizing the bonding process; S3. Combined with the data set in step S1, the multi-layer perceptron model constructed in step S2 is trained, and the weights and biases of the model are optimized using the stochastic steepest descent method, the adaptive moment estimation method, or the impulse algorithm; through iterative optimization, the loss function value of the model meets the preset requirements, and the loss function is used to characterize the error between the predicted value and the actual value of the process parameter; S4. According to the multi-layer perceptron model trained in step S3, an initial process parameter set is set, where the initial process parameter set includes the selected process parameter type and its value range; the process parameter set is input into the multi-layer perceptron model to obtain a corresponding real-time prediction result, and the error between the predicted value and the target value is calculated; S5. Based on the prediction error in step S4, the Bayesian optimization algorithm is used for iterative search to obtain the optimal process parameter combination in the process parameter set of step S4, and the combination is sent to the multi-layer perceptron model for error verification until the optimal process parameter combination that meets the requirements is found.

[0008] The waferless array module bonding method is specifically implemented in the following steps: 1) Cut the driver substrate wafer and Micro-LED wafer into a number of square pieces arranged in a matching pattern along the specified dicing lanes, where each square piece contains several single chips; 2) Using alignment marks, bond the square driver substrate cut in step 1) to the Micro-LED chip. The square bonding method is not limited to bump bonding and hybrid bonding. 3) Use a cutting wheel to cut the driver substrate according to the cutting knife of a single chip. This step only cuts the driver substrate and does not cut the Micro-LED chip; 4) Flip it over and use a laser to cut the sapphire substrate of the Micro-LED chip. The laser spot is placed at 1 / 3 and 2 / 3 of the sapphire thickness. This step only cracks the sapphire and does not crack the driver substrate. 5) The single chip is obtained after two cleavages for subsequent assembly and packaging.

[0009] Furthermore, in step 2), the bonding pressure range is 1-400 kN, the bonding temperature range is 20-300° C., and the bonding time range is 1-60 min.

[0010] The beneficial effects of the present invention are as follows: the non-wafer array module bonding method of the present invention greatly reduces the one-by-one operation steps of single-chip bonding by simultaneously performing batch bonding on multiple Micro-LED chips, thereby significantly improving the integration efficiency; batch bonding of multiple chips can reduce the impact of single-point defects on the overall yield, while avoiding interface problems caused by stress or uneven processing in wafer bonding, thereby improving integration reliability; combined with the subsequent scratching process, it optimizes the manufacturing process, reduces complex post-processing steps, reduces production costs, and provides economically feasible technical support for the large-scale application of Micro-LED; it can better meet the application requirements of Micro-LED in high-density and high-reliability scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a diagram of a multi-layer perceptron deep learning model in the waferless array module bonding method of the present invention; Figure 2 This is a process diagram of the waferless array module bonding method of the present invention. DETAILED DESCRIPTION

[0012] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0013] The present invention's waferless array module bonding method first slices the Micro-LED chip wafer and driver substrate wafer into multiple square wafers arranged in a matching number along specified dicing lines. Each slice contains several individual chips. Multiple Micro-LED chips are then bonded simultaneously to their corresponding driver substrates. After bonding, a cleaving process separates the bonded multi-chip module into individual chips for subsequent assembly and packaging. To further optimize the bonding process, the method incorporates a multilayer perceptron (MLP) model. By inputting key process parameters such as temperature, time, and pressure, the method predicts and adjusts the optimal bonding parameters in real time, ensuring a more accurate, stable, and efficient bonding process.

[0014] The waferless array module bonding method of the present invention comprises the following specific steps: Step 1: Collect a large amount of historical bonding process data, including input parameters such as temperature, pressure, and time, and corresponding bonding yield, brightness, and AOI (Automated Optical Inspection) results. This data is then cleaned, normalized, and feature-engineered to ensure data quality. A multilayer perceptron (MLP) is used to learn patterns in this historical process data and transform the relationship between bonding process parameters and bonding results into a mathematical model, providing data-driven guidance for process optimization.

[0015] The specific implementation steps are as follows: S1. Obtain process parameters related to bonding quality parameters and the value range of each process parameter, randomly combine process parameters within the corresponding value range, collect bonding quality results under different process parameter conditions through experiments, and generate a data set; S2. Construct a multi-layer perceptron model. By learning the rules between bonding process parameters and bonding quality in the data set of step S1, the relationship between bonding process parameters and bonding quality results is converted into a mathematical model to provide data-driven guidance for optimizing the bonding process, such as Figure 1 The multilayer perceptron deep learning model shown in the figure; 1) The multilayer perceptron consists of an input layer, a hidden layer, and an output layer; 2) The process parameters input into the input layer include but are not limited to bonding temperature, bonding pressure, bonding time, and other parameters; 3) The hidden layer contains several nodes, each representing a neuron, connected to the nodes of the input layer or the previous layer through weights and bias parameters, and the weights and bias parameters are continuously optimized and adjusted through training; 4) The output layer includes but is not limited to result parameters such as bonding yield, brightness, and AOI; S3. Based on the data set in step S1, the multilayer perceptron model constructed in step S2 is trained, and the weights and biases of the model are optimized using a stochastic steepest descent method, an adaptive moment estimation method, or an impulse algorithm; through iterative optimization, the loss function value of the model meets the preset requirements, and the loss function is used to characterize the error between the predicted value and the actual value of the process parameter; S4. According to the multi-layer perceptron model constructed in step S3, an initial process parameter set is set, where the initial process parameter set includes the selected process parameter type and its value range; the process parameter set is input into the multi-layer perceptron model to obtain a corresponding real-time prediction result, and the error between the predicted value and the target value is calculated; S5, based on the prediction error in step S4, adopt the Bayesian optimization algorithm to iteratively search, obtain the optimal process parameter combination in the process parameter set of step S4, and send it to the multi-layer perceptron model for error verification until the optimal process parameter combination that meets the requirements is searched; 1) Substitute the multiple errors corresponding to the multiple sets of initial process parameter sets obtained by calculation into the proxy function, and calculate the updated prior distribution of the initial sample or the previous sample; 2) Set the acquisition function, and optimize a new set of process parameter sets according to the prior distribution so that the acquisition function balances the proportion of exploration and utilization; 3) Input the new process parameters into the multi-layer perceptron model to predict the new quality parameters, and then judge whether the design requirements are met based on the target values of the quality parameters; if so, output the current process parameter combination as the optimal solution; if not, return to step S4 to update the initial process parameter set, repeat steps S4 and S5 until the error between the real-time prediction value and the target value meets the design requirements, and obtain the optimal solution process parameter combination.

[0016] A multi-layer perceptron deep learning model, in which the input layer includes but is not limited to parameters such as bonding temperature, bonding pressure, and bonding time; the hidden layer contains several nodes, each node represents a neuron, and is connected to the nodes of the input layer or the previous layer through weights and bias parameters; the output layer includes but is not limited to result parameters such as bonding yield, brightness, and AOI.

[0017] Use the model to predict bonding yield and provide parameter optimization suggestions, and continuously update the model with new data to adapt to the latest process conditions.

[0018] The new process steps generally involve slicing the wafer into squares containing several connected dies, which are then bonded together. After bonding, the dies are then cleaved into individual pieces. The idea behind cleaving is that after bonding, the silicon wafer and LEDs are a single unit. A cleaver wheel is used to cleave the wafer from the back, and then the LEDs are flipped over and cleaved with a laser, ultimately achieving a complete cleavage.

[0019] like Figure 2 The process diagram of the waferless array module bonding method is shown in the figure. The specific steps are as follows: Step 2: Cut the Micro-LED chip wafer and driver substrate wafer into multiple square slices arranged in a matching pattern along the specified dicing lanes. Each square slice contains several individual chips. You can choose to cut them in 2x2, 4x4, or 2x3 patterns, as long as the chips are connected and aligned with the driver substrate.

[0020] Step 3: Use alignment marks to bond the square driver substrate cut in Step 2 to the Micro-LED chip. The square bonding method is not limited to bump bonding and hybrid bonding. The above "multi-chip batch" does not refer to the arrangement of single chips. All single chips are integrated before bonding, so only one set of corresponding marking points is required. The bonding process includes parameter data such as pressure, temperature and time.

[0021] The bonding pressure range is 1-400kN, the bonding temperature range is 20-300℃, and the bonding time range is 1-60min.

[0022] Step 4: Use a cutting wheel to cut the driver substrate according to the cutting knife of a single chip. This step only cuts the driver substrate and does not cut the Micro-LED chip.

[0023] Step 5: Flip the sapphire substrate over again and use a laser to cut the Micro-LED chip substrate (corresponding to the cutting path of the driver substrate). The laser spot is placed at 1 / 3 and 2 / 3 of the sapphire thickness. This step only cracks the sapphire and does not crack the driver substrate.

[0024] Step 6: The single chip is obtained by two cleavages for subsequent assembly and packaging.

[0025] The above-described embodiments merely represent specific implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent. It should be noted that a person skilled in the art would be able to make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present invention. Therefore, the scope of protection of the patent for this invention shall be determined by the appended claims.

Claims

1. A non-wafer array module bonding method, characterized in that: First, the Micro-LED chip wafer and the driver substrate wafer are cut into multiple square wafers with matching numbers arranged in the upper and lower directions according to the specified cutting lanes. Each slice contains several single chips. Then, multiple Micro-LED chips are bonded to the corresponding driver substrate at the same time. After bonding is completed, the bonded multi-chip module is divided into independent single chip units through a cleavage process for subsequent assembly and packaging.

2. The waferless array module bonding method according to claim 1, wherein: The process parameters in the bonding are optimized using a multi-layer perceptron model.

3. The waferless array module bonding method according to claim 2, wherein: The specific method for optimizing the multi-layer perceptron model is as follows: S1. Obtain process parameters related to bonding quality parameters and the value range of each process parameter, randomly combine process parameters within the corresponding value range, collect bonding quality results under different process parameter conditions through experiments, and generate a data set; S2. Construct a multi-layer perceptron model, learn the patterns between bonding process parameters and bonding quality in the data set of step S1, and convert the relationship between bonding process parameters and bonding quality results into a mathematical model to provide data-driven guidance for optimizing the bonding process; S3. Combined with the data set in step S1, train the multi-layer perceptron model constructed in step S2, and optimize the weights and bias of the model using a stochastic steepest descent method, an adaptive moment estimation method, or an impulse algorithm; Through iterative optimization, the loss function value of the model meets the preset requirements. The loss function is used to characterize the error between the predicted value and the actual value of the process parameter; S4. According to the multi-layer perceptron model trained in step S3, an initial process parameter set is set, where the initial process parameter set includes the selected process parameter type and its value range; the process parameter set is input into the multi-layer perceptron model to obtain a corresponding real-time prediction result, and the error between the predicted value and the target value is calculated; S5. Based on the prediction error in step S4, the Bayesian optimization algorithm is used for iterative search to obtain the optimal process parameter combination in the process parameter set of step S4, and the combination is sent to the multi-layer perceptron model for error verification until the optimal process parameter combination that meets the requirements is found.

4. The waferless array module bonding method according to claim 1 or 2, wherein: The specific implementation steps are as follows: 1) Cut the driver substrate wafer and Micro-LED wafer into a number of square pieces arranged in a matching pattern along the specified dicing lanes, where each square piece contains several single chips; 2) Using alignment marks, bond the square driver substrate cut in step 1) to the Micro-LED chip. The square bonding method is not limited to bump bonding and hybrid bonding. 3) Use a cutting wheel to cut the driver substrate according to the cutting knife of a single chip. This step only cuts the driver substrate and does not cut the Micro-LED chip; 4) Flip it over and use a laser to cut the sapphire substrate of the Micro-LED chip. The laser spot is placed at 1 / 3 and 2 / 3 of the sapphire thickness. This step only cracks the sapphire and does not crack the driver substrate. 5) The single chip is obtained after two cleavages for subsequent assembly and packaging.

5. The waferless array module bonding method according to claim 4, wherein: In step 2), the bonding pressure range is 1-400 kN, the bonding temperature range is 20-300° C., and the bonding time range is 1-60 min.

Images