Apparatus and method for wafer-to-wafer bonding

Abstract

A method includes obtaining a first wafer bonding recipe and a wafer bonding process model, the model including an input indicative of physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based on the physical parameters of the first wafer; obtaining measurements of the first wafer to obtain the physical parameters of the first wafer; the model generating the first wafer bonding recipe from the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first post-bond wafer.

Description

[0001] Related applications

[0002] This application claims priority to the following non-provisional application: U.S. Patent Application Serial No. 16 / 924,847, filed July 9, 2020, entitled “Apparatus and Methods for Wafer Bonding,” which is incorporated herein by reference in its entirety. Technical Field

[0003] The present invention relates generally to substrate processing, and in specific embodiments to apparatus and methods for wafer-to-wafer bonding. Background Technology

[0004] Wafer-to-wafer bonding is a packaging technology used in the production of microelectromechanical systems (MEMS), nanoelectromechanical systems (NEMS), microelectronic devices, and optoelectronic devices. Fused bonding (often also called direct bonding) is a wafer-to-wafer bonding process that does not require any additional intermediate layers. In fused bonding, two wafers (e.g., a top wafer and a bottom wafer) are placed together, and as the wafer surfaces begin to contact, the two wafers begin to bond, forming a bonded wafer. Annealing the bonded wafer at an elevated temperature increases the bond strength between the two wafers and forms a fused-bonded wafer.

[0005] Figure 1A This demonstrates a portion of a wafer-to-wafer bonding process 100 that uses a conventional fused bonding process to bond two wafers. In fused bonding, a top wafer 105 and a bottom wafer 110 are bonded to form a bonded wafer 115. Fused bonding also includes annealing the bonded wafer to strengthen the bond between the two wafers, which... Figure 1A Not shown in the image.

[0006] The quality of wafer-to-wafer bonding can depend on a variety of factors, which can be divided into two distinct categories: wafer characteristics and process conditions. Examples of wafer characteristics include wafer flatness, wafer smoothness, wafer cleanliness, and wafer materials, while examples of process conditions include bonding temperature, environmental conditions in the bonding chamber where wafer-to-wafer bonding is performed, and the applied forces.

[0007] Figure 1B A side view of wafer 115 after bonding is shown. (As shown) Figure 1B As shown, the top wafer 105 and the bottom wafer 110 are well bonded together and there is a smooth interface between the two wafers. Figure 1C A side view of wafer 130 after bonding is shown, highlighting the poor bonding between the two wafers. (See image.) Figure 1CAs shown, the top wafer 105 has a concave profile, which results in a gap 135 when the top wafer 105 and the bottom wafer 110 are bonded. Summary of the Invention

[0008] According to an embodiment of the present invention, a method is provided. The method includes: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicating physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based on the physical parameters of the first wafer; obtaining measurement results of the first wafer to obtain the physical parameters of the first wafer; generating the first wafer bonding recipe based on the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first bonded wafer.

[0009] According to another embodiment, a method is provided. The method includes: obtaining a wafer bonding process model, the model including inputs indicating physical parameters of a first bonded wafer and configured to output a wafer bonding recipe based on the physical parameters of the first bonded wafer; bonding a first wafer to a second wafer according to the wafer bonding recipe to form the first bonded wafer; obtaining measurements of the first bonded wafer to obtain physical parameters of the first bonded wafer; generating the first wafer bonding recipe based on the physical parameters of the first bonded wafer in the model; and bonding a third wafer to a fourth wafer according to the first wafer bonding recipe to form a second bonded wafer.

[0010] According to another embodiment of the present invention, a processing system is provided. The processing system includes: a non-transitory computer-readable storage medium including instructions that, when executed, cause a processor of a computing device to perform operations according to a semiconductor wafer fabrication process. The instructions include: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicating physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based on the physical parameters of the first wafer; obtaining measurement results of the first wafer to obtain the physical parameters of the first wafer; generating the first wafer bonding recipe based on the physical parameters of the first wafer in the model; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first bonded wafer. Attached Figure Description

[0011] To gain a more complete understanding of the invention and its advantages, reference will now be made to the following description, taken in conjunction with the accompanying drawings, in which:

[0012] Figure 1A This demonstrates a portion of a wafer-to-wafer bonding process that uses conventional fused bonding to bond two wafers.

[0013] Figures 1B to 1C A side view of the bonded wafer is shown;

[0014] Figure 2 illustrates the traditional wafer fusion bonding process;

[0015] Figure 3 A wafer bonding system for adjusting wafer bonding formulations using wafer metrology data, according to an example embodiment presented herein, is demonstrated.

[0016] Figure 4 An example wafer fabrication tool is shown according to the example embodiments presented herein;

[0017] Figure 5 A wafer fusion bonding process with example feedback operation is demonstrated according to the example embodiments presented herein;

[0018] Figure 6 A wafer fusion bonding process with a first example feedforward operation is illustrated according to an example embodiment presented herein;

[0019] Figure 7 A wafer fusion bonding process with a second example feedforward operation is demonstrated according to an example embodiment presented herein;

[0020] Figure 8 A flowchart illustrating example operations occurring in a wafer bonding process with feedforward optimization of wafer bonding formulation, according to example embodiments presented herein;

[0021] Figure 9A A flowchart illustrating a first example process for creating a wafer bonding process model using a semi-empirical modeling-enhanced finite element modeling (FEM) technique, according to the example embodiments presented herein;

[0022] Figure 9B A flowchart illustrating a second example process for creating a wafer bonding process model using semi-empirical modeling and calibration-enhanced FEM techniques, according to the example embodiments presented herein;

[0023] Figure 10 A flowchart illustrating example operations that occur when creating a wafer bonding process model using a fingerprinting function according to the example embodiments presented herein;

[0024] Figure 11 A flowchart illustrating the operations that occur in the generation of a wafer bonding recipe with feedforward optimization for the wafer bonding recipe, according to the example embodiments presented herein;

[0025] Figure 12 A flowchart illustrating example operations occurring in a wafer bonding process with feedback optimization of wafer bonding formulation, according to example embodiments presented herein;

[0026] Figure 13 A flowchart illustrating an example process for creating a wafer bonding process model using feedback data according to the example embodiments presented herein; and

[0027] Figure 14 A flowchart illustrating the operations that occur in the generation of a wafer bonding recipe with feedback optimization for the wafer bonding recipe, according to the example embodiments presented herein.

[0028] Unless otherwise indicated, corresponding numbers and symbols in different figures generally refer to corresponding parts. The figures are drawn to clearly illustrate relevant aspects of the embodiments, and these figures are not necessarily drawn to scale. The edges of features drawn in the figures do not necessarily indicate the termination of the feature's extent. Detailed Implementation

[0029] The manufacture and use of various embodiments are discussed in detail below. However, it should be understood that the various embodiments described herein can be applied to a wide variety of specific situations. The specific embodiments discussed are merely illustrative of specific ways of making and using the various embodiments and should not be interpreted in a limited manner.

[0030] The various techniques described herein involve using wafer metrology data to dynamically control wafer bonding formulations to control post-bonding wafer deformation in wafer melt bonding. As an example, wafer metrology data is provided to a wafer bonding process model to determine process conditions for the wafer bonding process, which are predicted to produce bonded wafers that meet post-bonding deformation thresholds. These process conditions can then be applied to the actual bonding of the wafers. In other words, wafer metrology data is used to adjust the wafer bonding formulation to produce bonded wafers that meet post-bonding deformation thresholds.

[0031] Figure 2 illustrates a conventional wafer fusion bonding process 200. As shown in Figure 2, the wafer processing tool 205 uses a process record (POR) wafer bonding recipe to fusion bond a first top wafer 210 and a first bottom wafer 220 to form a first bonded wafer 230. Typically, multiple top wafers are bonded to multiple bottom wafers to form multiple bonded wafers. As an example, a second top wafer 212 is bonded to a second bottom wafer 222 to form a second bonded wafer 232, and an Nth top wafer 214 is bonded to an Nth bottom wafer 224 to form an Nth bonded wafer 234. The annealing stage of the conventional wafer fusion bonding process 200 is not shown in Figure 2.

[0032] In traditional wafer fusion bonding processes, the same POR wafer bonding formulation is used for all top and bottom wafers, regardless of whether the top and bottom wafers belong to the same corresponding wafer batch.

[0033] Wafers are typically processed in batches. For example, a batch might contain up to 25 wafers, one of which is a monitoring wafer that may or may not be used in the bonding process. Batch-to-batch wafer variation is a significant factor contributing to wafer deformation after bonding. Produced wafers often vary from batch to batch, which can be due to variations in process control from previous steps, inherent wafer deformation, or other less obvious factors affecting processing conditions. Variations between wafers are called shape variations. Shape variations can be localized, meaning the average surface profile of one batch of wafers looks completely different from that of another batch. In a 300mm wafer with logic devices, a standard deviation of shape variation ranges from 1 μm to 5 μm. Shape variations in wafers with memory devices can differ. Shape variations can also be global, meaning the peak-to-valence variation between batches can be on the order of 10 μm or larger. For memory applications, shape variations can be much larger. Intra-batch variation is typically small, except in cases of batch splitting. Batch splitting means that a subset of wafers within a batch is separated and processed differently. When not considering split batches, intra-batch variation is typically much smaller than 10% of inter-batch variation, but the actual value may vary depending on the wafer being considered.

[0034] Shape variation is directly related to in-plane deformation (or simply "deformation"). In-plane deformation of a wafer is defined as deviation along the principal plane of the wafer. As an example, high points (e.g., peaks) and low points (e.g., valleys) on the wafer surface are examples of in-plane deformation. Observations show that a 1 μm shape variation can be attributed to approximately 5 nm of in-plane deformation. Therefore, in-plane deformation is a key indicator of wafer bonding process performance.

[0035] One way to evaluate the quality of a wafer bonding process is to inspect the bonded wafers. For example, if the shape variation of the bonded wafer is large, the wafer bonding process may not be well-suited for the bonded wafers. Conversely, if the shape variation is small, the wafer bonding process can be well-suited for the bonded wafers. As another example, if the shape variation of the bonded wafer is low within a batch but high between batches, then the wafer bonding process can be well-suited for a specific batch of wafers but not for different batches.

[0036] When both wafers being bonded are patterned (i.e., both wafers contain devices and / or structures), alignment between the patterns on the two substrates along the bonding interface is critical. Alignment marks on each patterned layer can be measured using a scanner during a post-bonding step. This scanner measurement step does not necessarily have to be performed immediately after the bonding step. As an example, the post-bonded wafers can be annealed in a processing chamber at elevated temperatures (e.g., several hundred degrees Celsius) to improve the bond strength between the two layers. As another example, one of the wafers can be thinned by grinding to expose the bonding interface. Other, less obvious steps may also exist between the wafer bonding step and the post-bonding metrology step. Regardless, the absolute offset between the two wafers in the post-bonded wafer is critical. For example, a typical post-bonding alignment error under a one standard deviation variation might be approximately 30 nm.

[0037] When only one wafer is patterned, the post-bonding deformation of the patterned wafer is important. Post-bonding deformation can vary depending on the shapes of the two pre-bonding wafers and the wafer bonding process. Some important factors determining post-bonding deformation include wafer temperature distribution, clamping forces on the wafers, adhesion forces between the wafers, surface roughness of the bonding surfaces of the two wafers, chuck flatness, spacing between the wafers before the bonding process, pre-cleaning of the two target bonding surfaces, and many other factors not listed here. Post-bonding deformation can also be measured using a scanner after bonding, similar to the case where both wafers are patterned. However, in this case, post-bonding deformation is defined as the absolute deviation from the ideal deformation. The ideal deformation can be defined by the scanner. Scanners can employ various models to define the ideal deformation. Traditionally, scanners use a 6-term model for the ideal deformation. These 6 terms include magnification and tilting of the deformed surface map. Recently, models have become more complex; for example, 33-term models have been proposed. Some state-of-the-art technologies support free-form models (thousands of terms). The challenge of choosing one of the highly complex models is that the time and cost required to create such a model become impractical in a production environment.

[0038] Furthermore, post-bonding deformation is becoming an increasingly difficult production target to meet. For example, the current state-of-the-art post-bonding deformation threshold has a standard deviation of 10 nm or less, while customers require post-bonding deformation to continue to decrease in both the short and long term.

[0039] Because the conventional wafer bonding process shown in Figure 2 uses a single POR wafer bonding formulation to bond wafers (regardless of whether the wafers being bonded are within or between batches), bonding process-related factors affecting post-bonding deformation (i.e., factors controlled by the POR wafer bonding formulation, such as wafer temperature distribution, wafer clamping force, chuck flatness, wafer spacing before bonding, and pre-cleaning of the two target bonding surfaces) remain constant, while wafer shape variations differ between individual wafers and between wafer batches. Therefore, post-bonding deformation control between batches can be poor because the shape variations between batches can be significant, and a fixed wafer bonding formulation cannot compensate for these inter-batch shape variations.

[0040] According to one example embodiment, a wafer bonding system is provided that utilizes wafer metrology data to adjust the wafer bonding formulation. The wafer metrology data is provided to a wafer bonding process model, and the model adjusts the process conditions of the wafer bonding formulation to produce a bonded wafer that meets a post-bonding deformation threshold.

[0041] While directly measuring post-bonding deformation on the wafer is important, performing this measurement is costly for R&D purposes. For example, performing all the aforementioned bonding steps prior to the metrology step can be extremely time-consuming, potentially on the order of weeks. In embodiments, instead of direct post-bonding deformation measurement, physical parameters of the wafer, such as wafer shape data, can be measured against the bonded wafer and used as a representative estimate of the post-bonding deformation. For example, the wafer shape data can be in the form of a high-density mesh map of a freestanding wafer topology. The mesh can have, for example, a lateral resolution of 0.5 mm and can be measured across the entire wafer. For example, the resolution of wafer shape metrology in out-of-plane directions should be less than 1 nm, which can be defined as the deviation along a plane perpendicular to the wafer's principal plane. The wafer shape data can be directly correlated with the deformation using geometric formulas. Examples of geometric formulas include those that correlate wafer shape data with deformation based on elasticity theory, beam theory, or free-form wafer shape measurement.

[0042] In this embodiment, the wafer metrology data includes pre-bonding wafer metrology data (e.g., top wafer, bottom wafer, or top and bottom wafer metrology data), and this data is provided to a model to adjust the process conditions of the wafer bonding formulation to produce a bonded wafer that meets a post-bonding deformation threshold. In other words, given the pre-bonding wafer metrology data, the model is used to adjust the process conditions of the wafer bonding formulation to produce a bonded wafer that meets a post-bonding deformation threshold. The model predicts the post-bonding deformation of the wafer based on the pre-bonding wafer metrology data and adjusts the wafer bonding formulation to produce a bonded wafer that meets the post-bonding deformation threshold. This mode of operation is called feedforward operation. A detailed description of feedforward operation is provided below.

[0043] In this embodiment, the wafer metrology data includes metrology data of the bonded wafer, and this data is provided to a model to adjust the process conditions of the wafer bonding formulation to produce a bonded wafer that meets the post-bonding deformation threshold. In other words, the model is used to adjust the process conditions of the wafer bonding formulation based on the metrology data of the bonded wafer to produce a bonded wafer that meets the post-bonding deformation threshold. This mode of operation is called feedback operation. A detailed description of feedback operation is provided below.

[0044] In this embodiment, the wafer metrology data includes metrology data for both the pre-bonding and post-bonding wafers. This metrology data is provided to a model to adjust the process conditions of the wafer bonding formulation to produce a post-bonded wafer that meets the post-bonding deformation threshold. Since metrology data for both the pre-bonding and post-bonding wafers is available, the model can be used to predict the post-bonding deformation of the post-bonding wafer based on the pre-bonding wafer metrology data, and to adjust the process conditions of the wafer bonding formulation based on the metrology data of both the pre-bonding and post-bonding wafers to produce a post-bonding wafer that meets the post-bonding deformation threshold.

[0045] Figure 3 A wafer bonding system 300 is illustrated, which utilizes wafer metrology data to adjust the wafer bonding formulation according to the example embodiments presented herein. In the embodiments, the wafer bonding system 300 uses wafer metrology data of the pre-bonding wafer (top wafer, bottom wafer, or both top and bottom wafers) and / or the post-bonding wafer to adjust the process conditions of the wafer bonding formulation to produce a post-bonded wafer that meets the post-bonding deformation threshold.

[0046] The wafer bonding system 300 includes a wafer processing tool 305 that bonds pre-bonded wafers (i.e., top wafer 310 and bottom wafer 315) to produce a post-bonded wafer 320. The wafer processing tool 305 is widely available in the market. The wafer bonding system 300 includes a shape measurement tool 325 that measures physical parameters, such as wafer shape data, of the top wafer 310, the bottom wafer 315, or both the top and bottom wafers 310. An example of the shape measurement tool 325 is a surface profilometer. Another example of the shape measurement tool 325 is a photolithography surface scanner, such as an atomic force microscope tool or a critical size scanning electron microscope. These physical parameters can be provided to the wafer processing tool 305 to adjust the wafer bonding recipe used to bond the pre-bonded wafers so that the resulting post-bonded wafer 320 meets the post-bonding deformation parameters.

[0047] like Figure 3As shown, the wafer bonding system 300 is configured to bond a pre-bonding wafer in which the top wafer 310 is a patterned wafer and the bottom wafer 315 is a carrier wafer. Therefore, it may not be necessary to measure the physical parameters of the bottom wafer 315. Thus, the line between the bottom wafer 315 and the shape measurement tool 325 is shown as a dashed line. However, in different configurations, even when the bottom wafer 315 is a carrier wafer, the physical parameters of the bottom wafer 315 can be obtained and used to adjust the wafer bonding formulation. Furthermore, when the top wafer 310 is a carrier wafer and the bottom wafer 315 is a patterned wafer, the physical parameters of the bottom wafer 315 will be measured and provided to the wafer processing tool 305 to adjust the wafer bonding formulation, while measuring the physical parameters of the top wafer 310 may be optional. When both the top wafer 310 and the bottom wafer 315 are patterned, the physical parameters of both wafers should be measured and provided to the wafer processing tool 305.

[0048] The wafer bonding system 300 also includes a scanner 330 that scans the bonded wafer 320 to determine deformation data (e.g., post-bonding deformation data) of the bonded wafer 320. This deformation data can be provided to the wafer processing tool 305 to adjust the wafer bonding recipe, thereby helping to ensure that the bonded wafer meets the post-bonding deformation parameters. In some embodiments, the scanner 330 also functions as a shape measurement tool 325 to measure the physical parameters of the wafer before bonding.

[0049] In this embodiment, the physical parameters of the bonded wafer 320 are provided to the wafer processing tool 305 to adjust the wafer bonding formula used to bond the wafer before bonding. Because the physical parameters of the bonded wafer 320 must be measured in this particular embodiment, but not in all embodiments, the line between the bonded wafer 320 and the shape measurement tool 325 is shown as a dashed line.

[0050] Figure 4 An example wafer processing tool 400 according to an exemplary embodiment presented herein is shown. For example, the wafer processing tool 400 may be... Figure 3 An example implementation of the wafer processing tool 305.

[0051] The wafer processing tool 400 includes a processing chamber 405 in which the bonding of a top wafer and a bottom wafer is performed. The processing chamber 405 provides a controlled environment, such as temperature, atmospheric pressure, plasma power, etc. The processing chamber 405 includes a wafer holder 410. In an embodiment, the wafer holder 410 includes a top chuck for holding the top wafer and a bottom chuck for holding the bottom wafer. The top wafer 415 and the bottom wafer 420 can be conveyed into the processing chamber 405 via a top transfer mechanism 417 and a bottom transfer mechanism 422, respectively.

[0052] In this embodiment, as the top and bottom wafers are conveyed to the processing chamber 405, a shape measurement tool 325 measures these wafers to determine their physical parameters. The processor 425 adjusts the process conditions of the wafer bonding formulation based on these physical parameters to produce bonded wafers that meet the post-bonding deformation threshold. A detailed description of the adjustment of the process conditions is provided below.

[0053] In this embodiment, before the top and bottom wafers are loaded into the wafer processing tool 400, they are pre-measured by a shape measurement tool 325 or some other shape measurement tool, and the physical parameters of the wafers are stored in a database, such as memory 430. Then, when the top and bottom wafers enter the processing chamber 405, the processor 425 retrieves the physical parameters of the wafers and adjusts the process conditions of the wafer bonding formula according to the physical parameters to produce a bonded wafer that meets the post-bonding deformation threshold. A detailed description of the adjustment of the process conditions is provided below.

[0054] The transfer mechanism 437 removes the bonded wafer 435 from the processing chamber 405. The scanner 330 scans the bonded wafer 435 to determine deformation data (e.g., post-bonding deformation data). The deformation data may also be stored in a database stored in the memory 430.

[0055] Figure 5 A wafer fusion bonding process 500 with example feedback operation is illustrated according to an example embodiment presented herein. Figure 5 As shown, the wafer processing tool 505 uses a POR wafer bonding formulation to fused and bond a first top wafer 510 and a first bottom wafer 515 to form a first bonded wafer 520. Deformation data of the first bonded wafer 520 (e.g., measured by a scanner) is provided to the wafer processing tool 505 to optimize the wafer bonding formulation (in... Figure 5 (Illustrated as "OPT formulation"). Subsequent top wafers (e.g., second top wafer 512 and Nth top wafer 514) are bonded to subsequent bottom wafers (e.g., second bottom wafer 517 and Nth bottom wafer 519) according to the optimized wafer bonding formulation to produce subsequent bonded wafers (e.g., second bonded wafer 522 and Nth bonded wafer 524). In cases where subsequent wafers are part of the same batch as the corresponding first wafer, the wafer melt bonding process 500 with feedback operation may be able to produce bonded wafers using a wafer bonding formulation optimized for relatively small variations occurring within the same batch of wafers.

[0056] In an embodiment, deformation data of a subsequently bonded wafer (e.g., a second bonded wafer 522) can be measured and provided to a wafer processing tool 505 to further optimize the wafer bonding formula. As an example, deformation data of the second bonded wafer 522 is measured and provided to the wafer processing tool 505 to optimize the wafer bonding formula for bonding a third bonded wafer. Then, deformation data of the third bonded wafer is measured and provided to the wafer processing tool 505 to optimize the wafer bonding formula for bonding a fourth bonded wafer, and so on.

[0057] In one embodiment, when multiple wafer fabrication tools are used to perform fused bonding, the deformation data of the wafer 520 after the first bonding is provided to these multiple wafer fabrication tools to optimize the wafer bonding formula respectively. In another embodiment, when multiple wafer fabrication tools are used to perform fused bonding, the optimized wafer bonding formula is shared among these multiple wafer fabrication tools.

[0058] Figure 6 A wafer fusion bonding process 600 with a first example feedforward operation is illustrated according to an example embodiment presented herein. Figure 6 As shown, the wafer processing tool 605 uses the physical parameters of the first top wafer 610 and the first bottom wafer 615 to optimize the wafer bonding formula (in... Figure 6 (Illustrated as "OPT recipe"). The wafer processing tool 605 uses an optimized wafer bonding recipe to bond a first top wafer 610 and a first bottom wafer 615 to form a first bonded wafer 620. Physical parameters of the first top wafer 610 and the first bottom wafer 615 can be measured, and these physical parameters can then be used to optimize the wafer bonding recipe to maximize the time available for optimization. As an example, physical parameter measurements are performed when wafers are received at the wafer manufacturing facility (potentially a significant amount of time before they are actually bonded), and the measurement results are stored in a database. As another example, physical parameter measurements are performed when wafers are loaded into the wafer bonding tool, and the measurement results are stored in a database. In an embodiment, subsequent top wafers (e.g., second top wafer 612 and Nth top wafer 614) are bonded to subsequent bottom wafers (e.g., second bottom wafer 617 and Nth bottom wafer 619) according to an optimized wafer bonding formulation to produce subsequent bonded wafers (e.g., second bonded wafer 622 and Nth bonded wafer 624). The advantage of the wafer fusion bonding process 600 is that (within a wafer batch) all wafers are bonded using a wafer bonding formulation optimized for relatively minor variations occurring within the same batch of wafers.

[0059] In embodiments, when one of the two pre-bonding wafers is patterned (e.g., the top wafer is patterned and the bottom wafer is a carrier wafer, or the bottom wafer is patterned and the top wafer is a carrier wafer), the wafer bonding formulation is optimized using only the physical parameters of the patterned wafer. As an example, when the top wafer is patterned and the bottom wafer is a carrier wafer, the wafer processing tool 605 uses the physical parameters of the first top wafer 610 to optimize the wafer bonding formulation for bonding the first top wafer 610 and the first bottom wafer 615. As another example, when the bottom wafer is patterned and the top wafer is a carrier wafer, the wafer processing tool 605 uses the physical parameters of the first bottom wafer 615 to optimize the wafer bonding formulation for bonding the first top wafer 610 and the first bottom wafer 615.

[0060] In one embodiment, when multiple wafer fabrication tools are used for fused bonding, the physical parameters of the first top wafer 610 and the first bottom wafer 615 are provided to these multiple wafer fabrication tools to optimize the wafer bonding formula respectively. In another embodiment, when multiple wafer fabrication tools are used to perform fused bonding, the optimized wafer bonding formula is shared among these multiple wafer fabrication tools.

[0061] Figure 7 A wafer fusion bonding process 700 with a second example feedforward operation is illustrated according to an example embodiment presented herein. Figure 7 As shown, the wafer processing tool 705 uses the physical parameters of the first top wafer 710 and the first bottom wafer 715 to optimize the wafer bonding formula. Figure 7(Displayed as "OPT Recipe 1"). The wafer processing tool 705 uses an optimized wafer bonding recipe to bond a first top wafer 710 and a first bottom wafer 715 to form a first bonded wafer 720. Physical parameters of the first top wafer 710 and the first bottom wafer 715 can be measured, and these physical parameters can then be used to optimize the wafer bonding recipe to maximize the time available for optimization. As an example, physical parameter measurements are performed when wafers are received at the wafer manufacturing facility (potentially a significant amount of time before they are actually bonded), and the measurement results are stored in a database. As another example, physical parameter measurements are performed when wafers are loaded into the wafer bonding tool, and the measurement results are stored in a database. The physical parameters of subsequent top and bottom wafers are used to optimize the wafer bonding recipe for bonding subsequent bonded wafers. As an example, the physical parameters of a second top wafer 712 and a second bottom wafer 717 are used to optimize the wafer bonding recipe for bonding a second bonded wafer 722. Similarly, the physical parameters of the Nth top wafer 714 and the Nth bottom wafer 719 are used to optimize the wafer bonding formula for bonding the Nth bonded wafer 724. The advantage of the wafer fusion bonding process 700 is that it optimizes the wafer bonding formula individually for each bonded wafer, regardless of the wafer batch.

[0062] In embodiments where one of the two pre-bonding wafers is patterned (e.g., the top wafer is patterned and the bottom wafer is a carrier wafer), the wafer bonding formulation is optimized using only the physical parameters of the patterned wafer. As an example, the wafer processing tool 705 uses the physical parameters of the first top wafer 710 to optimize the wafer bonding formulation for bonding the first top wafer 710 and the first bottom wafer 715.

[0063] In this embodiment, a combination of feedforward and feedback optimization of the wafer bonding formulation is used. In such an embodiment, both the physical parameters of the wafer before bonding and the deformation data of the wafer after bonding are used to optimize the wafer bonding formulation.

[0064] Figure 8 A flowchart illustrating an example operation 800 occurring in a feedforward optimized wafer bonding process with a wafer bonding formulation, according to an example embodiment presented herein, is shown. Operation 800 may indicate operations occurring in a wafer processing tool when the tool uses a feedforward optimized wafer bonding process with a wafer bonding formulation to bond wafers. Operation 800 may describe wafer fusion bonding processes 600 and 700.

[0065] Operation 800 begins with the creation of a wafer bonding process model (box 805). The wafer bonding process model is a mathematical model of the wafer bonding process that considers the physical parameters of the wafer before bonding and the process conditions of the wafer bonding process, and correlates them with the simulated deformation of the wafer after bonding. The wafer bonding process model can utilize any combination of finite element analysis, linear regression, random forest algorithms, genetic programming algorithms, pattern search algorithms, neural network algorithms, deep learning algorithms, etc., to correlate the simulated deformation of the wafer after bonding with the physical parameters of the wafer before bonding and the process conditions of the wafer bonding process. In one embodiment, the model is created a priori by the wafer fabrication tool and stored in memory. In another embodiment, the model is created by a tool that does not directly participate in wafer bonding and is then provided to the wafer fabrication tool. Several approaches that can be used to create this model are discussed in detail below.

[0066] The process conditions for wafer bonding can include pre-cleaning time, pre-cleaning solution chemical composition, upper wafer chuck temperature, lower wafer chuck temperature, plasma pretreatment time, plasma pretreatment power, wafer idle time, bonding time, bonding initiation force, chamber atmospheric pressure, chamber average temperature, chamber humidity, chamber atmospheric composition, bottom wafer chuck vacuum pressure, bottom wafer chuck vacuum zone, bottom wafer chuck vacuum on / off time, top wafer chuck vacuum pressure, top wafer chuck vacuum zone, top wafer chuck vacuum on / off time, top wafer chuck height variation, bottom wafer chuck height variation, bonding gap, wafer temperature before clamping, chuck height, etc. Wafer bonding process models can consider any or all of the process conditions listed above.

[0067] The wafer processing tool measures the incoming top wafer (box 807). The wafer processing tool can use any of a variety of metrology tools (such as a surface profilometer) or a scanner to measure the top wafer. The measurement of the top wafer provides its physical parameters (such as wafer shape data). The wafer processing tool may optionally measure the incoming bottom wafer (box 809). The measurement of the incoming bottom wafer can follow a similar process to the measurement of the top wafer described above. In the case where the incoming bottom wafer is a carrier wafer (i.e., the bottom wafer is not patterned), the influence of the bottom wafer's physical parameters on the post-bonding wafer deformation may be negligible compared to the influence of the patterned wafer (e.g., the incoming top wafer). Therefore, it may not be necessary to measure the bottom wafer. However, if both wafers are patterned, the incoming bottom wafer should be measured. The deformation of the incoming wafer is also obtained from the measurement results (box 811). For example, the wafer deformation can be obtained using the same techniques used during the creation of the wafer bonding process model. For example, the same technique described in detail below for creating wafer bonding process models can also be used to obtain the wafer fingerprint coefficient.

[0068] A wafer bonding recipe (box 813) is generated for bonding the incoming top wafer to the incoming bottom wafer. The wafer bonding recipe is generated based on the physical parameters of the incoming top and bottom wafers (optionally) and a post-bonding deformation threshold for the bonded wafer. The wafer bonding recipe is generated using a wafer bonding process model, such as the model created in box 805. In an embodiment, the wafer bonding recipe is generated by optimizing the wafer bonding recipe based on the physical parameters of the incoming top and bottom wafers (optionally) and the process conditions of the initial wafer bonding recipe (e.g., a POR wafer bonding recipe for the wafer bonding process). As an example, the wafer bonding recipe is optimized by adjusting the process conditions until the estimated post-bonding deformation is reduced to a point where the estimated post-bonding deformation is less than a post-bonding deformation threshold. The optimization algorithm can be a linear programming algorithm, a genetic algorithm, a random forest algorithm, a regression algorithm, or other techniques.

[0069] The optimized wafer bonding formulation, including optimized process conditions, is the instruction for the wafer bonding process performed on the wafer pair before the wafer processing tool enters. Optimized process conditions may include pre-cleaning time, pre-cleaning solution chemical composition, upper wafer chuck temperature, lower wafer chuck temperature, plasma pretreatment time, plasma pretreatment power, wafer idle time, bonding time, bonding initiation force, chamber atmospheric pressure, chamber average temperature, chamber humidity, chamber atmospheric composition, bottom wafer chuck vacuum pressure, bottom wafer chuck vacuum zone, bottom wafer chuck vacuum on / off time, top wafer chuck vacuum pressure, top wafer chuck vacuum zone, top wafer chuck vacuum on / off time, top wafer chuck height variation, bottom wafer chuck height variation, bonding gap, wafer temperature before clamping, chuck height, etc.

[0070] In an embodiment, a wafer bonding process model (having physical parameters of the incoming top wafer and the incoming bottom wafer (optional), and an initial wafer bonding recipe (e.g., a POR wafer bonding recipe)) is used to estimate the post-bonding deformation of the bonded wafer. If the estimated post-bonding deformation meets a post-bonding deformation threshold, the initial wafer bonding recipe is selected as the wafer bonding recipe for bonding the incoming top wafer and the incoming bonding wafer. If the estimated post-bonding deformation does not meet the post-bonding deformation threshold, the wafer processing tool changes one or more process conditions of the wafer bonding recipe and re-estimates the post-bonding deformation. During the iteration process, the wafer processing tool may continue to change the process conditions until the estimated post-bonding deformation meets the post-bonding deformation threshold (or, for example, until a specified number of iterations is met).

[0071] In an embodiment, the first wafer bonding formulation that causes the estimated post-bonding deformation to meet the post-bonding deformation threshold is the wafer bonding formulation for the wafer to which the bonding is performed. In another embodiment, when multiple wafer bonding formulations cause the estimated post-bonding deformation to meet the post-bonding deformation threshold, the wafer bonding formulation with the fewest number of process condition changes (or the fewest amount of process condition changes, the easiest process condition changes to implement, etc.) is the wafer bonding formulation for the wafer to which the bonding is performed. As an example, if the estimated post-bonding deformations of two wafer bonding formulations are approximately equal, then the wafer bonding formulation with the easiest process condition changes to implement (e.g., the fewest number of changes, the smallest amount of change, changes to process conditions that are easier to change, etc.) is the wafer bonding formulation for the wafer to which the bonding is performed.

[0072] The wafer processing tool bonds the incoming top wafer to the incoming bottom wafer (box 815). For example, the wafer processing tool uses the wafer bonding recipe generated in box 813 to bond the incoming wafer.

[0073] A check is performed to determine if there are more wafers to bond (box 819). If there are additional wafers to bond, another check is performed to determine if the additional wafers come from the same wafer batch as the previously bonded wafers (box 821). For example, a same-batch comparison can be performed by comparing the wafer batch associated with the wafer bonding recipe generated in box 813 with the wafer batch of the incoming wafer pair. This check may help determine whether the wafer bonding recipe should be re-optimized. As previously discussed, intra-batch variation is significantly less than inter-batch variation. If the additional wafers come from the same wafer batch, it may not be necessary to re-optimize the wafer bonding recipe. If there are no additional wafers to bond, the wafer processing tool can stop the wafer bonding process and may perform additional processing on the bonded wafers (box 829). Additional processing on the bonded wafers may include annealing the bonded wafers, which involves holding the bonded wafers at an elevated temperature for a specified amount of time to strengthen the bonding between the wafers.

[0074] If the additional wafers come from the same wafer batch, the same wafer bonding recipe generated in box 813 can be used to bond the additional wafers. The wafer processing tool returns to box 815 to bond another pair of incoming wafers (as shown by dashed line 823). The wafer processing tool uses the same wafer bonding recipe to bond this pair of incoming wafers. The operation shown is when the wafer processing tool returns to box 815 to bond another pair of incoming wafers without measuring this pair of incoming wafers or potentially updating the wafer bonding recipe. Figure 6 An example of a wafer fusion bonding process is shown.

[0075] While it may not be necessary to re-optimize the wafer bonding recipe for the same batch of wafers, it is still possible to re-optimize the wafer bonding recipe for each incoming wafer pair. In this case, the wafer processing tool returns to frame 807 to measure the incoming wafer, obtain deformation, generate another wafer bonding recipe, and bond the incoming wafer (as shown by double-dotted line 825). The operation shown is as follows: When the wafer processing tool returns to frame 807 to bond another pair of incoming wafers and measures this pair of incoming wafers and potentially updates the wafer bonding recipe, Figure 7 The example shown is a wafer fusion bonding process. As a practical implementation, wafer measurements can be performed on each wafer pair before the wafers enter the wafer processing tool, such as during or before previous steps. This can provide additional time for generating an optimized process formulation. In this case, the model can include corrections to compensate for additional deformations introduced by the processing steps to be performed.

[0076] If the additional wafer is not from the same wafer batch, the same wafer bonding recipe generated in box 813 may not be able to bond the wafer to achieve the post-bonding deformation threshold. The wafer bonding recipe should be regenerated using the physical parameters of the additional wafer. The wafer processing tool returns to box 807 to measure the incoming wafer, obtain the deformation, generate another wafer bonding recipe, and bond the incoming wafer (as shown by solid line 827).

[0077] In this embodiment, measurements of the bonded wafers using the optimized wafer bonding formula were used to refine the wafer bonding process model, thereby improving the model's estimation accuracy. The model refinement can follow the steps outlined in Figure 9 or... Figure 10 The technology shown or some other technology.

[0078] Figure 9AA flowchart of a first example process 900 is shown in the process of creating a wafer bonding process model using semi-empirical modeling-enhanced finite element modeling (FEM) technology according to the example embodiments presented herein. Process 900 may indicate the operations that occur in a tool, such as a wafer fabrication tool or a tool dedicated to model creation, when using semi-empirical modeling-enhanced FEM technology to create a wafer bonding process model. In the embodiments, FEM technology is used to simulate wafer bonding processes under a wide range of incoming wafer physical properties and wafer bonding formulation process conditions. For example, wafers with different shape profiles can be simulated to obtain certain patterns and / or behaviors that can subsequently be represented in the wafer bonding process model. FEM simulations typically use finite element analysis techniques to solve stress-strain equations using a 2-D (two-dimensional) or 3-D (three-dimensional) representation of the wafer. As a practical implementation, wafer measurements can be performed on each wafer pair before the wafers enter the wafer fabrication tool, such as during or before previous steps. This can provide additional time for generating an optimized process formulation. In this case, the model may include corrections to compensate for additional deformations introduced by the fabrication steps to be performed. Furthermore, parallel FEM simulations can also simulate other physical processes that alter the stress and strain of the system during wafer bonding. Some examples include reactions within the wafer, material reflow with varying stress, and degassing of different material layers within the wafer.

[0079] In this embodiment, semi-empirical modeling is used to create and refine models using data generated by FEM technology. Process 900 can be... Figure 8 Example implementation of box 805.

[0080] Process 900 begins with the physical properties of the wafer before bonding and the process conditions of the wafer bonding formulation (box 905). The physical properties of the wafer before bonding and the process conditions of the wafer bonding formulation can be inputs to a wafer bonding process simulation. This input is provided to an FEM simulation of the physics involved in the wafer bonding process (box 907). The FEM simulation generates a simulated post-bonded wafer based on this input (i.e., the physical properties of the wafer before bonding and the process conditions of the wafer bonding formulation, such as applied pressure and temperature). The FEM simulation produces a simulated post-bonded wafer (box 907). Simulated deformation data of the simulated post-bonded wafer can be determined.

[0081] FEM simulations can use the physics involved in wafer bonding as input (i.e., the physical properties of the wafer before bonding and the process conditions of the wafer bonding formulation, such as applied pressure and temperature) to simulate the output (i.e., the deformation data of the wafer after bonding). FEM simulations can be computationally intensive. Therefore, while FEM simulations can be used directly in actual wafer bonding deployments, some embodiments may use semi-empirical models. In these embodiments, semi-empirical modeling is used to help reduce the computational requirements associated with creating wafer bonding process models.

[0082] Inputs and outputs can be provided to a semi-empirical modeling process (Box 911). Inputs can target a wide range of pre-bonding wafers and wafer batches, as well as a range of process conditions or wafer bonding formulations, while outputs include simulated deformation data of the simulated post-bonding wafers bonded based on the inputs. This semi-empirical modeling process is used to determine the relationship between the inputs and outputs. In other words, the semi-empirical modeling process utilizes empirical data (inputs and outputs) to develop a wafer bonding process model. The semi-empirical modeling process outputs the wafer bonding process model. The semi-empirical modeling process can use any of a variety of techniques, such as optimization techniques, search techniques, annealing techniques, machine learning techniques (including but not limited to neural network techniques, deep learning techniques, regression techniques, classification techniques, clustering techniques, dimensionality reduction techniques, ensemble methods, transfer learning techniques, reinforcement learning techniques, etc.).

[0083] In the embodiments, actual data (e.g., the actual physical properties of the wafer before bonding, the process conditions of the wafer bonding formulation, and the actual deformation of the wafer after bonding using the wafer bonding formulation) are also provided to the semi-empirical modeling process to help further refine the wafer bonding process model.

[0084] Figure 9B A flowchart of a second example process 950 is shown, illustrating the process of creating a wafer bonding process model using semi-empirical modeling and calibration-enhanced FEM techniques according to the example embodiments presented herein. Calibration can help refine the model to account for parameters that may be difficult to account for in FEM simulations, such as pre-cleaning time or the time to hold the wafer together in box 815 before annealing or the annealing time. Furthermore, many process knobs can be modeled similarly; for example, the top wafer chuck position and the bottom wafer chuck position can be controlled independently but can be modeled similarly because the physical model only uses relative distances. Therefore, calibration may be necessary to correlate various process conditions with the model.

[0085] Furthermore, although calibration is described below regarding semi-empirical modeling, a similar approach can be used if FEM simulation is used directly without a semi-empirical model.

[0086] Process 950 can indicate the operations that occur within a tool when it uses semi-empirical modeling-enhanced FEM technology to create a wafer bonding process model, such as a wafer fabrication tool or a tool specifically designed for model creation. Furthermore, calibration can be used to determine the contribution and impact of different process parameters on the wafer bonding process model. Process 950 can be... Figure 8 Example implementation of box 805.

[0087] like Figure 9B As shown, a wafer bonding process model (box 955) is created using semi-empirical modeling-enhanced FEM technology. This model can be created based on the physical properties of the wafer before bonding and the process conditions of the wafer bonding formulation. Figure 9A The process 900 can be an example of box 955.

[0088] Calibration can be performed after the model has been created, or during model creation (box 957). Calibration may involve setting one or more process conditions of the wafer bonding formulation to predetermined values ​​and then simulating the output based on the calibrated process conditions. Calibration can be used to determine the effect or contribution of individual process conditions on the deformation of the bonded wafer. As an example, one process condition is changed while the others remain unchanged, and these process conditions are used to simulate the resulting bonded wafer. The wafer bonding process model can be refined based on the simulation results.

[0089] Figure 10 A flowchart illustrating example operation 1000 that occurs during the creation of a wafer bonding process model using a fingerprinting function, according to the example embodiments presented herein. Operation 1000 may indicate the operations that occur within a tool, such as a wafer fabrication tool or a tool dedicated to model creation, when it uses the fingerprinting function to create a wafer bonding process model. Operation 1000 may be... Figure 8 Example implementation of box 805.

[0090] Operation 1000 begins with measuring the wafer to obtain physical parameters (box 1005). The wafer being measured includes the pre-bonding wafer (i.e., the top and bottom wafers). Metering tools (such as surface profilometers, scanners, etc.) can be used to measure the wafer. If both the top and bottom wafers are patterned, both should be measured. If only one wafer is patterned, it may not be necessary to measure the unpatterned wafer. Measurements can be performed on multiple pre-bonding wafers. For example, top and bottom wafers from different batches can be measured. The deformation is obtained from the wafer measurements (box 1007). For example, the deformation can be obtained from the wafer measurements through geometric calculations.

[0091] The fingerprint function is fitted to the deformation (box 1009). Typically, a fingerprint function is a mathematical model of the corresponding metric that preserves spatial information of the measurement result. Fitting the fingerprint function to the deformation may involve analysis (such as regression analysis) to select and adjust the parameters of the fingerprint function to fit the deformation within a specified threshold. As an example, the deformation can be mathematically represented as a finite series of the fingerprint function, such as a polynomial (e.g., a Zernike polynomial) or a function (e.g., a Fourier series or a Bessel function). Each polynomial or function is weighted with corresponding coefficients, which are fitting parameters that can be adjusted to fit the deformation. The fitting coefficients represent the unique deformation trend of a particular wafer. In this embodiment, the fingerprint function is fitted to the deformation of a wafer (e.g., a pre-bonded wafer and a post-bonded wafer).

[0092] The fitting coefficients of the fingerprint function are fitted to the wafer bonding process conditions (box 1011). A wafer bonding process model can be created by fitting the fitting coefficients of the fingerprint function to the wafer bonding process conditions used for bonding the wafers and the pre-bonding wafer deformation (which is also converted into fingerprint coefficients). This function can be linear, quadratic, exponential, or other nonlinear, with one or more of the process conditions discussed above. The function may also include terms that cause two or more process conditions to interact with each other in a linear, quadratic, exponential, or other nonlinear manner. The exact form of the function can be determined by computer algorithms such as linear regression, random forest, genetic programming, pattern search, neural networks, deep learning, etc. The model is constructed such that post-bonding wafer deformation can be estimated using any process conditions and pre-bonding wafer physical parameters within the limits of a database. The ability of the wafer bonding process model to estimate post-bonding wafer deformation based on the wafer bonding process conditions and pre-bonding wafer characteristics enables optimization of the wafer bonding formulation prior to the bonding of pre-bonded wafer pairs. A detailed description of fingerprint recognition functions and examples of their use is provided in commonly assigned U.S. Patent Application No. 16 / 666087, filed October 28, 2019, entitled “Systems and Methods for Manufacturing Microelectronic Devices,” which is incorporated herein by reference in its entirety.

[0093] Figure 11 A flowchart illustrating operation 1100 occurring during wafer bonding recipe generation with feedforward optimization, according to an example embodiment presented herein, is provided. Operation 1100 may indicate operations occurring in the wafer fabrication tool when the tool uses feedforward optimization with wafer bonding recipe to generate the wafer bonding recipe. Operation 1100 may be... Figure 8 Example implementation of box 813.

[0094] Operation 1100 begins with the model in box 805, where the wafer fabrication tool estimates the post-bonding deformation of the bonded wafer using the initial wafer bonding formulation (box 1105). The post-bonding deformation of the bonded wafer is estimated using the physical parameters of the incoming wafer and the process conditions of the initial wafer bonding formulation (e.g., a POR wafer bonding formulation), such as... Figure 8 As described in frames 807, 809, and 811.

[0095] A check is performed to determine whether the estimated post-bonding deformation of the wafer meets the post-bonding deformation threshold (box 1107). If the estimated post-bonding deformation meets the post-bonding deformation threshold, the wafer fabrication tool creates a wafer bonding recipe (box 1109). The wafer bonding recipe is created based on the process conditions of the wafer bonding recipe used in the estimation in box 1105. This wafer bonding recipe includes any changes to the process conditions of the initial wafer bonding recipe.

[0096] If the estimated post-bonding deformation of the wafer does not meet the post-bonding deformation threshold, one or more process conditions of the wafer bonding formulation are modified (box 1111). The selection of these one or more process conditions can be based on factors such as how far the estimated post-bonding deformation is from meeting the post-bonding deformation threshold, the impact of process conditions on post-bonding deformation, and the ease of modifying the process conditions. For example, if the estimated post-bonding deformation is close to meeting the post-bonding deformation threshold, the one or more process conditions to be modified can be those that are easily modified and have been shown to alter the estimated post-bonding deformation in a fine-grained manner. Conversely, if the estimated post-bonding deformation is far from meeting the post-bonding deformation threshold, the one or more process conditions to be modified can be those that have been shown to alter the estimated post-bonding deformation in a coarse-grained manner (e.g., reducing the number of times the wafer bonding formulation needs to be modified before the estimated post-bonding deformation meets the post-bonding deformation threshold). As another example, some process conditions are easier to modify. For example, the wafer chuck temperature may be easier to modify than the pre-cleaning solution chemistry. Therefore, such process conditions are a more feasible candidate for modification.

[0097] Examples of changes to the wafer bonding formulation may include changes to any one or more of the following process conditions: pre-cleaning time, pre-cleaning solution chemistry, upper wafer chuck temperature, lower wafer chuck temperature, plasma pretreatment time, plasma pretreatment power, wafer idle time, bonding time, bonding initiation force, chamber atmospheric pressure, chamber average temperature, chamber humidity, chamber atmospheric composition, bottom wafer chuck vacuum pressure, bottom wafer chuck vacuum zone, bottom wafer chuck vacuum on / off time, top wafer chuck vacuum pressure, top wafer chuck vacuum zone, top wafer chuck vacuum on / off time, top wafer chuck height variation, bottom wafer chuck height variation, bonding gap, wafer temperature before clamping, chuck height, etc. The wafer processing tool returns to box 1105 to re-estimate the post-bonding deformation of the bonded wafers bonded according to the modified wafer bonding formulation.

[0098] Figure 12 A flowchart illustrating example operation 1200 occurring in a feedback-optimized wafer bonding process with a wafer bonding formulation, according to an example embodiment presented herein, is shown. Operation 1200 may indicate operations occurring in a wafer processing tool when the tool uses a feedback-optimized wafer bonding process with a wafer bonding formulation to bond wafers. Operation 1200 may describe a wafer fusion bonding process 500.

[0099] Operation 1200 begins with the creation of a wafer bonding process model (box 1205). This wafer bonding process model can be created based on measurements of the bonded wafer and the process conditions of the wafer bonding formulation. The wafer bonding process model can utilize any combination of finite element analysis, linear regression, random forest algorithms, genetic programming algorithms, pattern search algorithms, neural network algorithms, deep learning algorithms, etc., to correlate the deformation of the bonded wafer with the process conditions of the wafer bonding formulation. In one embodiment, the model is created a priori by the wafer fabrication tool and stored in memory. In another embodiment, the model is created by a tool that does not directly participate in wafer bonding and is then provided to the wafer fabrication tool. A detailed discussion of example approaches that can be used to create the model is provided below.

[0100] The wafer processing tool bonds the incoming top wafer to the incoming bottom wafer (box 1207). In one embodiment, for example, the wafer processing tool uses, for example, a POR wafer bonding formulation to bond the incoming top and bottom wafers. In another embodiment, the wafer processing tool uses a recent wafer bonding formulation for bonding wafers of the same design to bond the incoming top and bottom wafers. The wafers being bonded may come from the same batch or different batches of wafers bonded using this recent wafer bonding formulation. The wafer processing tool measures the bonded wafer (box 1209). For example, it measures the deformation of the bonded wafer. Measurements of the bonded wafer can be performed using metrology tools such as a scanner.

[0101] Generate a wafer bonding recipe (box 1215). The wafer bonding recipe is generated based on measurements (e.g., deformation) of the bonded wafer. The wafer bonding recipe is generated using a wafer bonding process model, such as the model created in box 1205. In an embodiment, the wafer bonding recipe is generated by optimizing the wafer bonding recipe based on the deformation of the bonded wafer and the process conditions of the initial wafer bonding recipe (e.g., the POR wafer bonding recipe of the wafer bonding process). As an example, the wafer bonding recipe is optimized by adjusting the process conditions until the estimated post-bonding deformation is reduced to a point where the estimated post-bonding deformation is less than a post-bonding deformation threshold. The optimization algorithm may be a linear programming algorithm, a genetic algorithm, a random forest algorithm, a regression algorithm, or other techniques.

[0102] The wafer processing tool bonds the incoming wafer pairs (box 1217). The incoming wafer pairs are bonded according to the wafer bonding recipe generated in box 1215.

[0103] Perform a check to determine if there are more wafers to bond (box 1221). If there are additional wafers to bond, perform another check to determine if the additional wafers come from the same wafer batch as the previously bonded wafers (box 1223). For example, a batch comparison can be performed by comparing the wafer batch associated with the wafer bonding formulation generated in box 1215 with the wafer batch of the incoming wafer pair. This check may help determine if the wafer bonding formulation should be re-optimized. As previously discussed, intra-batch variation is significantly less than inter-batch variation.

[0104] If the additional wafers come from the same wafer batch, it may not be necessary to re-optimize the wafer bonding recipe. In this case, the wafer processing tool returns to box 1217 to bond the incoming wafer pairs (as shown in line 1225). In an embodiment, even if the additional wafers come from the same wafer batch, the wafer bonding recipe can still be re-optimized. In this case, the wafer processing tool returns to box 1209 to measure the bonded wafers and uses the measurement results to generate the wafer bonding recipe in box 1215. If there are no additional wafers to bond, the wafer processing tool can stop the wafer bonding process and may perform additional processing on the bonded wafers (box 1229). Additional processing on the bonded wafers may include annealing the bonded wafers, which involves holding the bonded wafers at an elevated temperature for a specified amount of time to strengthen the bonding between the wafers.

[0105] If additional wafers come from different wafer batches, the wafer processing tool returns to box 1207 (as shown in line 1227) to bond the initial wafer pairs from that different wafer batch. The bonded wafers, including the initial wafer pairs from different wafer batches, can be used to generate a new wafer bonding recipe that can be used to bond the remaining wafers from that different wafer batch.

[0106] In this embodiment, measurements of the bonded wafers bonded using the optimized wafer bonding formula were used to refine the wafer bonding process model to help improve the model's estimation accuracy.

[0107] Figure 13 A flowchart illustrating an example process 1300 in creating a wafer bonding process model using feedback data according to an example embodiment presented herein is provided. Process 1300 may indicate the operations that occur within a tool, such as a wafer fabrication tool or a tool dedicated to model creation, when the tool uses the feedback data to create a wafer bonding process model. In an embodiment, a fingerprinting function is used to determine the relationship between the deformation data of the bonded wafer and the process conditions of the wafer bonding recipe. The use of fingerprinting advantageously reduces the computational load required to generate an optimized process recipe for each wafer. Process 1300 may be... Figure 12 Example implementation of box 1205.

[0108] Process 1300 begins with inputs—such as measurements of the bonded wafer (box 1305) and process conditions of the wafer bonding formulation (box 1307)—provided to the fingerprinting process (box 1309). Measurements, such as deformation measurements, can be performed on the bonded wafers after they have been bonded (using a specific wafer bonding formulation) or after they have been annealed. The process conditions correspond to the wafer bonding formulation used to form the bonded wafer. The fingerprinting process may include fitting a fingerprinting function to the deformation measurement results. As previously described, a fingerprinting function is a mathematical model of a corresponding metric that preserves spatial information of the measurement results. The fitting coefficients represent the unique deformation tendency of a particular bonded wafer. In this embodiment, the fingerprinting function is fitted to the deformation of the bonded wafer.

[0109] The fitting coefficients of the fingerprint function are provided to the modeling process to create a wafer bonding process model (Box 1311). The fitting coefficients of the fingerprint function can be fitted to the process conditions of the wafer bonding process. The wafer bonding process model can be created by fitting the fitting coefficients of the fingerprint function to the process conditions of the wafer bonding process used for bonding the wafer and the deformation of the wafer after bonding. The function can be linear, quadratic, exponential, or other nonlinear, with one or more of the process conditions discussed above. The function can also include terms that cause two or more process conditions to interact with each other in a linear, quadratic, exponential, or other nonlinear manner. The exact form of the function can be determined by computer algorithms, such as linear regression, random forest, genetic programming, pattern search, neural networks, deep learning, etc. The model is constructed such that the process conditions can be estimated using any deformation of the wafer after bonding. The ability of the wafer bonding process model to estimate process conditions based on the deformation of the wafer after bonding enables the optimization of the wafer bonding formulation based on the deformation measurements of the wafer after bonding.

[0110] Figure 14 A flowchart illustrating example operation 1400 occurring during wafer bonding recipe generation with feedback optimization, according to an example embodiment presented herein, is provided. Operation 1400 may indicate operations occurring in a wafer fabrication tool when the tool uses feedback optimization with wafer bonding recipe to generate a wafer bonding recipe. Operation 1400 may be... Figure 12 The implementation of frame 1215.

[0111] Operation 1400 begins with the wafer processing tool identifying a portion of the bonded wafer that fails to meet a post-bonded wafer deformation threshold (box 1405). As an example, given a specific bonded wafer whose deformation fails to meet a post-bonded wafer deformation threshold, the wafer processing tool identifies one or more portions of the bonded wafer whose deformation measurements fail to meet the post-bonded wafer deformation threshold. In an embodiment, if multiple portions of the bonded wafer fail to meet the post-bonded wafer deformation threshold, the wafer processing tool may identify a subset of these portions, correct the identified subset, and repeat this process for other subsets of the multiple portions until all portions of the bonded wafer that fail to meet the post-bonded wafer deformation threshold have been resolved.

[0112] The wafer processing tool alters the deformation of the identified portion of the bonded wafer (box 1407). In one embodiment, the wafer processing tool only alters the deformation in the identified portion of the bonded wafer. In another embodiment, the wafer processing tool alters not only the deformation in the identified portion of the bonded wafer but also the deformation in adjacent and potentially nearby portions of the bonded wafer, so that the change in deformation is continuous. The definition of "nearby portions" can be a configuration parameter of the wafer processing tool, potentially sacrificing optimization complexity in exchange for bonding performance.

[0113] The wafer fabrication tool estimates the process conditions for the wafer bonding recipe, which will produce a bonded wafer with the aforementioned deformation measurements (box 1409). In other words, the wafer fabrication tool uses a wafer bonding process model to estimate the process conditions that will result in a bonded wafer with a measured deformation that matches the modified deformation. The wafer fabrication tool creates the wafer bonding recipe based on the estimated process conditions (box 1411).

[0114] Example 1. A method comprising: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicating physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe at least in part based on the physical parameters of the first wafer; obtaining measurements of the first wafer to obtain the physical parameters of the first wafer; generating the first wafer bonding recipe based at least in part on the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first bonded wafer.

[0115] Example 2. The method as described in Example 1 further includes: annealing the first bonded wafer to produce a molten-bonded wafer.

[0116] Example 3. The method as described in Example 1 or 2, wherein the physical parameters of the first wafer include the out-of-plane deformation of the first wafer.

[0117] Example 4. The method as described in any one of Examples 1 to 3 further includes obtaining the wafer bonding process model, comprising: obtaining measurement results of a third wafer and a fourth wafer to obtain physical parameters of the third wafer and the fourth wafer; simulating wafer bonding of the third wafer and the fourth wafer according to process conditions to estimate the physical parameters of the simulated bonded wafer; and creating the wafer bonding process model based on the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated bonded wafer.

[0118] Example 5. The method as described in any one of Examples 1 to 4, wherein creating the wafer bonding process model includes: comparing the physical parameters of the third wafer, the physical parameters of the fourth wafer, these process conditions, and the estimated physical parameters of the simulated bonded wafer; and determining the wafer bonding process model based on the comparison.

[0119] Example 6. The method as described in any one of Examples 1 to 5, wherein generating the first wafer bonding formulation includes: estimating the post-bonding deformation of the first bonded wafer based on the physical parameters of the first wafer; adjusting the process conditions of the first wafer bonding formulation to optimize the estimated post-bonding deformation of the first bonded wafer; and generating the first wafer bonding formulation based on the adjusted process conditions.

[0120] Example 7. The method as described in any one of Examples 1 to 6 further includes: obtaining measurement results of the second wafer to obtain physical parameters of the second wafer, wherein the bonding formulation of the first wafer is further generated based on the physical parameters of the second wafer.

[0121] Example 8. The method of any one of Examples 1 to 7, further comprising: bonding a third wafer to a fourth wafer according to the first wafer bonding formulation to produce a second bonded wafer.

[0122] Example 9. The method as described in any one of Examples 1 to 8, wherein the first wafer and the third wafer are part of a first wafer batch, and the second wafer and the fourth wafer are part of a second wafer batch, and wherein the first wafer batch and the second wafer batch are processed together in a semiconductor manufacturing process flow.

[0123] Example 10. A method comprising: obtaining a wafer bonding process model, the model including inputs indicating physical parameters of a first bonded wafer and configured to output a wafer bonding recipe based on the physical parameters of the first bonded wafer; bonding the first wafer to a second wafer according to the wafer bonding recipe to form the first bonded wafer; obtaining measurements of the first bonded wafer to obtain physical parameters of the first bonded wafer; generating the first wafer bonding recipe based on the physical parameters of the first bonded wafer; and bonding a third wafer to a fourth wafer according to the first wafer bonding recipe to form a second bonded wafer.

[0124] Example 11. The method of Example 10 further includes: annealing the first bonded wafer to produce a first fused-bonded wafer; and annealing the second bonded wafer to produce a second fused-bonded wafer.

[0125] Example 12. The method as described in either Example 10 or 11, wherein obtaining the measurement results includes scanning the first bonded wafer to obtain the physical parameters of the first bonded wafer.

[0126] Example 13. The method of any one of Examples 10 to 12, further comprising: obtaining measurement results of the second bonded wafer to obtain physical parameters of the second bonded wafer; generating a second wafer bonding formula based on the physical parameters of the second bonded wafer; and bonding a fifth wafer to a sixth wafer according to the second wafer bonding formula to form a third bonded wafer.

[0127] Example 14. A processing system comprising: a non-transitory computer-readable storage medium including instructions that, when executed, cause a processor of a computing device to perform operations in cooperation with a semiconductor wafer fabrication process, the instructions including: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicating physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based on the physical parameters of the first wafer; obtaining measurements of the first wafer to obtain the physical parameters of the first wafer; the model generating the first wafer bonding recipe based on the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first bonded wafer.

[0128] Example 15. The processing system of Example 14 further includes: a processing chamber; a substrate holder in the processing chamber configured to mechanically support the second wafer when the first wafer is bonded to the second wafer; and a batching system for holding multiple wafers outside the processing chamber.

[0129] Example 16. A processing system as described in any one of Examples 14 to 15, wherein the instructions further include obtaining the wafer bonding process model, comprising: obtaining measurement results of a third wafer and a fourth wafer to obtain physical parameters of the third wafer and the fourth wafer; simulating wafer bonding of the third wafer and the fourth wafer according to process conditions to estimate the physical parameters of the simulated bonded wafer; and creating the wafer bonding process model based on the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated bonded wafer.

[0130] Example 17. A processing system as described in any one of Examples 14 to 16, wherein creating the wafer bonding process model includes: comparing the physical parameters of the third wafer, the physical parameters of the fourth wafer, these process conditions, and the estimated physical parameters of the simulated bonded wafer; and determining the wafer bonding process model based on the comparison.

[0131] Example 18. A processing system as described in any one of Examples 14 to 17, wherein generating the first wafer bonding formulation includes: estimating the post-bonding deformation of the first bonded wafer based on the physical parameters of the first wafer; adjusting the process conditions of the first wafer bonding formulation to optimize the estimated post-bonding deformation of the first bonded wafer; and generating the first wafer bonding formulation based on the adjusted process conditions.

[0132] Example 19. A processing system as described in any one of Examples 14 to 18, wherein the instructions further include: bonding a third wafer to a fourth wafer according to the first wafer bonding formulation to produce a second bonded wafer.

[0133] Example 20. A processing system as described in any one of Examples 14 to 19, wherein the processing system includes a wafer bonding system.

[0134] Although the invention has been described with reference to illustrative embodiments, this description is not intended to be limiting. Various modifications and combinations of the description, illustrative embodiments, and other embodiments of the invention will be apparent to those skilled in the art. Therefore, the appended claims are intended to cover any such modifications or embodiments.

Claims

1. A method for wafer-to-wafer bonding, comprising: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicative of physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based at least in part on the physical parameters of the first wafer; obtaining measurements of the first wafer to obtain the physical parameters of the first wafer; the model generating the first wafer bonding recipe based at least in part on the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first post-bond wafer; and the method further comprising: obtaining the wafer bonding process model, including: obtaining measurements of a third wafer and a fourth wafer to obtain physical parameters of the third wafer and physical parameters of the fourth wafer; simulating wafer bonding of the third wafer and the fourth wafer according to process conditions to estimate physical parameters of a simulated post-bond wafer; and creating the wafer bonding process model according to the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated post-bond wafer. annealing the first post-bond wafer to produce a fused bonded wafer. the physical parameters of the first wafer include out-of-plane deformations of the first wafer.

2. The method of claim 1, further comprising: creating the wafer bonding process model includes:

3. The method of claim 1, wherein, comparing the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated post-bond wafer; and 4. The method of claim 1, wherein, determining the wafer bonding process model according to the comparison. generating the first wafer bonding recipe includes: the model estimating post-bond deformations of the first post-bond wafer according to the physical parameters of the first wafer; 5. The method of claim 4, wherein, adjusting process conditions of the first wafer bonding recipe to optimize the estimated post-bond deformations of the first post-bond wafer; and generating the first wafer bonding recipe according to the adjusted process conditions. obtaining measurements of the second wafer to obtain physical parameters of the second wafer, wherein the first wafer bonding recipe is generated further based on the physical parameters of the second wafer.

7. An apparatus for wafer-to-wafer bonding, comprising:

6. The method of claim 1, further comprising: a non-transitory computer-readable storage medium including instructions that, when executed, cause a processor of a computing device to perform operations in accordance with a semiconductor wafer fabrication process, the instructions including: obtaining a first wafer bonding recipe and a wafer bonding process model, the model including inputs indicative of physical parameters of a first wafer to be bonded to a second wafer and configured to output a wafer bonding recipe based on the physical parameters of the first wafer; obtaining measurements of the first wafer to obtain the physical parameters of the first wafer; the model generating the first wafer bonding recipe according to the physical parameters of the first wafer; and bonding the first wafer to the second wafer according to the first wafer bonding recipe to produce a first post-bond wafer; and wherein the instructions further include obtaining the wafer bonding process model, including: ​ ​ obtaining measurements of a third wafer and a fourth wafer to obtain physical parameters of the third wafer and physical parameters of the fourth wafer; simulating wafer bonding of the third wafer and the fourth wafer according to process conditions to estimate physical parameters of a simulated post-bond wafer; and creating the wafer bonding process model according to the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated post-bond wafer.

8. The apparatus of claim 7, further comprising: a processing chamber; a substrate holder in the processing chamber configured to mechanically support the second wafer when the first wafer is bonded to the second wafer; and a batch system to hold a plurality of wafers outside the processing chamber. creating the wafer bonding process model comprises:

9. The apparatus of claim 7, wherein, comparing the physical parameters of the third wafer, the physical parameters of the fourth wafer, the process conditions, and the estimated physical parameters of the simulated post-bond wafer; and determining the wafer bonding process model according to the comparison. generating the first wafer bonding recipe comprises:

10. The apparatus of claim 9, wherein, the model estimating post-bond distortions of the first post-bond wafer according to the physical parameters of the first wafer; adjusting the process conditions of the first wafer bonding recipe to optimize the estimated post-bond distortions of the first post-bond wafer; and generating the first wafer bonding recipe according to the adjusted process conditions. ​

Images