Bonded dummy grain material selection and localization

Abstract

In some embodiments, a method is provided that selects thermomechanical three-dimensional characteristics of dummy grains for bonding a wafer to vary thermal extraction based on obtaining a more uniform coefficient of thermal expansion or thermal conductivity control. In some embodiments, a method is provided that also selects characteristics and positioning of dummy dies based on subsequent processing that occurs after the dies are bonded to a bonded wafer. The placement of dummy grains may enable edge management control of chemical mechanical polishing. In some embodiments, metrology feedback may be used to allow dummy dies to be positioned based on non-functional dies bonded to a bonded wafer.

Description

Technical Field

[0001] The implementation of this principle generally involves semiconductor processing of semiconductor substrates. Background Technology

[0002] To form processing circuits or logic circuits and similar integrated chip structures, individual dies are diced and bonded to a bonding wafer. The bonding wafer can then undergo further processing, such as gap-filling processes requiring subsequent planarization. The inventors observed that subsequent processing is typically greatly influenced by the die height, die placement, and die density on the bonding wafer at the wafer center and wafer edges.

[0003] Therefore, the inventors provide a method for improving the performance of bonded wafers in post-bonding processes. Summary of the Invention

[0004] This paper presents a method for improving the performance of bonded wafers in post-bonding processes by utilizing dummy grains during hybrid bonding processes.

[0005] In some embodiments, a method for placing a dummy die on a bonding wafer may include receiving bonding wafer parameters of the bonding wafer, receiving component die size parameters and component die positioning parameters of at least one component die for the bonding wafer, receiving bonding wafer processing information for at least one subsequent post-bonding process of the bonding wafer, determining the dummy die size parameters, dummy die material composition, or dummy die positioning parameters on the bonding wafer, or the thermal behavior of the dummy die material or the wafer material of the bonding wafer based on the at least one subsequent post-bonding process, selecting at least one dummy die based on the dummy die size parameters, dummy die material composition, or dummy die positioning parameters, and bonding the at least one dummy die to the bonding wafer based on the dummy die positioning parameters.

[0006] In some embodiments, the method may further include bonding wafer parameters, including the size of the bonding wafer or the material composition of the bonding wafer; component die size parameters, including the width, length, and height of the component die; dummy die size parameters, including the width, length, and height of the dummy die; and dummy die material composition, including dielectric and metallic materials. A machine learning model is used to receive the bonding wafer parameters, component die size parameters, component die positioning parameters, and bonding wafer processing information, and to infer the dummy die size parameters, dummy die material composition, or dummy die positioning parameters of the bonding wafer. The machine learning model is used to infer the dummy die size parameters and the dummy die positioning parameters, which are in at least one... During subsequent post-bonding processes, edge uniformity of the bonded wafer is maintained. Machine learning models are used to infer dummy grain size parameters and dummy grain positioning parameters. The step height difference between at least one component grain bonded to the bonded wafer and the surrounding area of ​​at least one component grain bonded to the bonded wafer is reduced. The dummy grain material composition is determined based on the thermal expansion or conductivity characteristics of at least one component grain. At least one subsequent post-bonding process is performed, including annealing, chemical mechanical planarization (CMP), chemical vapor deposition (CVD) gap filling, or electroplating gap filling, and / or methods integrated into or performed outside the integrated hybrid bonding tool are performed.

[0007] In some embodiments, a method for placing a dummy die on a bonding wafer may include receiving bonding wafer parameters of the bonding wafer, wherein the bonding wafer parameters include the size of the bonding wafer or the material composition of the bonding wafer; receiving component die size parameters of at least one component die and component die positioning parameters for at least one component die on the bonding wafer, wherein the component die size parameters include the width, length, and height of the component die; receiving bonding wafer processing information for at least one subsequent post-bonding process on the bonding wafer; and using a machine learning model to infer the dummy die size parameters, dummy die material composition, or dummy die positioning parameters on the bonding wafer, wherein the machine learning model includes the bonding wafer parameters, component die size parameters, component die positioning parameters, and bonding wafer processing information, to select at least one dummy die and to bond at least one dummy die to the bonding wafer based on the dummy die positioning parameters.

[0008] In some embodiments, the method may further include dummy grain size parameters, including the width, length, and height of the dummy grain; dummy grain material composition, including dielectric and metallic materials; using a machine learning model to infer the dummy grain size parameters and dummy grain positioning parameters, which maintain edge uniformity of the bonded wafer during at least one subsequent post-bonding process; using a machine learning model to infer the dummy grain size parameters and dummy grain positioning parameters, which reduces the step height difference between at least one component grain bonded to the bonded wafer and the surrounding area of ​​at least one component grain bonded to the bonded wafer; and / or determining the dummy grain material composition based on the thermal expansion or conductivity characteristics of at least one component grain.

[0009] In some embodiments, a non-transitory computer-readable medium having instructions stored thereon, when executed, causes a method for placing a dummy die on a bonded wafer to be performed. The method may include receiving bonded wafer parameters for the bonded wafer, receiving component die size parameters and component die positioning parameters for at least one component die of the bonded wafer, receiving bonded wafer processing information for at least one subsequent post-bonding process of the bonded wafer, determining dummy die size parameters, dummy die material composition, or dummy die positioning parameters on the bonded wafer based on at least one subsequent post-bonding process or thermal behavior of the dummy die material or the wafer material of the bonded wafer, selecting at least one dummy die according to the dummy die size parameters, dummy die material composition, or dummy die positioning parameters, and bonding at least one dummy die to the bonded wafer according to the dummy die positioning parameters.

[0010] In some embodiments, the method may further include using a machine learning model that receives bonding wafer parameters, component grain size parameters, component grain positioning parameters, and bonding wafer processing information and infers dummy grain size parameters, dummy grain material composition, or dummy grain positioning parameters for the bonding wafer, and / or determines dummy grain size parameters, dummy grain material composition, or dummy grain positioning parameters on the bonding wafer based on the bonding wafer's metrology information.

[0011] Other and further implementation methods are disclosed below. Attached Figure Description

[0012] The embodiments of this principle, which have been briefly summarized above and discussed in more detail below, can be understood by referring to the illustrative embodiments of the principle depicted in the accompanying drawings. However, the drawings only show typical embodiments of this principle and should not be considered as limiting the scope, as other equally effective embodiments are permissible.

[0013] Figure 1The process flow for hybrid bonding based on some implementations of this principle is described.

[0014] Figure 2 The processing flow for post-bonding processing based on some implementations of this principle is described.

[0015] Figure 3 Top views and cross-sectional views of bonded wafers with component grains based on some embodiments of this principle are depicted.

[0016] Figure 4 Cross-sectional views of bonded wafers with component grains after gap-filling treatment are depicted, illustrating some embodiments based on this principle.

[0017] Figure 5 Top and cross-sectional views of bonded wafers with component grains and dummy grains, based on some embodiments of this principle, are depicted.

[0018] Figure 6 Cross-sectional views of bonded wafers having component grains and dummy grains after gap filling treatment are depicted, based on some embodiments of this principle.

[0019] Figure 7 Cross-sectional views depicting the thermal expansion of component grains and dummy grains in some embodiments based on this principle are shown.

[0020] Figure 8 A cross-sectional view of a dummy grain used to replace a non-functional bottom component grain is depicted in some embodiments based on this principle.

[0021] Figure 9 These are methods for selecting and positioning dummy grains based on some implementation methods of this principle.

[0022] Figure 10 A cross-sectional view depicts an automated dummy die placement system that communicates with a hybrid bonder and post-bonding process to achieve processing feedback, based on some implementations of this principle.

[0023] Figure 11 This is a top view of some implementations of hybrid bonding tools based on this principle.

[0024] For ease of understanding, the same reference numerals are used where possible to denote common elements in the figures. The figures are not drawn to scale and may be simplified for clarity. Elements and features of one embodiment may be advantageously incorporated into other embodiments without further description. Detailed Implementation

[0025] This method provides improved bonded wafer performance by allowing enhanced subsequent processing through the use of dummy dies during the die-to-wafer hybrid bonding process. These methods provide a more uniform environment for the bonded wafer compared to large die environments with step heights of 15 to 150 micrometers by using dummy dies of the same height, thus facilitating post-bonding integration. The uniform environment of the die also enables cheaper gap-filling processes due to the lower overhead. This technology also advantageously provides an edge management solution for subsequent substrate polishing processes (e.g., chemical mechanical polishing (CMP)) by positioning dummy dies for optimal edge control. These methods can also be incorporated into complex die-to-wafer integration processes within integrated hybrid bonding platforms. This automation provided by the method can be a key differentiator, offering lower integration costs and higher yields. This method can be further enhanced using integration software, dummy die memories or feeders, and multi-size hybrid bonders.

[0026] In traditional die-to-wafer bonding, the bonded wafers will have dies with a wide variation in height (e.g., 15 to 150 micrometers). Subsequent processing requires filling / planarizing these large step heights (the difference between the die height and the surrounding unfilled area) in a manner that allows for stacking and back-side processing. Any large gaps between dies, especially at the edges of the bonded wafers, are extremely challenging. Chip packaging manufacturers will not reduce die yield by sacrificing known good dies to fill edges (i.e., areas from approximately 2 mm to approximately 10 mm from the wafer edge), and no die can extend beyond the wafer edge. Post-bonding processing will struggle to handle gaps and exposed corners (e.g., post-processing such as, but not limited to, gap filling, chemical mechanical polishing, electroplating, through-silicon via (TSV) openings), increasing yield risk and cost.

[0027] As used herein, a component die is a die that forms part of an active or passive semiconductor element and may include processors, logic, memory, inserters, etc. The methods of this principle provide the ability to automatically position dummy dies (e.g., silicon-based dummy dies, etc.) to homogenize the area around each component die. These methods can be directly incorporated into manufacturing integration costs. These methods can calculate dummy requirements and placement, and balance the output of hybrid bonding tools for integration. This approach can also incorporate feedback and feedforward capabilities from other semiconductor processing tools to incorporate subsequent processing into the dummy die processing model, such as CMP edge management, chemical vapor deposition (CVD) performance, etc., to optimize dummy die utilization and placement. By leveraging knowledge of subsequent processing requirements, dummy die processing can be minimized to one or more optimized die sizes, which minimizes the impact on hardware, such as in die-to-wafer hybrid bonding integration tools (e.g., perhaps only one additional die bonder is needed for the dummy dies in the integration tool).

[0028] Figure 1 An example of a hybrid bonding process 100 is depicted, where target 118 is the bonding wafer to which a component die will be bonded, and source 102 is a wafer having component dies removed from source 102 to be bonded to target 118. Hybrid bonding process 100 is an example bonding process, and other such processes can be performed with fewer or more processes. Therefore, hybrid bonding process 100 is not intended to be limiting. In the bonding process, component dies (sources) and targets (bonding wafers) to be bonded are prepared prior to bonding to enhance bonding performance. In some cases, source 102 can be processed in parallel before or after target 118, and component dies from source 102 will be bonded to target 118. As used herein, a source can be a wafer or substrate that provides a frame, chiplet, top die, or component die for bonding to a target such as a substrate, base wafer, base die, or cell, respectively. For the sake of brevity and without limitation, the term “component die” will be used herein to refer to the film frame, chiplet, top die, or component provided by the source to be bonded to the target.

[0029] Source 102 may undergo other processing prior to the hybrid bonding process. These other processing may include upstream processes such as patterning, CMP, back-side polishing, dicing, etc. In some embodiments, for example, component dies may be separated (divided) using dicing tape and held together on the back side to create source 102. In some embodiments, component dies may be reconstructed (molded) on a carrier wafer to form source 102, from which component dies are selected for bonding. In the hybrid bonding process 100, in some embodiments, source 102 typically undergoes a first wet cleaning process 104, followed by a degassing process 106 to help remove moisture from source 102. Source 102 then undergoes a first plasma activation process 108 to increase bonding attraction, followed by a first hydration process 110. Source 102 is then subjected to a radiation treatment 112 (e.g., UV radiation, etc.) to loosen the adhesive bonds holding component dies to source 102 prior to bonding. In some embodiments, target 118 may undergo additional treatments prior to the hybrid bonding process 100. Target 118 is treated before, simultaneously with, or after the treatment of source 102. In some embodiments, target 118 first undergoes a second wet cleaning process 120, followed by a second plasma activation process 122. Target 118 then undergoes a second hydration process 124 to prepare for bonding.

[0030] Bonding is then performed by subjecting source 102 to an ejection and pick-up process 114, which allows for the selection and flipping of component dies in preparation for bonding. In bonding process 116, the die is placed on target 118 and the component die is bonded to target 118, resulting in a target or bonded wafer 126 with die-to-target bonding. Bonded wafer 126 may have multiple component dies bonded to the surface during one or more bonding processes. Figure 2 This is an example of post-bonding process 200. In some embodiments, a low-temperature annealing process 202 is performed on the bonded wafer 126 to reflow the component dies and the bonded wafer 126 for further bonding. The bonded wafer 126 may then undergo a gap-filling process 204. The gap-filling process 204 may use a CVD deposition process or electroplating process, etc., to deposit a gap-filling material on the bonded wafer 126. The gap-filling material fills the gaps between the component dies bonded to the bonded wafer 126. The capping layer or excess gap-filling material is then removed by a CMP planarization process 206 to form the completed bonded wafer 126A. In some embodiments, after planarization, the bonded wafer 126 is returned to the bonder, where additional component dies are then bonded to the previously bonded component dies in a second bonding process 116B to form a multi-component die stack bonded wafer 126B. The above post-bonding process is exemplary and is not intended to be limiting. Other processing, such as metering, can also be performed before or after bonding.

[0031] Figure 3 View 300A depicts a bonding wafer 302 having multiple component dies 304 bonded to it after a hybrid bonding process (such as the exemplary process discussed above). The component dies 304 are generally arranged in an ordered manner, with gaps 306 between them. The circular shape of the bonding wafer 302 combined with the typical rectangular shape of the component dies 304 excludes full use of the surface of the bonding wafer 302, leaving large areas 310 without component dies, particularly near the edges 308 of the bonding wafer 302. View 300B depicts a magnified portion of the bonding wafer 302, which better shows the gaps 306 between the component dies and the large areas 310 without component dies 304. Figure 3 View 300C depicts a cross-section of the bonded wafer 302, showing the gap 306 between component dies 304 and the height 316 of the component dies 340. In some cases, the gap 306 can be from about 10 micrometers to about 50 millimeters. In some cases, the height 316 of the component dies can range from about 15 micrometers to about 150 micrometers. The component dies 304 will also have a length 314 and a width 312.

[0032] The inventors observed that during post-bonding processing, the gap 306 and height 316 of the component grains 304 had a negative impact on the post-bonding process, such as... Figure 4 As shown in view 400. During gap-filling processes, such as electroplating or CVD deposition, a significant amount of gap-filling material 402 is required to fill gaps 306. This significant deposition necessitates the removal of a large amount of capping layer 404 during subsequent planarization (CMP) processes, resulting in reduced yield (requiring more time for planarization) and increased costs. The inventors also observed that a large region 310 near edge 308 causes thinning of the gap-filling material 406 near edge 308, even after planarization, leading to inhomogeneities in the bonded wafer 302. The inventors discovered that if such... Figure 5 As shown in View 500A, positioning dummy dies 502-508 in gaps and / or blank areas near the edges of the bonded wafer can significantly reduce the impact of the aforementioned problems. As used herein, the dummy die has a composition of one or more materials having a height similar to the component dies bonded to the bonded wafer. The dummy die can have different length and width dimensions than the component dies, allowing it to be placed in gaps between component dies and / or along the edges of the bonded wafer. In the example of View 500A, four different sizes (width × length) of dummy dies 502-508 are shown, but this does not imply a limitation on the size or number of dummy dies that can be used. Figure 5View 500B shows a cross-sectional view of the bonded wafer 302, wherein dummy dies 502 of a first size are located between component dies 304, and dummy dies 508 of a second size are located near the edge 308 of the bonded wafer 302. The dummy dies effectively reduce the gap 306 to a smaller gap 510, and also reduce the component die step height difference between the surrounding areas of the component dies.

[0033] In such Figure 6 During subsequent gap-filling processes of the gap-filling material 602 shown in view 600, the amount of capping layer 604 is significantly reduced by using dummy dies to reduce the gap 306 between component dies 304 to a smaller gap 510. The reduced capping layer increases yield (less planarization time) and lowers cost. Furthermore, using second-sized dummy dies 508 near the edge 308 of the bonded wafer 302 allows for a much higher uniformity 606 in the deposited gap-filling material 602 near the edge 308, thereby maintaining the integrity of the component dies (sufficient gap-filling material) near the edge 308 of the bonded wafer 302. The inventors also found that if the coefficients of thermal expansion (CTE or coefficient of thermal expansion) of the component dies and dummy dies are significantly different, it can affect the quality of subsequent processing of the bonded wafer 302. Figure 7 In view 700, a dummy grain 708 having a first coefficient of thermal expansion is positioned next to a component grain 706 having a second coefficient of thermal expansion. During subsequent post-bonding processes such as annealing, gap filling, and / or planarization, the bonded wafer 302 undergoes a temperature rise, which causes the dummy grain 708 to expand to a first expansion width 704 and the component grain 706 to expand to a second expansion width 702.

[0034] CTE is not the only temperature-related thermal issue. Thermal confinement, such as heat within the underlying grain, can be caused by poor heat extraction conditions. In some embodiments, dummy grain material selection can be used to tune the dummy grain to provide better thermal conductivity than dielectric materials. This tuning can significantly enhance heat transfer, homogenization, and thus heat dissipation. For example, silicon dummy grains can be used for thermal control rather than dielectric dummy grains because silicon has better heat dissipation properties. Although silicon is used as an example, the use of silicon materials for dummy grains does not imply a limitation to materials that provide enhanced thermal properties such as heat extraction. Therefore, in some embodiments, when determining dummy grain size parameters, dummy grain material composition, or dummy grain positioning parameters on the bonded wafer, material selection can be based on the thermal behavior of the dummy grain material (e.g., enhancing heat extraction or matching CTE, etc.).

[0035] If the first coefficient of thermal expansion of the dummy grain is greater than the second coefficient of thermal expansion of the component grain, the first expansion width 704 of the dummy grain 708 may be greater than the gap width between the dummy grain 708 and the component grain 706, resulting in compressive / tensile forces and potentially causing cracks or fragmentation between the component grains or between the dummy grain 708 and the bonded wafer 302. The inventors have found that the thermal expansion problem can be mitigated by forming a dummy grain 708 having a thermal coefficient substantially similar to that of the component grain 706. The coefficient of thermal expansion of the dummy grain can be selected based on the use of a single material, such as a dielectric for the dummy grain, or by using a material composition for the dummy grain. In some cases, the dummy grain may also include a metal content, such as, but not limited to, a metal content similar by weight or composition to that found in the component grain or the simulated redistribution layer, to ensure a metal distribution throughout the dummy grain similar to that of the component grain.

[0036] As described above, the bonding wafer 302 can undergo more than one bonding process to create multiple die stacks on the bonding wafer 302, such as Figure 8 As shown in view 800, the first layer 804 of the component die and the gap filling have been completed. However, for example, during subsequent metrology testing, it has been found that the first component die 304A meets the performance criteria, but the second component die 304B is found to be non-functional. Because the bonding wafer 302 undergoes more bonding layers to produce multiple component die stacks, placing the functional component die 802 on top of the non-functional second component die 304B would effectively increase production costs and reduce yield, as the second layer of component dies would be wasted. As an alternative to wasting the functional component die 802, metrology data can be fed back into dummy die processing, allowing a dummy die 806 to be placed on top of the non-functional die (second component die 304B) to avoid using and losing the functional die in the second layer.

[0037] exist Figure 9 In block 902 of method 900, the bonding wafer parameters are determined by, for example, a dummy die processor (e.g., Figure 10 The dummy die processor 1002 receives the data. The dummy die processor 1002 can be used in hybrid bonding tools (such as those described below). Figure 11The bonding wafer (as further described below) is external to or part of a hybrid bonding tool. In some embodiments, bonding wafer parameters may include, but are not limited to, dimensions (e.g., diameters such as 200 mm, 300 mm, 450 mm, etc.), thickness, and / or material composition. For example, the material composition of the wafer may affect the selection of the dummy die's material composition. Similarly, the dimensions of the bonding wafer may affect the dimensions and placement of the dummy die. In block 904, for example, the dummy die processor 1002 receives component die parameters. In some embodiments, component die parameters may include, but are not limited to, dimensional parameters (width, length, and height), bonding wafer positioning / layout parameters, and / or component die material composition. In some embodiments, as described above, the height of the component die has a direct impact on the selection of the dummy die, thereby selecting a dummy die with a height substantially similar to that of the component die. The positioning / layout parameters of the component die on the bonding wafer (gap between component dies and / or distance between the component die and the edge of the bonding wafer, etc.) also directly affect the possibility of selecting the width and length of the dummy die.

[0038] In block 906, for example, dummy die processor 1002 receives post-bonding processing information. In some embodiments, post-bonding processing information may include, but is not limited to, metering, annealing, gap filling, planarization, etc., performed after die bonding. Dummy die processor 1002 may receive knowledge from previous post-bonding processes about which processes to perform and / or actual feedback from post-bonding process 1004. In block 908, dummy die size parameters, dummy die material composition, and / or dummy die positioning parameters are determined based on at least one subsequent post-bonding process or thermal behavior of the dummy die material or the wafer material of the bonded wafer. In some embodiments, dummy die size parameters may include, but are not limited to, the width, length, and / or height of the dummy die. In some embodiments, the dummy die material composition may include dielectric materials or dielectric and metallic materials, etc. In some embodiments, the dummy die material composition is based on the thermal expansion and / or thermal conductivity characteristics of the component die.

[0039] Given the layout and size of component dies, the dummy die processor 1002 can determine the regions where dummy dies can be positioned and the size of the dummy dies that will fit within these regions. For a given post-bonding process, such as planarization, the dummy die processor 1002 can determine that the dummy dies should be positioned within 2 or 3 mm of the edge of the bonded wafer to ensure planarization uniformity, etc. Similarly, after knowing the component die parameters (e.g., coefficient of thermal expansion), dummy dies with appropriate material composition can be selected. In some embodiments, the number of dummy die sizes and / or material compositions can be reduced so that yield is not significantly affected. The more sizes used, the more bonders are required, and the more dummy die sizes / material sources are stored. In box 910, dummy grains are automatically selected based on dummy grain size parameters (e.g., suitable size and size that can be used for bonding), dummy grain material composition (e.g., desired coefficient of thermal expansion, thermal conductivity, compatibility with substrate material, etc.), and / or dummy grain positioning parameters (e.g., smaller dummy grains for edge positions, etc.).

[0040] In some implementations, the determination of dummy grain parameters and the selection of dummy grains can be based on dummy grain model 1006 (see dummy grain model 1006). Figure 10 The dummy grain model 1006 incorporates bonding wafer parameters, component grain parameters, and / or subsequent post-bonding processes that the bonded wafer will undergo after bonding. In some embodiments, machine learning is used in conjunction with the dummy grain model 1006 to infer dummy grain size parameters and dummy grain positioning parameters, which maintain edge uniformity of the bonded wafer during at least one subsequent post-bonding process. In some embodiments, machine learning is used in conjunction with the dummy grain model 1006 to infer dummy grain size parameters and dummy grain positioning parameters, which reduces the step height difference between component grains and the surrounding region of the component grains on the bonded wafer. The dummy grain model 1006 may also incorporate results from previous post-bonding processes and "lessons learned" from previous selections of dummy grain size, material, and / or positioning. While post-bonding processing is in progress or after post-bonding processing has been completed, information can be fed back in real time to the dummy die model 1006 and / or directly to the dummy die model via the dummy die processor 1002. Inferences can also be made to minimize the number of dummy die sizes compared to performance tradeoffs during subsequent post-bonding processing.

[0041] In block 912, one or more selected dummy dies are bonded to the bonded wafer based on the dummy die positioning parameters determined and / or inferred as described above. In some embodiments, the hybrid bonding integration tool may have two bonders. The first bonder can be used to bond component dies to the bonded wafer. The second bonder can then be used to bond dummy dies to the bonded wafer. In alternative embodiments, the selection and bonding of dummy dies may be based on metrological data from metrology tool 1008 (see [link to documentation]). Figure 10 If the metering tool 1008 determines that a component die is not functional, a dummy die can be selected to replace the stacked component dies during subsequent post-bonding processes. When a previously bonded component die is determined to be non-functional, replacing the stacked component die with a dummy die allows the process to eliminate waste of known component dies, thereby maintaining yield and improving productivity by reducing waste of known component dies.

[0042] The automatic virtual die placement system 1012 can be used with the hybrid bonder 1010 (see also...) Figure 11 Combined use to improve post-bonding processes of bonded wafers, such as Figure 10 As shown in view 1000. The hybrid bonder 1010 is typically controlled by controller 1180 (see below for more details). Figure 11 In some embodiments, the automated dummy die placement system 1012 may reside partially or entirely within the controller 1180 of the hybrid bonder 1010. In some embodiments, the automated dummy die placement system 1012 is located external to the hybrid bonder 1010. The controller 1180 receives input from the automated dummy die placement system 1012 regarding the selection of dummy dies and the positioning of dummy dies on the bonded wafer. In some embodiments, the automated dummy die placement system 1012 may include a dummy die processor 1002 configured to perform the methods disclosed herein. The dummy die processor 1002 automatically selects and positions dummy dies for the hybrid bonder 1010 based on inputs regarding the bonded wafer, component die parameters, and / or dummy die type and / or size availability. Input regarding component die functional status, etc., may also be obtained from the metrology tool 1008. As previously described, the dummy die processor 1002 also receives input data in real time or after processing, and / or directly from post-bonding processes, etc. In some embodiments, the automated dummy die placement system 1012 may further include a dummy die model 1006, which allows machine learning inferences about dummy die selection and positioning on the bonded wafer. The dummy die model 1006 may receive input directly from the metrology tool 1008, the post-bonding process 1004, and / or the hybrid bonder 1010. In some embodiments, a bonder chamber 1140 of the hybrid bonder 1010 (see...) Figure 11It can be dedicated to bonding dummy dies (dummy die bonder 1140A) selected by the automatic dummy die placement system 1012. The dummy die bonder 1140A can communicate directly with the dummy die placement system 1012 and / or indirectly through the controller 1180 of the hybrid bonder 1010.

[0043] Automatic dummy die selection and bonding processing can be independent of various hardware architectures or incorporated into them (dummy die processing will be discussed further above and below). For example, in... Figure 11 The diagram illustrates a schematic top view of an integrated hybrid bonding tool 1100 for bonding a die to a target, based on at least some embodiments. The integrated hybrid bonding tool 1100 can be used to perform the methods further described above and below. The integrated hybrid bonding tool 1100 typically includes a device front-end module (EFEM) 1102 and a plurality of automation modules 1110 coupled in series to the EFEM 1102. The plurality of automation modules 1110 are configured to transfer one or more types of substrates 1112 from the EFEM 1102 through the integrated hybrid bonding tool 1100 and perform one or more processing steps on the one or more types of substrates 1112 (e.g., a source having component dies, a source having dummy dies, a target for bonding the die to, or a bonding wafer, etc.). Each of the plurality of automation modules 1110 typically includes a transfer chamber 1116 and one or more processing chambers 1106 coupled to the transfer chamber 1116 to perform one or more processes. Multiple automation modules 1110 are coupled to each other via their respective transfer chambers 1116 to provide modular scalability and customization for the integrated hybrid bonding tool 1100. For example... Figure 11 As shown, the multiple automation modules 1110 include three automation modules, wherein the first automation module 1110a is coupled to EFEM 1102, the second automation module 1110b is coupled to the first automation module 1110a, and the third automation module 1110c is coupled to the second automation module 1110b.

[0044] EFEM 1102 includes a plurality of loading ports 1114 for receiving one or more types of substrates 1112. In some embodiments, the one or more types of substrates 1112 include 200mm wafers, 300mm wafers, 450mm wafers, framed substrates, carrier substrates with or without reconfigurable dies, silicon substrates, glass substrates, etc. In some embodiments, the plurality of loading ports 1114 includes at least one of one or more first loading ports 1114a for receiving a first type substrate 1112a or one or more second loading ports 1114b for receiving a second type substrate 1112b. In some embodiments, the first type substrate 1112a has a different size than the second type substrate 1112b. In some embodiments, the second type substrate 1112b includes a framed substrate or a carrier substrate. In some embodiments, the second type substrate 1112b includes a plurality of dies disposed on a frame or an additional circuit board. In some embodiments, the second type substrate 1112b may hold component dies or dummy dies of different types and sizes. Therefore, one or more second loading ports 1114b may have different sizes or be configured to load receiving surfaces of second-type substrates 1112b with different sizes. In some embodiments, multiple loading ports 1114 are arranged along the common side of EFEM 1102. Although Figure 11 A pair of first load ports 1114a and a pair of second load ports 1114b are depicted, but EFEM 1102 may include other combinations of load ports, such as one first load port 1114a and three second load ports 1114b. Additionally, the integrated hybrid bonding tool 1100 may incorporate a buffer 1190, which provides temporary storage or buffering for sources and targets, etc. The buffer 1190 helps allow dummy dies of different sizes to meet timing and other factors and / or constraints by making targets and / or sources (parts and / or dummy dies) readily available for processing without the need for external retrieval.

[0045] In some embodiments, EFEM 1102 includes a scanning station 1108 with a substrate ID reader for scanning one or more types of substrates 1112 to identify information. In some embodiments, the substrate ID reader includes a barcode reader or an optical character recognition (OCR) reader. An integrated hybrid bonding tool 1100 is configured to use any identification information from the scanned one or more types of substrates 1112 to determine processing based on the identification information, such as different processing and / or placement for a first type of substrate 1112a and a second type of substrate 1112b. In some embodiments, scanning station 1108 may also be configured for rotary motion to align the first type of substrate 1112a or the second type of substrate 1112b. In some embodiments, one or more of a plurality of automation modules 1110 include scanning station 1108. An EFEM robot 1104 is disposed in EFEM 1102 and configured to transport the first type of substrate 1112a and the second type of substrate 1112b between a plurality of loading ports 1114 to scanning station 1108. The EFEM robot 1104 may include a substrate-end actuator for manipulating a first type substrate 1112a and a second-end actuator for manipulating a second type substrate 1112b. The EFEM robot 1104 may rotate or move linearly.

[0046] The transfer chamber 1116 includes a buffer 1120 configured to hold one or more first-type substrates 1112a. In some embodiments, the buffer 1120 is configured to hold one or more first-type substrates 1112a and one or more second-type substrates 1112b. The transfer chamber 1116 includes a transfer robot 1126 configured to transfer the first-type substrates 1112a and second-type substrates 1112b between the buffer 1120, one or more processing chambers 1106, and buffers disposed in adjacent automation modules of a plurality of automation modules 1110. For example, the transfer robot 1126 in the first automation module 1110a is configured to transfer the first-type substrates 1112a and second-type substrates 1112b between the buffer 1120 in the first automation module 1110a and the second automation module 1110b. In some embodiments, the buffer 1120 is disposed within the internal volume of the transfer chamber 1116, advantageously reducing the overall footprint of the tool. In addition, the buffer 1120 can be opened to the internal volume of the transfer chamber 1116 to facilitate the entry and exit of the transfer robot 1126.

[0047] One or more processing chambers 1106 may include atmospheric chambers configured to operate at atmospheric pressure and vacuum chambers configured to operate at vacuum pressure. Examples of atmospheric chambers may typically include wet cleaning chambers, radiation chambers, heating chambers, metering chambers, bonding chambers, etc. Examples of vacuum chambers may include plasma activation chambers. If desired, the types of atmospheric chambers discussed above may also be configured to operate under vacuum. One or more processing chambers 1106 may be any processing chamber or module required to perform bonding processes, cleaning processes, radiation processes, etc. In some embodiments, each of the plurality of automation modules 1110 includes one or more processing chambers 1106 comprising at least one of a wet cleaning chamber 1122, a plasma activation chamber 1130, a degassing chamber 1132, a radiation chamber 1134, or a bonder chamber 1140, such that the integrated hybrid bonding tool 1100 includes at least one wet cleaning chamber 1122, at least one plasma activation chamber 1130, at least one degassing chamber 1132, at least one radiation chamber 1134, and at least one bonder chamber 1140. One or more processing chambers 1106 may be arranged at any suitable location within the integrated hybrid bonding tool 1100.

[0048] A wet cleaning chamber 1122 is configured to perform a wet cleaning process to clean one or more types of substrates 1112 using a fluid such as water. The wet cleaning chamber 1122 may include a first wet cleaning chamber 1122a for cleaning a first type of substrate 1112a or a second wet cleaning chamber 1122b for cleaning a second type of substrate 1112b. A degassing chamber 1132 is configured to perform a degassing process to remove moisture through, for example, a high-temperature baking process. In some embodiments, the degassing chamber 1132 includes a first degassing chamber 1132a and a second degassing chamber 1132b. A plasma activation chamber 1130 may be configured to perform an activation process on the substrate in preparation for hybrid bonding. Activation helps increase the bonding strength between surfaces. In some embodiments, the plasma activation chamber 1130 includes a first plasma activation chamber 1130a and a second plasma activation chamber 1130b. Radiation chamber 1134 is configured to perform radiation treatment to reduce adhesion between grains on a source, such as a framed substrate or a carrier substrate with reconstructed grains. For example, radiation chamber 1134 may be an ultraviolet radiation chamber configured to direct ultraviolet radiation to the source or a heating chamber configured as a heating source. Reduced adhesion between the grains and the source facilitates easier removal of the grains from the source. Bonder chamber 1140 is configured to transfer at least a portion of the grains from the source and bond them to a target. Bonder chamber 1140 typically includes a first support 1142 for supporting one of a first type substrate 1112a and a second support 1144 for supporting one of a second type substrate 1112b.

[0049] In some implementations, the last automation module among a plurality of automation modules 1110 (e.g. Figure 11 The third automation module 1110c) includes one or more bonder chambers 1140. Figure 11 (Two are shown in the diagram). In some embodiments, the first of the two bonder chambers is configured to remove and bond component dies, and the second of the two bonder chambers is configured to remove and bond dummy dies. In some embodiments, any of the plurality of automation modules 1110 includes a metrology chamber 1118, which is configured to measure one or more types of substrates. Figure 11 In the diagram, metering chamber 1118 is shown as part of the second automation module 1110b, which is coupled to transfer chamber 1116. However, metering chamber 1118 may be coupled to or within any transfer chamber 1116.

[0050] Controller 1180 controls the operation of any integrated hybrid bonding tool (including integrated hybrid bonding tool 1100) described herein. Controller 1180 may be used with direct control of integrated hybrid bonding tool 1100, or alternatively, through control of a computer (or controller) associated with integrated hybrid bonding tool 1100. In operation, controller 1180 enables the collection of data and feedback from integrated hybrid bonding tool 1100 to optimize the performance of integrated hybrid bonding tool 1100 and control the processing flow based on the methods described herein, such as selecting dummy dies and bonding them to a bonded wafer, while simultaneously bonding component dies to the bonded wafer. Controller 1180 typically includes a central processing unit (CPU) 1182, memory 1184, and support circuitry 1186. CPU 1182 may be any form of general-purpose computer processor that can be used in an industrial environment. Support circuitry 1186 is typically coupled to CPU 1182 and may include cache, frequency circuitry, input / output subsystems, power supply, etc. Software routines (such as those described herein) may be stored in memory 1184 and, when executed by CPU 1182, transform CPU 1182 into a dedicated computer (controller 1180). Software routines may also be stored and / or executed by a second controller (not shown) located remotely from the integrated hybrid bonding tool 1100.

[0051] Memory 1184 takes the form of a computer-readable storage medium containing instructions that, when executed by CPU 1182, facilitate semiconductor processing and device operation. The instructions in memory 1184 are in the form of a program product, such as a program implementing the methods of this principle. The program code can conform to any of a variety of different programming languages. In one example, this disclosure can be implemented as a program product stored on a computer-readable storage medium for use by a computer system. The program product defines the functionality of various aspects (including the methods described herein). Exemplary computer-readable storage media include, but are not limited to: non-writable storage media (e.g., read-only memory elements within a computer, such as CD-ROM disks readable by a CD-ROM disk drive, flash memory, ROM chips, or any type of solid-state non-volatile semiconductor memory), on which information is permanently stored; and writable storage media on which changeable information is stored (e.g., floppy disks or any type of solid-state random access semiconductor memory within a floppy disk drive or hard disk). This is an aspect of this principle when such a computer-readable storage medium carries computer-readable instructions that direct the functionality of the methods described herein.

[0052] Implementations based on this principle can be carried out in hardware, firmware, software, or any combination thereof. Implementations can also be carried out as instructions stored using one or more computer-readable media, which can be read and executed by one or more processors. The computer-readable media can include any mechanism for storing or transmitting information in a machine-readable form (e.g., a computing platform or a “virtual machine” operating on one or more computing platforms). For example, the computer-readable media can include any suitable form of volatile or non-volatile memory. In some implementations, the computer-readable media can include non-transitory computer-readable media.

[0053] While the foregoing describes an implementation of this principle, other and further implementations of this principle can be designed without departing from its basic scope.

Claims

1. A method for placing dummy grains on a bonded wafer, the method comprising: Receive the bonding wafer parameters of the bonding wafer; Receive component die size parameters and component die positioning parameters of the at least one component die for the bonding wafer; Receive bonding wafer processing information for at least one subsequent post-bonding process of the bonded wafer; Based on the at least one subsequent post-bonding process or thermal behavior of the dummy grain material or the wafer material of the bonded wafer, determine the dummy grain size parameters, the dummy grain material composition, or the dummy grain positioning parameters on the bonded wafer; Based on the virtual grain size parameters, virtual grain material composition, or virtual grain positioning parameters, select at least one virtual grain; and The at least one dummy die is bonded to the bonded wafer according to the dummy die positioning parameters.

2. The method of claim 1, wherein the bonding wafer parameters include the size of the bonding wafer or the material composition of the bonding wafer.

3. The method of claim 1, wherein the component grain size parameters include the width, length and height of the component grain.

4. The method of claim 1, wherein the dummy grain size parameters include the width, length and height of the dummy grain.

5. The method of claim 1, wherein the dummy grain material composition includes dielectric material and metallic material.

6. The method of claim 1, further comprising: Using a machine learning model, the machine learning model receives the bonding wafer parameters, the component grain size parameters, the component grain positioning parameters, and the bonding wafer processing information, and infers the dummy grain size parameters, the dummy grain material composition, or the dummy grain positioning parameters for the bonding wafer.

7. The method of claim 6, further comprising: The machine learning model is used to infer dummy grain size parameters and dummy grain positioning parameters, which maintain the edge uniformity of the bonded wafer during the at least one subsequent post-bonding process.

8. The method of claim 6, further comprising: The machine learning model is used to infer dummy die size parameters and dummy die positioning parameters, which reduces the step height difference between the at least one component die bonded to the bonded wafer and the surrounding region of the at least one component die bonded to the bonded wafer.

9. The method of claim 1, further comprising: The composition of the dummy grain material is determined based on the thermal expansion or conductivity characteristics of the grains of at least one component.

10. The method of claim 1, wherein the at least one subsequent post-bonding process includes annealing, chemical mechanical planarization (CMP), chemical vapor deposition (CVD) gap filling, or electroplating gap filling.

11. The method of claim 1, further comprising: Integrate into an integrated hybrid bonding tool by performing the method described in claim 1; or The method of claim 1 is performed outside of the integrated hybrid bonding tool.

12. A method for placing dummy dies on a bonded wafer, the method comprising: Receive bonding wafer parameters for the bonding wafer, wherein the bonding wafer parameters include the size of the bonding wafer or the material composition of the bonding wafer; Receive component die size parameters and component die positioning parameters of at least one component die for the bonding wafer, wherein the component die size parameters include the width, length and height of the component die; Receive bonding wafer processing information for at least one subsequent post-bonding process of the bonded wafer; A machine learning model incorporating the bonding wafer parameters, the component die size parameters, the component die positioning parameters, and the bonding wafer processing information is used to infer dummy die size parameters, dummy die material composition, or dummy die positioning parameters on the bonding wafer to select at least one dummy die. and The at least one dummy die is bonded to the bonded wafer according to the dummy die positioning parameters.

13. The method of claim 12, wherein the dummy grain size parameters include the width, length and height of the dummy grain.

14. The method of claim 12, wherein the dummy grain material composition includes dielectric material and metallic material.

15. The method of claim 12, further comprising: The machine learning model is used to infer dummy grain size parameters and dummy grain positioning parameters, which maintain the edge uniformity of the bonded wafer during the at least one subsequent post-bonding process.

16. The method of claim 12, further comprising: The machine learning model is used to infer dummy die size parameters and dummy die positioning parameters, which reduces the step height difference between the at least one component die bonded to the bonded wafer and the surrounding region of the at least one component die bonded to the bonded wafer.

17. The method of claim 12, further comprising: The composition of the dummy grain material is determined based on the thermal expansion or conductivity characteristics of the grains of at least one component.

18. A non-transitory computer-readable medium having instructions stored thereon, which, when executed, cause a method for placing a dummy die on a bonded wafer, the method comprising: Receive the bonding wafer parameters of the bonding wafer; Receive component die size parameters and component die positioning parameters for at least one component die for the bonding wafer; Receive bonding wafer processing information for at least one subsequent post-bonding process of the bonded wafer; Based on the at least one subsequent post-bonding process or thermal behavior of the dummy grain material or the wafer material of the bonded wafer, determine the dummy grain size parameters, the dummy grain material composition, or the dummy grain positioning parameters on the bonded wafer; Based on the dummy grain size parameters, dummy grain material composition, or dummy grain positioning parameters, at least one dummy grain is selected; and The at least one dummy die is bonded to the bonded wafer according to the dummy die positioning parameters.

19. The non-transitory computer-readable medium of claim 18, further comprising: Using a machine learning model, the machine learning model receives the bonding wafer parameters, the component grain size parameters, the component grain positioning parameters, and the bonding wafer processing information, and infers the dummy grain size parameters, the dummy grain material composition, or the dummy grain positioning parameters for the bonding wafer.

20. The non-transitory computer-readable medium of claim 18, further comprising: The dummy grain size parameters, dummy grain material composition, or dummy grain positioning parameters on the bonded wafer are determined based on the metrological information of the bonded wafer.

Images