Methods and systems to assemble semiconductor wafers from fragments
Machine vision and machine learning methods automate the reassembly of fragmented semiconductor wafers, addressing the inefficiencies and risks of manual reassembly by quickly identifying and reassembling missing fragments, thus reducing time and ensuring accuracy.
Patent Information
- Application Number
- PCT/US2026/026796
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-08
- Filing Date
- 2026-05-05
- Publication Date
- 2026-11-12
AI Technical Summary
The loss of semiconductor wafer fragments during cross-sectional imaging leads to the loss of sensitive intellectual property and increases the risk of financial repercussions and customer trust due to the inefficiency and error-prone manual reassembly process.
Employing machine vision, machine learning, and extended reality-guided reassembly methods to automate the reassembly of fragmented semiconductor wafers, utilizing generative adversarial networks and reinforcement learning to quickly identify and reassemble missing fragments, and employing robotic pick-and-place tools for precise reassembly.
Significantly reduces the time required for reassembling wafers, eliminates human error, and ensures the accurate identification and return of all fragments, thereby preserving intellectual property and maintaining customer trust.
Smart Images

Figure US2026026796_12112026_PF_FP_ABST
Abstract
Description
Docket No.: 11123-1WOMETHODSAND SYSTEMS TO ASSEMBLE SEMICONDUCTOR WAFERS FROM FRAGMENTSCLAIM FOR PRIORITY
[0001] This application claims priority to U.S. Provisional Patent Application No.63 / 802,349, filed on May 08, 2025, titled “‘METHODS AND SYSTEMS TO ASSEMBLE SEMICONDUCTOR WAFERS FROM FRAGMENTS,” which is incorporated by reference in its entirety for all purposes.BACKGROUND
[0002] As part of process development, process engineers submit wafer coupons for cross-sectional imaging using scanning electron microscopy (SEM) or transmission electron microscopy (TEM). To prepare a fragment of the coupon for imaging (or to extract a lamella for imaging from a smaller fragment of the coupon), the coupon is broken further to a desired dimension. After the imaging is complete, the fragments are collected to ensure the original coupon, and the broken coupon are the same. Loss of coupon fragments can result in loss of sensitive intellectual property (IP) for state-of-the-art technologies and have both financial repercussions as well as impact customer trust.BRIEF DESCRIPTION OF DRAWINGS
[0003] Material described herein is illustrated by way of example and not by way of limitation in accompanying figures. For simplicity and clarity of illustration, elements illustrated in figures are not necessarily drawn to scale. For example, dimensions of some elements may be exaggerated relative to other elements for clarity. Also, various physical features may be represented in their simplified “ideal” forms and geometries for clarity of discussion, but it is nevertheless to be understood that practical implementations may approximate illustrated ideals. For example, smooth surfaces and square intersections may be drawn in disregard of finite roughness, comer-rounding, and imperfect angular intersections characteristic of structures formed by nanofabrication techniques. Further, where considered appropriate, reference labels have been repeated among figures to indicate corresponding or analogous elements.
[0004] Fig. 1A illustrates a collection of individual untextured coupons arranged randomly, in accordance with at least one implementation.Docket No.: 11123-1WO
[0005] Fig. IB illustrates a depiction of an initial stage of a partially re-assembled wafer whereby machine vision, machine learning, and / or combinatorial approaches is utilized, in accordance with at least one implementation.
[0006] Fig. 1C illustrates a depiction of an advanced stage of a partially re-assembled wafer whereby machine vision, machine learning, and / or combinatorial approaches is utilized, in accordance with at least one implementation.
[0007] Fig. ID shows a flowchart that illustrates a high-level computation algorithm that a machine learning model may follow to characterize incoming coupons or whole wafers, or for re-assembly of whole wafers from coupons sectioned from a whole wafer for metrology7measurements, in accordance with at least one implementation.
[0008] Figs. 2A - 21 illustrate a progression of images depicting progressive reassembly of a semiconductor wafer from a collection of textured coupons, in accordance with at least one implementation.
[0009] Fig. 3 depicts a carrier containing pre-sectioned coupons, in accordance with at least one implementation.
[0010] Fig. 4A illustrates an initial capture image of the contents of the carrier shown in Fig. 3 as received from an originating source, in accordance with at least one implementation.
[0011] Fig. 4B illustrates a capture image of regrouped fragmented coupons placed in the carrier after metrology has been conducted, in accordance with at least one implementation.
[0012] Fig. 5 illustrates a flowchart that recaps a procedure for verification of coupons as received and of coupons for return to their source, in accordance with at least one implementation.
[0013] Fig. 6A illustrates an architecture for a machine vision network, in accordance with at least one implementation.
[0014] Fig. 6B illustrates a data flow sequence flow chart for the machine vision network shown in Fig. 6A, in accordance with at least one implementation.
[0015] Fig. 7 illustrates a high-level block diagram and a flow chart of the machine vision operations and interactions with the remote server presented in Fig. 6A and 6B, in accordance with at least one implementation.
[0016] Fig. 8 illustrates a hardware block diagram of a computer w orkstation shown in Fig. 6A for receiving and imaging incoming as well as outgoing coupons, in accordance with at least one implementation.Docket No.: 11123-1WO
[0017] Fig. 9A illustrates an isometric view of an imaging station that may be included in a wafer metrology facility for imaging incoming and outgoing semiconductor wafers or coupons, in accordance with at least one implementation.
[0018] Fig. 9B illustrates a frontal isometric view of an alternative imaging station for a wafer metrology facility in accordance with at least one implementation.
[0019] Fig. 9C illustrates a rear isometric view of the alternative imaging station shown in Fig. 9B in accordance with at least one implementation.
[0020] Figs. 10A-10F illustrate a sequence of an operation whereby an incoming wafer is received, imaged, sectioned into coupons, and coupons are imaged and stored in carriers having tags that are recorded into a database for later identification, in accordance with at least one implementation.
[0021] Figs. 11A-11B illustrate a sequence of an operation whereby coupons (and fragments thereof) are reassembled into the wafer of origin by positional mapping of the individual coupons sectioned from the wafer, in accordance with at least one implementation.
[0022] Fig. 12 illustrates a processor system with a machine-readable storage medium having machine-readable instructions that when executed, cause with a machine-readable storage medium having machine-readable instructions that when executed cause a processor to execute machine-readable instructions according to the method summarized by flowcharts shown in Fig. 5 or Fig. 7, as well as operations shown in Figs. 10A-F and 11A-B, in accordance with at least one implementation.DETAILED DESCRIPTION
[0023] One way to preserve the coupon of fragments, two photographs can be taken. A first photograph of the coupons can be taken before the imaging process is started (e.g., before the coupon is sectioned to generate appropriate fragments). A second photograph can be taken after the cross-sectional imaging is complete and fragments are collected. In this case, a user must use their best judgement to determine if any fragments are missing based on the first and second photographs. At times of high-volume production, the imaging system may not always operate consistently as the operators are different. In addition, small fragments (e.g., as small as 0.5 mm) missed may not be easily discernable to a human eye that will be looking at several coupons in a short period of time. Much time is spent zooming into the first and second photographs and assessing if such small pieces are missing, reducing productivity. Historically, in addition to lost productivity time, engineers may have missed detecting a small lost fragment for months, resulting in further loss of time searching forDocket No.: 11123-1WOpieces that should have been identified and collected at the time the coupon was imaged. Once the coupons from a given wafer are available, customers typically want those to be returned as a full wafer. This takes a significant amount of time as the engineers look for scribed information on the coupon as well as its location on a piece of paper tracked during the time the coupons were created. Solving this jig-saw puzzle of identical coupons with hard to discern patterns is slow, error prone, and may put customer IP at risk if coupons or fragments of coupons are lost during the assembly.
[0024] One or more apparatuses and methods are disclosed for machine vision, machine learning, or extended reality -guided reassembly of extensively fragmented semiconductor wafers, as well as detection of missing fragments of lost area that can occur during cleaving, or sectioning, of semiconductor wafers and coupons. The context of this disclosure is in semiconductor process development or quality control, whereby whole wafers, and / or wafer portions (e.g., coupons, including individual diced chips) may be provided to a wafer metrology7lab for imaging after various processing treatments or steps have been accomplished. Imaging of entire semiconductor wafers or wafer portions (e.g., coupons) is a primary part of process development for semiconductor device manufacture. Images may identify problems with various semiconductor wafer processing stages, thus helping to pinpoint and correct process parameter settings, as well as helping to identify faulty7equipment.
[0025] Submission of such samples to a metrology lab may be for purposes of process qualification, verification, or quality control. For example, whole wafers or wafer fragments may be submitted. In the latter case, samples may be submitted as whole wafers for fragmentation by cleavage into a plurality of non-uniformly shaped coupons (e.g., by a metrology lab). Samples may also be submitted pre-sectioned or diced into regular-shaped or irregular-shaped coupons. In many instances, coupons may be marked for identification purposes and for real or virtual mapping to an original wafer, to reassemble the fragments back into the whole wafer for returning to a customer. The disclosed methods include computer vision mechanism tied to deep learning and extended reality software to guide reassembly (e.g., manually by users or by robotic pick and place tools) of coupons into whole wafers to restore them in their entirety after fragmenting for imaging analysis. The combination of machine vision and trained deep learning models, with an option of extended reality7may aid in at least two ways. First, it may aid by directing reassembly of coupons to restore the original wafer prior to fragmentation by a metrology laboratory or a process lab. This computer-guided re-assembly can significantly reduce the time needed to reassemble theDocket No.: 11123-1WOfragments into whole wafers. Second, it may aid in identifying whether there are any missing pieces.
[0026] The disclosed methods include taking an initial photographic image of a whole wafer, a coupon (e.g., a portion of a wafer), group of coupons, as received from a customer or from an internal process development unit. For example, a process development unit may provide one or more whole wafers or wafer fragments (coupons) to a metrology unit for measurements of deposited film thicknesses and uniformity, surface roughness, etc. In at least one implementation, an entire wafer may be completely fragmented into a set of coupons following a processing step which is to be evaluated or validated by metrological data.
[0027] In at least one implementation, coupons include individual diced (singulated) chips (e.g.. integrated circuits) or groups of chips from a whole wafer after processing. This fragmentation of a whole wafer may be performed by engineers in the process development unit or post-facto by engineers in the metrology unit, whole wafer may be received for analysis taking. In at least one example, one or more of the coupons may be further fragmented into smaller pieces by imaging technicians at a metrology station for cross-sectional analysis by scanning electron microscopy (SEM) and / or by transmission electron microscopy (TEM). In some implementations, the metrology group may be directed to restore the original wafer by reassembly of coupons. This action is comparable to solving a jigsaw puzzle. If some coupons were further fragmented, these fragments are to be reassembled as well to restore the coupon. As some fragment pieces may be marginally visible to the naked eye, they are prone to being lost.
[0028] As a subsequent step, all the coupons and coupon fragments, if any, are collected and photographically imaged a second time as a collection of individual pieces after undergoing metrological scrutiny (e.g., SEM and / or TEM analysis). Simultaneous with the post-metrology imaging step, the image of disassembled coupons and fragments is analyzed robotically by computer vision software that includes machine learning models and extended reality to guide engineers in rapidly reassembling whole wafers from the collection of coupons, or individual coupons from a collection of coupon fragments. Here, significant time savings may be achieved for reassembly of coupons to wafers or coupon fragments to whole coupons. For example, current time estimates for restoration of a whole wafer by purely human assembly of a plurality of irregular and non-uniform coupons may take between 4 and 20 hours. The disclosed method also eliminates human error in the assembly process, and immediate identification of missing pieces.Docket No.: 11123-1WO
[0029] In at least one implementation, the machine learning model employs generative adversarial networks (GAN) and reinforcement learning to quickly find wafer or coupon reassembly solutions. Such an approach may generate solutions, akin to jigsaw puzzle solving, more rapidly than conventional combinatorial approaches. The GAN model may be trained with many datasets that include hundreds of randomly fragmented wafers, each fragmented into pluralities consisting of. for example, 30 to 1000 coupons, depending on wafer diameter, having small or large numbers of irregular shape patterns. In at least one example, machine learning model may be trained to match adjacent coupons based on coupon shape analysis that includes measurement of lengths of curvature arcs (by tangents) to identify outer pieces, edge length, and edge angles. In addition to assembling coupons based on coupon shape, coupon texture may also be included to enhance finding adjacent matches when many coupons in a plurality of coupons from a wafer have similar shapes. If metadata are present, such as scribed alphanumerical markings on the coupons, specific lithographic patterns, reflective or fluorescent coloration or patterns, etc., can also be considered.
[0030] In some implementations, the disclosed machine learning methods may be used in conjunction with either human or robotic assembly of coupons and / or fragments to restore a whole wafer. Here, the machine learning model may identify the individual coupons and / or fragments in a plurality of randomly placed unmatched coupons and / or fragments from a wafer, find matches of adjacent coupons to virtually reassemble the wafer, then generate an image of the fully re-assembled wafer by the methodology described above. The image shows where the individual pieces fit. A positional mapping between the unassembled fragments or coupons and corresponding positions of the individual pieces in a virtually reassembled wafer or coupon(s) may be generated by the machine learning model. The positional mapping data, which may comprise the disassembled and assembled images, may be fed to a robotic pick-and-place tool, for example, so that software may show' to the robot the positional mapping of the unmatched pieces to their respective positions in the assembled image. The robot may be commanded to pick up individual unmatched coupons starting with a key piece, then physically reassemble the wafer in a jigsaw fashion according to mapping directives that may be part of the software.
[0031] In some examples, augmented, virtual or extended reality may be employed for a user. In at least one implementation, piece-by-piece positional mapping may be displayed to a user in real time, showing a particular individual coupon in a collection of coupons in an image. Then in a separate window or frame, where in the growing partial assembly of coupons that coupon may fit is shown according to the positional mapping generated by theDocket No.: 11123-1WOmachine learning model. In at least one implementation, one or more machine vision cameras may monitor the collection of coupons and the wafer assembly simultaneously. The user may then manually add the piece to the growing assembly, as one would add a jigsaw piece to a puzzle.
[0032] In another implementation, an image of a virtually reassembled complete wafer may be shown to the user. In at least one example, individual coupons in the assembly are identified by alphanumeric markings or other insignia that are on the individual coupons. As one machine vision camera watches the group of disassembled coupons to observe which individual coupon the user selects, the user may be then shown on a computer monitor screen or with augmented reality goggles or glasses, where the individual coupon fits in the growing assembly, which is simultaneously monitored by the same or a second machine vision camera.
[0033] In some implementations, the disclosed method may also be employed to verity' identity of coupons or whole wafers. For example, the disclosed machine vision, machine learning, or extended reality -guided wafer reassembly method may also ensure that an outgoing (e.g., from a metrology lab) reassembled coupon(s) or wafer is the same as what was received as an incoming (e g., to a metrology lab) coupon(s) or whole wafer.
[0034] In at least one implementation, machine vision cameras having high pixel resolutions may be employed to capture texture details. Environmental conditions include mitigating reflections of nearby objects from the coupon surfaces. Additionally, positioning of machine vision cameras (e g., distance from coupons) may be adjusted for details acquired, such as close-in imaging to capture texture details, or positioning cameras further away if only shape data is to be captured. In at least one example, diffuse lighting may be employed to minimize surface reflections from coupons, as well as lighting from below a transparent table surface.
[0035] Fig. 1A illustrates a collection 100A of individual untextured coupons, which are wafer portions, arranged randomly. The coupons belong to a single wafer, from where they are formed by cleaving the wafer manually using a scribe, for example, breaking the wafer into a plurality of irregular pieces. In some implementations, a wafer may be diced into rectangular pieces by a laser or mechanical saw, producing uniform rectangular coupons. The coupons may be individual chips that are singulated from the wafer by the dicing operation.
[0036] Figs. IB and 1C illustrate depictions of partially reassembled wafers 100B and 100C, respectively, by implementation of the disclosed method, whereby machine vision, machine learning, and / or combinatorial approaches produce matching coupons. Fig. IBDocket No.: 11123-1WOdepicts an early stage and Fig. 1C depicts a more advanced stage of assembly. Here, a machine learning model may perform a progressive virtual re-assembly of the wafer by initially matching, grouping, associating, pairing, and / or linking neighboring coupons that have a curvature. The diameter of the wafer may be input to the model to have a perimeter calculation. The model may generate a virtual circle having that diameter. Pieces may be chosen in succession by the model that have best fits as adjacent neighbors. The model may choose a particular coupon having a curvature and place the curved edge on the virtual circle. Best fits for choosing successive pieces may be made according to curvature matches and estimated arclengths of the curved edges, as well as adj acent straight edge matches to find adjacent pieces.
[0037] All the peripheral pieces may be gradually placed about the perimeter of the circle, as shown in Fig. IB. As show n in the figure, individual coupons are delineated by outlines. Individual coupons are also numbered. The numbers may correspond to metadata that may have been provided with collection 100A for each coupon. For example, the numbers may have been laser scribed or manually scribed on the backside of each coupon. These numbers may be scribed for tracking purposes, that is, data obtained from each coupon is cataloged and referenced by coupon number. The coupons may be simply numbered in the order by which they are analyzed. In some implementations, the numbering of coupons may be related to other criteria, such as position on the wafer, particular lithographic patterns, or by groups. For the latter, some coupons may be grouped together and placed in containers having some means of positive identification, such as radio frequency identification (RFID) tags or visible alphanumeric information. The numbers on the coupons may reflect such metadata to keep track of the coupons and to ensure that that are returned to the originator as received.
[0038] Fig. 1C shows a more advanced stage of wafer re-assembly, depicted as partially reassembled wafer 100C. Here, most or all the coupons have been reassembled, and the software may create a mapping of the individual coupons to positions within the virtually assembled wafer for semi-manual or automated pick-and-place operations. Interior pieces may all have straight edges as shown. In some implementations, the machine learning model can be trained to detect straight edges and measure the length of the edges to identify other coupons as adjacent neighbors. While the interior pieces are rectangular having different edge lengths in the illustrative embodiment, some pieces in general may have oblique angles between edges. If all the coupons are regular in the sense that they are rectangular, the model may be trained to first identify a first coupon, referred to here as a nucleus or nuclear coupon.Docket No.: 11123-1WOThe model may then fit adjacent pieces together first by finding mating straight edges between the nuclear coupon and the remaining coupons, where mating edges are adjacent cleavage edges. Here, the matching criterion may be edge length matches. Once an initial neighbor is found for a nuclear coupon, remaining edges may be similarly matched by edge length criteria. For rectangular coupons, edge length may remain the only matching criterion to coalesce other coupons around the nucleus by abutting matched edges.
[0039] For irregular coupons having oblique angles between edges, some edges of adjacent coupons may also subtend oblique angles. To find coupon matches that fit such adjacent edges, the model may search for coupons having a complementary oblique angle between two of its edges and find matches by complementarity of edges, in accordance with some implementations.
[0040] For example, a first coupon has a non-rectangular neighbor, whereby the two matched coupons form a nuclear pair. Here, a nuclear pair may function as a “nucleus” that may accelerate assembly of other coupons or fragments around the two initially paired coupons, due to more facile edge and / or texture matching as the assembly grows. The nuclear pair has an angle of 146 degrees at one of its vertices between an edge of the first coupon and an edge of the non-rectangular neighbor. The model may search for a single unmatched non-rectangular coupon that has a similar angle between two of its edges, viz., 146 degrees. Alternatively, in at least one example, if the model cannot find a single unmatched coupon having a 146 degrees comer angle, it may then measure angles of all comers of all coupons to find two unmatched coupons having comer angles that sum to 146 degrees. These two coupons may then be matched to fit the 146 degrees vertex of the nuclear pair. While angles and edge lengths may be matched, some uncertainty of the match may still exist. For example, while a match is found to fit the oblique angle of the nuclear pair, now forming a three-coupon nucleus, some ambiguity may exist because there may be a fourth coupon that could have fit the 146 degrees vertex.
[0041] As noted above, the coupons shown in Fig. 1A are not textured, therefore matching proceeds by matching geometric features, such as edge length and vertex angles. Texturing may enhance matching, as lithographic patterns or features that are visible to the naked eye, for example, may help to positively identify neighboring coupons and lift most if not all any ambiguity. For example, textured coupons may contain discontinuous portions of a textured pattern that is broken up by a cleavage edge, forming discontinuous portions of textures on each coupon. In cases where coupons are formed by manual cleaving of the wafer, it may be likely that textures are sectioned at random positions. Thus, no two couponsDocket No.: 11123-1WOmay have identical portions of a texture. In the case of textured coupons, searching for texture matches by combining respective texture portions, in addition to edge lengths and vertex angles, may result in less ambiguous matches than geometnc features alone can provide.
[0042] While manual wafer cleavage methods may result in a unique texture on each coupon, coupons produced by automated dicing methods, such as laser dicing or mechanical sawing may form rectangular and uniform chips or groups of chips, which may constitute coupons. Here, textures may not be unique to each coupon. In such a case, metadata, such as lithographically formed alphanumerical data that is unique to each coupon, may be relied upon to match coupons to reassemble the wafer.
[0043] While a combinatorial model or a deep learning model may be used to reconstruct the wafer from the coupons, once a virtual reassembly of the wafer is performed, an automated robotic pick-and-place tool may be used to reassemble the coupons into the original wafer, or an engineer may do so semi-manually by extended reality approaches. In at least one implementation, a pick-and-place tool may be commanded to pick up the coupons and position them according to the mapping between the disassembled coupons as shown in Fig. 1A and the reassembled wafer 100C (a more complete version than that shown in Fig.1C) determined by the machine learning model. In at least one implementation, a user (engineer, robot, or technician) employing extended reality tools, such as a computer screen or virtual reality glasses or goggles, may be guided by the machine learning model to manually reassemble the wafer by showing the mapping to the user.
[0044] Fig. ID shows a flowchart 150 that illustrates a high-level computation algorithm that a machine learning model may follow to characterize incoming coupons or whole wafers, or for re-assembly of whole wafers from coupons sectioned from a whole wafer for metrology measurements, in accordance with at least one example. Image data is taken from coupon image input 152 and wafer image input 154.
[0045] At operation 156, an image of a collection of coupons may be input to the system as coupon image input 152. In some implementations, coupon image input 152 is received by a machine learning module or by a machine vision module. The image, which may be an image of one or more coupons submitted to a metrology lab, is captured by a digital camera coupled to a computer workstation at the metrology lab. In some implementations, coupon image input 152 may comprise an image of pre-formed coupons or comprise an image of an incoming whole wafer that is sectioned into a collection of coupons by metrology engineers after receiving the incoming whole wafer.Docket No.: 11123-1WO
[0046] The individual coupons are identified in the image by the machine vision software, the machine learning model (module), or a combination of both. In at least one implementation, each individual coupon image may be evaluated by machine learning and / or machine vision software to classify each coupon in terms of geometric and texture characteristics. The coupon images are passed to operation 158.
[0047] At operation 158. data is generated regarding coupon characteristics. As shown in flowchart 150, individual coupons are classified as peripheral pieces along the perimeter of the wafer. The coupon may be further classified as having contours or straight edges, whether the coupon is a simple polygon, such as a rectangle. The classification also considers the number of comers. Comers may also be regarded as vertices, where vertex angles are measured and recorded. If curvature is present, the tangent angles are also measured and recorded.
[0048] Edge lengths are measured and recorded, where each edge of a coupon may be identified and associated with a length measurement. Curved edges of coupons are arcs that follow the perimeter of a circle that is proportional to the diameter of the original parent wafer from which the coupon was extracted. Thus, coupons possessing curved edges are peripheral pieces that fit along the perimeter of an imaginary circle of the same diameter as the parent wafer. At operation 158, the arclengths of the curved surface may be measured to be used during re-assembly of the original wafer from the individual coupons.
[0049] If texture is present on the coupon, the texture may be recorded and matched to texture on other coupons. The texture may be shared across coupon cleavage lines, for example, and complex texture matches may be made by a well-trained machine learning model. Data related to matches may be used further downstream in flowchart 150 when reconstructing whole wafers.
[0050] At operation 162, matching functions may be invoked to reconstruct a whole wafer to reconstruct the wafer according to its incoming state image. The match functions may obtain the w afer image from wafer image input 154. An image of an incoming w afer may be a high-resolution image showing texture detail. The match functions may perform matching operations that include edge length matching, complementary edge angle matching and texture matching. Texture matching, if texture is present, may resolve ambiguities arising from multiple edge length or vertex angle matches, for example, when fitting coupons together for wafer reassembly.
[0051] In some examples, metadata 160 may optionally be utilized as input data to help resolve w afer assembly ambiguities in cases where texture is not visible, or coupons areDocket No.: 11123-1WOuntextured. Some metadata may include scribed alphanumeric markings on coupons, including numbers, codes, and comer or edge marks to help find adjacent neighbors. Other metadata, such as coloration or fluorescent markers, may be present on some or all coupons, and may also be utilized during the wafer reassembly process to assist in resolving ambiguous edge length matches, for example.
[0052] Operation 162 may comprise one or more of matching suboperations 164, 166, and 168. The matching functions contained in suboperations 164, 166, and 168 may import geometric and texture data for each coupon included in coupon image input 152 from coupon classification measurements and assessments having been performed at operation 158.Suboperation 164 enables cost matrix analysis and formulates decisions for fitting coupons together based on edge length and angle computations, as well as optional use of metadata 160 for resolution of match ambiguities where texture matches cannot be done or are also ambiguous. Some cost matrix decisions may include where to fit edge coupons having curvature. Here, a virtual circle having the same diameter as the parent wafer is constructed as a virtual guide to fit peripheral pieces together. Wafer reassembly may begin with the outermost coupons that fit along the circle perimeter.
[0053] Suboperation 166 comprises the functions that enable reassembly of the wafer. Here, individual coupons are fitted into the growing wafer subassembly according to matching calculations performed in suboperation 164.
[0054] Suboperation 168 comprises validation functions that perform checks on the last assembled coupon to ensure that edges and texture, if present, match neighboring coupons in the wafer subassembly. After validation is complete for a particular coupon, the match functions reiterate their sequence for the next coupon.
[0055] Fig. 2A shows a collection 200A of textured coupons that belong to a wafer. The coupons of collection 200A may be textured or blank. Collection 200A comprises 21 coupons having both regular and irregular shapes. In at least one implementation, the model may analyze each coupon for shape characteristics and other features, such as texturing of each coupon. In at least one implementation, a high-resolution imaging system is employed to record texture details of each coupon.
[0056] Fig. 2B shows a progressive step 200B in the virtual reassembly of the coupons of collection 200 A. In at least one implementation, coupons having a curvature, such as coupon 202, are identified and distinguished from the coupons having only straight edges by changing coloration or color shading of the coupons having curvature. In the illustrative example, the curved edges of the 12 coupons identified as having curvature are assessed forDocket No.: 11123-1WOarclength as well as straight edge lengths to determine overall shapes. In some implementations, the coupons may be virtually numbered or labeled in some other manner by the model.
[0057] Fig. 2C shows a progressive step 200C in the virtual reassembly of the coupons of collection 200A. In at least one implementation, coupons may be numbered virtually by the model. In some implementations, virtual labelling of each coupon may be performed by the model if other metadata had not been provided to track each coupon. In some implementations, virtual labelling, such as adding numbers to each coupon, may be performed to augment the metadata already present.
[0058] In at least one implementation, the model may begin virtual reassembly of the wafer by determining a circle having the diameter of the original wafer. In some examples, the model may perform an evaluation of best paths with a circle validation function. The best paths may be a determination of matching arcs and straight edges from the collection of coupons. The circle may be arranged by starting with a first piece that may be akin to a cornerstone for construction of a brick wall or building process. Other coupons follow in the assembly process according to arc matching and fitting the arcs along the circle.
[0059] In Fig. 2C, a completion of the virtual placement of the outer coupons having a curved edge around the circle is shown. The order of coupon assembly is not in a particular order of the identification numbers. For example, moving clockwise around the circle, coupons 2. 7, 14. 20. 10. 16. 9, 0, 3, 17, 13, and 19 have been matched and placed at the periphery. Unmatched coupons 1, 6, 4, 5, 8, 11, 12, 15, and 18 have all straight edges and are to be evaluated for edge matching to alreadymatched coupons at the periphery of the circle as well as the unmatched coupons.
[0060] Fig. 2D shows a progressive step 200D in the virtual reassembly of the coupons of collection 200 A. In at least one implementation, the model may assess available edges for matching. For example, the model may assess the edges available from the matched peripheral coupons (e.g., coupon numbers 0, 2, 3, 7, 9, 10, 14, 16, 17, 13, 19, and 20). In Fig.2D, edges that are as yet unmatched are identified by the machine learning model are indicated by the bars 204 and arrows 206 pointing away from the identified edges toward the interior of the circle. These edges are recognized as available for further matching. The model may create minimum cost paths from the identified edges (e.g., those edges identified by bars 204).
[0061] In at least one implementation, the model may evaluate candidate coupons by first finding close matches between the peripheral coupons anchored on the circle perimeter andDocket No.: 11123-1WOthe remaining unmatched coupons (e.g., coupon numbers 1, 4, 6, 5, 8, 11, 12, 15, and 18), by evaluation of edge lengths and adjacent angles, forming a shortlist. In at least one implementation, in addition to edge length and adjacent angle evaluations, the model evaluates the textures on the shortlisted coupons to help positively match coupons. The model may search for and find complementary textures between sets of coupons. Each set may include a peripheral coupon and one or more unmatched coupons. For example, a texture on a particular peripheral coupon may be contiguous with textures on one, two or more unmatched coupons, being separated by cleavage lines. The continuity of the textures across cleavage lines made between coupons to break up the original wafer can provide unambiguous fingerprint information to match adjacent coupons during the reassembly process.
[0062] In some implementations, textures may not be unique on each coupon but may be repeated on several coupons within collection 200A or repeated on all coupons. For example, the texture may be visible metallization pattens replicated on all chips on the original wafer. Where coupons are produced by a dicing apparatus, such as a laser or mechanical saw, the coupons may comprise individual whole chips or groups of whole chips. In such an instance, texture evaluation may not definitively match coupons, as textures remain intact and are not split up. Manual cleavage of the original wafer may produce coupons that are more amenable to matching by texture evaluation. Manual cleaving can produce cleavage lines that extend across individual textures, randomly breaking them across a pattern, producing coupons that have portions of the same pattern or texture that can be matched.
[0063] Fig. 2E shows a progressive step 200E in the virtual reassembly of the coupons of collection 200A In the illustrative example, the model evaluates edge lengths and textures as all remaining unmatched coupons are rectangular. The model measures edge lengths for unmatched edges of peripheral coupons 208, 210, and 212 (e.g., peripheral coupons 13, 17, and 19, respectively), shown by edge measurement bars 214, 216, and 218, respectively.
[0064] Subsequently, in at least one implementation, the model performs edge measurement of the unmatched coupons and finds a best match. For example, edge length measurement bars 220 and 222 of coupon 6. and edge length measurement bars 224 and 226 of coupon 4 match edge length measurement bar 218 of peripheral coupon 212 (coupon 19). A vertical match is made by chaining coupons 4 and 6, and summing edge lengths denoted by edge length measurement bars 216 and 228, respectively. Edge length measurement bars 216 and 228 sum to the combined edge length of edge length measurement bars 214 and 215 of peripheral coupons 208 and 210. respectively. In this example, the edge of coupon 4 associated with edge length measurement bar 216 may be matched to the edge of coupon 210Docket No.: 11123-1WOassociated with edge length measurement bar 215. The edge of coupon 6 associated with edge length measurement bar 228 may be matched to the edge of coupon 208 associated with edge length measurement bar 214. A certain ambiguity exists, as either coupon 4 or coupon 6, now chained together, may be placed in the slot directly below peripheral coupon 212 (coupon 19), followed by the second coupon to fill the slot. To resolve this ambiguity', the model may evaluate textures on peripheral coupons 208, 210, and 212, as well as unmatched coupons 4 and 6. The model may then find the best match based on the texture matches as well as edge length matches.
[0065] Still referring to Fig. 2E, the model proceeds to insert unmatched coupons in the open slots of the progressively assembled wafer 232. From edge length and texture evaluations, unmatched coupons are chained and inserted into open slots. For example, coupons 4 and 6 are inserted into the slot bordered by peripheral coupons 208, 210, and 212, where coupon 4 is placed directly below peripheral coupon 212 according to a texture match. Further matches are found by chaining together coupons 11 and 15, as well as coupons 1 and 5. These coupons are inserted into the open slots based on edge length matching and texture matching. As can be seen in the figure, because of texture matching, coupon 11 is placed directly below peripheral coupon 234 (coupon 2). While coupon 11 is found to fit below peripheral coupon 234, the orientation of coupon 11 is ambiguous. The orientation ambiguity' of coupons 11 and 5 may also be resolved by texture matching
[0066] Similarly, the chain comprising coupons 1 and 5 is placed below peripheral coupon 236 (coupon 7). Here, two edges of coupon 5 match the open edge of coupon 236, creating an ambiguity'. The ambiguity may be resolved by finding an exact match of the edges based on texture matching. Thus, the question of which of coupons 1 or 5 is matched to peripheral coupon 236 is resolved by the texture match, as well as is the orientation ambiguity of coupon 5. Wafer virtual reassembly has progressed to include all but wafers 12 and 18, as shoyvn in the progressive step 200F in Fig. 2F.
[0067] Figs.2G and 2H show progressive step 200G and 200H, respectively, in the virtual reassembly of the coupons of collection 200 A. Due to the width of coupon 234. a remaining open slot is adjacent to the chain comprising anchored coupons 11 and 15.Coupons 12 and 18 remain unmatched. As shown in Fig. 2G, the model may chain these two coupons together based on edge length measurements and texture matching. For example, edge “a” of unmatched coupon 12, indicated by edge length measurement bar 238, and edge “b” of unmatched coupon 18, indicated by edge length measure bar 240, may sum toDocket No.: 11123-1WOcombined edges of anchored coupons 11 and 15, indicated by edge length measurement bar 242.
[0068] As shown in Fig. 2H, coupons 12 and 18 are chained to form a chain 244 having a combined long edge comprising edges “a” and ‘"b”. Texture evaluation and verification determine the orientation of the induvial coupons 12 and 18 in the chain (e.g., determination of which edges of coupons 12 and 18 match textures). Subsequent texture evaluations may be conducted iteratively to resolve the orientation ambiguity to fit the chain into the open slot by iterative evaluation of textures on anchored coupons 1, 5, 11, and 15, as well as coupons 12 and 18 in the chain 244.
[0069] Fig. 21 shows a step 2001 in the virtual reassembly of the coupons of collection 200A. Here, progressively assembled wafer 232 is complete. Positions for all coupons have been found by the model, virtually restoring the original wafer. Depending on the number of coupons, this process may take seconds to minutes. The model performs the matching by first evaluating all available edges of coupons; creating minimum cost paths from edges of peripheral coupons (once anchored to the circle), and iterating matching of individual chains by edge length matching and texture matching.
[0070] For untextured coupons, the model may rely strictly on edge length (and angle) matching. In some implementations, other metadata may be provided, such as pre-mapping of coupons by scribing numbers or alphanumeric data on each coupon. To move matching ambiguities, edges, or comers of adjoining coupons may be labeled by scribing as well to indicate adjacent edges. These types of metadata indicia may be useful for coupons produced by automated dicing of a wafer, such as by singulation of individual chips, producing uniform coupons, or for groups comprising multiple individual chips that may serve also as coupons.
[0071] For metrology of pre-sectioned coupons, the deep learning model may be adapted to verily identity of coupons as received, as well as verily that any coupons fragmented after reception for purposes of SEM and TEM evaluations are returned intact by verification that there are no missing fragments of coupons fragmented for micrographic analysis. For example, coupons may be submitted to a metrology lab from a process development lab. A sequence of identification and the machine learning model
[0072] Fig. 3 depicts a carrier 300 containing pre-sectioned coupons 302, 304, and 306. Carrier 300 is to be submitted to a metrology7lab. Carrier 300 may have a radio frequency identification (RFID) tag 308 attached for rapid identification.
[0073] Fig. 4A illustrates an initial capture image 400 of the contents of carrier 300 as received by a metrology lab from an originating source, where capture image 400 may beDocket No.: 11123-1WOproduced by the machine vision software interfaced by a camera. In at least one implementation, machine vision software identifies pre-sectioned coupons 302, 304, and 306. For positive identification, the machine vision software may draw boxes 402, 404, and 406 around the coupons. The machine vision software may provide algorithms that measure edge dimensions and comer (e.g., vertex) angles, and compute the geometric area of each coupon to a high level of accuracy. In some implementations, capture image 400 is a high-resolution image that can capture texture details. An example of a high-resolution image includes images containing 30 to 50 megapixels.
[0074] In at least one implementation, the machine vision software may combine the measured data with metadata that may be input into a data management module to keep track of the individual coupons, such as by scanning encoded tag 308 or barcode on carrier 300. Other metadata may be provided. For example, coupons 302, 304, and 306 may have alphanumeric information scribed on their surfaces. The RFID or barcode and / or alphanumeric information may contain identifying and tracking data, such as coupon number, wafer number, process run number, date and time information, process chamber number, etc. The data may be encoded to be read only by the metrology lab and originating source.
[0075] Other metadata may also be present. For example, visible light coloration may be present on portions of the coupons, and this coloration may be captured in capture image 400. The coupons may have portions having tethered fluorophores, which fluoresce in the red, orange, yellow or green portion of the visible light spectrum, for example, under blue to ultraviolet light illumination. Such fluorescence may be captured by the machine vision software by illuminating the coupons under an appropriate wavelength of light.
[0076] Fig. 4B illustrates a capture image 420 of regrouped fragmented coupons placed in carrier 300 after metrology has been conducted. General locations of coupons 302, 304, and 306 are indicated by dashed boxes. Here, coupons 302, 304, and 306 have been fragmented into smaller pieces by manual cleavage for SEM, TEM, and other metrologies. For example, coupon 304 has been sectioned into fragment 408, fragment 410 and fragment 412 to investigate regions of the coupon. Coupon fragmentation may be useful to reduce the size of the coupon for mounting in a SEM or TEM. A second image is taken after the metrology is complete. The machine vision software is configured to identify' the fragments and determine if any fragments may be missing. In the illustrative example, the machine vision software has identified the individual fragments and drawn boxes around each one. The machine vision software may also measure the geometric areas of each fragment. As anDocket No.: 11123-1WOexample, areas of fragments 408 - 412 may be measured and summed to compare to the measured area of coupon 304 prior to cleavage.
[0077] As the fragmented coupons 302, 304, and 306 are to be returned to a process development lab, fragments, such as fragments 408 - 412 of coupon 304, are regrouped and placed in the general location of the original coupons. If any fragments are missing, the software is configured to alert users accessing the software. This process may be particularly useful if fragments are small and difficult to see by the naked eye. For example, fragments having edge dimensions of 3 mm or less (due to fracturing, for example) may be difficult to see. Such a system can provide a safeguard against incidences of loss of fragments and generate notification of fragment loss before the coupons are returned.
[0078] Fig. 5 illustrates a flowchart 500 that recaps the procedure for verification of coupons as received and for verification of coupons for return to their source, according to at least one implementation.
[0079] At operation 502, a carrier, such as carrier 300, is brought into a metrology lab. The carrier may have a RFID tag or barcode, such as encoded tag 308, attached. The RFID tag or barcode may contain a significant amount of metadata, such as origin, entity of origin (e g., name of unit, company or organization that originated the coupons), date and time, type of process, a process number, and other data that may be useful for verification purposes.
[0080] At operation 504. a metrology lab (or other receiving facility) records information and metadata from the RFID tag and / or barcode. Here, the originating entity, or the entity that supplies the carrier, is identified and recorded. The metrology lab may be an outsourced entity, separate from the originator of the coupons. For example, the metrology' lab may be a vendor of such services, whereas the originator may be an originating entity', such as a fabrication house or facility that contracts with the metrology lab to perform metrology’ services. In such a case, the metrology lab records the entity' information to keep track of the originating entity7of the coupons.
[0081] At operation 506, metadata contained within the RFID tag or barcode information or other metadata, such as scribed encoded data associated with the coupons may also be identified and stored for coupon verification. Other metadata may include non-encoded alphanumeric information scribed on the coupon, or colored and fluorescent films, as described above. The metadata may be stored on a local or networked server to be retrieved at any time by authorized parties.
[0082] At operation 508. an image of the coupons received is recorded by a camera linked to a machine vision software. The coupons may be imaged as they appear in theDocket No.: 11123-1WOcarrier, as shown in Fig.4A. During imaging, the machine vision software may isolate each coupon by drawing boxes around them, as also shown in Fig 4A. The image may be recorded and stored on a local or networked server to be used for identification purposes and a reference for verification of the coupons or fragments thereof after metrology is completed and the coupons are being prepared for return to the originator.
[0083] At operation 510. the machine vision software and / or other software modules may virtually identify each coupon and associate metadata with each. At this stage, the machine vision software may proceed to operation 512, where the machine vision software may scan each coupon, then measure and record its area, edge lengths, and any vertex angles. Such data may be stored on the local or networked server and may be used to compare with one or more images taken after metrology is completed. During metrology, the coupons may be fragmented to mount in a SEM and / or TEM. The machine vision software and / or other software modules may compare the fragments to the original coupons to verify fragment grouping and detect any missing fragments.
[0084] At operation 514. a post-metrology image is taken of the coupons or their fragments. Fragments may be regrouped and placed back in the earner in the locations of the original coupons. Fragments may be arranged to reconstruct the original coupons, as shown in Fig. 4B. A record of the post-metrology image may be stored on the local or netw orked server for reference.
[0085] At operation 516. the machine vision software may isolate each fragment by drawing boxes around them, as shown in Fig. 4B. The areas of the fragments may be determined in this operation, as well as edge lengths and angles. At operation 518, the machine vision software and / or other software modules may combine fragments to find fragment groupings based on fragment areas and areas of original coupons. At this stage, the machine vision software may determine if any fragments are missing based on area comparisons.
[0086] Fig. 6A illustrates an architecture for a machine vision network 600, in accordance with at least one implementation. The network may comprise a local computer workstation 602 having a machine vision software module stored in internal storage.Computer workstation 602 may be located within the metrology facility. In at least one implementation, computer workstation 602 is coupled to a local network 604 and a remote server 606 that may be a remote server located on cloud server 608 (or on a local server 607 located on local network 604). Remote server 606 may be a cloud-based artificial intelligence (Al) server that comprises the machine learning model. Local network 604 may also storeDocket No.: 11123-1WOmachine vision software on local server 607. In at least one implementation, computer workstation 602 is part of a workstation located in the metrology facility where a camera 610 is coupled to computer workstation 602. Camera 610 may be a high-resolution digital camera. For example, camera 610 may have a resolution of 30 to 50 megapixels. While such a resolution may not be necessary for capturing large features, such as the overall dimensions of coupons and fragments, a high pixel resolution may enable camera 610 to resolve fine texture detail.
[0087] In at least one implementation, the machine learning model employs standard computer vision and deep learning algorithms like convolutional neural networks (CNN), generative adversarial network (GAN), and reinforcement learning to quickly find wafer or coupon re-assembly solutions. Such an approach may generate solutions, akin to jigsaw puzzle solving, more rapidly than conventional combinatorial approaches. The GAN model may be trained with many datasets that include hundreds of randomly fragmented wafers, each fragmented into pluralities consisting of, for example, 30 to 1000 coupons, depending on wafer diameter, having small or large numbers of irregular shape patterns. The machine learning model may be trained to match adjacent coupons based on coupon shape analysis that includes measurement of lengths of curvature arcs (by tangents) to identify outer pieces, edge length, and edge angles. In addition to assembling coupons based on coupon shape, coupon texture may also be included to enhance finding adjacent matches when many coupons in a plurality of coupons from a wafer have similar shapes. If metadata are present, such as scribed alphanumerical markings on the coupons, specific lithographic patterns, reflective or fluorescent coloration or patterns, etc., these can also be considered.
[0088] In at least one implementation, the local network 604 may comprise a local server. In at least one implementation, remote server 606 comprises an Al server 612 that stores a machine learning model module, and database storage 614. Other cloud services may include module 616 specialized for interfacing a RFID reading device and / or barcode reading device, for example. In at least one implementation, a reading device 618 is coupled to computer workstation 602 and is configured to read reading encoded tags, such as reading encoded tag 308 configured as an RFID tag, can provide metadata to module 616. Reading device 618 may include a barcode reader. In some implementations, a separate barcode reader (not shown) may also be coupled with computer workstation 602. While RFID and barcode technologies, including QR-code tracking, are used in the examples of encoded tag readers described here, other classes of technology that are similar to RFID and barcode technologies, near-field communication (NFC) may also be included, such as Bluetooth LowDocket No.: 11123-1WOEnergy, wi-fi, ultra-wideband (UWB) and global positioning system (GPS) tracking may also be employed for reading encoded metadata.
[0089] In at least one implementation, cloud services also include a general storage 620 (e.g., a blob storage) based in the cloud that may store documents and metrology documentation concerning coupons and or wafers received and returned. For example, incoming identifying data for coupons received may be stored in general storage 620, including documentation sent by an external facility, such as a process development lab. Notes taken on coupons received by metrology personnel may also be stored in general storage 620. Other documents may be stored in general storage 620, including metrology7data, which may comprise SEM and TEM micrographs.
[0090] In at least one implementation, computer workstation 602 is coupled to a robotic pick-and-place tool 622. In at least one implementation, computer workstation 602 is optionally coupled to an extended reality system 624. In at least one implementation, extended reality system 624 comprises augmented reality, virtual reality or extended reality software tools to guide a human user, such as a technician or engineer, to envision a mapping made by machine learning module stored on Al server 612 between disjoined coupons and fragments and virtually assembled semiconductor wafer. The mapping may be visualized on a computer screen or by using extended (e.g., virtual or augmented) reality headset including goggles or glasses, for example. The user may point to an unmatched coupon or fragment that may be laid out on a table and an image of the virtually reassembled semiconductor wafer. For example, the user may point to a particular unmatched coupon, and the virtual reality' system 624 may draw an outline of the coupon in its assembled position in the growing wafer assembly.
[0091] Fig. 6B illustrates a data flow sequence flowchart 630 for machine vision network 600, in accordance with at least one example. Flowchart 630 summarizes a sequence by which data is moved from node to node within machine vision network 600.
[0092] At operation 632, an encoded tag containing metadata related to the wafer or coupons, such as Encoded tag 308, is read by an encoded tag reader such as an RFID reader (e.g., reading device 618 configured as an RFID tag reader) of an RFID tag on an incoming coupon or wafer container, such as carrier 300 (Fig.3). Alternatively, a barcode may be present on the container, either alone or with an encoded tag, for example. The reader may be at computer workstation 602, for example at a metrology lab or facility, where incoming coupon containers are scanned by the reader. The reader may communicate with reader module 616 on cloud server 608, where metadata read from an RFID tag may be collected. ADocket No.: 11123-1WObarcode reader or scanner (e g., such as barcode reader 808, see below) may be present at computer workstation 602 to read incoming barcodes. A similar dataflow sequence may be operational for a barcode reader. In some examples, other types of encoded tags and encoded tag readers, such as those mentioned above, may additionally or alternatively be used in a similar dataflow sequence.
[0093] At operation 634. coupon metadata read by reading device 618, may also be retrieved by computer workstation 602. Information retrieved may be used for verification purposes on outgoing containers. Thus, the dataflow is bidirectional between computer workstation 602 and module 616 on cloud server 608.
[0094] At operation 636. dataflow from computer workstation 602 includes uploading images of incoming coupons and / or wafers to general storage 620 that may be located on cloud server 608, as well as images of outgoing coupons and wafers after metrology sen-ices are finished. A state may be created for each group of incoming coupons that have arrived together in a container, or for a whole wafer. The state is a collection of metadata information that may have been extracted from module 616, containing the entity of origin (e.g., entity of origin may be an internal unit or external enterprise that is supplying the incoming coupons or wafer to the metrology lab), date and time stamps, process codes, metrology data and comments, incoming notes, coupon IDs, image identification numbers, and any other information that may be useful to identify coupons or original wafers.
[0095] At operation 638. updated state information is sent to database storage 614 in preparation for coupon image evaluation by the machine learning module stored on Al server 612. Here, the updated state information may include metrological data, new images of sectioned coupons, etc. The general storage 620 as well as database storage 614 may be accessed by the machine learning module to retrieve the images, both for the as received incoming coupons or whole wafer, and post-metrology image, for determining coupon areas, edge lengths, angles, etc. For reassemblv of whole wafers, as shown in Figs. 2A-2I, the machine learning model, embodied in machine learning module stored on Al server 612, may evaluate any geometric features of fragments. Fragments may be first analyzed by the machine learning module to find best matches of edges in conjunction with edge length and comer or vertex angle measurements. Vertex angle matches may be evaluated secondarily, and textures, if present, may be evaluated thirdly for potential matches between fragments.
[0096] Regarding texture, portions of a texture may span across a cleavage line(s) between two or more adjacent fragments. A texture match may be defined as a match between texture portions on two or more fragments that complete the entire texture. TextureDocket No.: 11123-1WOmatching may help to resolve ambiguities in edge length matches, for example, as multiple candidate fragments may match a particular fragment by edge length matching. However, these candidates may be false matches because other edges of these candidates may not match edges of neighboring fragments that are already matched to the particular fragment. Mismatches of vertex angles and textures between candidate fragments and already matched adjacent and near-adjacent neighboring fragments (adjacent to the particular fragment) increase certainty of finding a valid match by quickly eliminating many candidate fragments (e.g., only having an edge match with the particular fragment) from the pool of possible matches.
[0097] At operation 640. images that had been uploaded to the cloud-based general storage 620 are retrieved by the machine learning module stored on Al server 612. As noted previously, outgoing (e.g., post-metrology) images of coupons may be compared to incoming images and verified for positive coupon identification. Where coupons have been sectioned into fragments, these are measured for area, and regrouped fragments are verified as belonging to an original coupon. If any fragments are missing, the machine learning module may send back a report (see below) indicating the position, shape, and size of the missing fragment(s), so that personnel in the metrology unit may be notified and act. These actions may also apply to reassembled whole wafers, where the reassembled wafer may be assessed for missing coupons or fragments thereof.
[0098] At operation 642. state information is retrieved by the machine learning module from database storage 614. This information is to be used in a report (see below) that is to be prepared by the machine learning module on Al server 612.
[0099] At operation 644, the machine learning module generates the report referred to above that is to be sent back to the computer workstation 602 from Al server 612. The report may describe comparisons between incoming and outgoing coupon images, report discrepancies in area computations for incoming and outgoing images, indicating missing fragments, and other necessary information that may be archived and sent back to the originating entity as well as retained by the metrology facility.
[0100] At operation 646. the report is fetched as needed by computer workstation 602 from database storage 614. Again, the report may be retained by computer workstation 602.
[0101] Fig. 7 illustrates a high-level block diagram and flowchart 700 of the machine vision operations and interactions with Al server 612 presented in Fig. 6A and 6B, in accordance with at least one implementation. In the block diagram, machine vision 702 is represented as machine executable instructions that run on a workstation, such as computerDocket No.: 11123-1WOworkstation 602, that is local to the metrology facility. As shown in Fig. 6A, computer workstation 602 is in communication with remote server 606. Al server 612 stores and runs an Al software engine that drives the machine learning model (also stored on Al server 612). For increased security, machine vision 702 may be networked on a local area network (LAN) using ethemet.
[0102] Referring to flowchart 700, operation 704 comprises obtaining an image of incoming coupons (as received). The image may be taken by a camera having high pixel resolution. For example, the camera may possess 40-megapixel optics. In some implementations, an image of frontside and backside surfaces of coupons may be obtained. For example, if metadata has been scribed on the backside of the coupons, it may be captured by photographing the backside of wafers. To this end. the coupon carrier may be transparent, or coupons may be removed from the carrier and placed on transparent table. A first camera above and a second camera below the tabletop may capture images of both top and bottom.
[0103] At operation 706, the images are preprocessed to find shapes, where detection of coupons is enabled for coupon detection at operation 708. Here, coupons are detected by the machine learning module having been trained on many data sets to learn to recognize coupons of various shapes and sizes. In at least one implementation, the machine learning module located on Al server 612 is configured to detect and isolate coupons and obtain geometric data, such as area, edge lengths, and angles. The machine learning module may draw boxes around coupons, as shown in Figs. 4A and 4B, for example, to display to users for verification.
[0104] At operation 710, the machine learning module may compute areas of detected coupons. The geometric data may then be stored either in cloud storage (e.g., database storage 614) or general storage 620. Optionally, reports concerning geometric data analyses may be sent back to the computer workstation to be collected as a readable record that may be stored on local or cloud storage. The report data may be optionally displayed to personnel. The computed coupon areas may be used to compare with outgoing coupons after metrology is complete. Incoming coupons may be sectioned to produce smaller fragments for mounting within a SEM or TEM, for example. When regrouping fragments, the fragments are traced to particular coupons and regrouped, for example by placing the fragments of each coupon back in the approximate location of the incoming coupons within the carrier, as shown in Fig. 4B.Outgoing coupons or fragments may be imaged and assessed by the machine learning module to determine if any fragments are missing. This may be accomplished by recomputing theDocket No.: 11123-1WOarea of the ensemble of fragments and comparing to the original coupon area determined for the incoming coupons.
[0105] Fig. 8 illustrates a hardware block diagram 800 of computer workstation 602 for receiving and imaging incoming as well as outgoing coupons. In at least one implementation, processor 802 is configured to receive data inputs from a camera 804 and an RFID reader 806, which can include a barcode reader as well as other technologies enumerated above. Processor 802 may optionally be configured to receive data input from a barcode reader 808. Processor 802 may also be configured to output control signals to camera 804, RFID reader 806, and barcode reader 808 for operation of these devices.
[0106] In at least one implementation, processor 802 outputs command signals to a light emitting diode (LED) driver 810, which may be coupled to LED banks 812 and 814. which may be positioned over and under, respectively, a carrier containing coupons or coupons removed from carrier and placed on a transparent table. In at least one implementation, processor 802 is coupled to a display 816 (e.g., computer workstation 602) for displaying images of coupons and fragments. Other information, such as metadata, may be displayed as well to human users of the workstation for verification purposes.
[0107] Fig. 9A illustrates an isometric view of an imaging station 900 that may be included in a wafer metrology7facility7. Texture capturing during coupon and wafer imaging may be complicated by lighting artefacts. For example, imaging wafers and coupons may be prone to artefacts brought on by diffraction of light from wafer and coupon surfaces, producing colors or rainbows that may obscure texture. In addition, these artefacts may be view dependent. In at least one implementation, imaging station 900 can minimize or eliminate such artefacts.
[0108] Imaging station 900 comprises light diffusing table 902. overhead reflector 904, over light diffusing table 902, light source 906, and camera 908. In at least one implementation, light diffusing table 902 comprises a transparent material, such as acrylic, lucite, polycarbonate, soda-lime glass, or other suitable materials that may scatter light and conduct it internally, causing a diffuse glow off of light diffusing table 902. Light diffusing table 902 may interact with white light from light source 906, located below light diffusing table 902. Light emanating from light diffusing table 902 in the vertical direction may reflect from overhead reflector 904 to illuminate wafer 910 with diffuse light so that no reflections are made on the surface of wafer 910, that rests on light diffusing table 902. While a whole wafer is shown in the illustrative embodiment, it is understood that one or more coupons may be equally represented.Docket No.: 11123-1WO
[0109] Camera 908 is coupled to computer workstation 602 shown in Fig. 6A and described above. Camera 908 may be tilted at an angle to capture images of wafer 910 without being directly above wafer 910 causing some blockage of light and possibly reflecting from the surface of wafer 910 in captured images. A secondary light source 912 may be present as well.
[0110] Fig. 9B illustrates a frontal isometric view of an alternative imaging station 920 for a wafer metrology facility. In at least one implementation, imaging station comprises a working surface 922, such as a tabletop or desktop. A w afer imaging area 924 and coupon imaging area 926 may be arranged on w orking surface 922 for positioning under a w afer imaging camera 928 and a coupon imaging camera 930. In some implementations, both wafer imaging camera 928 and coupon imaging camera 930 are supported on shelf 932, which is held above working surface 922 by side panels 934 and 936. In some implementations, wafer imaging camera 928 separately or in combination with coupon imaging camera 930, is / are equivalent to camera 610 in Fig. 6A, and described above. In some implementations, imaging station 920 comprises RFID readers 938 and 940 are attached to side panels 934 and 936, respectively, and positioned in proximity of wafer imaging area 924 and coupon imaging area 926.
[0111] In at least one implementation, imaging station 920 comprises a display monitor 942 between wafer imaging camera 928 and coupon imaging camera 930 on shelf 932. In some implementations, display monitor 942 is a touch screen monitor for the operation of cameras and a projector 944. In at least one implementation, projector 944 is adjacent to wafer imaging camera 928, above wafer imaging area 924. Display monitor 942 may display images of whole wafers and coupons after cleaving. As will be shown below, display monitor 942 may display virtual wafer assembly images, metadata read from RFID tags on earners, and step-by-step instructions for cataloging wafers, coupons, fragments and how to reassemble wafers from pieces. In addition to projected wafer reassembly images displayed on display monitor 942, virtually reassembled wafers are projected from projector 944 onto wafer imaging area 924 to guide reassembly. It is noted that in addition to or in place of display monitor 942, in at least one implementation an extended reality (ER) headset, including ER goggles, may be employed by a user to visualize the data displayed. Extended reality may include any of virtual reality, augmented reality' and mixed reality software models.
[0112] Fig. 9C illustrates a rear isometric view of alternative imaging station 920. In the illustrative implementation, a controller 946 may be mounted on the rear of display monitorDocket No.: 11123-1WO942. Controller 946 may be a compact computer system that is embodied by computer workstation 602 in Fig. 6A and described above.
[0113] Figs. 10A - 10F illustrate a sequence of an operation 1000, encompassing operations 1000A-1000F whereby an incoming wafer is received, imaged, sectioned into coupons, and coupons are imaged and stored in carriers having RFID tags that are recorded into a database for later identification. The identification is to be used for reassembling the wafer, as will be shown in a sequence illustrated in Figs. 11A and 11B. Operations 1000A-1000F of operation 1000 are shown to employ hardware and software of imaging station 920, which is illustrated in a top-frontal 3D view in the figures.
[0114] At operation 1000A shown in Fig. 10A, wafer 1002, which may be an incoming whole wafer to a metrology facility, for example, is placed in wafer imaging area 924 after transfer from wafer carrier 1003. In at least one implementation, wafer carrier 1003 may carry a RFID tag that is read by RFID reader 938. Metadata from the RFID tag may be displayed on display monitor 942 as identification data in region 1005. Wafer 1002 is imaged by wafer imaging camera 928 from above, capturing a plan view of wafer 1002. The field of view 1004 of wafer imaging camera 928 is indicated by the dashed circle. In at least one implementation, imaging of wafer 1002 may be initiated by tapping the region 1006 on the touch screen of display monitor 942, shown in an enlarged view', whereby region 1006 is a virtual button indicating the acquisition of an image is performed by pressing on region 1006. In some implementations, a mouse or other pointing device may be employed to activate region 1006 and other portions of the touchscreen of display monitor 942 in addition to (or in lieu of) the activation by touch. In at least one implementation, an image 1008 of wafer 1002 may appear on display monitor 942. While not shown, it is understood that the same visual data may be displayed in extended reality tools including an extended reality headset.
[0115] At operation 1000B shown in Fig. 10B, wafer 1002 is sectioned into multiple coupons 1010. Coupons 1010 may be imaged by wafer imaging camera 928 by pressing again on region 1006 on the touchscreen of display monitor 942. In at least one implementation, an image 1012 of coupons 1010 is displayed on the screen of display monitor 942, confirming for the user that the image 1012 has been recorded by the software.
[0116] At operation 1000C shown in Fig. 10C, a coupon carrier 1014 is transferred from a stack 1016 of coupon carriers to coupon imaging area 926. In some implementations, coupon carrier 1014 is equipped with an RFID tag (not shown) containing metadata, for example, as described above. The RFID tag may be read by RFID reader 940. Such metadataDocket No.: 11123-1WOmay be used for identification purposes. The metadata may or may not be displayed on display monitor 942.
[0117] At operation 1000D shown in Fig. 10D, groups of coupons may be stored in carriers, such as carrier 1014. In the illustrative example, a first coupon group 1018 is shown in the process of transfer from wafer imaging area 924 to carrier 1014.
[0118] At operation 1000E shown in Fig. 10E, coupon group 1018 has been placed in coupon carrier 1014. A user may activate coupon imaging camera 930 by touching or clicking on region 1006 of display monitor 942. A field of vision 1020 of coupon imaging camera 930 is also shown by the dashed circle around coupon carrier 1014. In at least one implementation, an image 1022 of coupon group 1018 in coupon carrier 1014 may be shown on display monitor 942.
[0119] At operation 1000F shown in Fig. 10F, individual coupons 1010 or groups of coupons 1010 may be transferred to coupon carriers such as coupon carrier 1024, taken from coupon carrier stack 1016. In some implementations, an RFID tag on coupon carrier 1024 may be read by RFID reader 940 to garnish metadata on the RFID tag. Such metadata may be an identification indicator, for example. For instance, coupon group 1026 is transferred from its position 1028 on wafer imaging area 924 to coupon carrier 1024. In at least one implementation, a user may capture an image of coupon group 1026 seated in coupon carrier 1024 by activating coupon imagining camera 930, as described above (e.g., by tapping or clicking on region 1006). An image 1030 of coupon group 1026 may appear on the screen of display monitor 942, indicting to the user that the image is recorded by the system software.
[0120] Operations 1000A-F may be repeated until all of coupons 1010 are placed in coupon carriers and cataloged. Metadata from RFID tags or barcodes on the coupon carriers may be associated with images, such as images 1022 and 1030 for identification purposes that may be later used to reconstruct the original wafer (e.g., wafer 1002), as shown in Figs.11A-B
[0121] Figs. 11A - 11B illustrate a sequence of an operation 1100 encompassing operations 1100A and 1100B, whereby coupons (and fragments) are reassembled into the wafer of origin (e.g., wafer 1002) by positional mapping of the individual coupons sectioned from the wafer.
[0122] At operation 1100A shown in Fig. 11A, a coupon carrier 1102 is placed on coupon imaging area 926. RFID reader 940 reads metadata from an RFID tag (not shown) that is attached to coupon carrier 1102. Metadata gleaned from the RFID tag may contain identification data to identify the coupon group stored within. The metadata can be cross-Docket No.: 11123-1WOchecked by system software to associate the carrier with an image taken during an operation such as operation 1000E (see Figs. 10A-F), for example, as coupons taken from an incoming wafer (e.g., wafer 1002) are stored and cataloged during operation 1000. In at least one implementation, an image 1104 of coupon carrier 1102 is captured by coupon imaging camera 930 and displayed on display monitor 942. In some implementations, image 1104 may be a previously captured image stored in the system database (e.g., database storage 614, Fig. 6A), such as images 1022 or 1030 of stored coupons (e.g., see Figs. 10E and 10F).
[0123] In at least one implementation, a virtual reconstruction of the original wafer (e.g., wafer 1002) is assembled by, e.g., the machine learning module. In at least one implementation, an image 1106 of the virtually reassembled wafer from the individual coupons may be displayed on the screen of display monitor 942, where original positions of the coupon set are mapped. In at least one implementation, image 1106 of the virtually reassembled wafer is created by a machine learning model (e.g., machine learning module on remote sen' er 606). In at least one implementation, projector 944 projects image 1106 onto wafer imaging area 924, also showing a positional mapping of all coupons and fragments that have been imaged in operation 1000. Projected image 1106 may have the actual size of the original wafer, showing locations of all individual coupons and fragments, if any, are indicated by boundary' lines 1108. Locations of the individual coupons may be determined based on image 1012, for example, captured in operation 1000B and stored on the system database.
[0124] Still referring to operation 1100A in Fig. 11 A, according to at least one implementation, an arrow 1110 may be displayed on display monitor 942, drawn between coupon 1112 in image 1104 and location 1114 shown in image 1106, pointing to location 1114. In at least one implementation, arrow 1110 may be displayed to guide a user in a physical reassembly process, indicating where to manually (or robotically) place coupon 1112 on the coupon map shown in image 1106. Projected image 1106 also displays location 1114 on the coupon map to show' a user exactly where to physically place coupon 1112 to reassemble the original wafer. For implementations involving fully automated reassembly, arrow 1110 may also be displayed for informational purposes to users, showing positions of coupons and fragments and progress made by a robotic pick-and-place tool, for example, in the physical reassembly of the original w afer.
[0125] At operation 1100B shown in Fig. 11B, image 1106 continues to be projected by projector 944 onto wafer imaging area 924, showing the coupon map for manual or automated reassembly of the original wafer. In the illustrative example, a second coupon, aDocket No.: 11123-1WOlocation 1118 coupon 1120 is indicated in image 1106. In at least one implementation, an arrow 1116 may be drawn between coupon 1120 and location 1118 as a guide to a user for physical transfer of coupon 1120 from coupon carrier 1102 to location 1118 displayed in wafer imaging area 924.
[0126] Following the above operations, the wafer may be fully reassembled from the coupons sectioned therefrom during operation 1000. This reassembly process may be done after metrology is complete, for example. Once the wafer is fully reconstructed, the reassembled coupons may be placed back into wafer carrier 1003. It is noted that any missing coupons or fragments may be identified during the reassembly process.
[0127] Fig. 12 illustrates a processor system 1200 with a machine-readable storage medium having machine-readable instructions that when executed cause a processor (e.g., processor 802) to execute machine-readable instructions according to the method summarized by flowcharts 630 or 700, as well as operations 1000 and 1100 shown in Figs. 10A-F and 11A-B, respectively, in accordance with at least one implementation. In at least one example, processor system 1200 comprises memory 1201, processor 1202, machine-readable storage medium 1203 (also referred to as tangible machine-readable medium), communication interface 1204 (e.g., wireless or wired interface), and network bus 1205, coupled together as show n. In at least one example, processor system 1200 may be part of a computing system associated with block diagram 800, illustrating computer workstation 602, where processor 1202 is equivalent to process 802 shown in Fig. 8. In at least one implementation, the processes described herein may be stored in machine readable medium 1203 as computerexecutable instructions. In at least one implementation, a machine-readable storage medium may be random access memory' (RAM).
[0128] In at least one example, processor 1202 is a digital signal processor (DSP), an application specific integrated circuit (ASIC), a general-purpose central processing unit (CPU), or a low power logic implementing a simple finite state machine to perform various processes described herein.
[0129] In at least one example, various logic blocks of processor system 1200 are coupled together via netw ork bus 1205. Any suitable protocol may be used to implement network bus 1205. In at least one implementation, machine-readable storage medium 1203 includes instructions (also referred to as program software code / instructions) for actuating valves of the process gas delivery system, and heating portions of delivery lines, for example, coded into software stored in machine-readable storage medium 1203.Docket No.: 11123-1WO
[0130] In at least one implementation, machine-readable storage media 1203 is a machine-readable storage media with instructions for operation of computer workstation 602 executed by processor 802, or any cloud-based server functions, such as the Al server 612 associated with machine learning module. In at least one implementation, machine-readable medium 1203 has machine-readable instructions, that when executed, cause processor 1202 to perform the method discussed herein.
[0131] In at least one example, program software code / instructions associated with various implementations may be implemented as part of an operating system or a specific application, component, program, object, module, routine, or other sequence of instructions or organization of sequences of instructions referred to as "program software code / instructions," "operating system program software code / instructions," "application program software code / instructions," or simply "software" or firmware embedded in processor. In some implementations, program software code / instructions associated with processes of various implementations are executed by processor system 1200.
[0132] In at least one example, machine-readable storage media 1203 is a computer executable storage medium. In at least one implementation, program software code / instructions associated with various implementations are stored in computer executable storage medium 1203 and executed by processor 1202. Here, computer executable storage medium 1203 is a tangible machine-readable medium 1203 that can be used to store program software code / instructions and data that, when executed by a computing device, causes one or more processors (e.g., processor 1202) to perform a process.
[0133] In at least one implementation, tangible machine-readable medium 1203 may include storage of executable software program code / instructions and data in various tangible locations, including for example, ROM, volatile RAM, non-volatile memory, and / or cache, and / or other tangible memory as referenced in present application. Portions of this program software code / instructions and / or data may be stored in any one of these storage and memory devices. In some implementations, program software code / instructions can be obtained from other storage, including, e.g., through centralized servers or peer to peer networks and the like, including the Internet. Different portions of software program code / instructions and data can be obtained at different times and in different communication sessions or in the same communication session.
[0134] In at least one example, software program code / instructions associated with various implementations can be obtained in their entirety prior to execution of a respective software program or application. Alternatively, portions of software programDocket No.: 11123-1WOcode / instructions and data can be obtained dynamically, e.g., just in time, when needed for execution. Alternatively, some combination of these ways of obtaining software program code / instructions and data may occur, e.g., for different applications, components, programs, objects, modules, routines, or other sequences of instructions or organization of sequences of instructions, by way of example. Thus, it is not required that data and instructions be on a tangible machine-readable medium 1203 in entirety at a particular instance of time.
[0135] In at least one implementation, tangible machine-readable medium 1203 includes but is not limited to recordable and non-recordable type media, such as volatile and nonvolatile memory devices, read only memory (ROM), random access memory (RAM), flash memory devices, floppy and other removable disks, magnetic storage media, optical storage media (e.g., Compact Disk Read-Only Memory (CD ROMs). Digital Versatile Disks (DVDs), etc.), among others. In at least one implementation, software program code / instructions may be temporarily stored in digital tangible communication links while implementing electrical, optical, acoustical, or other forms of propagating signals, such as carrier waves, infrared signals, digital signals, etc., through such tangible communication links.
[0136] In some instances, well-known methods and devices are shown in block diagram form, rather than in detail, to avoid obscuring at least one implementation. Reference throughout this specification to “an implementation,’" “one implementation,’" “in at least one implementation,” or “some implementations” means that a particular feature, structure, function, or characteristic described in connection with implementation is included in at least one implementation. Thus, appearances of phrase “in an implementation,” “in at least one implementation,"’ or “in one implementation,” or “some implementations” in various places throughout this specification are not necessarily referring to same implementation of disclosure. Furthermore, particular features, structures, functions, or characteristics may be combined in any suitable manner in one or more implementations. For example, a first implementation may be combined with a second implementation anywhere particular features, structures, functions, or characteristics associated with two implementations are not mutually exclusive.
[0137] As used in herein, singular forms “a,” “an,” and "‘the” are intended to include plural forms as well, unless context clearly indicates otherwise. It will also be understood that term “and / or"’ as used herein refers to and encompasses all possible combinations of one or more of associated listed items.Docket No.: 11123-1WO
[0138] Here, “coupled” and “connected,” along with their derivatives, may be used herein to describe functional or structural relationships between components. These terms are not intended as synonyms for each other. Rather, in particular implementations, “connected” may be used to indicate that two or more elements are in direct physical, optical, or electrical contact with each other. “Coupled” may be used to indicate that two or more elements are in either direct or indirect (with other intervening elements between them) physical, electrical, or in magnetic contact with each other, and / or that two or more elements co-operate or interact with each other (e.g., as in a cause an effect relationship). Coupled may also have the meaning of non-mechanical contact or connection. Coupling may also have the meaning of thermal connectivity, where one object may be a heat source and another object may be a heat sink, either in thermal equilibrium with each other or subject to a common conductive, convective, or radiative heat flow between them; electrically coupled, where objects may be connected electrically in an electric or electronic circuit and a current flow may be induced by application of a voltage between the electrically interconnected objects or by an electric field between mechanically coupled or isolated objects; magnetically, where two mechanically coupled or isolated objects mutually share a common magnetic field flux; and fluidically, where objects, such as vessels and conduits may share a common gas or liquid fluid that is static or flowing.
[0139] Here, “over,” “under.” “between,” and “on” as used herein refer to a relative position of one component or material with respect to other components or materials where such physical relationships are noteworthy. For example, in context of materials, one material or material disposed over or under another may be directly in contact or may have one or more intervening materials. Moreover, one material disposed between two materials may be directly in contact with two layers or may have one or more intervening layers. In contrast, a first material “on” a second material is in direct contact with that second material / material. Similar distinctions are to be made in context of component assemblies. As used throughout this description, and in claims, a list of items joined by term “at least one of’ or “one or more of can mean any combination of listed terms.
[0140] Here, “adjacent” generally refers to a position of a thing being next to (e.g., immediately next to or close to with one or more things between them) or adjoining another thing (e.g., abutting it).
[0141] Unless otherwise specified in explicit context of their use, terms “substantially equal,” “about equal,” and “approximately equal” mean that there is no more than incidentalDocket No.: 11123-1WOvariation between two things so described. Such variation is typically no more than + / -10% of a predetermined target value.
[0142] Here, '‘left;’ ‘'right,” "front,” '‘back,” “top,” “bottom,” “over,” “under,” and similar terms are used for descriptive purposes and not necessarily for describing permanent relative positions. For example, terms “over,” “under,” “front side,"’ “back side,” “top,"’ “bottom,'’ '‘over,” “under,” and '‘on” as used herein refer to a relative position of one component, structure, or material with respect to other referenced components, structures, or materials within a device, where such physical relationships are noteworthy. These terms are employed herein for descriptive purposes only and predominantly within context of a device z-axis and therefore may be relative to an orientation of a device. Hence, a first material “over” a second material in context of a figure provided herein may also be “under” second material if device is oriented upside-down relative to context of figure provided. Similar distinctions are to be made in context of component assemblies.
[0143] Here, a device that is “configured to” perform a task or function may be configured (e.g., programmed and / or hardwired) at a time of manufacturing by a manufacturer to perform the function. In at least one example, the device may be configurable (or reconfigurable) by a user after manufacturing to perform the function and / or other additional or alternative functions. In at least one example, the configuring may be through firmware and / or software programming of the device, through a construction and / or layout of hardware components and interconnections of the device, or a combination thereof.
[0144] Here, “between” may be employed in context of z-axis, x-axis, ory-axis of a device. A material that is between two other materials may be in contact with one or both of those materials. In another example, a material that is between two or other material may be separated from both of other two materials by one or more intervening materials. A material ■‘between” two other materials may therefore be in contact with either of other two materials. In another example, a material “between” two other materials may be coupled to other two materials through an intervening material. A device that is between two other devices may be directly connected to one or both of those devices. In another example, a device that is between two other devices may be separated from both of other two devices by one or more intervening devices.
[0145] Here, “semiconductor wafer” may generally refer to a silicon wafer, or a wafer of another semiconductor material used in the electronics industry. A wafer is a thin slice of a grown single crystal of semiconductor material, such as silicon. Wafers can be sliced in anyDocket No.: 11123-1WOcrystal orientation. Wafers are generally circular in cross section, having diameters ranging from 10 cm to 30 cm.
[0146] Here, '‘wafer portion” may generally refer to a fragment of a semiconductor wafer, referred to as a “coupon”, that is supplied to a metrology facility by a process development laboratory, where personnel at the process development laboratory7may section a wafer to produce the coupons. A coupon may be sectioned from a wafer by scribing the wafer with a sharp or pointed object forming an outline of the coupon and then cleaving the coupon from the wafer along the scribe lines. In at least one example, coupons may be sectioned from an experimental wafer by process engineers and are sent from a process laboratory' to a metrology7facility7, where metrology engineers may section the coupons into fragments, defined below.
[0147] Here, “fragment” may generally refer to a piece of a coupon in the context of this disclosure. Fragments may be made by cleaving coupons into smaller pieces in the same manner that coupons are sectioned from wafers. This may be done to create pieces that are of appropriate size for mounting within an electron microscope (e.g., SEM or TEM). While “fragments” are used interchangeably with “coupons” within this disclosure, particularly in the claims, fragments are mostly7discussed as portions of coupons sent from a process laboratory7and may be generally larger than fragments. Coupons may be sectioned into fragments in a metrology facility, where personnel at the metrology facility may section coupons sent by a process laboratory7for analysis.
[0148] Here, “edge length” may generally refer to the length of a straight edge of a coupon or fragment. The length is defined as the distance of a straight edge between any two vertices of a coupon or fragment. Vertices may include comers, such as the four comers of a rectangle.
[0149] Here, '‘vertex angle” may generally refer to an angle a vertex of a coupon or fragment subtends. For example, a comer of a rectangular coupon has a vertex angle of 90 degrees. Vertex angles may be acute, or being less than 90 degrees, or obtuse, being greater than 90 degrees.
[0150] Here, “complementary vertex angle” may generally refer to a vertex angle on a first coupon or fragment that has a first angle that sums to 180 degrees or 360 degrees with a second angle on a second coupon or fragment. For example, a first vertex angle on the first coupon may be 43 degrees, whereas a second vertex angle on the second coupon may be 137 degrees. The two coupons may7fit together as two pieces of a jigsaw puzzle by abutting the two complementary vertices together, summing to 180 degrees. In another example, a firstDocket No.: 11123-1WOvertex may have an angle of 90 degrees on a first coupon, and a second vertex on a second coupon may have an angle of 270 degrees, summing to 360 degrees in this case. By abutting the two complementary vertices together, the two coupons fit together provided the edge lengths match and if any texture is present, that the texture matches as well.
[0151] Here, “texture” may generally refer to lithographic patterning on a surface of a semiconductor wafer, coupon or fragment that is visible to the naked eye. Textures may be metallization patterns or active layer patterning, such as transistor banks and similar structures. In at least one example, textures may be recorded on a whole semiconductor wafer or on individual coupons, so that fragments that are sectioned from as received coupons may be reassembled by matching textures on neighboring fragments that are derived from the same coupon. A texture may be shared by neighboring coupons, where a first texture on a first coupon matches a second texture portion on a second coupon across a boundary between the two coupons. When joining the two coupons, the texture spans the cleavage line which defines the boundaries of the two coupons.
[0152] Here, “curvature” may generally refer to a curved edge of a coupon or fragment, where the fragment edge has no straight portions. The edge is curved, having curvature that follows an arc that is part of the circumference of an imaginary circle having the diameter of a semiconductor wafer from which the fragment (e.g., a peripheral coupon). The arc curvature of the edge may be detected and measured or computed by fitting tangent lines to points along the arc.
[0153] Here, “match” may generally refer to finding two or more coupons or fragments that are adjacent neighbors when a wafer or coupon, respectively, is reassembled. The match may be based on matching edge lengths on edges of two coupons, matching complementary vertex angles on two coupons or fragments, matching textures on two coupons or fragments, or any combination of such matches. The verb “matching” may be defined as the process of finding such matches. The term “match” as applied to this disclosure may not indicate that any two fragments are identical.
[0154] Here, “image” may generally refer to a photograph captured by a camera coupled to a machine vision system, as in the context of this disclosure. The image may be a digital photograph or a digitized photograph.
[0155] Here, “ambiguity” may generally refer to a match having two or more solutions. For example, an edge length on one coupon may be matched by several other coupons having the same edge length. An ambiguity may be resolved by matching texture or metadata, which helps to identify individual coupons.Docket No.: 11123-1WO
[0156] Here, “computer workstation” may generally refer to a computer that is used to manage wafer and coupon imaging tools, as well as wafer or coupon reassembly tools by intermediary machine vision and machine learning modules. In at least one example, the machine learning module may be accessed via a cloud network.
[0157] Here, “radio frequency identification, or RFID” may generally refer to “tags” or microchips containing identification data pertaining to object to which the tags are associated. The data may be obtained by reading the tags or microchips by very short-range wireless microwave signals that query the microchip from a reader. The reader signal causes the microchip to send the requested data over the microwave link to the reader. The reader signal provides enough power to the microchip to activate the circuitry' to send a weak signal back to the reader, where the data stream contains is received by the reader.
[0158] Here, “RFID tag” may generally refer to a tag comprising a RFID microchip, described above.
[0159] Here, “cloud storage” may be generally refer to remote memory storage for digital data that is accessed by remote networking capabilities. “Cloud” generally refers to dispersed remote servers at unknown locations and distances that may share stored data.
[0160] Here, “artificial intelligence server, or Al server” or “cloud-based artificial intelligence server” may generally refer to a server storing software of a machine learning module, where the machine learning module may be based on a neural network. The server may be a cloud server such as one or more remote servers that communicate with one another.
[0161] Here, “server” may generally refer to any computing station that is remote from a client computer executing software to accomplish a task. The server may store the software or store data used by or generated by the software that is executed on a client computer.
[0162] Here, “area” may generally refer to the geometric area of a semiconductor wafer, a coupon or a fragment.
[0163] Here, “machine learning module” or “machine learning model” may generally refer to a software program that is based on artificial intelligence and is capable of deep learning. A specific machine learning model may be chosen as the kernel or engine of the machine learning module. A module may be a self-contained piece of software or is part of a larger software system that is overseen by a managing or overseeing software module.
[0164] Here, “virtual reality or VR” may generally refer to computer generated three-dimensional environment that simulates reality to give a user illusions that they are immersed in the created reality. Normally, goggles or glasses are used to enable viewing of images thatDocket No.: 11123-1WOthe virtual and augmented reality software create. Virtual reality' may be combined with augmented reality (AR) to create extended reality. Extended reality (ER) combines computer generated illusionary imagery of virtual and augmented realities, to create an illusion that the user is immersed in that reality'. The user may wear a sensor suit or have sensors on a few parts of the body to detect movement and bodily positions, so that the VR, AR, and ER software are responsive to the user’s reactions, and adaptively changes the imagery the user sees in response to the user’s reaction. The visible environment seen by the user may be enhanced, where virtual objects are included in the actual environment to make the user feel that the real environment includes those virtual objects. Extended reality may include virtual reality', augmented reality' and mixed reality, which is a combination of virtual and augmented realities. In at least one example, virtual reality, augmented reality and extended reality’ may be employed to assist human users in re-assembly of semiconductor wafers or coupons by comparison of the original state of the wafer or coupons with the state of the wafer or coupons after metrology evaluation. For the latter, the wafer or coupons may be sectioned into small fragments. Without assistance, reassembly of the fragments may take hours of time. With VR assistance, a machine learning module can direct a human to put together separate fragments that are laid out on a surface, such as a table, and show the user through the virtual or augmented reality' vision system, where to place a fragment in a growing re -assemblage of a wafer or coupon.
[0165] Here, "“visualization device” may generally refer to a computer screen display, a VR goggle set, or glasses. “‘Visualization device” is used in conjunction with an extended reality or virtual reality' tool. Generally, an extended reality or VR goggle set or computer screen may be employed for physical reassembly of the original semiconductor wafer or wafer portion from the corresponding fragments.
[0166] Here, “‘machine vision” may generally refer to computer vision based on visual input from a digital camera to a computer system executes code in a machine vision module to interface with the camera to process digital images taken by the camera. The digital images may be video images as well as still images. Machine vision modules may include code that measures geometric features of objects captured by the digital images.
[0167] Here, “human user” may generally refer to an engineer or technician in a facility, such as a metrology facility, that is using the virtual reality or extended reality tools for reassembly of the semiconductor wafer or coupons.Docket No.: 11123-1WO
[0168] Here, “user” may generally refer to an engineer, technician, or a robot in a facility, such as a metrology facility, that is using the virtual reality or extended reality tools for reassembly of the semiconductor wafer or coupons.
[0169] Here, “virtual image” may generally refer to an image that is generated by software such as a machine learning model or a machine vision module. The virtual image may be visually displayed on a computer display or remain in memory without a visual display as binary data representing an array of pixels. In at least one example, a virtual image is a binary representation of a visual image, such as a semiconductor wafer, that is captured by a digital camera. The virtual image may contain binary representations of coupons, having geometric features such as edges, angles and textures. A machine learning model may be capable of analyzing and measuring these geometric features from the virtual image by¬ invoking mathematical algorithms.
[0170] Here, “virtual assembly” may generally refer to a non-physical assembly of fragments or coupons, where the virtual assembly may be a “reassembled semiconductor wafer or wafer portion” as defined below. The virtual assembly may be displayed on a computer display or be confined to memory as a binary array of pixels stored for analysis by the machine learning model and not be displayed.
[0171] Here, “virtually reassembled semiconductor wafer” or “wafer portion” may generally refer to a virtual image showing an assembly of individual fragments and / or portions of a semiconductor wafer or a coupon, where the individual pieces are arranged to re-create the whole wafer or the coupon.
[0172] Here, “unassembled fragment” may generally refer to a fragment that is not phy sically part of a reassembled wafer or a coupon (w afer portion).
[0173] Here, “unassembled matched fragment” or “matched fragment” may generally refer to a fragment that is not part of a physical or virtual wafer assembly, but has been virtually matched with other such fragments by the machine learning model. This definition may be germane to a physical collection of fragments and w afer portions, where the individual fragments are to be physically assembled to recreate the original wafer from which the fragments are derived according to the virtual matching by the machine learning model (e.g., by cleavage of the original wafer).
[0174] Here, “collection of unassembled fragments” may generally refer to an ensemble of fragments or wafer portions. The collection of unassembled fragments is a physical collection of unassembled fragments that may be physically reassembled into a wafer or wafer portion according to mapping generated by the machine learning model, where theDocket No.: 11123-1WOmapping associates individual fragments in the collection to positions within a virtually reassembled wafer or wafer portion.
[0175] Here, '‘positional mapping” may generally refer to a virtual mapping (e.g., generated by the machine learning model) between individual fragments in a collection of unassembled fragments and corresponding positions of each individual fragment in the virtually reassembled semiconductor wafer or wafer portion.
[0176] Here, “robotic” may generally refer to a tool or device that is remotely controlled by automation. The automation is computer generated. In some implementations, the machine learning model may generate the code or call specific subroutines that drive the robotic automation.
[0177] Here, “robotic pick-and-place tool” may generally refer to a robotic device that picks up small objects from one location and precisely places them at another location. Pick-and-place tools are commonly used in semiconductor manufacturing facilities.
[0178] Structures of various examples described herein can also be described as method(s) of forming those structures or apparatuses, and method(s) of operation of these structures or apparatuses. The following examples are provided that illustrate the various examples of the disclosure. The examples can be combined with other examples. As such, various examples can be combined with other examples without changing the scope of the invention.
[0179] Example lis a method for reassembly of a fragmented semiconductor wafer, comprising capturing a first image of a semiconductor wafer or of one or more wafer portions; sectioning the semiconductor wafer or the one or more wafer portions into a plurality of fragments; capturing a second image of the plurality of fragments; determining areas of individual fragments of the plurality of fragments from the second image; determining one or more edge lengths, one or more curvatures, and one or more vertex angles of each of the individual fragments of the plurality of fragments from the second image; and identifying one or more groups of neighboring fragments from one or more unassembled fragments in a collection of unassembled fragments to enable the reassembly of the semiconductor wafer or the one or more wafer portions from the one or more groups of neighboring fragments, where an identification is based on any combination of the one or more edge lengths, , the one or more curvatures, the one or more vertex angles, and comparison to the first image.1. Example 2 is a method as in any of the examples, in particular example 1, wherein identify ing the one or more groups of neighboring fragments from the one or moreDocket No.: 11123-1WOunassembled fragments in the collection of unassembled fragments comprises determining a texture on a surface of each of the individual fragments of the plurality of fragments from the second image and matching a first texture portion on a first fragment of the plurality of fragments with a second texture portion on a second fragment of the plurality of fragments.1. Example 3 is a method as in any of the examples, in particular example 1, wherein identifying the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments comprises identifying a match for at least the one or more edge lengths, the one or more curvatures, and the one or more vertex angles comprises matching a first texture portion on a first fragment and a second texture portion on a second fragment.
[0180] Example 4 is a method as in any of the examples, in particular example 1, wherein the one or more curvatures include a first curvature and a second curvature, wherein identify ing the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments comprises identifying a match for the first curvature of a first fragment and the second curvature of a second fragment.
[0181] Example 5 is a method as in any of the examples, in particular example 1, wherein identify ing the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments based on least the one or more edge lengths, the one or more curvatures, and the one or more vertex angles comprises matching a first vertex angle of a first fragment to a second vertex angle of a second fragment, wherein the first vertex angle is complementary to the second vertex angle.
[0182] Example 6 is a method as in any of the examples, in particular example 1, further comprising reassembling the fragmented semiconductor wafer or the one or more wafer portions by forming a virtual assembly of all matched fragments of the plurality of fragments wherein matches between two or more fragments are based on a comparison of edge lengths between the two or more fragments of the plurality of fragments, a comparison of vertex angles between the two or more fragments of the plurality of fragments, and a comparison of textures between the two or more fragments of the plurality of fragments.
[0183] Example 7 is a method as in any of the examples, in particular example 6, wherein reassembling the fragmented semiconductor wafer or the one or more wafer portions by forming the virtual assembly of all the matched fragments of the plurality of fragments comprises: forming a virtual image of a reassembled semiconductor wafer or the one or more wafer portions indicating positions of all the matched fragments in a virtually reassembledDocket No.: 11123-1WOsemiconductor wafer or the one or more wafer portions: and mapping a position of each matched fragment in the virtually reassembled semiconductor wafer in the virtual image of the reassembled semiconductor wafer or the one or more wafer portions to a corresponding unassembled fragment in the collection of unassembled fragments belonging to the semiconductor wafer or to the one or more wafer portions.
[0184] Example 8 is a method as in any of the examples, in particular example 7, further comprising directing a robotic pick-and-place tool to physically reassemble the semiconductor wafer or the one or more wafer portions, wherein each unassembled fragment in the collection of unassembled fragments is placed by the robotic pick-and-place tool in a corresponding mapped position in a growing assembly of two or more matched fragments according to the virtual image of the reassembled semiconductor wafer or the one or more wafer portions.
[0185] Example 9 is a method as in any of the examples, in particular example 7, further comprising displaying the virtual image of the reassembled semiconductor wafer or the one or more wafer portions in an extended reality visualization device, wherein the extended reality visualization device may include an augmented reality, a virtual reality or a mixed reality algorithm; and displaying a mapping of each matched fragment in the virtually reassembled semiconductor wafer or the one or more wafer portions to a corresponding fragment in the collection of unassembled fragments.
[0186] Example 10 is a method as in any of the examples, in particular example 6. further comprising determining a first area of the semiconductor wafer or of the one or more wafer portions from the first image; capturing a third image of a reassembled semiconductor wafer or of one or more reassembled wafer portions from a collection of fragments belonging to the semiconductor wafer or to the one or more wafer portions, determining, from the third image, a second area of the reassembled semiconductor wafer or of the one or more reassembled wafer portions, wherein the second area is a sum of areas of each of the matched fragments included in the reassembled semiconductor wafer; and comparing the second area of the matched fragments to the first area.
[0187] Example 11 is a method as in any of the examples, in particular example 10, further comprising determining if one or more fragments are missing from the reassembled semiconductor wafer or from the one or more reassembled wafer portions.
[0188] Example 12 is a system, comprising a computer workstation; a camera coupled to the computer workstation; an artificial intelligence server coupled the computer workstation; and a machine learning module on the artificial intelligence server, wherein the machineDocket No.: 11123-1WOlearning module is configured to recognize one or more unassembled fragments belonging to a semiconductor wafer or to one or more wafer portions in a collection of unassembled fragments captured in an image taken by the camera, and to virtually reassemble the semiconductor wafer or the one or more wafer portions by identifying one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments, and wherein the machine learning module is configured to form a virtually reassembled semiconductor wafer or one or more virtually reassembled portions from the one or more groups of neighboring fragments.
[0189] Example 13 is a system as in any of the examples, in particular example 12, wherein the machine learning module is configured to measure one or more edge lengths of the one or more unassembled fragments of the semiconductor wafer or of the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer edge length data to a storage coupled to the computer workstation.
[0190] Example 14 is a system as in any of the examples, in particular example 13, wherein the machine learning module is configured to measure one or more vertex angles of the one or more unassembled fragments of the semiconductor wafer or of the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer vertex angle data to the storage coupled to the computer workstation.
[0191] Example 15 is a system as in any of the examples, in particular example 13, wherein the machine learning module is configured to detect a texture on surfaces of the one or more unassembled fragments of the semiconductor wafer or the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer texture data to the storage coupled to the computer workstation.
[0192] Example 16 is a system as in any of the examples, in particular example 12, wherein the machine learning module is configured to virtually reassemble the semiconductor wafer or the one or more wafer portions from the collection of unassembled fragments, wherein the machine learning module is configured to match two or more unassembled fragments from the collection of unassembled fragments as adjacent neighbors to reassemble the semiconductor wafer or the one or more wafer portions based on at least one of matching edge lengths of the two or more unassembled fragments, finding complementary vertex angles of the two or more unassembled fragments, and finding matching texture portions of the two or more unassembled fragments from the image captured by the camera.Docket No.: 11123-1WO
[0193] Example 17 is a system as in any of the examples, in particular example 16, wherein the machine learning module is configured to create a positional mapping that relates each matched fragment of the collection of unassembled fragments to a corresponding position in a virtual reassembly of the semiconductor wafer or of the one or more wafer portions.
[0194] Example 18 is a system as in any of the examples, in particular example 17, further compnsing a robotic pick-and-place tool that is coupled to the computer workstation, wherein the robotic pick-and-place tool is configured to reassemble the semiconductor wafer or the one or more wafer portions by commands sent by the machine learning module according to the positional mapping created by the machine learning module, wherein the commands cause the robotic pick-and-place tool to transfer a matched fragment from the collection of unassembled fragments to the corresponding position of the matched fragment in the virtual reassembly of the semiconductor wafer or the one or more wafer portions.
[0195] Example 19 is a system as in any of the examples, in particular example 17, further comprising an extended reality tool that is coupled to the computer workstation, wherein the extended reality tool is configured to display the positional mapping to a human user to enable manual reassembly of the semiconductor wafer or the one or more wafer portions by visual signals sent to a visualization device of the extended reality tool, wherein the visual signals display the positional mapping created by the machine learning module to the human user, and wherein the extended reality tool includes one or more augmented reality, virtual reality’ or mixed reality algorithms.
[0196] Example 20 is a system as in any of the examples, in particular example 12, further comprising a reader which is configured to read a set of metadata related to the semiconductor wafer or to the one or more wafer portions from an encoded tag attached to a carrier for the semiconductor wafer or the one or more wafer portions that are sectioned from the semiconductor wafer or from the one or more wafer portions.
Claims
Docket No.: 11123-1WOCLAIMSWhat is claimed is:
1. A method for reassembly of a fragmented semiconductor wafer, comprising:capturing a first image of a semiconductor wafer or of one or more wafer portions; sectioning the semiconductor wafer or the one or more wafer portions into a plurality of fragments;capturing a second image of the plurality of fragments;determining areas of individual fragments of the plurality of fragments from the second image;determining one or more edge lengths, one or more curvatures, and one or more vertex angles of each of the individual fragments of the plurality of fragments from the second image; andidentifying one or more groups of neighboring fragments from one or more unassembled fragments in a collection of unassembled fragments to enable the reassembly of the semiconductor wafer or the one or more wafer portions from the one or more groups of neighboring fragments, where an identification is based on any combination of the one or more edge lengths, , the one or more curvatures, the one or more vertex angles, and comparison to the first image.
2. The method of claim 1, wherein identifying the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments comprises determining a texture on a surface of each of the individual fragments of the plurality of fragments from the second image and matching a first texture portion on a first fragment of the plurality of fragments with a second texture portion on a second fragment of the plurality of fragments.
3. The method of claim 1, wherein identifying the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments comprises identifying a match for at least the one or more edge lengths, the one or more curvatures, and the one or more vertex angles comprises matching a first texture portion on a first fragment and a second texture portion on a second fragment.Docket No.: 11123-1WO4. The method of claim 1, wherein the one or more curvatures include a first curvature and a second curvature, wherein identifying the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments comprises identifying a match for the first curvature of a first fragment and the second curvature of a second fragment.
5. The method of claim 1, wherein identifying the one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments based on least the one or more edge lengths, the one or more curvatures, and the one or more vertex angles comprises matching a first vertex angle of a first fragment to a second vertex angle of a second fragment, wherein the first vertex angle is complementary to the second vertex angle.
6. The method of claim 1 , further comprising reassembling the fragmented semiconductor wafer or the one or more wafer portions by forming a virtual assembly of all matched fragments of the plurality of fragments wherein matches between two or more fragments are based on a comparison of edge lengths between the two or more fragments of the plurality of fragments, a comparison of vertex angles between the two or more fragments of the plurality of fragments, and a comparison of textures between the two or more fragments of the plurality of fragments.
7. The method of claim 6, wherein reassembling the fragmented semiconductor w afer or the one or more wafer portions by forming the virtual assembly of all the matched fragments of the plurality of fragments comprises:forming a virtual image of a reassembled semiconductor wafer or the one or more w afer portions indicating positions of all the matched fragments in a virtually reassembled semiconductor wafer or the one or more wafer portions; and mapping a position of each matched fragment in the virtually reassembled semiconductor wafer in the virtual image of the reassembled semiconductor wafer or the one or more w afer portions to a corresponding unassembled fragment in the collection of unassembled fragments belonging to the semiconductor wafer or to the one or more w afer portions.Docket No.: 11123-1WO8. The method of claim 7, further comprising directing a robotic pick-and-place tool to physically reassemble the semiconductor wafer or the one or more wafer portions, wherein each unassembled fragment in the collection of unassembled fragments is placed by the robotic pick-and-place tool in a corresponding mapped position in a growing assembly of two or more matched fragments according to the virtual image of the reassembled semiconductor wafer or the one or more wafer portions.
9. The method of claim 7, further comprising:displaying the virtual image of the reassembled semiconductor wafer or the one or more wafer portions in an extended reality visualization device, wherein the extended reality visualization device may include an augmented reality, a virtual reality or a mixed reality algorithm; anddisplaying a mapping of each matched fragment in the virtually reassembled semiconductor wafer or the one or more wafer portions to a corresponding fragment in the collection of unassembled fragments.
10. The method of claim 6, further comprising:determining a first area of the semiconductor wafer or of the one or more wafer portions from the first image;capturing a third image of a reassembled semiconductor wafer or of one or more reassembled wafer portions from a collection of fragments belonging to the semiconductor wafer or to the one or more wafer portions; determining, from the third image, a second area of the reassembled semiconductor wafer or of the one or more reassembled wafer portions, wherein the second area is a sum of areas of each of the matched fragments included in the reassembled semiconductor wafer; andcomparing the second area of the matched fragments to the first area.
11. The method of claim 10, further comprising determining if one or more fragments are missing from the reassembled semiconductor wafer or from the one or more reassembled wafer portions.
12. A system, comprising:a computer workstation;Docket No.: 11123-1WOa camera coupled to the computer workstation;an artificial intelligence server coupled the computer workstation: anda machine learning module on the artificial intelligence server,wherein the machine learning module is configured to recognize one or more unassembled fragments belonging to a semiconductor wafer or to one or more wafer portions in a collection of unassembled fragments captured in an image taken by the camera, and to virtually reassemble the semiconductor wafer or the one or more wafer portions by identifying one or more groups of neighboring fragments from the one or more unassembled fragments in the collection of unassembled fragments, and wherein the machine learning module is configured to form a virtually reassembled semiconductor wafer or one or more virtually reassembled portions from the one or more groups of neighboring fragments.
13. The system of claim 12, wherein the machine learning module is configured to measure one or more edge lengths of the one or more unassembled fragments of the semiconductor wafer or of the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer edge length data to a storage coupled to the computer workstation.
14. The system of claim 13, wherein the machine learning module is configured to measure one or more vertex angles of the one or more unassembled fragments of the semiconductor wafer or of the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer vertex angle data to the storage coupled to the computer workstation.
15. The system of claim 13, wherein the machine learning module is configured to detect a texture on surfaces of the one or more unassembled fragments of the semiconductor wafer or the one or more wafer portions from the image captured by the camera, and wherein the machine learning module is configured to transfer texture data to the storage coupled to the computer workstation.
16. The system of claim 12, wherein the machine learning module is configured to virtually reassemble the semiconductor wafer or the one or more wafer portions from theDocket No.: 11123-1WOcollection of unassembled fragments, wherein the machine learning module is configured to match two or more unassembled fragments from the collection of unassembled fragments as adjacent neighbors to reassemble the semiconductor wafer or the one or more wafer portions based on at least one of matching edge lengths of the two or more unassembled fragments, finding complementary vertex angles of the two or more unassembled fragments, and finding matching texture portions of the two or more unassembled fragments from the image captured by the camera.
17. The system of claim 16, wherein the machine learning module is configured to create a positional mapping that relates each matched fragment of the collection of unassembled fragments to a corresponding position in a virtual reassembly of the semiconductor wafer or of the one or more wafer portions.
18. The system of claim 17, further comprising a robotic pick-and-place tool that is coupled to the computer workstation, wherein the robotic pick-and-place tool is configured to reassemble the semiconductor wafer or the one or more wafer portions by commands sent by the machine learning module according to the positional mapping created by the machine learning module, wherein the commands cause the robotic pick-and-place tool to transfer a matched fragment from the collection of unassembled fragments to the corresponding position of the matched fragment in the virtual reassembly of the semiconductor wafer or the one or more wafer portions.
19. The system of claim 17, further comprising an extended reality tool that is coupled to the computer workstation, wherein the extended reality tool is configured to display the positional mapping to a human user to enable manual reassembly of the semiconductor wafer or the one or more wafer portions by visual signals sent to a visualization device of the extended real ity tool, wherein the visual signals display the positional mapping created by the machine learning module to the human user, and wherein the extended reality’ tool includes one or more augmented reality, virtual reality or mixed reality algorithms.
20. The system of claim 12, further comprising a reader which is configured to read a set of metadata related to the semiconductor wafer or to the one or more wafer portions from an encoded tag attached to a carrier for the semiconductor wafer or the one or more waferDocket No.: 11123-1WOportions that are sectioned from the semiconductor wafer or from the one or more wafer portions.
