Method and device for identifying a defective electrical connection, and method for generating a trained computation model
Through electron beam charging and machine learning model recognition technology, the problem of electrical connection defect identification of complex microelectronic devices is solved, and fast and reliable defect detection is achieved, suitable for advanced packaging and panel-level packaging substrates.
Patent Information
- Application Number
- CN202280094236.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-05
- Publication Date
- 2024-11-19
AI Technical Summary
The prior art is difficult to quickly and reliably identify defects in electrical connections in complex microelectronic devices, especially on advanced packaging and panel-level packaging substrates, where traditional testing methods fail due to reduced size and increased complexity.
Using electron beam combined with machine learning technology, secondary electronic signals are detected by contactless charging of substrate surface contacts, and a trained machine learning model is used to identify defect information of electrical connections based on input data.
It realizes fast and reliable identification of electrical connection defects of complex microelectronic devices, improves testing efficiency and accuracy, and is suitable for densely arranged surface contacts and multiple electrically connected substrates.
Smart Images

Figure CN118985042A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to methods and apparatus for identifying defective electrical connections extending through a substrate, particularly an advanced packaging (AP) substrate or a panel level packaging (PLP) substrate. More particularly, embodiments described herein relate to non-contact testing of electrical connections in a substrate using an electron beam in combination with machine learning techniques and artificial intelligence, particularly for identifying and characterizing defects such as shorts, opens, and / or leaks. Background Art
[0002] In many applications, it is necessary to inspect the substrate to monitor the quality of the substrate. Since defects may occur during processing of the substrate, such as during coating of the substrate, it may be beneficial to inspect the substrate to detect defects and monitor the quality of the display.
[0003] Semiconductor substrates and printed circuit boards used to manufacture complex microelectronic or micromechanical components are often tested before, during, and / or after manufacturing to identify defects (e.g., "short defects" or "open defects") in conductive paths and interconnect structures on or extending through substrate layers. For example, a substrate used to manufacture a complex microelectronic device may include a plurality of interconnect paths for connecting semiconductor chips to be mounted on the substrate.
[0004] Various methods for testing such components are known. For example, contact pads of the component to be tested can be mechanically contacted with a contact probe to determine whether the component is defective. However, as components and contact pads become increasingly smaller due to the continued miniaturization of components, contacting the contact pads with a contact probe can be difficult and even risk damaging the device under test during testing.
[0005] The complexity of package substrates is increasing, while design rules (feature size) are decreasing dramatically. Within such substrates, surface contacts (for later flip-chip or other chip mounting) are connected to other surface contacts on the package substrate to interconnect semiconductor (or other) devices. Standard methods of electrical testing, such as electro-mechanical probing, cannot meet the requirements of high-volume production test due to reduced throughput (higher number of test points) and reduced contact reliability (smaller contact size). In addition to the issues of reduced size and possible damage to contact pads, the topography of the package substrate causes difficulties for other test methods, such as those using capacitive detectors or electric field detectors, which advantageously have small mechanical spacing.
[0006] Therefore, it would be beneficial to provide a testing method and a testing apparatus suitable for reliably and quickly identifying defects in electrical connections of complex microelectronic devices. Summary of the invention
[0007] In view of the above, according to the independent claims, a method and apparatus for identifying defective electrical connections of a substrate are provided. Further provided is a method of generating a trained computational model for identifying defective electrical connections, and a computer readable storage medium storing the trained computational model. Further aspects, advantages and beneficial features are apparent from the dependent claims, the description and the drawings.
[0008] According to one aspect, a method for identifying a defective electrical connection of a substrate is provided, the substrate having a first surface contact and a first electrical connection extending from the first surface contact. The method comprises the following steps: (a) charging the first surface contact by directing an electron beam onto the first surface contact; (b) detecting a first secondary electron signal as a function of time during the charging of the first surface contact; (c) inputting input data to a trained computational model, in particular a trained machine learning model, the input data comprising or based on the first secondary electron signal; and (d) receiving defect information about the first electrical connection as an output from the trained computational model.
[0009] The input data may further include position information about the first surface contact on the substrate. The trained computational model may be configured to provide defect information as an output based on the input data, the input data including the first secondary electronic signal generated during charging of the first surface contact and the position information of the first surface contact. The input data may optionally include other information or other input parameters, such as information about the substrate and / or information about the first electrical connection, such as material information and / or electrical properties of the first electrical connection that may affect the first secondary electronic signal generated during charging.
[0010] In some embodiments, the substrate is an advanced packaging substrate (AP substrate), a panel level packaging (PLP) substrate, or a wafer level packaging (WLP) substrate.
[0011] According to another aspect, a method of generating a trained computational model (particularly a trained machine learning model) for identifying defective electrical connections of a substrate is provided. The trained computational model can then be used to identify defective electrical connections according to any of the methods described herein. The method includes the following steps: charging a first surface contact by directing an electron beam onto the first surface contact; detecting a secondary electron signal as a function of time during the charging of the first surface contact; determining associated defect information about a first electrical connection extending from the first surface contact; providing a first training data set including input data and associated defect information, the input data including or based on the secondary electron signal; and generating a trained computational model by training the computational model with the first training data set and a plurality of other training data sets, the other training data sets being similarly provided for a plurality of other surface contacts, the plurality of other surface contacts having corresponding electrical connections extending from the plurality of other surface contacts.
[0012] The computational model may be a machine learning model trained based on a machine learning algorithm using a plurality of training data sets.
[0013] In some embodiments, a first training data set includes: (i) a secondary electron signal as a function of time generated (directly or in processed form) during charging of a first surface contact; (ii) position information regarding the position of the first surface contact; and (iii) associated defect information specifying whether a first electrical connection extending from the first surface contact includes a defect. The associated defect information can be determined via one or more voltage contrast measurements. A plurality of other training data sets generated in a similar manner by testing a plurality of surface contacts having defective and / or non-defective electrical connections extending therefrom can be used to train a computational model, such that a trained computational model capable of identifying defects based on input data including a SE signal as a function of time is provided.
[0014] According to another aspect, there is provided an apparatus configured to identify defective electrical connections of a substrate according to any of the methods described herein.
[0015] An apparatus for identifying defective electrical connections of a substrate as described herein may include: a vacuum chamber housing a platform for placing a substrate; an electron source configured to generate an electron beam; a scanning deflector for directing the electron beam onto a first surface contact to charge the first surface contact; a secondary electron detector configured to provide a first secondary electron signal as a function of time during charging of the first surface contact; and a data processing unit having a memory storing a trained computational model, the trained computational model being configured to receive input data including or based on the first secondary electron signal and to provide as output defect information about a first electrical connection extending from the first surface contact.
[0016] According to another aspect, a computer readable storage medium storing a trained computing model, particularly a trained machine learning model, is provided. The trained computing model is configured to: receive input data including or based on a secondary electronic signal detected as a function of time during charging of a surface contact; and in response thereto, provide defect information about an electrical connection extending from the surface contact.
[0017] Embodiments also relate to apparatus for implementing the disclosed methods, and include apparatus parts for performing each described method aspect. These method aspects can be performed by hardware components, a computer programmed by appropriate software, any combination of the two, or in any other manner. In addition, embodiments according to the present disclosure also relate to methods for operating the apparatus described and methods for manufacturing the apparatus and devices described herein. The method for operating the apparatus includes method aspects for implementing each function of the apparatus. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to be able to understand the above features of the present disclosure in detail, a more detailed description of the present disclosure briefly summarized above can be obtained by referring to the embodiments. The accompanying drawings are related to the embodiments of the present disclosure and are described as follows:
[0019] FIG. 1 shows a schematic diagram of a device for identifying a defective electrical connection according to an embodiment described herein;
[0020] 2A-2C schematically illustrate a method of testing an electrical connection according to an embodiment described herein;
[0021] FIG2D shows some examples of secondary electron signals detected as a function of time during charging of corresponding surface contacts;
[0022] FIG3 schematically illustrates a method of identifying a defective electrical connection according to an embodiment described herein;
[0023] FIG. 4 schematically illustrates a method of identifying a defective electrical connection according to an embodiment described herein;
[0024] 5 is a block diagram illustrating a method for identifying defective electrical connections using a trained machine learning model according to embodiments described herein;
[0025] FIG. 6 is a flow chart illustrating a method of identifying a defective electrical connection according to an embodiment described herein; and
[0026] 7 is a flow chart of a method for generating a trained machine learning model for identifying defective electrical connections according to an illustration of embodiments described herein. DETAILED DESCRIPTION
[0027] Reference will now be made in detail to various exemplary embodiments, one or more examples of which are illustrated in each of the accompanying drawings. Each example is provided by way of explanation and is not meant to be limiting. For example, features illustrated or described as part of one embodiment may be used in other embodiments or used with other embodiments to produce further embodiments. The present disclosure is intended to include such modifications and variations. In the following descriptions of the drawings, the same reference numerals refer to the same parts. Only the differences associated with the various embodiments are described. The structures shown in the drawings are not necessarily depicted to true scale, but are used to better understand the embodiments.
[0028] Over the years, the complexity of package substrates has been increasing in order to reduce the space requirements of semiconductor packages. Traditional semiconductor packages are manufactured from semiconductor wafers, which are then cut and packaged. The semiconductor package can then be mounted on a printed circuit board (PCB) along with other microelectronic components.
[0029] In order to reduce manufacturing costs and space requirements, packaging technology has been further developed in recent years, and technologies such as 2.5D IC, 3D-IC, wafer-level packaging (WLP) (such as fan-out WLP or fan-in WLP) and panel-level packaging (PLP) have been proposed. In WLP technology, the integrated circuit is packaged before cutting, while it is still part of the wafer. Therefore, the resulting package is actually the same size as the wafer.
[0030] "2.5D integrated circuits" (2.5D ICs) and "3D integrated circuits" (3D ICs) combine multiple dies in a single integrated package. Here, two or more unpackaged dies are placed on a packaging substrate, such as a silicon interposer. In 2.5D ICs, the dies are placed side by side on the packaging substrate, while in 3D ICs, at least some of the dies are placed on top of other dies. The assembly can be packaged as a single part, which can reduce cost and size compared to traditional 2D circuit board assemblies.
[0031] The package substrate typically includes a plurality of inter-device electrical interconnect paths for providing electrical connections between the die to be placed on the package substrate. The inter-device electrical interconnect paths may extend vertically (perpendicular to the surface of the package substrate) and / or horizontally (parallel to the surface of the package substrate) through the body of the package substrate in a complex network of connections, with endpoints exposed at the substrate surface (referred to herein as surface contacts).
[0032] To reduce manufacturing costs, the panel-level substrates manufactured are configured to integrate multiple devices (e.g., chips / die that may be heterogeneous (e.g., may have different sizes and configurations)) in a single integrated package. The panel-level substrate typically provides chip sites for multiple chips / die placed on its surface (e.g., on one or both sides thereof), as well as multiple inter-device electrical interconnection paths extending through the body of the package substrate. It is worth noting that the size of the panel-level substrate is not limited to the size of the wafer. For example, the panel-level substrate may be rectangular or have another shape. Specifically, the panel-level substrate may provide a larger surface area than the surface area of a typical wafer, for example, 1000 cm² or more. For example, the size of the panel-level substrate may be 30 cm x 30 cm or more, 60 cm x 30 cm or more, 60 cm x 60 cm or more, or larger than the above.
[0033] Advanced packaging (AP) substrates provide electrical interconnect paths between devices on or within a wafer such as a silicon wafer. For example, an AP substrate may include through silicon vias (TSVs) (which are provided, for example, in a silicon interposer) or other conductor lines extending through the AP substrate. Panel-level packaging substrates are typically provided by composite materials, such as printed circuit board (PCB) materials or other composite materials, including, for example, ceramic and glass materials.
[0034] Conventional test apparatus may be inappropriate or unsuitable for testing advanced packaging substrates or panel-level packaging substrates due to the geometry and density of the surface contacts, and / or because the size of the packaging substrate may be different from the size of a conventional bare die or printed circuit board. The present disclosure relates to methods and apparatus for identifying defects in substrates having a plurality of densely arranged surface contacts and a plurality of electrical connections extending between two or more of the surface contacts, respectively. In particular, the methods and apparatus described herein may be applicable to testing packaging substrates that are configured for integrating a plurality of devices in an integrated package and may include at least one inter-device electrical interconnect path extending between a first surface contact and at least one second surface contact.
[0035] As used herein, "surface contacts" may be understood as endpoints of electrical interconnection paths (also referred to herein as "electrical connections") that are exposed at the substrate surface so that an electron beam may be directed onto the surface contacts for contactless charging or probing of the surface contacts. The surface contacts may be used to make electrical contact with a chip / die to be placed on the substrate surface, such as via soldering. For example, the surface contacts may be configured as solder bumps.
[0036] 1 schematically shows an apparatus 100 for identifying defects in electronic connections (e.g., interconnect paths and / or vias extending through a substrate 10) according to embodiments described herein. The apparatus 100 may include a vacuum chamber 101, which may be a test chamber specifically configured for testing, or the vacuum chamber 101 may be a vacuum chamber of a larger vacuum system (e.g., a processing chamber of a substrate manufacturing or processing system). For example, the apparatus may be configured as an in-line inspection apparatus integrated in a substrate processing system.
[0037] The vacuum chamber 101 may house a stage 105, such as a movable stage, for placing the substrate 10 thereon. The apparatus 100 further includes an electron source 120 for generating an electron beam, a scanning deflector 130 for deflecting the electron beam to a predetermined position on the substrate, and an electron detector 140 for detecting secondary electrons emitted from the substrate after the electron beam strikes.
[0038] As schematically shown in FIG1 , the substrate 10 comprises a first surface contact 21 and a first electrical connection 20, the first electrical connection 20 extending from the first surface contact 21 through the substrate, for example extending to one or more other surface contacts that may be located on the same substrate surface as the first surface contact 21 or may be located on an opposite substrate surface (not shown in the figure). The substrate 10 may comprise a plurality of surface contacts and a plurality of electrical connections extending from the plurality of surface contacts, for example 10.000 or more electrical connections, in particular 100.000 or more electrical connections, or even 1.000.000 or more electrical connections. According to the method described herein, the plurality of electrical connections extending from the plurality of surface contacts may be inspected to identify defective connections.
[0039] The apparatus 100 comprises an electron beam column 110, such as a thermal field emitter, having an electron source 120 configured to generate an electron beam 111 propagating along an electron beam path 115 toward a substrate. The electron beam 111 can be directed, in particular focused, onto a predetermined position on the substrate. In particular, the apparatus 100 can comprise a scanning deflector 130 configured to direct the electron beam 111 onto a surface contact, such as onto the first surface contact 21 as described in FIG. 1 . By directing the electron beam 111 onto the surface contact, the surface contact and an electrical connection extending from the surface contact can be charged, in particular negatively charged.
[0040] In some embodiments, the electron beam 111 can be focused on the surface of the substrate 10, for example, by a focusing lens 125, in particular a magnetic and / or electrostatic focusing lens. Specifically, the focusing lens 125 can be configured to focus the electron beam 111 onto the first surface contact 21 so as to charge the first surface contact 21 in a targeted manner.
[0041] Further beam-optical components 171 may optionally be provided along the electron beam path 115 to influence the electron beam 111 , such as a focusing lens and / or an aberration corrector, such as a stigmator and / or a chromator.
[0042] In some embodiments, the electron energy of the electron beam 111 (i.e., the landing energy of the electrons of the electron beam 111 on the substrate surface) can be higher than the neutral charge point. As used herein, "neutral charge point" refers to the electron energy of the electron beam that does not change the charge on the uncharged surface contact when the electron beam impinges on the uncharged surface contact, because the amount of signal electrons emitted from the substrate after the impingement substantially corresponds to the amount of electrons transferred to the surface contact by the electron beam. In some embodiments, the neutral charge point can correspond to an electron energy of the electron beam 111 between 1.5 keV and 3 keV, particularly about 2 keV.
[0043] The probability of electrons striking the substrate with a landing energy above the neutral charge point producing secondary electrons emitted from the substrate may be reduced, so that the substrate is negatively charged when struck by the electron beam with electron energies above the neutral charge point. The probability of electrons striking the substrate with a landing energy below the neutral charge point producing secondary electrons leaving the substrate may be increased, so that the substrate may be discharged when struck by the electron beam with electron energies below the neutral charge point.
[0044] In some embodiments, the electron energy of the electron beam 111 can be 5 keV or greater, in particular about 10 keV, in particular an electron energy above the neutral charge point. Therefore, the amount of signal electrons emitted from the substrate after impact is typically less than the amount of electrons transmitted to the substrate by the electron beam 111. Therefore, a negative charge can be transmitted to the surface contact by the electron beam 111, so that the surface contact together with the electrical connection extending therefrom can be negatively charged. As used herein, "charging" can particularly relate to applying a negative charge (i.e., electrons) to the surface contact to result in a predetermined (negative) potential of the electrical connection extending from the surface contact.
[0045] The device 100 further includes an electron detector 140, which is configured to detect secondary electrons 113 emitted from the substrate 10, in particular during the impact of the electron beam 111. The electron detector 140 can be configured to detect secondary electrons (SE) emitted during the charging of the first surface contact 21 with the electron beam 111 to provide a secondary electron signal 114 as a function of time during the charging period. As used herein, a "secondary electron signal" or "SE signal" can refer to the number of secondary electrons emitted by the surface contact as a function of time during the charging of the surface contact. The number of SEs depends on the surface voltage. Specifically, the secondary electron signal 114 includes information about the time dependence of the surface voltage during charging. Figure 1 schematically shows the secondary electron signal 114 as a graph showing the detected SE yield (x axis) as a function of time (y axis) during charging. During charging, the SE production may be measured at discrete time periods (eg every 100 nm until a saturation value is reached) to provide a secondary electron signal 114 as a (discontinuous) function of time.
[0046] In some embodiments, the electron detector 140 includes an Everhard-Thornley detector. An energy filter for signal electrons can be arranged in front of the electron detector 140, in particular in front of the Everhard-Thornley detector. The energy filter can include, for example, a grid electrode configured to be set at a predetermined potential. The energy filter can enable low-energy signal electrons to be suppressed. The energy filter can be set to optimize voltage contrast detection. Therefore, the signal intensity detected by the electron detector 140 can depend on the energy of the signal electrons, which indicates whether the surface contact is at a predetermined potential.
[0047] It is worth noting that if an image of a surface area is taken by a conventional scanning electron microscope, one is generally only interested in the SE yield of each pixel, rather than the time dependence of the SE signal during the charging process, which distinguishes the embodiments disclosed in this article from conventional imaging scanning electron microscopes.
[0048] The secondary electron signal 114 generated during the charging with the electron beam 111 is time-dependent because during the charging, the amount of negative charge on the first surface contact increases, which increases the production of secondary electrons due to the increasingly negative potential of the first surface contact. Further, the time course of the secondary electron signal 114 depends on the electrical characteristics of the first electrical connection 20, such as the capacitance of the first electrical connection and / or the cross-capacitance relative to adjacent electrical connections. Specifically, electrical connections with small capacitance charge quickly, causing the secondary electron signal 114 to rise rapidly over time, while electrical connections with large capacitance charge slowly, causing the secondary electron signal 114 to rise slowly over time. The time-varying secondary electron signal 114 during the charging of the surface contact can generally follow an S-shaped curve, that is, it rises slowly at the beginning when the electron beam initially strikes, rises sharply in the center portion, and slowly rises toward a saturation value, which corresponds to the potential of the first electrical connection that is no longer increased by the impact of the electron beam (because there are almost no secondary electrons that can leave the surface). However, in reality, the time dependence of the secondary electron signal may show deviations from such an S-shaped curve, for example due to cross-capacitance effects, interactions between adjacent surface contacts, charge accumulation effects, defects and / or other effects.
[0049] If the electrical connection is interrupted due to a defect (i.e., if there is an "open" defect), then the capacitance of the corresponding electrical connection is smaller than expected, which will cause the secondary electron signal to rise unexpectedly quickly over time. If the electrical connection is shorted to another electrical connection due to a defect (i.e., if there is a "short" defect), then the capacitance of the corresponding electrical connection is larger than expected, which will cause the secondary electron signal to rise unexpectedly slowly over time during charging. Therefore, if it is known what the SE signal of the corresponding electrical connection is in a non-faulty situation, then the secondary electron signal generated as a function of time during charging of the surface contact can provide information about the defects of the electrical connection connected to the surface contact.
[0050] Reliably determining defect information about an electrical connection based on a secondary electron signal generated during charging can be challenging because the time dependence of the secondary electron signal generated during charging can depend on multiple factors, for example on the defect class of the defect that may be present and the location of the corresponding surface contact on the substrate. For example, a surface contact located near a corner or edge of the substrate may typically generate a different secondary electron signal when charging than a surface contact located in a central area of the substrate, for example, due to density variations of the surrounding surface contacts, which variations may result in cross-capacitance effects that change the temporal behavior of the SE yield. In particular, the electrical characteristics of the electrical connection and therefore the SE signal generated when charging may depend on the location of the corresponding surface contact from which the electrical connection extends.
[0051] In some embodiments, each of the plurality of surface contacts of the substrate, or at least a subset of the surface contacts having corresponding "position identifiers", if free of defects, can generate a secondary electronic signal of a corresponding characteristic when charged. The position dependence of the time dependence of the secondary electronic signal can make it particularly challenging to reliably determine defect information about the electrical connection.
[0052] In view of the foregoing, the apparatus and method described herein provide a method for reliably identifying defective electrical connections of a substrate even if the substrate has a large number of surface contacts that may each generate characteristic secondary electronic signals during their charging.
[0053] As schematically depicted in FIG. 1 , the apparatus 100 comprises a data processing unit 160 (e.g., a calculator), the data processing unit 160 comprising a trained computational model 500 (e.g., stored in a memory of the data processing unit), the trained computational model being configured to receive input data 510 comprising or based on the first secondary electronic signal 114 as a function of time, and the trained computational model being configured to provide defect information about the first electrical connection 20 as an output.
[0054] In some embodiments, the trained computing model is a machine learning model that has been trained with a corresponding training data set (as described below), in particular a model based on deep learning. In some embodiments, the machine learning model can be based on a neural network, in particular a deep neural network (DNN) having at least one hidden layer, for example a deep neural network having three or more layers. A neural network with fewer layers can also be used, depending on the complexity of the substrate and the number of parameters that affect the SE signal during charging. The description of the deep neural network (DNN) mentioned herein can also be applied to general neural networks (NN) and other artificial intelligence (AI) computing models, in particular machine learning (ML) computing models. Specifically, the computing model may include an input layer, an output layer, and one or more hidden layers between the input layer and the output layer. However, the computing model used according to the embodiments of this article is not limited to being based on a DNN with a hidden layer, and in some embodiments, the relationship between the input data (in particular including the secondary electronic signal, and optionally, the position of the corresponding surface contact) and the output data (in particular including defect information) may be simpler and clearer. For example, the input data and the defect information can be linked via an if-else based relationship that provides a mapping between one or more characteristics obtained from the SE signal and the position information of the corresponding electrical contact on the one hand and the corresponding defect information on the other hand. In other words, the computational model is not limited to a deep learning based model with multiple layers.
[0055] The trained computational model 500 is configured to receive input data 510, the input data 510 including or based on the first secondary electronic signal 114. For example, the input data 510 may include the secondary electronic signal 114 in an unprocessed form or a processed form, such as one or more portions of the secondary electronic signal 114 (e.g., an initial portion until a predetermined threshold potential of the first surface contact is reached; or a change or steepness value of the secondary electronic signal 114 at one or more predetermined times during charging). In some embodiments, the input data 510 may include the secondary electronic signal 114 as a function of time from the start of charging to a certain point in time (e.g., until a predetermined potential is reached).
[0056] The input data 510 may optionally include other information, in particular position information about the position of the first surface contact 21 on the substrate. For example, the input data 510 may include the (absolute) position of the first surface contact 21 on the substrate (e.g., the x-coordinate and the y-coordinate on the substrate surface) or the relative position of the first surface contact 21 relative to one or more other surface contacts (e.g., "corner position", "edge position", "center position", the number of adjacent contacts, etc.). In some embodiments, the position information may include a position identifier that characterizes the position of a first surface contact among a plurality of surface contacts, and / or characterizes the position of a first electrical connection among a plurality of electrical connections. In one example, for a surface contact in the nth column and the mth row of surface contacts in a two-dimensional surface contact array, its position identifier is "n, m", and / or the position identifier is "x", where x identifies a subgroup of surface contacts with common electrical characteristics to which the measured surface contact belongs, and n, m, and x are integers.
[0057] In some embodiments, both the SE signal as a function of time during charging and the position information of the corresponding surface contact can be used as input parameters (i.e., as part of the input data) to reliably obtain defect information as an output from the computational model. As described above, each surface contact position can result in a SE signal with corresponding characteristics, or a corresponding subgroup of surface contacts can be characterized by a corresponding typical SE signal, so that if both the "SE signal" and the "position information:" are used as input parameters of the computational model, defect information can be reliably obtained.
[0058] Optionally, the input data input to the calculation model may include any one or more of the following information: (1) information about the substrate, in particular the substrate type, substrate material, one or more substrate layer materials, substrate design rules and / or substrate identifier. For example, the substrate identifier may define the electrical connections and / or surface contacts present on a specific substrate type, such as the number and arrangement of the electrical connections. If a substrate with the same substrate identifier has been previously tested multiple times and the calculation model has been trained with a training data set obtained by the tests, then the substrate identifier as part of the input data may improve the reliability of the defect information obtained. (2) information about the first electrical connection, in particular a capacitance value, a cross-capacitance value providing information about the cross-capacitance relative to the adjacent electrical connection, the number of surface contacts connected to the first electrical connection, the number of adjacent surface contacts, and / or a connection identifier characterizing the type, arrangement and / or electrical characteristics of the electrical connection. For example, if an electrical connection with the same connection identifier has been previously tested multiple times and the calculation model has been trained with a training data set obtained by the tests, then the connection identifier as part of the input data may improve the reliability of the defect information obtained.
[0059] The trained computational model 500 can be configured to provide defect information about the first electrical connection 20 as an output. In some embodiments that can be combined with other embodiments described herein, the defect information provided by the trained computational model includes information about whether the first electrical connection appears to be defective. In particular, the defect information can include a binary defect value (e.g., "0" = "no defect"; "1" = "defective"). Optionally, if a defect is detected, the defect information can further include a defect category, such as a short defect, an open defect, or a leakage defect. Alternatively or additionally, if a defect is detected, the defect information can include a defect location that locates the location of the defect at or in the substrate. In some embodiments, the defect location can specify the location of the defective electrical connection at or in the substrate (e.g., a "location identifier" that identifies the defective electrical connection among multiple electrical connections of the substrate). In some embodiments, the defect location can further identify the location of the defect along the defective electrical connection (e.g., locating the defect location between two specific surface contacts, or locating the defect at the intersection between two specific electrical connections, such as in the case of a short defect). Alternatively or additionally, the defect information may further include reliability information characterizing the reliability of the defect information. For example, in some cases, the trained computational model may detect defects and / or specific defect categories with a high reliability of 90% or greater. In other cases, for example, if the secondary electronic signal 114 deviates only slightly from the expected secondary electronic signal of a non-defective connection, the trained computational model may not be able to identify the defect with high reliability. For example, the trained machine learning model may provide defect information including a reliability value (which is expressed, for example, in %) indicating the reliability that the detected defect and / or the detected defect category or defect location actually exists.
[0060] In some embodiments that may be combined with other embodiments described herein, a substrate has a plurality of surface contacts (with corresponding electrical connections extending therefrom), in particular 10.000 or more, 100.000 or more, or even 1.000.000 or more surface contacts, which are tested sequentially according to the test method described herein. In particular, each of the plurality of surface contacts can be tested as follows: the surface contact is charged by directing an electron beam onto the surface contact (e.g., by deflecting with a scanning deflector 130) (stage (a)). The SE signal generated during the charging of the corresponding surface contact is detected as a function of time with an electron detector 140 (stage (b)). Input data based on the SE signal, and optionally based on the position of the corresponding surface contact, is generated and input into the trained computational model 500 (stage (c)). Defect information about the electrical connection extending from the corresponding surface contact is received as an output from the trained computational model 500 (stage (d)). Therefore, by deflecting the electron beam onto surface contacts connected to the corresponding electrical connections, by detecting the SE signal as a function of time during charging, and by analyzing the SE signal using a trained computational model 500 (in particular, a trained machine learning model, more particularly a trained deep learning-based model), multiple electrical connections can be tested sequentially.
[0061] In some embodiments, the substrate 10 is an advanced packaging substrate or a panel-level packaging substrate configured to provide multi-device in-package interconnection, and the first electrical connection is an inter-device electrical interconnection path. For example, the substrate can be a panel-level packaging (PLP) substrate, a wafer-level packaging (WLP) substrate, or a micro-LED substrate. The packaging substrate can include a plurality of 1.000.000 or more surface contacts having corresponding electrical connections extending therefrom, all of which can be tested in turn.
[0062] In some embodiments, the plurality of surface contacts are distributed over a substrate surface area or a substrate surface sub-area of 16 cm² or more, particularly 25 cm² or more, more particularly 100 cm² or more, or even 225 cm² or more. The method may include the following steps: sequentially deflecting the electron beam 111 onto the plurality of surface contacts using a scanning deflector 130 to sequentially charge and test electrical connections extending from the plurality of surface contacts using the trained computational model.
[0063] In some embodiments, the scanning deflector 130 can provide a deflection area (i.e., an area of the substrate surface that can be reached by deflecting the electron beam with the scanning deflector 130 without moving the platform 105) of 16 cm² or more (particularly 100 cm² or more, more particularly 225 cm² or more). The large deflection area can quickly and accurately test a large number of electrical connections because the substrate does not need to be moved to test multiple electrical connections. In particular, at least one complete chip site on the substrate surface can be tested without moving the platform by deflecting the electron beam onto the surface contacts of the chip site.
[0064] The trained computational model 500 may be generated in a previous training phase as follows: the computational model (particularly the machine learning model) is trained with a plurality of training data sets by a machine learning algorithm. Each training data set comprises input data (which comprises or is based on a SE signal detected as a function of time during charging of a surface contact) and defect information about an electrical connection extending from the surface contact associated with the input data. The input data may optionally include any other input parameters mentioned above, in particular position information about the position of the corresponding surface contact.
[0065] After the trained computational model 500 is generated, the trained computational model 500 model can be used to reliably and quickly check whether a specific electrical connection is defective based on the SE signal generated during charging.
[0066] In some embodiments that may be combined with other embodiments described herein, the apparatus further comprises a discharge device for discharging at least a portion of the substrate, in particular for discharging the first surface contact before and / or after charging the first surface contact. For example, the first surface contact may be discharged before charging to ensure that charging starts from a predetermined potential of the first surface contact. For example, if multiple charges are already present on the first surface contact before charging, the secondary electron signal 114 may appear different than expected due to the existing charges affecting the SE yield. Therefore, it may be beneficial to discharge the first surface contact before charging and testing. In particular, each of the multiple surface contacts may be discharged before charging and testing thereof.
[0067] Alternatively or additionally, the first surface contact can be discharged after charging and testing the first surface contact. The discharge after the test reduces or avoids the accumulation of charge on the substrate surface, which may distort the results of subsequent test measurements. In particular, the charge on the substrate surface may deflect the electron beam and / or may affect the SE signal detected by the electron detector, for example, if the charged surface contact is arranged near the first surface contact currently being tested. Therefore, it may be beneficial to discharge the first surface contact after charging and testing. In particular, each of the multiple surface contacts can be discharged after charging and testing thereof. In particular, each of the multiple surface contacts can be discharged before and after testing.
[0068] In some embodiments, the discharge device may include any of the following: a second electron source configured to generate a second electron beam for discharge, an electron flood gun configured to discharge a large area of the substrate surface, and / or a UV discharge lamp. The second electron source may be configured to generate a second electron beam having a second electron energy that is different from the electron energy of the electron beam 111. In particular, the second electron energy of the second electron beam may be lower than the neutral charge point, such as 2 keV or less, such as about 1.5 keV, so that the second electron beam can be used to remove charge from the substrate.
[0069] 2A-2C schematically illustrate a method for identifying a defective electrical connection. FIG. 2A shows a test of a first electrical connection 20 extending from a first surface contact 21, wherein the first electrical connection 20 is free of defects. An electron beam 111 is directed onto the first surface contact 21 to charge the first surface contact 21, and secondary electrons 113 emitted from the first surface contact are detected by an electron detector 140 during charging. A first secondary electron signal 114 corresponding to the SE yield as a function of time during charging is detected and input (directly or in a processed form) to a trained computational model as part of the input data. The input data may further include a location identifier or another location information about the first surface contact 21 or the first electrical connection 20 under test. In the example of FIG. 2A, the first electrical connection 20 is free of defects, resulting in a characteristic time dependency of the first secondary electron signal 114 (note: the time dependency of the SE yield generally depends on the location of the electrical connection under test), so that the trained computational model can determine that the first electrical connection is free of defects ("defect-free") based on the input data.
[0070] In particular, the trained computational model may have been previously trained using a training data set obtained by testing a plurality of non-defective (and defective) electrical connections having the same location identifier as the first electrical connection, such that the trained computational model can identify whether the first electrical connection 20 is defective based on the input data. In some embodiments, the absolute SE yield at one or more times during charging, the steepness of the SE curve at one or more times during charging, the endpoint potential, and / or the characteristic shape of the SE signal over time can provide the trained computational model with an indication of whether the first electrical connection is defective and what type of defect may be present.
[0071] In some embodiments, if the secondary electronic signal 114 is partially and / or completely outside of a predetermined range 224 that is characteristic of a SE signal for a non-defective electrical connection extending from a surface contact having the same location identifier and / or the same electrical characteristics, then the trained computational model can identify that the first electrical connection 20 is defective. The predetermined range 224 associated with a particular location identifier can be defined by the computational model by training with multiple training data sets having the same location identifier. For example, the predetermined range 224 can be defined by the computational model as a range around a typical (e.g., average) secondary electronic signal that varies over time for non-defective electrical connections having the same location identifier.
[0072] FIG. 2B shows a test of the first electrical connection 20 extending from the first surface contact 21 if the first electrical connection 20 is defective. Specifically, the first electrical connection 20 depicted in FIG. 2B is interrupted, i.e., includes an open defect 31. As a result, the capacitance of the first electrical connection is reduced, which causes the detected secondary electronic signal 114' to rise faster because the first electrical connection is charged faster. The trained computational model can identify that the first electrical connection 20 is defective and includes an open defect 31 based on input data including the secondary electronic signal 114'. Further, the reliability of the defect information and the defect location can optionally be included in the output data provided by the trained computational model.
[0073] The secondary electronic signal 114' has an unexpectedly high steepness or gradient, which may indicate to the trained computational model that the first electrical connection is defective, in particular, interrupted. Alternatively or additionally, the trained computational model may recognize that the secondary electronic signal 114' is outside a predetermined range 224 defined by the computational model for non-defective electrical connections associated with the same location identifier based on training with a plurality of training data sets.
[0074] FIG2C shows a test of a first electrical connection 20 extending from a first surface contact 21 if the first electrical connection has another defect. Specifically, the first electrical connection 20 depicted in FIG2C is shorted to the second electrical connection 24, i.e., includes a short defect 32. As a result, the capacitance of the first electrical connection increases, which causes the detected secondary electronic signal 114'' to rise more slowly because the first electrical connection is charged more slowly. The trained computational model can identify that the first electrical connection 20 is defective and includes a "short" defect 31 based on input data including the secondary electronic signal 114''.
[0075] The secondary electronic signal 114'' has an unexpectedly slow steepness or gradient in an initial portion, which may indicate to the trained computational model that the first electrical connection is defective, in particular a short circuit. Alternatively or additionally, the trained computational model may recognize that the secondary electronic signal 114''' is outside a predetermined range 224 defined by the computational model for non-defective electrical connections associated with the same location identifier based on training with multiple training data sets.
[0076] If a short defect is detected, then one or more surface contacts of adjacent electrical connections may be subsequently charged with the electron beam 111 to further characterize the defect, for example, to determine the defect location. For example, if the secondary electron signal generated when the second surface contact 22 is subsequently charged begins with an unexpectedly high SE yield, then the short defect 31 may be identified as being located between the first electrical connection 20 and the second electrical connection 24 extending from the second surface contact 22. Alternatively or additionally, a reliability value may be included in the output data provided by the trained computational model. For example, if the SE signal is far outside the predetermined range 224, or has a gradient that is strongly different from the expected gradient of the SE signal for an electrical connection without a defect, then the reliability value of the detected defect may be high (e.g., close to 100%).
[0077] FIG2D shows some other examples of secondary electron signals 214 detected as a function of time t during charging of surface contacts, which are respectively connected to defective electrical connections. Different defects lead to different SE signals during charging. In some embodiments, if the secondary electron signal as a function of time exceeds a predetermined range 224 (which is represented by the gray area in FIG2D), the trained computational model can identify the corresponding electrical connection as defective. The predetermined range 224 can be defined by the computational model based on training the computational model with multiple training data sets.
[0078] In some embodiments, the trained computational model may define a corresponding predetermined range for each of a plurality of location identifiers characterizing a surface contact location. The predetermined range may be defined by the computational model as a result of training with a corresponding plurality of training data sets, wherein each training data set includes an SE signal, location information (e.g., a location identifier) identifying a location of a corresponding surface contact on a substrate, and defect information specifying whether the corresponding electrical connection is defective.
[0079] FIG3 schematically illustrates a method of testing a second electrical connection 24 extending from a second surface contact 22. The second surface contact 22 has a different position identifier than the first surface contact 21 tested in FIG2A. Specifically, the second surface contact 22 is located (in the cross section of FIG3) between two adjacent surface contacts, which results in an increased cross-capacitance experienced by the second electrical connection 24 compared to the first electrical connection 20. Therefore, assuming that both the first electrical connection and the second electrical connection are not defective, the secondary electronic signal 314 generated as a function of time during the charging of the second surface contact 22 is different from the secondary electronic signal 114 (see FIG2A) generated as a function of time during the charging of the first surface contact 21.
[0080] The secondary electronic signal 314 is provided as part of the input data together with a corresponding position identifier characterizing the position of the second surface contact 22 to a trained computing model, which provides as output defect information (here: "no defect") about the second electrical connection 24. In particular, if the secondary electronic signal 314 is within a predetermined range defined for the surface contact with the corresponding position identifier, then the trained computing model can identify that no defect is present.
[0081] In some embodiments that may be combined with other embodiments described herein, the computational model is configured to, based on a training data set for each of a plurality of location identifiers that respectively characterize the location of one or more surface contacts, define a corresponding predetermined range that can be compared to a secondary electronic signal measured during charging of a surface contact having the corresponding location identifier to determine whether the surface contact is connected to a defective electrical connection.
[0082] FIG4 schematically illustrates a method for identifying defective electrical connections of a substrate 10 according to an embodiment described herein. The substrate 10 may be a PLP substrate including a plurality of chip sites 401, 402, 403, each chip site being configured to place a corresponding chip, and 1.000 or more inter-device electrical interconnection paths may extend from each chip site to one or more other chip sites through the PLP substrate. For example, the substrate may include two, three or more pairs of associated chip sites, and 1.000 or more inter-device electrical interconnection paths (which may optionally have exactly two surface contact points, i.e., one in each chip site in a pair) may extend through the package substrate between each pair of chip sites, as schematically described for the upper paired chip sites in FIG4.
[0083] Each of the plurality of surface contacts located at the first chip site 401 may have a corresponding location identifier. Surface contacts arranged at corresponding locations at different chip sites may have corresponding location identifiers. For example, three surface contacts respectively arranged at the upper left corners of corresponding chip sites, shown as black circles in FIG. 4 , may have corresponding location identifiers because the electrical properties of the electrical connections extending therefrom may be similar or substantially the same. Alternatively, each surface contact of a particular substrate type may have a corresponding location identifier. The behavior of the secondary electronic signals generated when charging the surface contacts having the same location identifier as a function of time during their charging may be substantially similar or the same.
[0084] The surface contacts and the corresponding electrical connections are grouped according to the corresponding electrical properties that cause similar or identical SE signals when charged, and each group has a corresponding position identifier, which facilitates training of the computational model and improves the reliability of the defect information provided by the computational model as an output. Specifically, the trained computational model can define a characteristic time behavior (specifically, the "predetermined range" explained above) of the SE signal associated with the corresponding position identifier, which indicates a specific defect or a non-defective electrical connection.
[0085] 5 is a block diagram illustrating a method of identifying defective electrical connections using a trained computational model 500 according to embodiments described herein. The trained computational model 500 may be stored in a computer-readable storage medium, such as a computer memory.
[0086] According to embodiments described herein, the trained computational model 500 is configured to receive input data 510 comprising or based on a secondary electronic signal 114 detected as a function of time during charging of a surface contact, and is configured to provide as an output, in response thereto, defect information 511 regarding an electrical connection extending from the surface contact.
[0087] As schematically depicted in FIG5 , during charging of the surface contact with the electron beam, secondary electrons 113 are detected by the electron detector 140, and the secondary electron signal 114 detected as a function of time during the charging is forwarded to an input data generator 515, which provides input data 510 based on the secondary electron signal 114. The input data generator 515 may generate the input data 510 based on the secondary electron signal 114 and optionally based on other input parameters, in particular based on position information about the position of the detected surface contact, in particular a position identifier. The input data 510 may optionally include other information, such as information about the substrate and / or information about the electrical connections, which may facilitate the trained computational model 500 to identify defects.
[0088] The trained computational model 500 provides as output defect information 511, which indicates whether the electrical connection is defective. The defect information may also include defect category, defect location, and / or reliability value.
[0089] The trained computational model 500 may be generated by a previous training phase. Training may include generating a plurality of training data sets 551 and training the computational model 501 with the plurality of training data sets 551, for example in a training process 550 utilizing a machine learning algorithm that may depend on the computational model 501 to be trained.
[0090] Each training data set in the plurality of training data sets 551 may include input data and associated defect information determined in other ways. The input data may include or be based on secondary electron signals detected during charging of the corresponding surface contact, and may further include position information. The associated defect information may be determined, for example, by performing a generally known voltage contrast measurement.
[0091] For example, associated defect information about a first electrical connection extending from a first surface may be determined via one or more voltage contrast measurements, in particular by probing with an electron beam or a second electron beam any of: (i) a first surface contact; (ii) one or more second surface contacts that should be electrically connected to the first surface contact; and (iii) one or more third surface contacts that should be electrically disconnected from the first surface contact. Alternatively or additionally, the associated defect information may be determined "manually" by operator classification.
[0092] The input data of each training data set may optionally further comprise position information about the position of the respective surface contact on the substrate, in particular a position identifier. The input data of each training data set may optionally comprise further information about the substrate and / or about the respective electrical connection under test, such as a substrate identifier and / or a connection identifier.
[0093] In some embodiments that can be combined with other embodiments described herein, a plurality of corresponding training data sets are provided for surface contacts having corresponding position information or corresponding position identifiers. In particular, several surface contacts arranged at corresponding positions of different sub-areas of a substrate (e.g., different chip sites) or arranged at corresponding positions of different substrates of the same type are tested, thereby providing a plurality of training data sets for surface contacts that are positionally related.
[0094] It may be beneficial to group surface contacts according to corresponding position identifiers, because surface contacts with corresponding position identifiers may generally have similar electrical properties, which may result in similar or identical SE signals when charged. If the training data set includes not only corresponding SE signals but also corresponding position information, so that when used for testing, the position information about the tested electrical connection can also be used as an input parameter of the trained computational model, then training can be facilitated, and the trained computational model can provide more reliable defect information in the future.
[0095] The computational model 501 may be a machine learning model, in particular a model based on deep learning. In some embodiments, the machine learning model may be based on a neural network, in particular a deep neural network (DNN) having at least one hidden layer, such as a deep neural network having three or more layers. A neural network with fewer layers may also be used.
[0096] Each layer of a DNN can include multiple basic computational elements (CEs), which are often referred to in the art as dimensions, neurons, or nodes. The computational elements of a given layer are connected to the CEs of subsequent layers through connections. Each connection between a CE of the previous layer and a CE of the next layer can be associated with a weighted value.
[0097] A given hidden CE can receive input from a previous layer of CE via a corresponding connection, and each given connection is associated with a weighted value that can be applied to the input of the given connection. The weighted value can determine the relative strength of the connection, thereby determining the relative influence of the corresponding input on the output of the given CE. A given hidden CE can be configured to calculate an activation value (e.g., a weighted sum of the inputs) and further derive the output by applying an activation function to the calculated activation. The activation function can be, for example, an identity function, a deterministic function (e.g., linear, sigmoid, threshold or similar function), a random function, or other suitable function. The output from a given hidden CE can be transmitted to a CE of a subsequent layer via a corresponding connection. Similarly, each connection at the output of a CE can be associated with a weighted value that can be applied to the output of the CE, which is then received as an input to a CE of a subsequent layer. In addition to the weighted value, there can also be a threshold (including a limit function) associated with the connection and the CE.
[0098] The weights and / or thresholds of the neural network can be preliminarily selected before training, and can be further iteratively adjusted or modified during training to achieve an optimal set of weights and / or thresholds in the trained computational model. A set of DNN input data used to adjust the weights / thresholds of the computational model is referred to herein as a training data set.
[0099] It should be noted that the teachings of the test methods disclosed herein are not constrained by the number of hidden layers and / or the DNN architecture. As a non-limiting example, the layers in the DNN can be convolutional, fully connected, locally connected, pooled / sub-sampled, cyclic, and the like. In some examples, the computational model is an if-else based model. For example, the computational model can define a predetermined range for each location identifier, and if the corresponding electrical connection is not defective, then the corresponding SE signal is expected to be within the range. If during the test, the measured SE signal is within the predetermined range, then the trained computational model identifies it as "non-defective", otherwise, the trained computational model identifies it as "defective".
[0100] 6 is a flow chart illustrating a method of identifying a defective electrical connection according to embodiments described herein.
[0101] In block 601, a first surface contact is charged by directing an electron beam onto the first surface contact under test.
[0102] At block 602, a first secondary electronic signal is detected as a function of time during a charging period.
[0103] In block 603, input data is generated based on the detected first electronic signal. Specifically, the input data may include not only the first electronic signal (or a portion or characteristic thereof), but also other information, such as position information about the first surface contact, such as a position identifier characterizing the position of the first surface contact on the substrate surface. The input data is input into a trained computing model, in particular a trained machine learning model.
[0104] In block 604, defect information about the first electrical connection is received as an output from the trained computational model. The defect information may indicate whether the first electrical connection is defective, and if so, may further include any of a defect category, a defect location, and / or a reliability value.
[0105] In block 605, the method may continue by deflecting the electron beam to impact the second surface contact to charge and test a second electrical connection extending from the second surface contact. The inspection of the second electrical connection may be similar to the first electrical connection.
[0106] Optionally, any surface contacts may be discharged before and / or after charging to avoid charge accumulation on the substrate that could negatively impact defect detection.
[0107] 7 is a flow chart illustrating a method of generating a trained computational model for identifying defective electrical connections according to embodiments described herein.
[0108] In block 701, a first surface contact is charged by directing an electron beam onto the first surface contact.
[0109] In block 702, a first secondary electronic signal is detected as a function of time during charging.
[0110] In block 703, associated defect information about a first electrical connection extending from a first surface contact is determined. The associated defect information can be obtained via one or more voltage contrast measurements. For example, the first surface contact, one or more second surface contacts that should be electrically connected to the first surface contact, and / or one or more third surface contacts that should be electrically separated from the first surface contact can be detected, for example, using an electron beam or a second electron beam. If the first surface contact and the one or more second surface contacts are not at the same electrical potential, then an "open circuit" defect can be determined. If the first surface contact and the one or more third surface contacts are at the same electrical potential, then a "short circuit" defect can be determined.
[0111] In block 704, a first training data set is provided, which includes input data and associated defect information determined in block 703. The input data includes or is based on the secondary electronic signal and optionally other information, such as position information about the first surface contact.
[0112] In block 705 , a trained computing model is generated by training the computing model with the first training data set provided in block 704 (particularly by using a machine learning algorithm).
[0113] Multiple other training data sets can be provided similarly to the first training data set by testing multiple other surface contacts having corresponding electrical connections extending therefrom, and the machine learning model can be trained with the multiple other surface contacts. In particular, multiple surface contacts having the same position identifier can be tested to provide corresponding training data sets including the position identifier as a parameter. Using the position identifier as an input parameter of the trained computational model during testing improves the reliability of the defect information, because the electrical characteristics of the electrical connection (and therefore the time dependence of the SE yield) may strongly depend on the position of the tested surface contact on the substrate.
[0114] For example, using a self-learning algorithm, a trained machine learning model can be further improved over time by training with additional training data sets that may be generated during defect inspection and classification using the trained computational model.
[0115] According to the embodiments described herein, methods are provided for quickly and reliably identifying defective electrical connections of substrates, which use trained computational models, in particular trained machine learning models and deep learning-based algorithms. The secondary electron signal measured as a function of time during the charging of the corresponding surface contact is used as an input parameter of the trained computational model. Other input parameters (e.g., position information about the contact of the tested surface) can be given to the trained machine learning model. Artificial intelligence and machine learning techniques are used to detect defective electrical connections, which improves the detectability of defects and provides higher defect classification accuracy and purity compared to traditional electron beam inspection techniques based on voltage contrast measurement. By training with other training data sets, the computational model can be improved over time, so that a self-learning and self-improving defect detection algorithm can be provided. The methods described herein can be applied to testing complex substrates with a large number of electrical connections, in particular AP substrates and / or PLP substrates, where the SE signal generated during charging can depend on multiple parameters, including the position of the tested surface contact on the substrate.
[0116] While the foregoing is directed to certain embodiments, other and further embodiments may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow.
Claims
1. A method of identifying a defective electrical connection of a substrate (10), the substrate having a first surface contact (21) and a first electrical connection (20) extending from the first surface contact, the method comprising the following steps: (a) charging the first surface contact by directing an electron beam (111) onto the first surface contact; (b) detecting a first secondary electron signal (114) as a function of time during said charging of said first surface contact; (c) inputting input data (510) into the trained computational model (500), the input data comprising or based on the first secondary electronic signal (114); as well as (d) receiving defect information (511) about the first electrical connection (20) as an output from the trained computational model (500).
2. The method of claim 1, wherein the input data (510) input to the trained computational model (500) further comprises position information about the first surface contact (21).
3. The method of claim 2, wherein the position information comprises at least one of an absolute position of the first surface contact on the substrate, a relative position of the first surface contact relative to one or more other surface contacts, and a position identifier.
4. The method according to any one of claims 1 to 3, wherein the input data input to the trained computing model further includes any one or more of the following information: information about the substrate (10), in particular substrate type, substrate material, one or more substrate layer materials, substrate design rules and / or a substrate identifier; and Information about the first electrical connection (20), in particular a capacitance value, a cross-capacitance value, a number of surface contacts connected to the first electrical connection, a number of adjacent surface contacts, and / or a connection identifier.
5. The method according to any one of claims 1 to 4, wherein the defect information about the first electrical connection comprises any one or more of the following: information about whether the first electrical connection (20) comprises a defect, Defect category, The defect location, and Reliability value.
6. The method of any one of claims 1 to 5, wherein the substrate has a plurality of surface contacts, the plurality of surface contacts having corresponding electrical connections extending from the plurality of surface contacts, and wherein steps (a), (b), (c), (d) are performed on each of the plurality of surface contacts to obtain defect information about the corresponding electrical connections. 7 . The method of claim 6 , wherein the substrate is a panel level packaging substrate or an advanced packaging substrate, and the plurality of surface contacts comprises 1.000.000 or more surface contacts.
8. The method according to claim 6 or 7, wherein the plurality of surface contacts are distributed over a surface area of 16 cm² or more of the substrate, the method further comprising the steps of: The electron beam is deflected onto the plurality of surface contacts by a scanning deflector (130) to charge and test the plurality of surface contacts in sequence.
9. The method according to any one of claims 1 to 8, further comprising generating the trained computational model (500) by the following steps: The computational model (501) is trained with a plurality of training data sets (551), the plurality of training data sets respectively comprising input data and associated defect information about respective electrical connections, wherein the input data respectively comprise or are based on secondary electron signals detected as a function of time during charging of a surface contact from which the respective electrical connection extends. 10 . The method according to claim 9 , wherein the input data further comprises position information about the surface contact, respectively.
11. A method of generating a trained computational model (500) for identifying defective electrical connections of a substrate, the method comprising the steps of: charging the first surface contact (21) by directing an electron beam (111) onto the first surface contact; detecting a secondary electron signal as a function of time during charging of the first surface contact (114); determining associated defect information about a first electrical connection (20) extending from the first surface contact; providing a first training data set including input data including or based on the secondary electronic signal and the associated defect information; as well as The trained computational model is generated by training the computational model with the first training data set and a plurality of other training data sets similarly provided for a plurality of other surface contacts having corresponding electrical connections extending therefrom.
12. A method according to claim 11, wherein the associated defect information about the first electrical connection is determined via one or more voltage contrast measurements, in particular by using an electron beam (111) or a second electron beam to detect one or more of the following items: the first surface contact, one or more second surface contacts that should be electrically connected to the first surface contact, and one or more third surface contacts that should be electrically separated from the first surface contact.
13. A method according to claim 11 or 12, wherein the input data of the first training data set further includes position information about the position of the first surface contact on the substrate, in particular, wherein the input data of each training data set of the multiple other training data sets further includes position information about the position of the corresponding surface contact on the substrate.
14. A method according to any one of claims 11 to 13, wherein corresponding multiple training data sets are provided for surface contacts having corresponding position information or corresponding position identifiers, in particular, wherein several surface contacts arranged at corresponding positions of different sub-areas of the substrate or at corresponding positions of different substrates are tested, thereby providing multiple training data sets for positionally related surface contacts.
15. The method according to any one of claims 1 to 14, wherein the trained computational model is a trained machine learning model, in particular a trained deep learning based model based on a deep neural network having at least one hidden layer.
16. A device (100) for identifying a defective electrical connection of a substrate (10), the device comprising: A vacuum chamber (101), the vacuum chamber (101) accommodating a platform (105) for placing the substrate; an electron source (120), the electron source (120) being configured to generate an electron beam (111); a scanning deflector (130) for directing the electron beam onto a first surface contact (21) to charge the first surface contact; an electron detector (140) configured to detect a first secondary electron signal (114) as a function of time during said charging of said first surface contact; as well as A data processing unit (160) having a memory storing a trained computational model (500), the trained computational model being configured to receive input data (510) comprising or based on the first secondary electronic signal (114) and to provide as output defect information about a first electrical connection (20) extending from the first surface contact (21).
17. The apparatus of claim 16, wherein the scanning deflector (130) is configured to provide a deflection area of 16 cm² or more.
18. The apparatus according to claim 16 or 17, further comprising a discharge device for discharging at least a portion of the substrate, in particular a second electron source, a flood electron gun or a UV discharge lamp configured to generate a second electron beam with a second electron energy.
19. The apparatus according to any one of claims 16 to 18, configured to perform the method according to any one of claims 1 to 15.
20. A computer-readable storage medium storing a trained computing model (500), wherein the trained computing model is configured to: receiving input data (510) comprising or based on a secondary electronic signal (114) detected as a function of time during charging of the surface contact, and In response thereto, defect information regarding an electrical connection extending from the surface contact is provided as an output.
21. The computer readable storage medium of claim 20, wherein the trained computational model (500) is generated according to any one of claims 11 to 15.
