Test system and test method
By using an automated test system in integrated circuit manufacturing and using neural networks to identify probe marks, the quality problems caused by human errors are solved, testing efficiency and productivity are improved, and costs are reduced.
Patent Information
- Application Number
- CN202311297136.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-31
- Filing Date
- 2023-10-09
- Publication Date
- 2024-12-03
AI Technical Summary
During the manufacturing process of integrated circuits, probe marking inspection is prone to introduce human errors, resulting in poor test quality, long time to consume and high production costs, which affects time to market and product quality.
Design a test system that includes an evaluation subsystem, a neural network subsystem and a program control processor, receives test images through automated mode, uses a neural network to identify the image specifications marked by probes, generates analysis data, and compares it with quality thresholds to generate Check the results and reduce manual intervention.
It significantly reduces manual errors, improves testing efficiency and productivity, improves product quality and customer service levels, and reduces production costs.
Smart Images

Figure CN119064750A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to a testing system and a testing method, and more particularly to a testing system and a testing method for identifying probe marks generated during a testing process through a neural network. Background Art
[0002] In integrated circuit manufacturing technology, testing is the final step to detect defects generated during the integrated circuit manufacturing process and determine the root cause of these defects. Before the packaging process, circuit probe inspection is performed between wafers to verify whether each die meets product specifications.
[0003] Since probe mark inspection is performed by operators, human errors may be introduced into the process, which affects the test quality and is time-consuming. For example, test cycle time and yield are affected. Detecting probe marks based on customer complaints also affects the manufacturer's overall service quality, and the need for operators and engineers to deal with it also increases production costs. As a result, time to market, scrap rate, and product quality are all affected. Therefore, there is a huge demand for methods and systems to automatically analyze probe marks and take corrective actions in order to improve customer service quality. Summary of the invention
[0004] One embodiment of the present disclosure is related to a test system, including an evaluation subsystem, a neural network subsystem, and a program control processor. The evaluation subsystem receives a test image of a tested wafer from a probe device. In response to the probe device obtaining the test image, the program control processor controls the evaluation subsystem to perform an evaluation operation to transmit the test image to the neural network subsystem in an automated mode. The neural network subsystem identifies the image specifications of the probe mark in the test image and generates analysis data of the test image to the evaluation subsystem. The evaluation subsystem also generates a first probe mark inspection result to the program control processor based on the analysis data to generate a test result.
[0005] In some embodiments, the evaluation subsystem is further configured to generate training data for the neural network subsystem by comparing the test image with the analyzed image in the analysis data. The training data includes the test image and the corresponding identification data.
[0006] In some embodiments, the neural network subsystem is further configured to be trained by training data to update a plurality of weight values in a neural network model in the neural network subsystem, the neural network model operating in the neural network processor.
[0007] In some embodiments, the image specification includes coordinates of probe marks, a number of probe marks, inspection of the substrate of at least one pad in the test wafer, or a combination thereof.
[0008] In some embodiments, the evaluation subsystem is configured to compare the analysis data with a plurality of quality thresholds and generate a first probe mark inspection result indicating a first failure result when at least one value in the analysis data fails to satisfy a corresponding threshold in the quality thresholds.
[0009] In some embodiments, the program-controlled processor is further configured to generate a test result indicating a second fault result in response to the first probe mark inspection result indicating a first fault result.
[0010] In some embodiments, the analysis data includes a plurality of distances between the probe marks and a plurality of edges of a plurality of pads in the tested wafer.
[0011] In some embodiments, the test system further includes a database subsystem. The database subsystem includes a data server, and the data server is used to store the test image, the analysis data, and the first probe mark inspection result in the list data. The evaluation subsystem is also used to access the list data and generate the second probe mark inspection result based on the list data and the quality threshold.
[0012] In some embodiments, the neural network subsystem is also used to access inventory data and be trained by training data to update the neural network model operating in the neural network subsystem.
[0013] In some embodiments, the evaluation subsystem is further configured to perform evaluation operations on a plurality of batches of images according to a work queue provided by the program-controlled processor.
[0014] Another embodiment of the present disclosure is related to a testing method, comprising the following operations: obtaining multiple test images of a batch of multiple wafers through a probe device; marking the test images based on the corresponding image specifications of multiple probe marks of each of the test images by a neural network subsystem to generate multiple marked images in multiple analysis data; comparing the test images and the marked images by an evaluation subsystem to generate a comparison result for each of the wafers in the batch; generating a first probe mark inspection result in response to the comparison operation; and updating the record corresponding to the batch of wafers in the program control processor according to the first probe mark inspection result.
[0015] In some embodiments, the corresponding image specifications of the probe mark include coordinates of the probe mark, the number of probe marks, inspection of the substrate of multiple pads in a batch of wafers, multiple distances between the probe mark and multiple edges of pads in the batch of wafers, or a combination thereof.
[0016] In some embodiments, the testing method further includes determining whether the analysis data meets a plurality of quality thresholds to generate a second tag inspection result.
[0017] In some embodiments, when the comparison result indicates an accurate result, the operation of generating the first probe mark inspection result includes: generating the first probe mark inspection result indicating a pass result in response to the comparison result.
[0018] In some embodiments, when at least one mark on the mark image does not match the correction mark, the corresponding comparison result indicates an inaccurate result.
[0019] In some embodiments, the testing method further includes generating a plurality of training data according to the calibration mark and the test image; and training a neural network model in the neural network subsystem according to the training data.
[0020] In some embodiments, the corresponding image specification of the probe mark includes the number of probe marks on the pad in one of the wafers of the batch in the test image. The test method also includes: when the number of probe marks on the pad in one of the analysis data is greater than a threshold, generating a second probe mark inspection result indicating a failure result; and updating the record according to the second probe mark inspection result.
[0021] In some embodiments, the corresponding image specification of the probe mark includes inspection of the substrate of multiple pads in the batch of wafers and the number of probe marks on the first pad of the pads corresponding to one of the batch of wafers in the test image. The test method also includes determining whether the number of probe marks on the first pad is greater than a first threshold when the inspection of the substrate indicates that no material of the substrate is present on at least one of the pads of the substrate; and generating a second probe mark inspection result indicating a failure result when the number of probe marks on the first pad is greater than the first threshold.
[0022] In some embodiments, the testing method further includes generating a second probe mark inspection result indicating a pass result when the number of probe marks on the first pad is less than a first threshold; and scheduling the lot of wafers from the probe device to a process stage.
[0023] In some embodiments, the testing method further includes determining whether multiple distances between the probe marks on the first pad and multiple edges of the first pad are within a second threshold when the number of probe marks on the first pad is less than a first threshold; and generating a second probe mark inspection result indicating a failure result when one of the distances is outside the range of the second threshold. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] An aspect of an embodiment of the present disclosure is described in the accompanying Figure 1 The following detailed description is best understood when read together. It should be noted that, according to standard practice in the industry, various features are not drawn to scale. In fact, the sizes of various features may be arbitrarily increased or reduced for clarity of discussion.
[0025] Figure 1 is a schematic diagram of a test system according to some embodiments of the present disclosure
[0026] Figure 2 is a schematic diagram of a test image corresponding to a portion of a tested wafer according to some embodiments of the present disclosure.
[0027] Figure 3 is a schematic diagram of a testing method according to some embodiments of the present disclosure.
[0028] Figure 4 According to another embodiment of the present disclosure Figure 1 Schematic diagram of the control interface in the test system.
[0029] Figure 5 is a block diagram of a processing device of a test system according to some embodiments of the present disclosure. DETAILED DESCRIPTION
[0030] The spirit of the present disclosure will be discussed in the following drawings and detailed description, and those skilled in the art will be able to change and modify the disclosure of the present disclosure without departing from the spirit and scope of the present disclosure.
[0031] It should be understood that, in this document and in the claims that follow, when an element is considered to be "connected" or "coupled" to another element, it may refer to being directly connected or coupled to another element, and there may be other components in between. Conversely, when an element is considered to be "directly connected" or "directly coupled" to another element, there will be no other elements in between. In addition, "electrically connected" or "connected" can be used to indicate that two or more elements operate or act on each other.
[0032] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used to distinguish one element from another element. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element without departing from the scope of the embodiment.
[0033] It should be understood that the terms "comprising," "including," "having," "having," etc. used herein are open ended and mean "including, but not limited to."
[0034] It should be understood that “and / or” used herein includes any and all combinations of one or more associated items.
[0035] Reference now Figure 1 , Figure 11 is a schematic diagram of a test system 10 according to some embodiments of the present disclosure. In some embodiments, the test system 10 is used as one of multiple test stages to evaluate the tested wafers before shipping. The test system 10 is a test stage that uses optical testing to indicate surface defects on the wafers. In some embodiments, the test system 10 is an automated system to generate test data of the tested wafers for sorting the tested wafers for shipment.
[0036] The test system 10 includes a probe device 110, a data server 120, a program control processor 130 and a plurality of subsystems, the subsystems including an evaluation subsystem 140, a database subsystem 150, a neural network subsystem 160 and a test result distribution subsystem 170. The probe device 110 is connected to the data server 120, the program control processor 130 and the evaluation subsystem 140 for transmitting the test data of the wafer under test. The program control processor 130 is further connected to control the evaluation subsystem 140, the database subsystem 150 and the test result distribution subsystem 170 to generate the test results corresponding to the wafer under test for transportation. The neural network subsystem 160 is trained based on the test image provided by the probe device 110, and the neural network subsystem 160 provides the inspection result of the instant test image to the evaluation subsystem 140 to automatically generate the test result.
[0037] In some embodiments, the probe device 110 includes a camera device for capturing test images of a large number of wafers under test that are transmitted to the test system 10 in the test phase. The test image includes a surface image of a device under test (DUT) on the wafer under test. For example, in some embodiments, a pad on the DUT of the wafer under test contacts the probe in a previous test phase and the test image specifically includes probe marks left by the pins on the pads on the DUT. In some embodiments, the camera can be a complementary metal-oxide semiconductor (CMOS) camera, a charge-coupled device (CCD) camera, a video camera, or another suitable type of camera.
[0038] Reference now Figure 2 . Figure 2 FIG. 2 is a schematic diagram of a test image 200 corresponding to a portion of a wafer 201 under test according to some embodiments of the present disclosure. For illustration, the test image 200 corresponding to the portion of the wafer 201 under test includes several pads 202, such as Figure 2As shown in the embodiment of FIG. 2 , the probe mark 203 is located on some portions of the pad 202, wherein one of the probe marks 203 is spaced a distance Dx in the x direction and a distance Dy in the y direction from the edge of the pad 202. A crack portion 204 on the pad 202 exposes a substrate region of the tested wafer 201. In some embodiments, the crack portion 204 is generated by heavy contact between the probe and the pad 202, which scrapes off the metal cover material from the pad and exposes the substrate region under the pad.
[0039] Reference now Figure 3 . Figure 3 300 is a schematic diagram of a test method 300 according to some embodiments of the present disclosure. It should be understood that additional embodiments of the test method 300 may be Figure 3 Additional operations are provided before, during, and after the process shown, and some of the operations described below may be replaced or eliminated for additional embodiments of the test method 300. The test method 300 includes the following references Figures 1 to 4 In some embodiments, the test method 300 includes the following steps: Figure 1 to Figure 2 and Figure 4 The test system 10 shown performs.
[0040] In operation 301 , the probe device 110 obtains test images of a plurality of tested wafers, for example Figure 2 1 , wherein the batch is transferred to the probe device 110 from a previous test stage, for example, a test stage using electrical specifications of the devices on the tested wafer.
[0041] In operation 302, the evaluation subsystem 140 receives the test image 200 from the probe device 110. In some embodiments, the probe device 110 also transmits the test image 200 and corresponding information (eg, lot number, number of wafers in the lot, test time, etc.) to the data server 120.
[0042] In operation 303 , the program control processor 130 controls the evaluation subsystem 140 to perform an evaluation operation to transmit the test image 200 to the neural network subsystem 160 in an automated mode in response to the probe device 110 acquiring the test image 200 .
[0043] In some embodiments, the test method 300 further includes an operation in which the evaluation subsystem 140 sends the test image 200 and corresponding information to the data server 151 of the database subsystem 150 in response to a command sent by the program control processor 130 .
[0044] In some other embodiments, when there are multiple batches of wafers to be tested, the program control processor 130 is further used to provide a work queue for the evaluation subsystem 140. Accordingly, the evaluation subsystem 140 performs evaluation operations in batches based on the work queue. Figure 4 , Figure 4 According to another embodiment of the present disclosure Figure 1 1 is a schematic diagram of a control interface 40 in the test system 10 in FIG. In some embodiments, the output unit 142 of the evaluation subsystem 140 provides the control interface 40 according to the operation / code / data received from the data processor 141 in the evaluation subsystem 140. A work queue is shown in a box 410 of the control interface 40. In addition, information about the corresponding wafer under test is shown in a box 430, and values in the analysis data (e.g., coordinates of the probe mark 203, the number of probe marks 203, coordinates and sizes of the mark 205, and / or other values related to the test) are shown in a box 440.
[0045] When the neural network subsystem 160 receives the test image 200, the neural network subsystem 160 identifies the image specifications of the probe mark 203 in the test image 200, and further generates analysis data of the test image 200 to the evaluation subsystem 140 in operation 304, the analysis data including data corresponding to the image specifications of the probe mark. In some embodiments, the image specifications of the probe mark in the test image 200 are test data for checking the quality of the tested wafer, including data such as information about Figure 2 The discussion may include the number of probe marks, the coordinates of the probe marks relative to reference points on the wafer being tested, the inspection of the substrate of the pad, the location of the probe marks relative to the edge of the pad, and / or other suitable specifications of the image.
[0046] In some embodiments, the testing method 300 further includes an operation of labeling the test image 200 by the neural network subsystem 160 based on the corresponding image specifications of the probe markers 203 in each test image to generate a labeled image in the analysis data. Figure 4 , a marked image 210 (also referred to as an analysis image) in the analysis data is shown in the control interface 40 with marks 205. Specifically, one of the marks 205 circles a crack portion 204 of the substrate of the pad 202. In some embodiments, the coordinates, size and / or related parameters of the crack portion 204 are included in the analysis data.
[0047] After the neural network subsystem 160 analyzes the test image 200, in operation 305, the evaluation subsystem 140 generates a probe mark check result based on the analysis data to the program control processor 130. In some embodiments, the evaluation subsystem 140 compares the analysis data with the quality threshold to generate the probe mark check result. For example, the evaluation subsystem 140 determines whether all values in the analysis data meet the corresponding quality threshold to generate the probe mark check result.
[0048] Specifically, in some embodiments, when the inspection of the substrate in the analysis data meets a quality threshold indicating that material of the substrate is not present on the pad 202 , the evaluation subsystem 140 generates a probe mark inspection indicating a pass result to the program control processor 130 .
[0049] In some embodiments, after checking the test of the substrate, the method further includes determining whether the amount of probe marks 203 on the pad in the corresponding analysis data meets a threshold. For example, when the amount of probe marks 203 is less than the threshold of 5, the evaluation subsystem 140 generates a probe mark inspection result indicating a pass result. In various embodiments, when the amount of probe marks 203 is greater than the threshold of 5, the evaluation subsystem 140 generates a probe mark inspection result indicating a failure result.
[0050] In addition, in some embodiments, after checking the number of probe marks 203, the method further includes determining whether the distance between the probe mark 203 and the edge of the pad 202 in the corresponding analysis data is within a threshold. For example, when the distance Dx or Dy is within a specific range, the evaluation subsystem 140 generates a probe mark inspection result indicating a pass result, wherein the specific range indicates that the probe mark 203 does not overlap with the edge of the pad 202. In various embodiments, when the distance Dx or Dy is out of range, that is, indicating that the probe mark 203 overlaps with the edge of the pad 202, the evaluation subsystem 140 generates a probe mark inspection result indicating a fail result.
[0051] In various embodiments, for certain quality criteria, the evaluation subsystem 140 determines whether a value in the analysis data (e.g., one of the factors in the image specification) meets a quality threshold to generate a probe mark inspection result. For example, when the number of probe marks 203, the inspection of the substrate, or the distance between the probe mark 203 and the edge of the pad 202 meets the corresponding threshold as discussed above, the evaluation subsystem 140 generates a probe mark inspection indicating a pass result to the program control processor 130. Conversely, when a value in the analysis data fails to meet the corresponding quality threshold, the evaluation subsystem 140 generates a probe mark inspection indicating a fail result to the program control processor 130.
[0052] In operation 306, the record in the program control processor 130 is updated according to the probe mark check. For example, when the probe mark check indicates a pass result, the macro test value of the test result in the record corresponding to the lot under test in the test system 10 is updated from a first value (e.g., none) to a second value (e.g., pass) different from the first value. Conversely, when the probe mark check indicates a fail result, the macro test value in the record corresponding to the lot under test in the test system 10 is updated from a first value (e.g., none) to a third value (e.g., fail) different from the first value.
[0053] In some embodiments, the method further includes an operation of scheduling the batch of test wafers from the probe device 110 to a process stage (not shown, for example, a marking stage before an outgoing quality control (OQC) stage) by the test system 10 controlled by the program control processor 130 after the test system 10. Specifically, the program control processor 130 sends a command to the electronic product / process abnormal system (EPAS) processor 171 of the test result distribution subsystem 170, and the EPAS processor 171 communicates with the probe device 110 to schedule the batch to the next process stage.
[0054] In some embodiments, the test image is stored in a storage unit 162 in the neural network subsystem 160 and is further used to optimize the neural network model operating in the neural processor 161 of the neural network subsystem 160.
[0055] Specifically, in some embodiments, after the neural network subsystem 160 analyzes the test image, the test method 300 further includes an operation in which the evaluation subsystem 140 compares the test image 200 with the marked image 210 to generate a corresponding comparison result for each wafer in the batch to the data server 152 in the database subsystem 150. Figure 4 As shown in , the test image 200 and the marked image 210 are displayed in box 420. In some embodiments, the evaluation subsystem 140 also compares the image specifications between the test image 200 and the analyzed image 210 based on the input signal and the analysis data. For example, the number of probe marks 203 in the analysis data provided by the neural network subsystem 160 is 3, which matches the number of probe marks 203 on the pad 202 in the test image 200 based on the input signal. Accordingly, the evaluation subsystem 140 generates a comparison result indicating an accurate result, and generates a probe mark inspection result indicating a pass result to the program control processor 130 based on the comparison result.
[0056] In some embodiments, the input signal includes process data from a previous process stage (e.g., the number of probe marks performed in the process, the location of the probe marks, etc.). For example, three electrical tests in the previous process stage resulted in three probe marks. In various embodiments, the input signal is generated by an input unit 143 of the evaluation subsystem 140 controlled by an operator. For example, the operator can control the evaluation subsystem 140 to perform image processing to mark, enlarge, reduce, scale items, rotate, adjust brightness and contrast, or perform other suitable operations on the test image 200 or the analysis image 210 based on the image processing tools shown in the box 450 of the control interface 40.
[0057] In contrast, when the image specifications of the test image 200 do not match the image specifications of the analysis image 210, the test method 300 further includes the operation of the evaluation subsystem 140 generating a comparison result indicating an inaccurate result and further correspondingly generating training data including the test image and the corresponding identification data to the neural network subsystem 160. For example, Figure 4 As shown, there are two marks 205 in the analysis image 210, while only one area of the pad 202 exposes the substrate. Accordingly, the operator controls the input unit 143 to generate the calibration mark 206. The evaluation subsystem 140 compares the test image 200 and the analysis image 210 to determine that the mark 205a does not match the calibration mark 206, and further generates a corresponding comparison result, which indicates an inaccurate result.
[0058] In some embodiments, the calibration mark is referred to as identification data and is stored in inventory data in the data server 152 of the database subsystem 150 along with the test image 200 , analysis data, and probe mark inspection results corresponding to the lot of tested wafers.
[0059] In some embodiments, the test method 300 further includes the neural network subsystem 160 accessing the inventory data to train and update the operation of the neural network model operating therein according to the training data in the inventory data. Specifically, the parameters such as the weight value in the neural network model associated with the wrong mark 205a are adjusted (e.g., reduced), thereby training the neural network model.
[0060] In some embodiments, when the quality standard of the shipment is adjusted based on the product specification, the test method 300 further includes the operation of the evaluation subsystem 140 accessing the inventory data in the data server 152 and generating another probe mark inspection result based on the inventory data and the adjusted threshold value. For example, the threshold value of the number of probe marks 203 is adjusted from 5 to 10. The evaluation subsystem 140 determines that the number of probe marks is equal to 6 (and corresponds to the failure result of the probe mark inspection result) is less than the adjusted threshold value, and accordingly, the evaluation subsystem 140 generates a probe mark inspection result indicating a pass result to update the record of the corresponding batch in the program control processor 130.
[0061] In some embodiments, Figure 1 As shown, the test method 300 also includes an operation of storing the test image 200, analysis data, and corresponding information from the evaluation subsystem 140 and the data server 151 in the backup data server 153, and an operation of generating a process / product report by the report processor 172 in the test result distribution subsystem 170 based on the data stored in the backup data server 153.
[0062] Figures 1 to 4 The configuration is given for illustrative purposes. Various implementations are within the contemplated scope of the present disclosure. For example, in some embodiments, the number of probe marks 203 is the sum of the probe marks on all pads 202 in the test image 200.
[0063] It should be noted that Figure 1 The test system 10 depicted in FIG. 1 may also include a method for implementing Figures 1 to 4 A processing device that implements one or more of the described tools, subsystems, methods, or operations.
[0064] Figure 5 is a block diagram of a processing device 50 of a test system 10 according to some embodiments of the present disclosure. In other words, the probe device 110, the data server 120, the program control processor 130, the evaluation subsystem 140, the database subsystem 150, the neural network subsystem 160, and the test result distribution subsystem 170 are composed of components with respect to, for example, Figure 5 The processing device 50 shown is implemented by a device configured as shown.
[0065] The processing device 50 may include a processor 510, a network interface (I / F) 520, an input / output (I / O) device 530, a storage device 540, and a memory 550 communicatively coupled via a bus 560 or other interconnection communication mechanism. In some embodiments, the memory 550 includes a random access memory (RAM), other dynamic storage devices, read-only memory (ROM) or other static storage devices coupled to the bus 560, and the above devices are used to store data or instructions to be executed by the processor 510, such as a user space 551, a kernel 552, a portion of the kernel or the user space and its elements. In some embodiments, the memory 550 is also used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processor 510.
[0066] In some embodiments, a storage device 540 (such as a disk or optical disk) is coupled to the bus 560 for storing data or instructions, for example, the kernel 552, the user space 551, etc. The input / output device 530 includes an input device, an output device, or a combination of input / output devices for implementing user interaction with the test system 10. The input device includes, for example, a keyboard, a keypad, a mouse, a trackball, a touchpad, or a cursor arrow key, for transmitting information and commands to the processor 510. The output device includes, for example, a display, a printer, a speech synthesizer, etc., for transmitting information to the user. In some embodiments, reference is made to Figures 1 to 4 One or more operations or functions of the described tool or system are implemented by processor 510, which is programmed to perform such operations and functions. One or more of memory 550, network interface 520, storage device 540, input / output device 530, and bus 560 are operable to receive instructions, data, design rules, netlists, layouts, models, and other parameters processed by processor 510.
[0067] In some embodiments, the present invention is implemented by specially configured hardware (e.g., by including one or more application-specific integrated circuits (ASICs)). Figures 1 to 4One or more of the operations, functions, and systems described. Some embodiments include more than one of the operations or functions described in a single ASIC. In some embodiments, the operations and functions are implemented as functions of a program stored in a non-transitory computer-readable recording medium. Examples of non-transitory computer-readable recording media include (but are not limited to) external / removable or internal / built-in storage or memory units, such as one or more of optical disks (e.g., DVDs), magnetic disks (e.g., hard disks), semiconductor memories (e.g., ROM, RAM, memory cards, etc.).
[0068] In some embodiments, the neural network model described herein is based on a deep learning architecture. For example, the model can be a convolutional neural network (CNN) model, which can use deep learning concepts to solve the problem of representation in the table. The model can have any CNN configuration known in the art. For example, the model can be a super resolution CNN (SRCNN), which can use deep learning concepts to convert low-resolution images into high-resolution images. The model can have any SRCNN configuration known in the art. The neural network model of the neural network subsystem 160 is trained using classified test images. Before the probe marking inspection, the CNN model is trained by providing a classified test image to the neural processor 161. The classified test image may be pre-stored in the storage unit 162. In some embodiments, the classified test image is provided from the data server 152. The neural processor 161 may implement a convolutional neural network for identifying image specifications in the test image 200.
[0069] Through the operation of the above embodiments, the test system and test method provided by the present disclosure integrates the optical image inspection tool into the neural network system to classify the wafers according to the probe mark inspection, thereby providing an automatic processing method to improve the productivity of the test wafers. The present disclosure significantly reduces manual labor and the errors caused thereby, and further improves the overall productivity.
[0070] Although the present disclosure has been described by way of examples and preferred embodiments, it should be understood that the present disclosure is not limited thereto. Those skilled in the art may make various changes, substitutions, and modifications to the present disclosure without departing from the spirit and scope of the present disclosure. In view of this, the present disclosure is intended to cover various modifications and variations of the present disclosure, as long as such modifications and variations fall within the following claims.
[0071]
Explanation of symbols
[0072] 10: Test system
[0073] 110: Probe device
[0074] 120:Data Server
[0075] 130: Program control processor
[0076] 140:Evaluation Subsystem
[0077] 141: Data Processor
[0078] 142: Output unit
[0079] 143: Input unit
[0080] 150: Database subsystem
[0081] 151-153: Data Server
[0082] 160:Neural Network Subsystem
[0083] 161:Neural Processor
[0084] 162: Storage unit
[0085] 170:Test result distribution subsystem
[0086] 171:EPAS Processor
[0087] 172: Report Processor
[0088] 200: Test image
[0089] 201: Wafer under test
[0090] 202:Pad
[0091] 203: Probe labeling
[0092] 204: Cracked part
[0093] 205: Mark
[0094] 205a: Marking
[0095] 206: Calibration mark
[0096] Dx: distance
[0097] Dy: Distance
[0098] x: direction
[0099] y: direction
[0100] 300:Test Method
[0101] 301-306: Operation
[0102] 40: Control interface
[0103] 410,420,430,440,450:frame
[0104] 50: Processing device
[0105] 510: Processor
[0106] 520: Network interface
[0107] 530: Input / output device
[0108] 540: Storage device
[0109] 550: Memory
[0110] 551: User Space
[0111] 552: Kernel
[0112] 560: bus.
Claims
1. A testing system, characterized in that: Include: An evaluation subsystem for receiving a test image of a wafer under test from a probe device; Neural network subsystem; as well as a program-controlled processor for controlling the evaluation subsystem to perform an evaluation operation to transmit the test image to the neural network subsystem in an automated mode in response to the probe device acquiring the test image, The neural network subsystem is used to identify the image specifications of multiple probe marks in the test image and generate analysis data of the test image to the evaluation subsystem, The evaluation subsystem is further configured to generate a first probe mark inspection result to the program-controlled processor based on the analysis data to generate a test result.
2. The test system of claim 1 , wherein the evaluation subsystem is further configured to generate training data to the neural network subsystem by comparing the test image with the analyzed image in the analysis data, The training data includes the test image and corresponding identification data.
3. The testing system according to claim 2, wherein the neural network subsystem is also used to be trained by the training data to update multiple weight values in a neural network model in the neural network subsystem, and the neural network model operates in a neural network processor. 4 . The test system of claim 1 , wherein the image specification comprises coordinates of the plurality of probe marks, a number of the plurality of probe marks, inspection of a substrate of at least one pad in the test wafer, or a combination thereof.
5. The test system of claim 1 , wherein the evaluation subsystem is configured to compare the analysis data with a plurality of quality thresholds, and to generate the first probe mark inspection result indicating a first failure result when at least one value in the analysis data fails to satisfy a corresponding threshold in the plurality of quality thresholds. 6 . The test system of claim 5 , wherein the program-controlled processor is further configured to generate the test result indicating a second fault result in response to the first probe mark inspection result indicating the first fault result. 7 . The test system of claim 5 , wherein the analysis data comprises a plurality of distances between the plurality of probe marks and a plurality of edges of a plurality of pads in the tested wafer.
8. The test system according to claim 1, wherein: Also includes: a database subsystem, comprising a data server, wherein the data server is used to store the test image, the analysis data and the first probe mark inspection result in a list data, The evaluation subsystem is further configured to access the list data and generate a second detection mark inspection result based on the list data and a quality threshold.
9. The test system of claim 8, wherein the neural network subsystem is further used to access the inventory data and be trained by training data to update the neural network model operating in the neural network subsystem.
10. The testing system of claim 1, wherein the evaluation subsystem is further configured to perform the evaluation operation on a plurality of batches of images according to a work queue provided by the program control processor.
11. A testing method, characterized in that: Include: obtaining a plurality of test images of a batch of a plurality of wafers by a probe device; labeling, by a neural network subsystem, the plurality of test images based on corresponding image specifications of the plurality of probe labels for each of the plurality of test images to generate a plurality of labeled images in a plurality of analysis data; comparing, by an evaluation subsystem, the plurality of test images and the plurality of marked images to generate a comparison result for each of the plurality of wafers in the lot; In response to the comparison operation, generating a first probe label inspection result; as well as The records of the plurality of wafers in the batch are updated in the program-controlled processor according to the first probe mark inspection result.
12. The testing method according to claim 11, wherein the corresponding image specifications of the multiple probe marks include coordinates of the multiple probe marks, the number of the multiple probe marks, inspection of the substrate of the multiple pads in the multiple wafers in the batch, multiple distances between the multiple probe marks and multiple edges of the multiple pads in the multiple wafers in the batch, or a combination thereof.
13. The testing method according to claim 12, wherein: Also includes: It is determined whether the plurality of analysis data satisfy a plurality of quality thresholds to generate a second tag inspection result.
14. The testing method according to claim 11, wherein when the comparison result indicates an accurate result, the operation of generating the first probe mark inspection result comprises: The first probe marker inspection result indicating a pass result is generated in response to the comparison result.
15. The testing method of claim 14, wherein when at least one mark on the plurality of mark images does not match a calibration mark, the corresponding comparison result indicates an inaccurate result.
16. The testing method according to claim 15, wherein: Also includes: generating a plurality of training data based on the calibration mark and the plurality of test images; and The neural network model in the neural network subsystem is trained according to the multiple training data.
17. The testing method according to claim 11, wherein the corresponding image specification of the plurality of probe marks comprises the number of the plurality of probe marks on the pads in one of the plurality of wafers of the lot in the plurality of test images, The test method also includes: generating a second probe mark inspection result indicating a failure result when the number of the plurality of probe marks on the pad in one of the plurality of analysis data is greater than a threshold value; and The record is updated according to the second probe marker inspection result.
18. The testing method according to claim 11, wherein the corresponding image specifications of the plurality of probe marks include inspection of a substrate of a plurality of pads in the plurality of wafers in the batch and the number of the plurality of probe marks on a first pad of the plurality of pads corresponding to one of the plurality of wafers in the batch in the plurality of test images, The test method also includes: When the inspection of the substrate indicates that no material of the substrate is present on at least one of the plurality of pads of the substrate, determining whether the number of the plurality of probe marks on the first pad is greater than a first threshold; and When the number of the plurality of probe marks on the first pad is greater than the first threshold, a second probe mark inspection result indicating a failure result is generated.
19. The testing method according to claim 18, wherein: Also includes: generating the second probe mark inspection result indicating a pass result when the number of the plurality of probe marks on the first pad is less than the first threshold value; as well as The plurality of wafers of the lot are dispatched from the probe device to a process stage.
20. The testing method according to claim 18, wherein: Also includes: When the number of the plurality of probe marks on the first pad is less than the first threshold, determining whether a plurality of distances between the plurality of probe marks on the first pad and a plurality of edges of the first pad are within a second threshold; as well as When one of the plurality of distances is outside the range of the second threshold, the second probe mark inspection result indicating a failure result is generated.
