KR102883473B1 - Deep Learning Based Image Sensor Wafer Test System - Google Patents

Abstract

The present invention has the advantage that, when using image sensor wafer test equipment using deep learning, test result data is provided to a deep learning-based discriminator and automatically updated, and the accuracy of determining whether there is a defect is continuously increased through learning using deep learning, thereby ensuring the reliability of the wafer inspection equipment.
In addition, the present invention relates to an image sensor wafer tester system using deep learning, which can prevent major failures in advance by finding minute changes in wafer test equipment, illuminants, illuminance meters, calibrators, etc. by analyzing data classes determined to be defective using deep learning.
In order to solve the above problem, the image sensor wafer inspection system using deep learning according to the present invention is characterized by including: a prober for mounting an image sensor wafer; a probe card for electrically connecting to an image sensor chip, a Pogo Tower, a probe interface board; wafer test equipment for performing a wafer test and determining whether the wafer is good or bad based on the test result; an illuminant for projecting light for the wafer test; an illuminance meter; a calibrator for controlling the projection light intensity of the illuminant; and a deep learning-based discriminator for receiving wafer test result data from the wafer test equipment and analyzing the data provided using a deep learning technique to determine whether each image sensor chip constituting the wafer is good or bad.

Description

Deep Learning-Based Image Sensor Wafer Test System

The present invention relates to a wafer test system for an image sensor using deep learning, and more particularly, to a wafer test system for an image sensor using deep learning that efficiently analyzes EDS (Electrical Die Sorting) test result data measuring various electrical characteristics of a plurality of image sensor chips manufactured on a silicon wafer in a semiconductor manufacturing process through an artificial intelligence (AI) deep learning technique and determines whether the chips are in a good or bad state.

Typically, in the semiconductor manufacturing process, multiple semiconductor dies or chips are manufactured on a single wafer, and the EDS (Electrical Die Sorting) test results data is analyzed to measure various electrical characteristics of each semiconductor chip manufactured on a specific wafer using wafer testing equipment, and the process of determining whether it is pass or fail is performed based on pre-determined criteria.

In this process, semiconductor chips determined to be defective are visually marked on the wafer surface at their corresponding locations for separate management, or information for identification is recorded by including it in the measurement data.

In addition, for EDS testing, an electrical connection must be made between the wafer test equipment and the semiconductor chips that make up the wafer. The wafer test equipment is connected to a probe card through a probe interface board (PIB) and a pogo tower, and is connected to the semiconductor chips through a number of needles provided on the probe card.

At this time, the same number of needles as the number of connection pins of the image sensor chip must be provided, and when connecting multiple image sensor chips at the same time, the total number of needles is proportional to the number of image sensor chips to be connected, so the probe card is customized according to the image sensor chip.

Meanwhile, in order to test the image sensor chips that make up the wafer, an illuminator that injects light, a lux meter that is the subject of regular calibration, and a calibrator that adjusts to the desired illuminance, or a computer with software that performs the calibration function, are additionally equipped.

In addition, as a preparatory step for wafer testing for image sensors, after removing the prober, probe card, and Pogo Tower from the wafer test system according to the planned procedure, light of the required intensity is projected onto the illuminator, and if the illuminance measured by the illuminance meter is different from the set reference illuminance, the calibrator is operated to correct the light intensity of the illuminator. This preparatory work is usually performed once at a set periodic interval.

Meanwhile, a calibration step is performed once a day using a separate illuminance meter installed inside the lighting device to adjust the brightness of the lighting device's projection light if it falls outside the acceptable range.

After that, the prober, Pogo Tower, and probe card are installed in the wafer test system, the wafer for the image sensor is inserted, the needle is connected, and the EDS test is performed according to the plan.

As a prior art of the semiconductor wafer test system as described above, a wafer vision inspection device and method (hereinafter referred to as “patent document”) is disclosed in Patent Publication No. 2022-0111061.

The above patent document comprises a display unit; a camera for obtaining wafer images of wafers accommodated in each slot; a processor for extracting images of specific areas of wafers accommodated in each slot using the wafer images, detecting wafer defects based on a correlation between the specific area images and specific area images for learning corresponding to external environmental factors, and outputting the detection results through the display unit.

However, the above patent document detects wafer defects by acquiring images of the wafer through a camera, and thus has limitations in accurately detecting microscopic defects that occur irregularly and in various ways.

In addition, properly monitoring semiconductor wafer defects has a significant impact on defect analysis and the overall yield in the manufacturing process. However, the existing method of monitoring wafer defects relies on experts in some manual inspection fields, which is not suitable in terms of accuracy and efficiency. In addition, as the size of semiconductor chips gradually decreases, it is becoming more difficult for people to manually monitor wafer defects.

Therefore, in semiconductor wafer inspection equipment, the development of an innovative deep learning-based wafer test system for image sensors is required to increase the efficiency and reliability of wafer defect detection by automating learning and analysis of irregularly occurring micro-defect data and good data using artificial intelligence techniques.

KR 10-2022-0111061 A (2022. 08. 09.) KR 10-2372147 B1 (2022. 03. 03.) KR 10-2021-0108597 A (2021. 09. 03.) KR 10-2239051 B1 (2021. 04. 06.)

The present invention has been devised to solve the problems of the conventional wafer test system for image sensors as described above, and the problem to be solved by the present invention is to provide a wafer test system for image sensors using deep learning, which can improve the efficiency and accuracy of image sensor wafer testing by using EDS test result data for each image sensor chip constituting a wafer to learn using a deep learning technique including an autoencoder, and automatically determines whether the data includes various types of defects that may occur and analyzes the data to suggest the cause of the defect, such as whether it is a defect in the image sensor chip, a defect in the illuminator, or a defect in the wafer test equipment.

In order to solve the above problem, the wafer test system for an image sensor using deep learning according to the present invention comprises: an image sensor wafer test equipment including a prober for mounting an image sensor wafer to be tested, a probe card having a plurality of needles for electrically connecting to one or more image sensor chips constituting the wafer, a Pogo Tower for electrically connecting to the probe card, a probe interface board for electrically connecting to the Pogo Tower and for bidirectionally transmitting and receiving digital data and analog signals between the wafer test equipment and the wafer, and a computer system for performing a wafer test through a program for the wafer test, receiving, storing and analyzing test result data, and determining whether the image sensor test result is good or bad; an illuminator for projecting light for the wafer test; an illuminance meter for measuring the brightness of light projected from the illuminator at a planned location; a corrector for controlling the projection brightness of the illuminator to be adjusted when the illuminance measured by the illuminance meter deviates from the planned illuminance and an allowable error range; And it is characterized by including a deep learning-based discriminator that receives wafer test result data from the wafer test equipment and analyzes the data provided using a deep learning technique to determine whether each image sensor chip constituting the wafer is good or bad.

And the present invention includes an autoencoder technique as the deep learning technique, and the autoencoder technique inputs multi-layer data by dividing each test item of each image sensor chip constituting a wafer into a separate layer, and considering the imbalanced data problem in which the image sensor wafer test result data has a large number of data sets of good classes but a relatively small number of data sets of bad classes, the present invention is characterized in that the wafer test result data is provided from the image sensor wafer test equipment, and only the data determined to be good classes are selected as much as possible for learning, and the classes determined to be bad are excluded from the learning target.

In addition, in the present invention, the deep learning-based discriminator classifies the pattern of defective pixels in the image for each image sensor chip determined to be defective among the test result data provided from the image sensor wafer test equipment into three types: Random failure, Systematic failure, and Hybrid failure. First, in the case of Random failure, it examines by comparing similar defective data sets whether the problem is due to a failure in the internal circuit of the image sensor chip or a problem due to the environment or carelessness, such as dust or scratches on the micro lens of the corresponding pixel of the image sensor chip during the test process. Next, in the case of Systematic failure, it examines whether the problem occurred due to a slight change in the performance of the illuminator, illuminance meter, or calibrator over time after precise calibration, or due to a problem with the wafer test equipment, based on the pattern of defective pixels. If an error that caused a good chip to be determined to be defective due to a change in the inspection environment, such as a failure or deterioration of the performance of the wafer test equipment, illuminance meter, or calibrator, is automatically discovered in advance during the post-processing process, the yield of good determination can be increased and time can be saved through re-inspection. It is characterized by the ability to detect even minor changes early by utilizing techniques.

In addition, among the data classes judged as good, there may be chips that satisfy the good judgment criteria but have abnormal data as a result of the inspection. In this case, in the past, after the post-process wafer inspection, a batch re-inspection had to be performed on all chips included in the wafer that was judged as good to find out the cause of the abnormality. However, the present invention, which applies deep learning techniques in parallel with the post-process inspection, can automatically detect abnormalities caused by minute changes in wafer test equipment, illuminators, illuminance meters, calibrators, etc. in advance, and can prevent major failures in advance, thereby enhancing the competitiveness of post-process companies.

According to the present invention, when using image sensor wafer test equipment using deep learning, test result data is provided to a deep learning-based discriminator in near real time and automatically updated, and the accuracy of determining whether or not something is defective is continuously increased through learning using deep learning, thereby ensuring the reliability of the wafer inspection equipment.

In addition, if errors that cause good chips to be judged as defective due to changes in the inspection environment, such as failures or performance degradation of wafer test equipment, illuminators, illuminance meters, and calibrators, are automatically identified in advance through analysis using deep learning for data classes judged as defective, the yield of good judgments can be increased through re-inspection, and time can also be saved.

In addition, among the data classes judged as good, there may be chips that satisfy the good judgment criteria but are in an abnormal state. In this case, in the past, after undergoing post-process wafer inspection, a batch re-inspection had to be performed on all wafers judged as good to identify the cause of the abnormal state. However, the present invention can automatically detect abnormal states caused by slight changes in wafer test equipment, illuminators, illuminance meters, and calibrators in advance by applying deep learning techniques in parallel with post-process inspection, thereby preventing major failures in advance.

FIG. 1 is a schematic diagram showing an example of a wafer test system for an image sensor using deep learning according to the present invention.
Figure 2 is a schematic diagram showing an example of an autoencoder algorithm according to the present invention.

Hereinafter, a preferred embodiment of the present invention will be described in detail with reference to the attached drawings.

Various descriptions are provided herein to provide an understanding of the present disclosure. However, it will be apparent that these embodiments may be practiced without these specific descriptions.

Additionally, the terms “part,” “technique,” “algorithm,” “component,” “module,” “system,” and the like, as used herein, refer to computer-related entities, hardware, firmware, software, a combination of software and hardware, or an execution of software. For example, a component may be, but is not limited to, a procedure running on a processor, a processor, an object, a thread of execution, a program, and/or a computer.

For example, both an application running on a computing device and the computing device can be a component. One or more components can reside within a processor and/or a thread of execution. A component can be localized within a single computer. A component can be distributed between two or more computers. Furthermore, these components can execute from various computer-readable media having various data structures stored therein. The components can communicate via local and/or remote processes, for example, by signals comprising one or more data packets (e.g., data from one component interacting with another component in a local system, a distributed system, and/or data transmitted to another system via a network such as the Internet).

Furthermore, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or", and thus, unless otherwise specified or clear from context, "X employs A or B" is intended to mean either of the natural inclusive permutations.

Also, if X uses A, or X uses B, or X uses both A and B, then "X uses A or B" can apply to any of these cases.

Additionally, the term "and/or" as used herein should be understood to refer to and include all possible combinations of one or more of the relevant items listed.

Additionally, the terms "comprises" and/or "comprising" should be understood to mean the presence of the features and/or components, provided that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, components and/or groups thereof.

Additionally, unless otherwise specified or clear from context to refer to the singular form, the singular in this specification and claims should generally be construed to mean “one or more.”

And the term "at least one of A or B" should be interpreted to mean "if it includes only A", "if it includes only B", or "if it is combined in the composition of A and B".

Furthermore, those skilled in the art should additionally recognize that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both.

And to clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality; however, whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system, and skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

In addition, the description of the presented embodiments is provided to enable a person skilled in the art to use or practice the present invention, and various modifications to these embodiments will be apparent to a person skilled in the art.

The general principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure, and thus the present invention is not limited to the embodiments set forth herein, but is to be accorded the widest scope consistent with the principles and novel features set forth herein.

In the present disclosure, terms expressed as first, second, or third, N, are used to distinguish at least one entity. For example, entities expressed as first and second may be the same or different.

FIG. 1 illustrates an image sensor wafer test equipment (110) including a prober (150) for mounting an image sensor wafer (160) to be tested, a probe card (140) equipped with a plurality of needles for electrically connecting to one or more image sensor chips constituting the mounted wafer (160), a Pogo Tower (170) for electrically connecting to the probe card (140), a probe interface board (180) for electrically connecting to the Pogo Tower (170) and for transmitting and receiving digital data and analog signals in both directions between the wafer test equipment (110) and the wafer (160), and a computer system for performing a wafer test through a program for the wafer test, receiving, storing and analyzing test result data, and determining whether the image sensor test result is good or bad; an illuminant (120) for projecting light for the wafer test; an illuminant meter (132) for measuring the brightness of light projected from the illuminator at a planned location; It includes a corrector (131) that controls the projection light intensity of a lighting device when the illuminance measured by the illuminance meter is outside the planned illuminance and the allowable error range; and a deep learning-based determiner (150) that receives wafer test result data from the wafer test equipment (110) and analyzes the data provided using a deep learning technique to determine whether each image sensor chip constituting the wafer is good or bad.

Meanwhile, among the test result data provided from the image sensor wafer test equipment, the pattern of defective pixels in the image is divided into three types: random failure, systematic failure, and hybrid failure for each image sensor chip determined to be defective. First, in the case of random failure, it is examined by comparing similar defective data sets to see whether the problem is caused by a malfunction in the internal circuit of the image sensor chip or a problem caused by the environment or carelessness, such as dust or scratches on the micro lens of the corresponding pixel of the image sensor chip during the test process. Next, in the case of systematic failure, it is examined whether the problem occurred because the performance of the illuminator, illuminance meter, or calibrator changed slightly over time after precise calibration according to the pattern of defective pixels, or whether it was caused by a problem with the wafer test equipment. It has the function of detecting even slight changes in images determined to be defective in the test results at an early stage by utilizing deep learning techniques.

Figure 2 shows a configuration of an autoencoder included in a deep learning technique, including an encoder and a decoder. Each image sensor chip constituting a wafer is input as a multi-layer image (210) by dividing it into separate layers for each test item, and the encoder compresses the input image data and converts it into output image data (220, 230) with a lower dimension than the input data, and the decoder takes the output (230) of the encoder as input and converts it inversely into image data (240, 250) with a higher dimension.

Meanwhile, in the present invention, the deep learning-based discriminator classifies the pattern of defective pixels in the image for each image sensor chip determined as defective among the test result data provided from the image sensor wafer test equipment into three types: random failure, systematic failure, and hybrid failure. First, in the case of random failure, it examines whether the problem is due to a malfunction in the internal circuit of the image sensor chip, or whether it is due to an environmental or careless problem such as dust or scratches on the micro lens of the corresponding pixel of the image sensor chip during the test process by collecting and comparing similar defective data sets. Next, in the case of systematic failure, it determines whether the problem occurred due to a slight change in the performance of the illuminator, illuminance meter, or calibrator over time after precise calibration, or whether it is due to a problem with the wafer test equipment. It has the function of detecting even slight changes in images determined as defective in the test results by utilizing deep learning techniques.

As described above, the present invention automatically updates data determined as a defective class of wafer test equipment for image sensors between uses of the wafer inspection equipment, thereby continuously increasing the accuracy of the defective test, thereby ensuring the reliability of the wafer test equipment.

In the above, for the convenience of explanation, a preferred embodiment has been described by giving drawing numbers and names to the components shown in the drawings, but this is only one embodiment according to the present invention, and the scope of the rights should not be interpreted as being limited to the shapes shown in the drawings and the names given, and it will be readily apparent that changes to various shapes predictable from the description of the invention and simple substitutions with components that perform the same function are within the scope of changes that can be easily performed by a person skilled in the art.

110: Wafer inspection equipment 120: Illumination equipment
130: Corrector 131: Lighting controller
132: Light meter 140: Probe card
150: Prober 160: Wafer
170: Pogo Tower 180: Probe Interface Board
190: Deep Learning Discriminator 210: Input Image
220: 1st encoded image 230: Latent space
240: 1st decoded video 250: Playback video

Claims (4)

A wafer test system for image sensors using deep learning,
An image sensor wafer test equipment comprising a prober for mounting an image sensor wafer, a probe card for electrically connecting with one or more image sensor chips constituting the wafer, a Pogo Tower for electrically connecting with the probe card, a probe interface board for electrically connecting with the Pogo Tower and transmitting and receiving digital data and analog signals in both directions between the wafer test equipment and the wafer, and a computer system for performing a wafer test through a program for the wafer test, receiving, storing and analyzing test result data, and determining whether the image sensor test result is good or bad;
An illuminator for projecting light for the above wafer testing;
A illuminance meter that measures the brightness of light projected from the illuminant at a planned location; a corrector that controls the projection brightness of the illuminant when the illuminance measured by the illuminance meter is outside the planned illuminance and the allowable error range; and
A deep learning-based discriminator that receives wafer test result data from the above wafer test equipment, learns by utilizing the data provided using a deep learning technique, and analyzes the data provided using the learned deep learning technique to determine whether each image sensor chip constituting the wafer is good, defective, or requires re-inspection;
Including,
The above deep learning is,
Includes an autoencoder,
The above autoencoder is,
A wafer test system for an image sensor using deep learning, characterized in that the wafer test result data is input as multi-layer data by dividing each test item of each image sensor chip constituting the wafer into a separate layer, and only data determined to be a good class among the wafer test result data is selected and learned, and the class determined to be a bad class is excluded from the learning target.
delete In claim 1,
The above deep learning-based discriminator is,
A wafer test system for an image sensor using deep learning characterized by having a function of classifying the pattern of defective pixels in an image for each image sensor chip determined to be defective among the test result data provided from the image sensor wafer test equipment into three categories: random failure, systematic failure, and hybrid failure, analyzing whether the problem is due to a failure in the internal circuit of the image sensor chip or a problem due to the environment or carelessness, such as dust or scratches on the image sensor chip during the test process, and, in the case of a systematic failure, analyzing whether the problem is due to changes in the performance of an illuminator, illuminance meter, or calibrator over time or a problem with the wafer test equipment, depending on the pattern of defective pixels.
In claim 1 or claim 3,
The above deep learning-based discriminator is,
A wafer test system for an image sensor using deep learning, characterized in that it has a function of analyzing whether there is abnormal data although the good judgment criteria are satisfied for each image sensor chip determined to be good from the test result data provided from the image sensor wafer test equipment.