Slicer semiconductor chip cleaning control system

Abstract

The present invention discloses a semiconductor chip cleaning control system for a dicing machine, and relates to the field of semiconductor technology. The system includes: obtaining target organic pollution information; extracting the first target organic matter, and calling the first plan; performing dynamic monitoring to obtain the first dynamic cleaning record; acquiring a module, performing cleaning timing analysis, and obtaining the first cleaning timing feature; predicting the first predicted cleaning time, and issuing a first stop instruction; and entering the cleaning link of the second target organic matter based on the first stop instruction. The present invention solves the technical problems that the cleaning time is difficult to accurately control when using a plasma cleaning method to treat organic pollution after semiconductor chip cutting, the over-cleaning phenomenon is serious, and the degree of automation is insufficient, which leads to the impact on chip performance. It achieves a more efficient and intelligent cleaning control in the process of plasma cleaning organic pollution, and significantly improves the technical effect of chip cleaning quality and performance protection.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to a semiconductor chip cleaning control system for a dicing machine. Background Art

[0002] At a time when semiconductor technology is booming, semiconductor chips occupy a vital position in various electronic devices. In the manufacturing process of semiconductor chips, dicing machine cutting is one of the key links. However, the cutting process will inevitably produce organic pollution on the surface of the chip. If not treated, these organic pollution will have a great adverse effect on the performance of the chip. At present, plasma cleaning has been widely used in the field of semiconductor chip cleaning. Plasma cleaning uses active particles in plasma to react physically and chemically with pollutants on the surface of the object, thereby decomposing, vaporizing and discharging the pollutants. It has many advantages such as good cleaning effect, fast cleaning speed and low environmental pollution. However, when the existing technology uses plasma cleaning to treat organic pollution after semiconductor chip cutting, it is difficult to accurately control the cleaning time. In the plasma cleaning process, the reaction rate of active particles and organic pollution is affected by multiple factors such as plasma power, cleaning gas type and flow rate, and cleaning time. This makes it extremely difficult to determine the cleaning time. If the cleaning time is too long, it will cause over-cleaning and damage the chip surface; if the cleaning time is too short, it will lead to incomplete cleaning, which will also affect the chip performance.

[0003] The existing technology has the problem that it is difficult to accurately control the cleaning time when using plasma cleaning to treat organic contamination after semiconductor chip cutting, the over-cleaning phenomenon is serious, and the degree of automation is insufficient, which leads to technical problems that affect chip performance. Summary of the invention

[0004] The present application provides a semiconductor chip cleaning control system for a dicing machine, which is used to solve the technical problems in the prior art that when using plasma cleaning to treat organic contamination after semiconductor chips are cut, the cleaning time is difficult to accurately control, the over-cleaning phenomenon is serious, and the degree of automation is insufficient, which leads to technical problems affecting chip performance.

[0005] In view of the above problems, the present application provides a semiconductor chip cleaning control system for a dicing machine.

[0006] The present application provides a semiconductor chip cleaning control system for a dicing machine, the system comprising:

[0007] An organic pollution information acquisition module, the organic pollution information acquisition module is used to acquire target organic pollution information, the target organic pollution information refers to the cutting organic pollution information of the target semiconductor chip; a first target organic extraction module, the first target organic extraction module is used to extract the first target organic in the target organic pollution information, and call the first cleaning control plan to match the first target organic; a first dynamic cleaning record acquisition module, the first dynamic cleaning record acquisition module dynamically monitors the process of cleaning the first target organic based on the first plan to obtain a first dynamic cleaning record; a first cleaning timing feature acquisition module, the first cleaning timing feature acquisition module is used to perform cleaning timing analysis on the first dynamic cleaning record to obtain a first cleaning timing feature; a first stop instruction acquisition module, the first stop instruction acquisition module is used to predict a first predicted cleaning time according to the first cleaning timing feature, and issue a first stop instruction according to the first predicted cleaning time; a cleaning link entry module, the cleaning link entry module enters the cleaning link of the second target organic in the target organic pollution information based on the first stop instruction.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] An organic pollution information acquisition module is used to acquire target organic pollution information; a first target organic extraction module is used to extract the first target organic in the target organic pollution information, and call the first cleaning control plan to match the first target organic; a first dynamic cleaning record acquisition module is used to dynamically monitor the process of cleaning the first target organic based on the first plan to obtain a first dynamic cleaning record; a first cleaning timing feature acquisition module is used to perform cleaning timing analysis on the first dynamic cleaning record to obtain a first cleaning timing feature; a first stop instruction acquisition module is used to predict the first predicted cleaning time according to the first cleaning timing feature, and issue a first stop instruction according to the first predicted cleaning time; a cleaning link entry module is used to enter the cleaning link of the second target organic in the target organic pollution information based on the first stop instruction. A more efficient and intelligent cleaning control is achieved in the process of plasma cleaning organic pollution, and the technical effect of significantly improving chip cleaning quality and performance protection is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A schematic diagram of the structure of a semiconductor chip cleaning control system for a dicing machine is provided for an embodiment of the present application;

[0012] Figure 2 A schematic diagram of a first cleaning timing characteristic acquisition process of a semiconductor chip cleaning control system for a dicing machine is provided for an embodiment of the present application.

[0013] Explanation of the reference numerals: organic pollution information acquisition module 10 , first target organic matter extraction module 20 , first dynamic cleaning record acquisition module 30 , first cleaning timing feature acquisition module 40 , first stop instruction acquisition module 50 , cleaning phase entry module 60 . DETAILED DESCRIPTION

[0014] The present application provides a semiconductor chip cleaning control system for a dicing machine, which is used to solve the technical problems in the prior art that when using plasma cleaning to treat organic contamination after semiconductor chips are cut, the cleaning time is difficult to accurately control, the over-cleaning phenomenon is serious, and the degree of automation is insufficient, which leads to technical problems affecting chip performance.

[0015] The technical solutions in the embodiments of the present application will be described clearly and completely below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0016] Embodiment 1, as Figure 1 As shown, the present application provides a semiconductor chip cleaning control system for a dicing machine, the system comprising:

[0017] The organic pollution information acquisition module 10 is used to acquire target organic pollution information, and the target organic pollution information refers to the cutting organic pollution information of the target semiconductor chip.

[0018] Specifically, the organic pollution information acquisition module 10 accurately obtains the target organic pollution information. When it comes to the target semiconductor chip, especially during the cutting process, various factors lead to the generation of organic pollution. These organic pollutions come from the lubricant and coolant residues in the cutting process, or from contact with organic substances in the surrounding environment. The organic pollution information acquisition module 10 uses a variety of technical means to detect and collect these pollution information. Through high-precision chemical sensors, these sensors can produce sensitive responses to specific organic molecules on the surface of the chip. When there is organic pollution on the chip, the sensor will convert it into an electrical signal, and the module will then process and analyze these signals, using spectral analysis technology to irradiate the chip surface and detect the reflected or transmitted spectrum to determine the type and concentration of organic matter. In addition, the image recognition technology is combined to observe the microscopic morphology of the chip surface and identify the existing organic pollution areas. Through these comprehensive methods, the organic pollution information acquisition module 10 can comprehensively and accurately obtain the cutting organic pollution information of the target semiconductor chip.

[0019] The first target organic matter extraction module 20 is used to extract the first target organic matter in the target organic matter pollution information, and call a first cleaning control plan to match the first target organic matter.

[0020] Specifically, the first target organic matter extraction module 20 first accurately extracts the first target organic matter from the target organic matter pollution information. The target organic matter pollution information contains a variety of organic impurities, and identifies the specific organic matter that has an important impact or needs to be treated first. Once the first target organic matter is determined, the cleaning control plan will be quickly retrieved. This plan is a pre-established database or strategy set, which contains cleaning solutions for different types of organic matter. The first plan that matches the first target organic matter is found in it. The first plan includes specific cleaning reagents, cleaning process parameters, cleaning time and other detailed information. By retrieving this plan, a specific action plan is formulated for cleaning the first target organic matter to ensure the efficiency and pertinence of the cleaning process. For example, if the first target organic matter is a specific oily substance, the first plan will specify the use of a specific organic solvent for cleaning, and specify parameters such as the cleaning temperature, time and stirring speed to remove this organic contamination to the greatest extent while ensuring that other parts of the semiconductor chip are not damaged.

[0021] The first dynamic cleaning record acquisition module 30 dynamically monitors the process of cleaning the first target organic matter based on the first plan to obtain a first dynamic cleaning record.

[0022] Specifically, when the first target organic matter is cleaned according to the first plan, the first dynamic cleaning record acquisition module 30 continuously monitors the cleaning process dynamically, and obtains key data in the cleaning process in real time by connecting various sensors and monitoring equipment. For example, the temperature, concentration, flow rate and other parameters of the cleaning solution are monitored, which have an important influence on the cleaning effect. The change of temperature will affect the activity of the cleaning reagent, the fluctuation of concentration will lead to the instability of the cleaning ability, and the size of the flow rate is related to the contact time and impact force between the cleaning solution and the chip surface. At the same time, the module also monitors the physical changes in the cleaning process, such as the morphological changes on the chip surface, the degree of removal of organic matter, etc. Through a high-resolution microscope or other imaging equipment, the changes in the chip surface during the cleaning process are observed in real time, and the process of gradual removal of organic matter is recorded. In addition, the progress and effect of cleaning are indirectly judged by detecting the changes in the organic matter content in the cleaning solution. After continuous dynamic monitoring, the first dynamic cleaning record acquisition module 30 finally obtains the first dynamic cleaning record, which contains various detailed information in the cleaning process. By analyzing the first dynamic cleaning record, the effectiveness of the cleaning process is evaluated, potential problems are found, and the cleaning parameters are adjusted in time to ensure the efficient cleaning work.

[0023] The first cleaning timing feature acquisition module 40 is used to perform cleaning timing analysis on the first dynamic cleaning record to obtain a first cleaning timing feature.

[0024] Specifically, the first cleaning timing feature acquisition module 40 undertakes the key analysis task. After the first dynamic cleaning record is generated, the first cleaning timing feature acquisition module 40 begins to perform an in-depth cleaning timing analysis on it. This module carefully combs the various changes that occur over time in the first dynamic cleaning record, marks different time nodes in the cleaning process, and divides the entire cleaning process into several stages. Then, for each stage, the changing trends of various parameters are analyzed. For example, observe the changes in the temperature of the cleaning solution at different time points, whether it gradually increases, remains stable, or gradually decreases; whether the concentration of the cleaning solution changes as the cleaning proceeds, and the impact of this change on the cleaning effect. At the same time, the module will also pay attention to the performance of the removal progress of organic matter on the chip surface on the time axis, and through image analysis technology, compare the morphological changes on the chip surface at different time points to determine the removal rate of organic matter at each stage. In addition, other relevant parameters, such as pressure changes during the cleaning process, adjustment of the stirring speed, and other temporal distribution characteristics, will also be analyzed. Through comprehensive analysis of these time series data, the first cleaning time series features can be extracted. These features include key time nodes in the cleaning process, main change trends in different stages, cleaning speed and other information, so that adjustments and optimizations can be made when necessary.

[0025] The first stop instruction acquisition module 50 is used to predict a first predicted cleaning duration according to the first cleaning timing characteristics, and issue a first stop instruction according to the first predicted cleaning duration.

[0026] Specifically, the first cleaning time series feature data is collected, including the cleaning liquid temperature, concentration, and flow rate change data at different time points during the cleaning process, the change data of the organic residue on the chip surface over time, and the pressure change data during the cleaning process. These data are preprocessed to remove obvious outliers. For example, if the temperature data at a certain time point deviates significantly from the normal range, it is removed or replaced by interpolation with the data at adjacent time points. At the same time, the data is normalized so that the data of different features are in a similar numerical range. Key features are extracted from the preprocessed first cleaning time series features. For example, the average rate of change of the cleaning liquid temperature, the average rate of reduction of the organic residue, etc. are calculated, and the features with higher correlation are selected by analyzing the correlation between different features and the cleaning time. For example, the Pearson correlation coefficient is used to measure the linear correlation between the feature and the cleaning time, and the feature with a larger absolute value of the correlation coefficient is selected as the input feature. Random forest is an ensemble learning algorithm composed of multiple decision trees. First, the first cleaning time series feature is extracted from the historical cleaning data as the input variable, and the corresponding actual cleaning time is used as the output variable. A part of the data is randomly selected as the training set, and the remaining data is used as the test set. In the training set, for each decision tree, some features and samples are randomly selected for training. The decision tree is split at each node according to the selected features, so that the target value (cleaning time) of the sample in the child node after the split is purer. Repeat the above process to construct multiple decision trees to form a random forest. When predicting, the new first cleaning time series feature is input into the random forest. Each decision tree will give a predicted value of the cleaning time. The final prediction result is the average of the predicted values ​​of all decision trees. The first cleaning time series feature in the current cleaning process is input into the trained random forest regression model. The model will calculate the first predicted cleaning time according to the input feature value. Since the random forest is an integration of multiple decision trees, its prediction result is relatively more stable and accurate. Set up a timer and start timing from the beginning of cleaning. Continuously compare the current elapsed time with the first predicted cleaning time. When the current elapsed time is close to the first predicted cleaning time, start to evaluate the cleaning effect in real time. For example, observe whether the amount of organic residue on the surface of the chip reaches a predetermined threshold through a high-resolution microscope, and monitor whether the parameters of the cleaning solution are stable. If the cleaning effect reaches the expected standard, a first stop instruction is issued to stop the cleaning process. If the cleaning effect does not reach the expected standard, monitoring continues until the condition is met. Through the random forest regression algorithm, the first stop instruction acquisition module 50 can effectively predict the first predicted cleaning time according to the first cleaning time series feature, and accurately issue the first stop instruction to achieve intelligent control of the cleaning process.

[0027] The cleaning phase enters the module 60, and the cleaning phase enters the cleaning phase of the second target organic matter in the target organic matter pollution information based on the first stop instruction.

[0028] Specifically, the cleaning link entry module 60 plays a key role in connection. When the first stop command is issued, the cleaning link entry module 60 starts to act. The main task of this module is to smoothly enter the cleaning link of the second target organic matter in the target organic matter pollution information based on the first stop command. During the entire cleaning process, in various types of organic matter pollution, the cleaning link entry module 60 performs step-by-step cleaning of each organic matter type in turn according to a pre-set order to ensure that each organic matter can be treated in a targeted manner. In order to achieve efficient cleaning, the module needs to timely predict the time to achieve the required cleaning effect. Through real-time monitoring and analysis of the cleaning process, the module can accurately judge the progress of the current cleaning and when the expected cleaning effect can be achieved. Once it is determined that the current first target organic matter cleaning meets the requirements, the module will immediately switch to the cleaning of the second target organic matter to avoid over-cleaning. Over-cleaning will not only waste time and resources, but also cause unnecessary damage to the semiconductor chip. For example, during the cleaning process, the module will comprehensively judge the cleaning state of the first target organic matter based on factors such as the parameter changes of the cleaning solution, the residual amount of organic matter on the chip surface, and the cleaning time. If the concentration of the cleaning solution gradually stabilizes, the residual amount of the first target organic matter on the chip surface is reduced to an acceptable range, and the cleaning time is close to the predicted duration, then the module will consider that the cleaning of the first target organic matter has been completed. At this time, the cleaning process of the second target organic matter is quickly started to ensure the continuity and efficiency of the entire cleaning process. The cleaning link enters module 60, through precise control and timely switching, to achieve orderly cleaning of different target organic matter.

[0029] In a possible implementation, the first target organic matter extraction module 20 further includes:

[0030] Establish a semiconductor chip cleaning database, the semiconductor chip cleaning database refers to chip cleaning record data after a dicing machine cuts a semiconductor; extract a first cleaning data group from the semiconductor chip cleaning database, the first cleaning data group includes first organic contamination information of the first chip, first organic cleaning information and first cleaning effect information; if the first cleaning effect information reaches a cleaning threshold constraint, use the first organic cleaning information as an initialization cleaning strategy for the first organic contamination information; establish the cleaning control plan based on the initialization cleaning strategy.

[0031] Specifically, building a semiconductor chip cleaning database is an important basic work. This database specifically collects chip cleaning record data after the dicing machine cuts the semiconductor, and stores a large amount of valuable information about chip cleaning. From this database, the first cleaning data group is extracted, and the first cleaning data group covers the first organic pollution information, the first organic cleaning information and the first cleaning effect information of the first chip. The first organic pollution information describes in detail the type, degree and distribution of the specific organic pollution existing on the first chip. The first organic cleaning information records the parameters such as the cleaning method adopted for this organic pollution, the cleaning reagent used, the cleaning time and temperature. The first cleaning effect information reflects the cleanliness of the first chip after cleaning, the performance recovery and whether the expected cleaning goal is achieved. If the first cleaning effect information reaches the cleaning threshold constraint, it means that the cleaning result meets the preset standards and requirements. In this case, the first organic cleaning information is used as the initialization cleaning strategy of the first organic pollution information. Based on this initialization cleaning strategy, a cleaning control plan is further established. The cleaning control plan is a comprehensive solution that integrates various cleaning strategies and methods to deal with different types of organic pollution and cleaning needs. By incorporating the initial cleaning strategy, the accuracy and reliability of cleaning can be improved. For example, if similar organic contamination is encountered during the subsequent cleaning process, the cleaning control plan can be directly called to quickly determine the appropriate cleaning method and parameters, thereby saving time and resources and improving cleaning efficiency.

[0032] In a possible implementation, the first target organic matter extraction module 20 further includes:

[0033] Reading a predetermined cleaning evaluation index; performing traversal matching on the first cleaning effect information based on the predetermined cleaning evaluation index to obtain first cleaning index data; performing normalized weighted calculation on the first cleaning index data to obtain a first cleaning index; and quantitatively representing the first cleaning effect information using the first cleaning index.

[0034] Specifically, reading the predetermined cleaning evaluation index is a key step. These predetermined cleaning evaluation indexes are pre-set according to the requirements and standards of semiconductor chip cleaning, including the amount of organic matter residue on the chip surface after cleaning, the residual degree of cleaning liquid, the performance recovery of the chip and other parameters.

[0035] It is a crucial link to traverse and match the first cleaning effect information based on the predetermined cleaning evaluation index. The predetermined cleaning evaluation index represents the expected requirements for the state of the semiconductor chip after cleaning. These indicators cover multiple aspects, such as the residual degree of organic matter on the chip surface, the residual amount of cleaning liquid on the chip, the recovery of the electrical performance of the chip, etc. When the first cleaning effect information is traversed and matched, each data in the first cleaning effect information is compared with the predetermined cleaning evaluation index one by one. First, for the indicator of the residual degree of organic matter on the chip surface, the data on organic matter detection in the first cleaning effect information is checked to determine whether the actual amount of organic matter residual on the chip surface after cleaning is within the predetermined acceptable range. If the residual amount is too high, it means that the cleaning effect is not good and the cleaning strategy needs to be further adjusted. Then, for the indicator of the residual amount of cleaning liquid, the data related to the cleaning liquid in the first cleaning effect information is analyzed to determine whether the cleaning liquid has been fully removed to avoid adverse effects on the subsequent use of the chip. At the same time, for the indicator of the recovery of the electrical performance of the chip, the electrical parameters of the chip before and after cleaning are compared to determine whether the cleaning process effectively restores the performance of the chip. Through such a comprehensive and detailed traversal matching process, the specific performance of the first cleaning effect on each predetermined cleaning evaluation index can be accurately determined, thereby obtaining the first cleaning index data, better evaluating the cleaning effect, optimizing the cleaning process, and ensuring that the cleaning quality of the semiconductor chip reaches the best level.

[0036] It is a key analysis step to perform normalized weighted calculation on the first cleaning index data to obtain the first cleaning index. The first cleaning index data usually contains information of multiple different dimensions, which have different dimensions and numerical ranges. In order to be able to integrate these data to accurately evaluate the cleaning effect, normalization processing is required. Normalization converts data of different dimensions into a unified range. For example, the data of indicators such as organic residue and cleaning liquid residue are converted into values ​​between 0 and 1. After normalization, weighted calculation is performed. Weighted calculation is to assign different weights according to the importance of each indicator. Different cleaning indicators have different degrees of influence on the cleaning effect. For example, for some high-precision semiconductor chips, organic residue is a more critical indicator, so it is assigned a higher weight; while for other application scenarios, the impact of cleaning liquid residue is relatively small, and the weight is correspondingly lower. By performing weighted calculation on the normalized first cleaning index data, a comprehensive value, namely the first cleaning index, is obtained. This index can more comprehensively and accurately reflect the overall level of cleaning effect. If the first cleaning index is high, it means that the cleaning effect is good and all indicators are within a relatively ideal range; conversely, if the index is low, it means that there are problems in the cleaning process and the cleaning strategy or parameters need to be further adjusted.

[0037] Finally, the first cleaning effect information is quantitatively characterized by the first cleaning index. This index provides a specific numerical representation for the cleaning effect, making the cleaning effect more intuitive and easy to compare. Whether the cleaning has met the expected requirements can be judged based on this index.

[0038] In a possible implementation, the first target organic matter extraction module 20 further includes:

[0039] If the first cleaning effect information does not reach the cleaning threshold constraint, extract a second cleaning data group from the semiconductor chip cleaning database; the second cleaning data group includes second organic pollution information, second organic cleaning information and second cleaning effect information of a second chip, wherein the second chip is a chip of the same type as the first chip, and the second organic pollution information is the same as the first organic pollution information; determine whether a second cleaning index obtained by analyzing the second cleaning effect information reaches the cleaning threshold constraint, and establish the cleaning control plan based on the determination result.

[0040] Specifically, if the first cleaning effect information does not reach the cleaning threshold constraint, it means that the current cleaning result for the first chip fails to meet the preset standard. At this time, the second cleaning data group is extracted from the semiconductor chip cleaning database. The semiconductor chip cleaning database stores a large number of cleaning record data of different chips. When the second cleaning data group needs to be extracted, it will be screened according to specific conditions. Since the first cleaning effect is not good, chip cleaning data with similar characteristics to the first chip will be found. The second chip in the second cleaning data group is determined to be a similar chip to the first chip, which means that there are similarities in material, manufacturing process, use, etc. At the same time, the second organic pollution information is required to be the same as the first organic pollution information, so as to ensure that the extracted data is highly correlated with the current cleaning problem. By extracting the second cleaning data group, the cleaning experience of similar chips and the same organic pollution situation can find solutions or improvement directions, which will help improve the cleaning effect so that it can eventually meet the requirements of the cleaning threshold constraint.

[0041] The second cleaning data group plays a key reference role. This data group covers the second organic contamination information, second organic cleaning information and second cleaning effect information of the second chip. First, the second chip is determined to be a similar chip to the first chip, and has similarities in many aspects, such as the chip material, structural design, manufacturing process and environmental conditions during use. This similarity makes the experience and information obtained from the cleaning data of the second chip more applicable to the first chip. Secondly, the second organic contamination information is the same as the first organic contamination information. This is very important because the same organic contamination means that the cleaning problems faced are highly consistent. For example, if both the first chip and the second chip are contaminated by a specific type of organic adhesive residue, then the cleaning strategy for the second chip has direct reference significance for the first chip.

[0042] It is a key decision-making process to judge and analyze the second cleaning effect information to determine whether the second cleaning index reaches the cleaning threshold constraint, and to form a cleaning control plan based on this. First, the second cleaning effect information is judged and analyzed in depth. This process includes comparing and evaluating each data in the second cleaning effect information with the predetermined cleaning evaluation index one by one. For example, check whether the residual amount of organic matter on the chip surface is within an acceptable range, whether the residual degree of the cleaning liquid meets the standard, and whether the performance recovery of the chip meets expectations. Through comprehensive analysis of these data, the second cleaning index is obtained by normalized weighted calculation. This index is a quantitative representation of the cleaning effect of the second chip, which can intuitively reflect the quality and level of cleaning. Next, it is judged whether the second cleaning index reaches the cleaning threshold constraint. If the second cleaning index reaches the constraint condition, it means that the cleaning strategy for the second chip is effective to a certain extent. At this time, according to the specific situation of the cleaning information and the second cleaning index of the second chip, combined with the characteristics of the first chip, a cleaning control plan is formed. For example, if the second chip achieves a good cleaning effect under the use of a specific cleaning reagent and cleaning parameter combination, these cleaning reagents and parameters are applied to the first chip, and appropriate adjustments are made according to the difference between the two. However, if the second cleaning index does not reach the cleaning threshold constraint, the cleaning strategy needs to be reviewed, and the cleaning data of other similar chips needs to be further extracted from the database for analysis, or different cleaning methods and parameter combinations need to be tried to find a more effective cleaning solution. In this process, continuous attempts and optimizations are made until a solution that meets the cleaning threshold constraint is found, thereby ensuring the effectiveness and reliability of the cleaning control plan, coping with cleaning problems of different chips, and improving cleaning efficiency and quality.

[0043] In one possible implementation, Figure 2 As shown, the first cleaning timing feature acquisition module 40 also includes:

[0044] Extract a first record from the first dynamic cleaning record, where the first record corresponds to a first monitoring time; calculate and analyze the first record based on the predetermined cleaning evaluation index to obtain a first real-time cleaning index; establish a first mapping relationship between the first real-time cleaning index and the first monitoring time; construct a first cleaning sequence based on the first mapping relationship, and collect time domain features of the first cleaning sequence to obtain the first cleaning sequence features.

[0045] Specifically, extracting the first record in the first dynamic cleaning record is a key step. The first dynamic cleaning record is a data set obtained by continuous dynamic monitoring of the process of cleaning the first target organic matter based on the first plan, and contains rich information at various time points in the cleaning process. When it is determined to extract the first record, this record corresponds to a specific first monitoring time. This time point is a key node in the cleaning process, or a time point selected according to specific needs. For example, it is a certain time interval after the start of cleaning, the moment when the temperature of the cleaning liquid reaches a certain value, or the time point when the progress of organic matter removal reaches a certain level. By extracting this specific first record, the cleaning state at this time point is analyzed more focused. The first record contains parameter information of the cleaning liquid, such as temperature, concentration, flow rate, etc.; state information of the chip surface, such as organic matter residue, surface morphology, etc.; and other relevant data in the cleaning process, such as pressure, stirring speed, etc. For example, by comparing records at different time points, analyze the development trend of the cleaning process; or evaluate whether the current cleaning effect meets expectations based on the data in the first record, so as to adjust the cleaning strategy in time.

[0046] The predetermined cleaning evaluation index is a set of carefully designed standards used to measure the effect and quality of the cleaning process. These indicators cover multiple aspects, such as the residual amount of organic matter on the chip surface, the residual degree of the cleaning liquid, the performance recovery of the chip, etc. When analyzing the first record, first compare the data in the first record with the predetermined cleaning evaluation index one by one. If the first record contains data on the residual amount of organic matter on the chip surface, according to the requirements of the residual amount of organic matter in the predetermined index, it is judged whether the current residual amount is within an acceptable range. Similarly, similar analysis is performed on the data on the residual degree of the cleaning liquid and the chip performance recovery. Then, calculations are performed based on the comparison results. The weighted average method is used to comprehensively calculate the results of different indicators. For example, if the importance of the residual amount of organic matter index is high, a larger weight will be given in the calculation; while the importance of the residual degree of the cleaning liquid index is relatively low, a smaller weight will be given. Through such a calculation and analysis process, the first real-time cleaning index is finally obtained. This index is a quantitative value that can intuitively reflect the real-time status of the cleaning process at the time point corresponding to the first record. If the first real-time cleaning index is high, it means that the cleaning effect is good and all indicators are close to or reach the predetermined requirements; conversely, if the index is low, it means that there are problems in the cleaning process and the cleaning strategy or parameters need to be further adjusted.

[0047] The first monitoring time represents a specific moment in the cleaning process, while the first real-time cleaning index reflects the effect state of the cleaning at that moment. By establishing this mapping relationship, the cleaning effect is clearly linked to the specific time point. From the perspective of practical operation, when the first record corresponding to the first monitoring time is extracted from the first dynamic cleaning record and the first real-time cleaning index is calculated, this index is associated with the first monitoring time and stored. The corresponding cleaning index can be queried at any time according to the time point to understand the progress of the cleaning process at different times. For example, as the cleaning proceeds, the first monitoring time at different time points will continuously generate new first real-time cleaning indexes. Through this mapping relationship, a curve of the cleaning index changing over time is drawn to intuitively show the dynamic change trend of the cleaning effect. If it is found that the cleaning index rises slowly or decreases within a certain period of time, the reason is analyzed in time, such as improper cleaning parameter settings, changes in organic pollution, etc. By observing the change pattern of the cleaning index over time, it is inferred whether the cleaning effect can meet the expected requirements at a certain time point in the future, so as to decide when to issue a stop command to avoid over-cleaning or incomplete cleaning.

[0048] The first cleaning time series is constructed by using the first mapping relationship between the first real-time cleaning index and the first monitoring time. This time series is a series of first real-time cleaning index values ​​arranged in chronological order, which intuitively shows the change of the cleaning effect over time during the cleaning process. For example, from the start of cleaning to each different time point, the corresponding first real-time cleaning index is arranged in sequence to form a dynamic change curve. Next, the time domain features of the first cleaning time series are collected. This means carefully analyzing the various characteristics of this time series data in the time domain. On the one hand, observe the changing trend of the cleaning index over time, whether it is gradually rising, indicating that the cleaning effect is getting better and better, or fluctuating or even declining. If it is gradually rising, further analyze the rising speed to determine whether the cleaning process is proceeding stably. On the other hand, pay attention to the key time nodes in the cleaning time series, such as the moment when the cleaning index changes significantly, or the time point when a certain threshold is reached. In addition, analyze the stability of the cleaning process, and measure its fluctuation degree by calculating the variance or standard deviation of the cleaning index over a period of time. By collecting comprehensive time domain features of the first cleaning time series, the first cleaning time series features are finally obtained. For example, the time required for cleaning to achieve the expected effect can be predicted based on these characteristics, or the cleaning strategy can be adjusted in time according to changes in characteristics during the cleaning process to ensure the optimization of the cleaning effect.

[0049] In a possible implementation, the first stop instruction acquisition module 50 further includes:

[0050] Extract a second record from the first dynamic cleaning record, where the second record corresponds to the first monitoring time; read a predetermined cleaning control index to perform calculation and analysis on the second record to obtain a first real-time cleaning intensity; establish a second mapping relationship between the first real-time cleaning intensity and the first monitoring time; construct a second cleaning sequence according to the second mapping relationship, and perform a trend analysis on the second cleaning sequence to obtain a cleaning intensity change ratio; when the cleaning intensity change ratio is within a ratio threshold, issue a prediction instruction, and perform a prediction analysis on the first cleaning sequence feature based on the prediction instruction to obtain the first predicted cleaning duration.

[0051] Specifically, the first dynamic cleaning record is a rich data set generated by dynamic monitoring of the cleaning process, which contains information from different aspects. When it is decided to extract the second record, this record is selected because it corresponds to a specific first monitoring time. The first monitoring time is of great significance in the cleaning process, is a key time node, or is a moment selected according to specific analysis needs. For example, the first monitoring time is when the temperature of the cleaning liquid reaches a certain value during the cleaning process, or when the progress of organic matter removal reaches a certain proportion. By extracting the second record corresponding to this time point, specific data related to that moment is obtained. The information contained in the second record is different from that in the first record, focusing on other aspects of the cleaning process, such as the operating parameters of the cleaning equipment, changes in the chemical properties of the cleaning liquid, etc.

[0052] Reading the predetermined cleaning control index and calculating and analyzing the second record to obtain the first real-time cleaning intensity is a key link. The predetermined cleaning control index is a set of pre-set standards and parameters used to measure and evaluate the intensity of the cleaning process. These indicators include factors such as the flow rate, pressure, temperature, and concentration of the cleaning liquid, as well as parameters such as the power and stirring speed of the cleaning equipment. By reading these indicators, the expected values ​​and ranges of various factors in the cleaning process are determined. When calculating and analyzing the second record, the data in the second record is compared with the predetermined cleaning control index. For example, if the second record contains the flow rate data of the cleaning liquid, it is compared with the predetermined flow rate range to determine whether the current flow rate meets the requirements. At the same time, other factors such as pressure and temperature are combined to calculate the cleaning intensity. Calculating the cleaning intensity involves algorithms and formulas. For example, the cleaning intensity is calculated based on the weighted average of factors such as the flow rate, pressure and temperature of the cleaning liquid. By calculating and analyzing the second record, the first real-time cleaning intensity is finally obtained. This intensity value reflects the actual intensity level of the cleaning process at a specific time point and determines whether the cleaning is strong enough to remove organic pollution. For example, if the cleaning intensity is too low, adjust the cleaning parameters, such as increasing the flow rate or pressure of the cleaning liquid, to improve the cleaning effect. On the contrary, if the cleaning intensity is too high, it will cause damage to the chip, so reduce the cleaning intensity.

[0053] A second mapping relationship between the first real-time cleaning intensity and the first monitoring time is established. Like the first mapping relationship, this relationship links the cleaning intensity to a specific time point, so that the changes in the cleaning intensity at different time points can be understood at any time. Through this mapping, a second cleaning time sequence is constructed, that is, the first real-time cleaning intensity values ​​arranged in chronological order.

[0054] The second cleaning time series is constructed by using the second mapping relationship between the first real-time cleaning intensity and the first monitoring time. This time series shows the change of the cleaning intensity during the cleaning process in chronological order. Through this time series, the specific values ​​of the cleaning intensity at different time points and its dynamic changes over time can be intuitively seen. Next, the second cleaning time series is trend analyzed, which means carefully observing the change trend of the cleaning intensity over time. Ideally, the cleaning intensity of each link should be gradually weakened to avoid over-cleaning. Over-cleaning will cause unnecessary damage to the chip and waste resources and time. In order to determine whether the cleaning intensity is gradually weakening and reaching the predetermined weakening rate, a series of calculations and analyses are required to calculate the difference in cleaning intensity between adjacent time points, and then calculate the change rate according to the time interval. By comparing the change rate of different time intervals, it is determined whether the weakening trend of the cleaning intensity is in line with expectations. If the change ratio of the cleaning intensity is within a reasonable range, that is, it is gradually approaching the predetermined weakening rate, it means that the cleaning process is moving in the right direction. At this time, continue to monitor the cleaning process to ensure that it achieves the expected cleaning effect while avoiding over-cleaning. If the change ratio does not meet the requirements, it is necessary to adjust the cleaning parameters, such as reducing the flow rate, pressure or temperature of the cleaning liquid, so as to reduce the cleaning intensity according to the predetermined trend. By constructing the second cleaning sequence and performing trend analysis to obtain the cleaning intensity change ratio, the cleaning process can be effectively controlled to ensure that the cleaning intensity gradually decreases to avoid over-cleaning, thereby improving the cleaning quality and efficiency.

[0055] Collect historical cleaning data and extract a large number of cleaning records from the semiconductor chip cleaning database, including the first cleaning timing characteristics of different chips under different cleaning conditions, the cleaning intensity change ratio, and the actual cleaning time. Preprocess the collected data, including removing outliers, processing missing values, and normalizing data, so that the data is more standardized and convenient for algorithm processing. Build a random forest model, randomly select features and samples, and the random forest is an integrated learning algorithm composed of multiple decision trees. When building each decision tree, randomly select a part of the features and samples from the historical cleaning data for training. Decision tree construction, for each sample, the decision tree is split at each node according to the selected features, so that the target value (cleaning time) of the sample in the child node after the split is purer. The splitting criteria are information gain, Gini index, etc. A random forest is formed, and the above process is repeated to construct multiple decision trees to form a random forest. When the cleaning intensity change ratio is at the ratio threshold, a prediction instruction is issued to input the first cleaning timing characteristics of the current cleaning process into the trained random forest model. Each decision tree will give a predicted value of the cleaning time based on the input features. The final first prediction cleaning time is the average of all decision tree predictions.

[0056] In a possible implementation, the first stop instruction acquisition module 50 further includes:

[0057] The first cleaning time series feature is used as input information of a cleaning time prediction model to obtain an output result, wherein the output result includes the first predicted cleaning time.

[0058] Specifically, the first cleaning time series feature is prepared as input information to be input into the random forest model. The first cleaning time series feature contains many key information of the cleaning process in the time dimension, such as the changing trend of the cleaning index over time, the fluctuation of the cleaning process, etc. Random forest is an integrated learning model composed of multiple decision trees. In the training phase, the corresponding cleaning time series features and actual cleaning time are extracted from a large amount of historical cleaning data to construct each decision tree in the random forest. During the construction process of each decision tree, a part of features and samples will be randomly selected for split growth. When the first cleaning time series feature is input, each decision tree in the random forest will predict the cleaning time according to the rules it has learned. Since the growth process of each tree is random, it is judged from different angles and feature combinations. Finally, the random forest combines the prediction results of all decision trees to obtain the output results, including the first predicted cleaning time. The average method is used, that is, the average of the predicted time of all decision trees is taken as the final prediction value.

[0059] In a possible implementation, the first stop instruction acquisition module 50 further includes:

[0060] The cleaning time prediction model is an intelligent model obtained by training a training data set based on machine learning principles. The training data set includes cleaning timing features and cleaning time. The cleaning time refers to the total cleaning time to achieve a predetermined cleaning effect.

[0061] Specifically, the cleaning time prediction model is an intelligent model built on the principle of machine learning, which plays a vital role in the cleaning process. The training data set includes cleaning timing features and cleaning time. Cleaning timing features refer to various features that change over time during the cleaning process, such as cleaning intensity, cleaning liquid temperature, cleaning liquid concentration, etc. Cleaning time refers to the total cleaning time required to achieve the predetermined cleaning effect. By monitoring and recording a large number of cleaning processes, a rich training data set is collected for training the random forest model. After sufficient training, the model can accurately predict the new cleaning timing features and give the corresponding first predicted cleaning time. In this way, in the actual cleaning process, the cleaning time is reasonably arranged according to the prediction results of the model to avoid over-cleaning or incomplete cleaning, and improve the cleaning efficiency and quality. Using the cleaning timing features and cleaning time information of the training data set, an intelligent time prediction capability is provided for the semiconductor chip cleaning process, which helps to achieve more efficient and accurate cleaning control.

[0062] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0063] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0064] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, a person skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. A semiconductor chip cleaning control system for a dicing machine, characterized in that: include: An organic pollution information acquisition module, the organic pollution information acquisition module is used to acquire target organic pollution information, the target organic pollution information refers to the cutting organic pollution information of the target semiconductor chip; A first target organic matter extraction module, the first target organic matter extraction module is used to extract the first target organic matter in the target organic matter pollution information, and call a first cleaning control plan to match the first target organic matter; a first dynamic cleaning record acquisition module, which dynamically monitors a process of cleaning the first target organic matter based on the first plan to obtain a first dynamic cleaning record; A first cleaning timing feature acquisition module, the first cleaning timing feature acquisition module is used to perform cleaning timing analysis on the first dynamic cleaning record to obtain a first cleaning timing feature; a first stop instruction acquisition module, the first stop instruction acquisition module being used to predict a first predicted cleaning duration according to the first cleaning timing feature, and to issue a first stop instruction according to the first predicted cleaning duration; a cleaning link entry module, wherein the cleaning link entry module enters a cleaning link of a second target organic matter in the target organic matter pollution information based on the first stop instruction; The first predicted cleaning duration is obtained according to the first cleaning timing feature, which includes: Extracting a second record from the first dynamic cleaning record, where the second record corresponds to a first monitoring time; Reading a predetermined cleaning control index, performing calculation and analysis on the second record, and obtaining a first real-time cleaning intensity; Establishing a second mapping relationship between the first real-time cleaning intensity and the first monitoring time; constructing a second cleaning time sequence according to the second mapping relationship, and performing a trend analysis on the second cleaning time sequence to obtain a cleaning intensity change ratio; When the cleaning intensity change ratio is within the ratio threshold, issuing a prediction instruction, and performing a prediction analysis on the first cleaning timing feature based on the prediction instruction to obtain the first predicted cleaning duration; Wherein, obtaining the first predicted cleaning time includes: The first cleaning timing feature is used as input information of a cleaning time prediction model to obtain an output result, wherein the output result includes the first predicted cleaning time. The cleaning time prediction model is an intelligent model obtained by training a training data set based on machine learning principles. The training data set includes cleaning timing features and cleaning time. The cleaning time refers to the total cleaning time to achieve a predetermined cleaning effect.

2. The semiconductor chip cleaning control system for a dicing machine according to claim 1, characterized in that: include: Establishing a semiconductor chip cleaning database, wherein the semiconductor chip cleaning database refers to chip cleaning record data after a semiconductor is cut by a dicing machine; Extracting a first cleaning data group from the semiconductor chip cleaning database, wherein the first cleaning data group includes first organic contamination information, first organic cleaning information, and first cleaning effect information of the first chip; If the first cleaning effect information reaches the cleaning threshold constraint, using the first organic matter cleaning information as an initialization cleaning strategy for the first organic matter pollution information; The cleaning control plan is established based on the initial cleaning strategy.

3. The semiconductor chip cleaning control system for a dicing machine according to claim 2, characterized in that: include: Reading predetermined cleaning evaluation indicators; Based on the predetermined cleaning evaluation index, the first cleaning effect information is traversed and matched to obtain first cleaning index data; Performing normalized weighted calculation on the first cleaning index data to obtain a first cleaning index; The first cleaning effect information is quantitatively represented by the first cleaning index.

4. The semiconductor chip cleaning control system for a dicing machine according to claim 2, characterized in that: If the first cleaning effect information does not reach the cleaning threshold constraint, extracting a second cleaning data group from the semiconductor chip cleaning database; The second cleaning data group includes second organic pollution information of a second chip, second organic cleaning information, and second cleaning effect information, wherein the second chip is a chip of the same type as the first chip, and the second organic pollution information is the same as the first organic pollution information; It is determined whether the second cleaning index obtained by analyzing the second cleaning effect information reaches the cleaning threshold constraint, and the cleaning control plan is established according to the determination result.

5. The semiconductor chip cleaning control system for a dicing machine according to claim 3, characterized in that: include: Extracting a first record in the first dynamic cleaning record, where the first record corresponds to a first monitoring time; Calculating and analyzing the first record based on the predetermined cleaning evaluation index to obtain a first real-time cleaning index; Establishing a first mapping relationship between the first real-time cleaning index and the first monitoring time; A first cleaning sequence is constructed according to the first mapping relationship, and time domain features of the first cleaning sequence are collected to obtain the first cleaning sequence features.

Images