substrate processing system

Abstract

This invention enables substrate processing apparatuses to be less labor-intensive, more energy-efficient, and / or less costly. The invention comprises: a sensor disposed in a substrate processing apparatus that detects a physical quantity of a target substrate during substrate processing; and a prediction unit that, by inputting time-series data of the physical quantity detected by the sensor, or time-series data of the physical quantity differentiated by time, into a machine learning model after learning, and outputting a grinding endpoint timing sequence, wherein the machine learning model is a model that performs machine learning using a learning dataset that takes past time-series data of the physical quantity, or time-series data of the past physical quantity differentiated by time, as input and past grinding endpoint timing sequences as output.

Description

[Technical Field] This invention relates to a substrate processing system. [Previous Technology] Various substrate processing devices are used in the manufacture of semiconductor devices. Among these, polishing (CMP) devices are representative examples of substrate processing devices. The wiring structure of a semiconductor device is formed by forming a metal film (such as a copper film) on an insulating film that forms grooves along the wiring pattern. Then, the unwanted metal film is removed by a polishing device. The polishing device supplies polishing slurry (slurry) to a polishing pad on a polishing table and polishes the substrate surface by moving the substrate and the polishing pad relative to each other. Previous polishing equipment included a polishing endpoint detection device to detect the polishing endpoint of the substrate. This polishing endpoint detection device monitored the polishing of the substrate based on polishing index values ​​that display film thickness (e.g., stage torque current, output signal of eddy current film thickness sensor, output signal of optical film thickness sensor), and determined the polishing endpoint as the time when the metal film was removed. Previously, the acquisition, analysis, and troubleshooting of operational data for substrate processing equipment (such as polishing equipment) were carried out by visiting the maintenance personnel of that equipment. In this case, it was done, for example, by communicating with the design or development department by telephone or email. For example, Patent Document 1 discloses a system capable of remotely monitoring multiple grinding endpoint detection devices and operating them remotely, comprising multiple grinding endpoint detection devices and a host computer connected to the multiple grinding endpoint detection devices via a network. Furthermore, Patent Document 1 describes the host computer having memory for storing grinding endpoint detection data received from the multiple grinding endpoint detection devices and a display screen for displaying the grinding endpoint detection data; the host computer transmits a new grinding endpoint detection scheme to at least one selected grinding endpoint detection device from the multiple grinding endpoint detection devices, and rewrites the grinding endpoint detection scheme of that selected at least one grinding endpoint detection device. [Prior Art Documents] [Patent Documents] [Patent Document 1] Japanese Patent Application Publication No. 2013-176828 [Summary of the Invention] (The problem that the invention aims to solve) However, since rewriting the polishing endpoint detection scheme still requires manpower, there is a need for labor-saving measures and automation of equipment, units (etc.), and factories. In addition, there is a need to shorten the downtime of substrate processing equipment, reduce the time and cost of personnel movement, analysis, and the formulation of countermeasures for anomalies, and to save labor, energy and / or reduce costs, as well as to automate the operation of equipment, units (etc.) and / or factories. The present invention addresses the aforementioned problems by providing a substrate processing system that reduces the effort, energy consumption, and / or cost of substrate processing apparatus. (Means for solving the problem) The first type of substrate processing system of the present invention includes: a sensor disposed in a substrate processing apparatus to detect a physical quantity of an object in the processing of the object substrate; and a prediction unit that inputs time-series data of the physical quantity detected by the aforementioned sensor or time-series data of the physical quantity differentiated by time into a machine learning model that has completed learning, and outputs a grinding endpoint timing sequence of the grinding end timing sequence; the aforementioned machine learning model is a model that performs machine learning using learning data that takes time-series data of the aforementioned physical quantity in the past or time-series data of the physical quantity differentiated by time as input and takes past grinding endpoint timing sequences as output. When this configuration is adopted, the grinding endpoint timing can be automatically predicted, thus reducing the time and cost spent on predicting the grinding endpoint timing, achieving labor saving, energy saving, and / or cost reduction. Furthermore, in the past, when using time-series data of the current value of the rotary stage motor differentiated by time, multiple minimum or maximum points would occur, leading to the problem of not being able to immediately identify which minimum or maximum point represents the grinding endpoint timing. On the other hand, since the learned machine learning model learns from past physical quantity time-series data or time-series data of past physical quantity differentiated by time as input and past grinding endpoint timings as output, even when inputting time-series data of unknown physical quantities or time-series data of time-series data of unknown physical quantities differentiated by time, the probability of outputting the correct grinding endpoint timing is still increased. The substrate processing system of the second aspect of the present invention, as in the first aspect, further comprises: a determination unit that compares the time series data of the physical quantity detected by the aforementioned sensor with past time series data to determine whether there is an abnormality in the time series change of the physical quantity; a decision unit that, when the determination unit determines that there is an abnormality, re-determines the processing conditions; and an update control unit that controls the update to be performed according to the processing conditions determined by the decision unit. When this configuration is adopted, the time and cost of predicting the grinding endpoint can be reduced because it can automatically predict the grinding endpoint timing. When there are abnormal changes in the time series of physical quantities, the grinding endpoint timing is automatically corrected by updating the processing conditions (scheme). Therefore, since the scheme can be updated without being on-site, labor saving, energy saving, and / or cost reduction can be achieved. If on-site operation is required, it can be completed with less work than before. Specifically, the grinding endpoint timing can be accurately determined from waveform changes, and whether the grinding is operating normally can be determined from the time series changes of physical quantities. Even if the grinding is not operating normally, the scheme can still be automatically updated. The third embodiment of the substrate processing system of the present invention, such as the first or second embodiment of the substrate processing system, wherein the physical quantity of the aforementioned object is the current value of the table rotary motor of the aforementioned substrate processing apparatus, the current value of the upper annular turntable rotary motor of the aforementioned substrate processing apparatus, or the torque of the table of the aforementioned substrate processing apparatus, and further comprises: a selection unit, which selects the time series data of the current value detected by the aforementioned sensor by time differentiation; and a learning unit, which generates the aforementioned learned machine learning model by using a learning dataset that takes the time series data of the current value selected by the aforementioned selection unit as input and the grinding end time sequence as output for machine learning. When this configuration is adopted, since only the desired minimum or maximum point can be selected from the time series data of the time series data of the time differential current value, the prediction accuracy of the grinding end time can be improved. The fourth embodiment of the substrate processing system of the present invention, such as the third embodiment of the substrate processing system, wherein when the aforementioned selection unit cannot detect a minimum or maximum point that meets the set criteria in the aforementioned time series data that is differentiated by time, it selects the aforementioned time series data of current value by excluding the time series data of current value before differentiation from the dataset used for learning. When this configuration is adopted, if no minimum or maximum point that meets the set reference cannot be detected, the time series data of the current value before the derivative can be excluded from the learning dataset, thereby improving the prediction accuracy of the grinding endpoint timing. The fifth embodiment of the substrate processing system of the present invention comprises: a sensor disposed in a substrate processing apparatus for detecting a target physical quantity in the processing of a target substrate; a storage unit for storing at least one batch of substrates in association with time-series data of past physical quantities in the processing of the substrate; an extraction unit for extracting time-series data of past physical quantities corresponding to the batch of target substrates being processed, with reference to the aforementioned storage unit; and a determination unit for comparing the time-series data of the physical quantity detected by the aforementioned sensor with the past time-series data extracted by the aforementioned extraction unit to determine whether there is an abnormality in the time-series change of the physical quantity. When this configuration is adopted, since anomalies in the time series data of physical quantities of the substrate processing device can be automatically detected, the detection time and cost of such anomalies can be reduced, thereby achieving labor saving, energy saving and / or cost reduction. The sixth embodiment of the substrate processing system of the present invention, such as the fifth embodiment of the substrate processing system, includes: a determination unit that determines the processing conditions again when the determination unit determines that there is an abnormality; and an update control unit that controls the system to update the processing conditions determined by the determination unit. When this configuration is adopted, if there are anomalies in the time series data of the processing volume of the substrate processing device, the processing conditions (scheme) can be updated. Therefore, the time and cost of formulating countermeasures for anomalies can be reduced, and labor saving, energy saving and / or cost reduction can be achieved. The seventh embodiment of the substrate processing system of the present invention comprises: at least one sensor disposed in a substrate processing apparatus to detect a target physical quantity in the processing of a target substrate; a first storage unit that stores at least one time-series data of past physical quantities in the substrate processing, correlated with a batch of substrates; an extraction unit that extracts the time-series data of past physical quantities corresponding to the batch of target substrates being processed, referring to the first storage unit; a maintenance requirement determination unit that compares the time-series data of the physical quantities at the time of the abnormality detected by the aforementioned sensor with the time-series data of the past physical quantities extracted by the aforementioned extraction unit to determine whether maintenance is required; a second storage unit that stores a combination of at least one physical quantity with and without abnormality, along with the cause of the abnormality and / or the solution to the abnormality; and a cause analysis unit that, when the maintenance requirement determination unit determines that maintenance is required, outputs the cause of the abnormality and / or the solution to the abnormality based on the combination of the physical quantity with and without abnormality, referring to the aforementioned second storage unit. When this configuration is adopted, maintenance personnel of the substrate processing equipment can immediately grasp the cause of the malfunction and / or the solution. Therefore, they can go to the local polishing equipment and quickly resolve the malfunction. In addition, the time and cost of detecting the cause of the malfunction and / or developing a solution can be reduced, achieving labor saving, energy saving, and / or cost reduction. The eighth embodiment of the substrate processing system of the present invention includes: an information processing device connected to a plurality of substrate processing devices via communication lines; and a fog computer or terminal connected to the aforementioned information processing device via communication lines; the aforementioned information processing device collects data from the aforementioned plurality of substrate processing devices, processes the collected data, and transmits the processing result to the aforementioned fog computer 2b or the aforementioned terminal; when the aforementioned fog computer or the aforementioned terminal receives the aforementioned processing result, it controls the system by outputting the processing result. When this configuration is adopted, the fog computer or terminal can output the results of the information processing device processing the data collected from multiple board processing devices. The substrate processing system of the ninth aspect of the present invention, such as the substrate processing system of the eighth aspect, wherein the aforementioned information processing device has the following means: extracting substrate processing conditions, substrate processing stage status, and / or parameters related to trimming uniformity reaching a reference or above from the aforementioned collected data; and comparing the aforementioned extracted parameters between substrate processing devices, and updating at least one parameter in the aforementioned data according to the comparison result. When this configuration is adopted, because it allows for close access to substrate processing conditions (e.g., polishing conditions), substrate processing stage states (e.g., polishing stage states), and / or finishing uniformity, variations in substrate processing (e.g., polishing) between substrate processing apparatuses (e.g., polishing apparatuses) can be reduced. (Effects of the Invention) When employing one embodiment of the present invention, the time and cost spent predicting the grinding endpoint timing can be reduced because the grinding endpoint timing can be automatically predicted. Furthermore, the scheme can be automatically updated when abnormalities occur during grinding, thus achieving labor saving, energy saving, and / or cost reduction. In addition, when using time-series data of the current value of the rotating stage motor in the past, multiple minimum or maximum points may occur, making it difficult to immediately identify which minimum or maximum point marks the grinding endpoint timing. This problem is difficult to detect due to both the waveform shape of the time-series data and the presence of noise in the waveform. AI, such as machine learning, can solve this problem through waveform analysis, noise removal, and trend analysis. Specifically, the learned machine learning model uses time-series data of past physical quantities, or time-series data of past physical quantities differentiated by time, as input, and past grinding endpoint times as output. Therefore, even when inputting time-series data of unknown physical quantities or time-series data of physical quantities differentiated by time, the probability of outputting the correct grinding endpoint times is still increased. In another embodiment of the invention, since anomalies in the time-series data of the physical quantities of the substrate processing device can be automatically detected, the time and cost spent detecting such anomalies can be reduced, achieving labor saving, energy saving, and / or cost reduction. In another embodiment of the invention, since maintenance personnel of the substrate processing device can immediately grasp the cause of the anomaly and / or the solution, they can go to the local grinding device to quickly resolve the anomaly. Furthermore, the time and cost of detecting the cause of the anomaly and / or formulating a solution can be reduced, achieving labor saving, energy saving, and / or cost reduction.

Implementation Method

Claims

1. A substrate processing system comprising: a sensor disposed on a substrate processing apparatus to detect a physical quantity of an object in substrate processing; and a prediction unit that inputs time-series data of the physical quantity detected by the aforementioned sensor, or time-series data obtained by differentiating the time-series data of the physical quantity over time, into a machine learning model to complete learning, and outputs a grinding endpoint timing sequence of the grinding end timing sequence; the aforementioned machine learning model is a model that uses learning data that takes time-series data of the aforementioned physical quantity in the past, or time-series data of the aforementioned physical quantity in the past, as input, and takes past grinding endpoint timing sequences as output for machine learning, wherein the aforementioned physical quantity of the object is the table rotation speed of the aforementioned substrate processing apparatus, the rotation speed of the upper annular turntable of the aforementioned substrate processing apparatus, the torque of the table of the aforementioned substrate processing apparatus, or the torque of the upper annular turntable, and the aforementioned substrate processing system further comprises: The selection unit selects the time series data of the physical quantity by excluding the time series data of the physical quantity before differentiation when no minimum or maximum point satisfying the set criteria is detected in the time series data of the physical quantity detected by the aforementioned sensor by differentiating the time series data of the physical quantity over time; and the learning unit performs machine learning using a learning dataset that takes the time series data of the physical quantity selected by the aforementioned selection unit as input and the grinding end time sequence as output, and generates the aforementioned machine learning model that has completed learning.

2. The substrate processing system of claim 1, further comprising: a determination unit that compares the time series data of the physical quantity detected by the aforementioned sensor with past time series data to determine whether there is an abnormality in the time series change of the physical quantity; a decision unit that, when the determination unit determines that there is an abnormality, re-determines the processing conditions; and an update control unit that controls the system to update according to the processing conditions determined by the decision unit.

Images