WO2025229526A1

WO2025229526A1 - Method and system for cleaning containers configured and adapted to store wafers or reticles

Info

Publication number: WO2025229526A1
Application number: PCT/IB2025/054451
Authority: WO
Inventors: Jagdish SARASWATULA
Original assignee: Brooks Automation Germany GmbH
Current assignee: Brooks Automation Germany GmbH
Priority date: 2024-04-30
Filing date: 2025-04-29
Publication date: 2025-11-06
Anticipated expiration: 2026-10-30

Abstract

Method and system for cleaning a container configured and adapted to hold wafers or reticles, comprising generating at least one digital twin simulation model configured and adapted to predict cleaning requirements of the container, and implementing at least one cleaning procedure based on the at least one digital twin simulation model.

Description

Method and system for cleaning containers configured and adapted to store wafers or reticles

The present invention relates to the field of cleaning of semiconductor containers configured and adapted to store wafers or reticles, referred to in the following also as semiconductor carrier devices. A typical example is a Front Openable Unified Pod (FOUP), for storing and transporting wafers in a semiconductor manufacturing environment, the so called fab environment. Another example is a reticle container, which carries reticles used in semiconductor lithography.

Introduction

Manufacturing a modem day computer chip requires a complex manufacturing process comprised of many hundreds of steps. Semiconductor manufacturers conduct manufacturing in large facilities referred to as “fabs”. The fabs house equipment from many different suppliers, that largely do not interact with one another. One of the main metrics that fabs use to measure the quality of production is yield, which is the amount of viable computer chips or product volume compared to that not meeting standards. The yield of a fab is a critical component to the profitability and viability of a fab’s commercial existence.

Semiconductor contamination control is an important aspect of semiconductor fab operations in order to maintain a mini environment. A carrier device mini environment plays a very critical and important role in the whole manufacturing flow and contamination and cleanliness of such can greatly affect the yield of the fab. For example, FOUPs are used to carry wafers that have been treated with concentrated chemicals and have been exposed to extreme environments, repeatedly and over time. Contamination of the FOUPs can dramatically affect the ultimate yield in a semiconductor manufacturing process. It is therefore extremely crucial to control and monitor and clean the FOUPs in order to maintain yield and reduce wafer detectivity. At the same time, it is valuable to limit the amount of cleaning a FOUP is exposed to and avoid unnecessary or ineffective FOUP cleaning. Those are two contravening goals, and very difficult to manage. As a result, fabs often resort to cleaning recipes and processes that are set to the “lowest common denominator”, meaning that they use recipes and methodologies that will clean the most contaminated semiconductor container, and apply those even to containers that do not require such an intense cleaning to reach adequate cleanliness. As mentioned, FOUPs are exposed to many different chemicals and impurities over the course of time and usage, and in many ways, each FOUR has a unique contamination profile. Similarly, reticle carriers suffer similar experiences and face similar challenges. All comments herein are applicable to reticle carriers too.

Purpose of or problem solved by Invention

Currently, semiconductor container (such as reticle carriers and/or FOUPs) cleaning is a standardized process that is not applied and adjusted individually to each and every semiconductor carrier that is cleaned, even though each semiconductor carrier has its own contamination profile and physical characteristics, e.g. a wear condition. An example of a cleaning system is a commercial FOUP cleaning system is the PuroMaxx line of cleaners offered commercially by Brooks Automation. While adjustments to the cleaning process and methodology are occasionally made, it is not consistently done on an individual semiconductor carrier basis, and it is not done in real-time. One reason is that, to the extent the process or methodology is set or adjusted, it is done by a human operator and cannot be done in truly real-time for each individual semiconductor container. One of the reasons that cleaning recipes and cleaning methodologies are not customized for each semiconductor container is the lack of a cleaning system that is capable of referencing, in real-time, historical exposure and contamination data of the semiconductor container (e.g., what chemicals the particular container was exposed to and when, for example, in the prior 3 months or 12 months), model the cleaning process on the particular semiconductor container given the particular contamination profile, while correlating that to historical and predicted cleanliness results, and further, optionally correlating that to historical ultimate fab yield on a completed product in the semiconductor fabrication facility, in which those semiconductor containers are handled. The present invention provides a comprehensive system and methodology that accounts for a semiconductor container’s contamination profile, its physical condition and configuration, and uses a digital twin of the cleaning system and semiconductor container to generate and model an optimized recipe and cleaning methodology that is then implemented for that semiconductor container. The contamination profile may include contemporary and past chemical exposure, physical attributes, historical cleaning recipes and methodologies and resultant cleanliness and ultimately yield of the fabrication facility.

The present invention suggests a method for cleaning a container configured and adapted to hold wafers or reticles comprising the features of claim 1 and a cleaning system for cleaning a container configured and adapted to hold wafers or reticles comprising the features of claim 9. Advantageous embodiments are the subject matter of the dependent claims and the further description.

The present invention uses a digital twin of the cleaning system to conduct a simulation model that constructs the container and the cleaning system, accounts for the contamination profile of the semiconductor container, and generates, models and implements an optimized cleaning recipe and methodology for that semiconductor container. The digital twin can be housed and utilized in a computing system contained with and connected to the semiconductor carrier cleaning system or can be remotely connected thereto.

A digital twin is a virtual representation of a physical object, system, or process. It uses real-time data, historical data, and simulations to mirror the physical counterpart's behaviour and performance. The digital twin allows for monitoring, analysis, and optimization by providing insights into how a real life application will perform.

The generation and modelling of the optimized cleaning recipe and methodology by using the digital twin may be done in real-time, so that when a semiconductor container (e.g., a FOUR) is entering the cleaning system, the contamination profile and various historical data are utilized in the digital twin model to generate an optimized cleaning method and recipe. Expediently, the application of the digital twin simulation model as well as the subsequent implementation of an optimised cleaning recipe and methodology is conducted as an automated process which may be in real-time, or near real-time. The simulation model could, for example, be fed with information or data on simulation models of various containers, especially FOUPs, from different vendors and/or manufacturers. This information or data could also be obtained for example by measurements of the containers, especially their dimensions, their surface characteristics and of the materials they are made of. The simulation model could derive information or data from CAD designs and other structural information (about the semiconductor container, for example FOUR or reticle holder). The information could also include material information relating to the material from which the container is made. The information or data can also comprise data obtained by sensors measuring the container (such as cameras, lidar or other 3D sensors), contamination profile data of the semiconductor container and/or its surroundings within the fab environment.

The digital twin simulation model as utilised according to the invention can especially be adapted to generate, model, test and optimize possible cleaning recipes based on data received relating to a container to be cleaned, and the cleaning system, in real time. The simulation model can also be used to test and optimize through an iterative process to generate a recommended cleaning recipe and methodology. The simulation model, which can especially be provided in the form of a so called digital twin, as mentioned, can be connected with or be provided as part of the cleaning system of the invention, to develop in real time an optimised cleaning method to be applied to such a container.

For example, in an advantageous embodiment of the invention, a container such as a FOIIP is introduced into a cleaning chamber of the cleaning system. Data for this container, the contamination profile, data obtained by sensors provided for example within the cleaning system, especially the cleaning chamber and/or data otherwise available, such as FOIIP identification data, are input into the digital twin simulation model. Based on these input data, the simulation model predicts and generates an optimised cleaning recipe and methodology for the particular container, which is then modelled by way of the digital twin and applied. Using such a simulation model or digital twin, the process for generating, modelling and implementing the optimised cleaning recipe is automatic and does not require human generation of the optimized cleaning recipe and methodology. In other words, the cleaning system with the digital twin performs a different methodology than what would be performed by a human user. The digital twin performs generation, modelling and optimization in realtime, instead of what would have been mental guessing, that an experienced engineer would otherwise need to perform in order to arrive at a hopefully suitable cleaning process. Further, the digital twin performs the modelling and optimization that generates the cleaning recipe and methodology in real-time and for the individual semiconductor container.

An advantage of the digital twin model-based approach is an ability to calculate and/or minimise the required quantity of consumables (such as purging gases and/or cleaning agents) more accurately and on a container by container basis, thereby helping to arrive at a better estimation and a reduced consumption of cleaning agents. For example, one FOUR may be determined to require heavy use of consumables to obtain adequate cleanliness, based on its contamination profile, whereas another FOUR may require less consumables based on its different contamination profile. Even a small amount of savings per cleaning process would result in substantial sustainability gains, as the containers, especially the FOUPs, are cleaned perennially during the manufacturing process.

Such models can also provide information about benefits and drawbacks of specific types of containers, such as FOUP(s), for example from selected suppliers, that the model shows to be more economical and hence ‘greener’ and sustainable.

The following information or data is/are especially valuable in connection with container or FOUP model development according to the teaching of the invention, although information or data processable by such a simulation model is not limited thereto: One of the key aspects in connection with the present invention is the digital twin’s provision and development of air flow models, indicating for example how a purge gas used to clean a semiconductor container, especially a FOUR, will flow through or around the container. For example, FOUPs for which a simulation model can show that there is intensive interaction between purge gas and surfaces result in the digital twin developing optimized cleaning recipes and methodologies that use a lower amount of purge gas in order to achieve the desired cleaning effect. Such flow information can be simulated/modelled based for example on CAD models and structural information of the FOIIP. Such flow information is used by the digital twin generate and model and implement an optimized cleaning recipe and methodology for the FOIIP cleaning.

Advantageously, the simulations conducted by the digital twin models as described may be compared to real cleaning statistics or data in order to calibrate and improve digital twin models and to ascertain the most efficient way of cleaning in terms of time and consumables.

It is envisaged that prior models thus generated, and the resulting cleanliness data and semiconductor fabrication yield, are used to continuously update digital twin models to produce the optimal cleaning recipe, as well as to predict time and efficiency of the cleaning process or process steps.

Air/gas based cleaning, also known as purging, is one of the most common modes of cleaning containers as discussed above. The mechanism or flow mechanics of gas transport within the FOIIP provides details regarding areas where cleaning is effected optimally and areas where it is marginal. A thorough understanding of this process enables an understanding of the impact of cleaning on various other parameters like humidity, airborne molecular contamination AMC/ volatile organic compounds VOC, temperature distribution etc. Digital twin modelling enables estimation of the time and quantity of the cleaning agents needed for the process.

In connection with surface modelling, the material of the container or FOIIP plays a very important role, especially in connection with its interaction with the cleaning agents and AMC/VOC. The material also plays a very important role in the absorption of outgassing from wafers and later secondary outgassing that for example depends on the material the FOUR walls are made of. Different vendors or suppliers use different types of materials for containers/FOUP(s) and hence their behaviour, such as their outgassing characteristics, differs.

The present invention constitutes a significant departure from existing technologies:

Existing cleaning methods involve cleaning of semiconductor containers/FOUP(s) with fixed cleaning recipes and methodologies, including e.g., fixed time schedules, and fixed recipes and methodologies, by using cleaning agents at a predefined rate, e.g. an N2 purge at the rate of 6 litres/m in. A main disadvantage of such a rule-based method of cleaning is the inability to generate optimized cleaning recipes and methodologies . For example, most cleaning recipes and methodologies are biased to “over clean” to ensure that even the least effective applications of the recipes and methodologies are adequate. With use of a digital twin model, a particular semiconductor container cleaning can be modelled to generate a cleaning recipe and method that according to the model is adequate, while utilizing less consumables and/or energy. It is advantageous to have a digital twin semiconductor cleaning system that is flexible and tuneable for individual semiconductor containers, e.g., taking into account the structural model of each particular FOIIP (such as an air flow model) and/or the material properties and geometry of the FOIIP surface, as well as the historical contamination profile.

Based on such data, both contemporary and historical, overall fabrication yield, and also on other data, if available, a digital twin simulation model of the cleaning system and/or the semiconductor container of a cleaning system, that mirrors the real-world cleaning system, especially in real time, is provided. It can use data from sensors, loT connectivity, and machine learning to simulate the behaviour and performance of the physical cleaning system. This allows for monitoring, predictive maintenance, testing, and optimization without directly interacting with the cleaning system. For example a digital twin of a cleaning system can predict performance and maintenance requirements, improving efficiency and reducing downtime.

Obtaining an effective digital twin can advantageously involve firstly determining its purpose, i.e. is it to assist in one or more of predictive maintenance, real-time monitoring, operational optimisation or fault detection. Then, key performance indicators such as usage of cleaning agent, wear and tear and energy efficiency can be determined.

Using for example CAD software, a 3D modelling of the cleaning system, especially of a cleaning chamber and a FOUR can be provided. The cleaning chamber and the FOUR are advantageously provided with sensors, for example loT sensors to enable data collection relating to relevant data, for example airflow, pressure, humidity or temperature over time within the FOUP, physical dimensions, temperature, AMC levels and other contaminant levels, and levels of such over time. Such sensors can also be provided in other sections or components of the cleaning system, i.e. outside of the cleaning chamber.

Following data collection, the dynamics, for example mechanical or fluid dynamics, as well as the cleaning rate within the FOUP (e.g. the reduction of contamination levels over time) can be replicated using the digital twin simulation model. In this connection, Al models adapted to detect anomalies and predict maintenance needs could be trained in order to enhance predictive maintenance. Also, historical data relating to different kinds of FOUPs or cleanings systems could be integrated or made use of.

It is advantageous to provide a continuous data flow between cleaning system and the digital twin. In this connection, it is expedient to set up alerts/warnings in case of system failures, leakages or efficiency changes.

In order to provide an effective digital twin based hereon, simulation advantageously runs under various conditions, for example varying the type of FOUP, the type of cleaning agent, temperature, pressure, humidity, should be performed. Expediently, simulation or digital twin predictions should be compared with real-world results, all in all leading to more robust models. Also, Al based feedback loops should be provided to further optimize wash cycles, cleaning agent usage and maintenance timings.

All these concepts and/or information/data combined or individually can enable an effective digital twin simulation in order to be able to choose or determine the optimal cleaning process and for example help estimate or determine the optimal time needed for cleaning vs the quantity of cleaning agents required. Especially, the invention suggests generating a digital twin of a container to be cleaned based on a structural model, such as an airflow model, and the material, of which the container is made and/or a material based model of the container. Such a digital twin allows optimisation of cleaning procedures, especially in relation to cleaning time and amount of consumables such as cleaning agents, as outlined above.

It is especially advantageous in connection with the invention to use artificial intelligence Al and/or machine learning ML based algorithms to generate the simulation mode, especially the digital twin, and/or to develop cleaning recipes based on historical recipes and data.

An embodiment of the invention will now be further described with reference to the appended Figures. Herein

Figure 1 shows a schematic diagram of a cleaning system according to a preferred embodiment of the invention,

Figure 2 shows a schematic diagram of a cleaning unit as part of a cleaning system according to a preferred embodiment of the invention, and

Figure 3 shows a schematic flow diagram of a method for generating a digital twin for use in cleaning a container configured and adapted to hold wafers or reticles, such as a FOUR. A preferred embodiment of a cleaning system 100 according to the invention is shown in Figures 1 and 2 in a purely schematic manner. The following description refers to both Figures simultaneously. The cleaning system comprises a housing 101 including a cleaning unit 120 comprising a cleaning chamber 121 , a contamination detection and analysing unit 140 and a vacuum unit 160 comprising a vacuum chamber 161 .

The cleaning system 100 comprises at least one load port 102, configured to introduce a container, such as a FOUR 112, which is to be cleaned, into the cleaning system, an Equipment Front End Module EFEM 104 adapted to cooperate with the at least one load port 102, a gate opening section 106 connected to the EFEM 104, and a gate closing section 110 connected to the EFEM 104. The system can comprise a buffer station 108 between the gate opening section 106 and the cleaning chamber 121 . The buffer station 108 is adapted to transport the FOUR 112 into the cleaning chamber 121 . The buffer station may be, in part, arranged between the analysing unit 140 and the gate closing section 110. The buffer station may be adapted to transport the FOIIP to the gate closing section 110, e.g. if the measurement result of a certain contaminant is not higher than a predetermined threshold value. A typical movement of a FOIIP within such a cleaning system is indicated by arrows in Figure 1 .

As shown in Figure 2, a preferred embodiment of the cleaning unit 120 comprises a cleaning chamber 121 and a nozzle 118, by means of which a cleaning fluid 127, such as deionised water or a suitable solvent, for cleaning a FOIIP 112 positioned within the cleaning chamber 121 can be introduced into and/or around the FOIIP, in order to interact with the inside and/or outside walls of the FOIIP. The cleaning unit further comprises a nozzle 122 for introducing a gas, especially a purge gas, 129 for drying and/or purging the FOIIP by means of a suitable gas, for example CDA, nitrogen or an inert gas. Both nozzles can be adapted to be moveable relative to the FOIIP 112. Providing the nozzles to be moveable is especially advantageous in the context of the present invention, as they can be effectively brought into the vicinity of areas of the FOIIP which a simulation model, as will be explained in the following, may determine to require enhanced cleaning treatment as compared to other areas of the FOIIP. The cleaning chamber 121 comprises at least one inlet and at least one outlet for controlled introduction and removal of cleaning fluid and purge gas (not explicitly shown).

The cleaning unit 120 is also provided with a heater 124 adapted to heat the FOUR 112 within cleaning chamber 121.lt can be adapted to heat the container to 50 to 70 C, especially to around 60 C. The FOUR may be positioned on a holder 114, which may be rotatable be means of a driving unit 116.

The cleaning unit 120 may be provided with a particle removal unit, a chemical treatment unit, a drying unit, a vacuum unit and the contamination detection unit, although at least some of these units may be provided separately from the cleaning unit.

The analyzing unit 140 is adapted to analyse air or gas within the FOIIP while it is in the cleaning chamber 121 or after it exits the cleaning chamber. The analyzing unit 140 can, for example, be adapted to measure AMC and/or humidity.

The vacuum unit 160 comprising vacuum chamber 161 is adapted to expose the FOIIP to a vacuum in order to enhance outgassing, for example if a contamination measurement result exceeds a certain threshold even after cleaning within the cleaning chamber 121 . However, the FOIIP can also be exposed to such a vacuum in a vacuum chamber independently of such a contamination measurement result in order to enhance the cleanliness state of the FOIIP.

According to a preferred embodiment of the invention, a plurality of sensors 150 can be positioned throughout the system 100, especially in at least one of the load port 102, the EFEM 104, the gate opening section 106, the buffer station 108, the cleaning unit 120, the analyser 140, the vacuum chamber 161 , or the gate closing section 110. Only some of these sensors are schematically shown in Figure 1 and Figure 2. These sensors are adapted to provide various data for a simulation unit 170, which can be provided as a part of a control unit 180. The simulation unit is adapted to provide a digital twin of the cleaning system and the FOIIP 112 to be cleaned. The data utilized by the simulation unit 170 in this connection can, for example, comprise identification data of the FOUR including design data, visual image date, lidar data, , any other contamination profile date, temperature, material data, surface data, airflow data, provided for example by a sensor reading an identification tag provided on the FOUR, as well as cleanliness data (such as AMC contamination) provided by various sensors positioned throughout the system.

A preferred embodiment of a simulation model or digital twin for use in the simulation unit 170 of a cleaning system according to the invention can be generated as explained in the following, with reference to Figure 3.

In a first step 300, the objectives of the simulation model, especially of the digital twin are defined. For example, it may be instructed to optimize energy efficiency, or consumable use, or cleanliness, or cleaning time, depending on the fab’s priorities.

In a second step 320, data relating to the cleaning system and/or a type of container, especially a type of FOIIP, to be cleaned, are collected. These data can especially comprise data obtained in connection with a specific first container of this type, especially a FOIIP, which is to be cleaned. For example, sensors within the FOIIP and/or the cleaning system track at least one of usage of cleaning agent or purge gas, airflow patterns within the FOIIP, pressure within the FOIIP, temperature within the FOIIP, cleanliness levels of specific contaminants within the FOIIP, for example AMC levels. Also, geometrical data of the FOIIP, such as volume, structures on the inside walls, as well as data relating to the material the FOIIP is made of can be collected.

In a third step 340, a digital simulation model is generated based on the data thus collected. For example, if a certain cleanliness level is measured after a cleaning of the FOIIP with purge gas in the cleaning chamber over a certain length of time, over which certain airflow patterns, pressures and temperatures have been measured in the FOIIP comprising certain geometrical and material features, a simulation model may be provided to predict that any FOIIP exhibiting these or at least similar features will achieve similar cleanliness levels after a corresponding cleaning activity or purge. Such a simulation model may be further enhanced by obtaining similar data for further containers, especially FOUPs, which are either identical to the container, especially FOUR, for which the simulation model has been generated, or differ from it, for example in their geometry and/or materials. Advantageously, a calibration of the simulation model based on a comparison of generated simulation data with real measurement data and/or cleaning statistics data or historical cleaning data can be performed. By thus expanding a simulation model by using such or other kinds of data, especially historical data, predictions as to the cleaning behaviour of various kinds of containers, especially FOUPs, can be robustly generated.

In a fourth step 360, simulation tools are used to replicate the dynamics within the cleaning system, especially the mechanical or fluid dynamics within the FOUP. This step may comprise training Al models adapted to detect anomalies and predict maintenance requirements. This step may be included in the third step 340.

A further step 380 may include real time data synchronisation to ensure a continuous or at least regular data flow between cleaning system and simulation model, especially digital twin, as well as running of simulations under various conditions to test the performance of resilience of the digital twin, and comparing predictions made by the digital twin with real-world results.

Based on or at least taking into account a simulation model thus generated by simulation unit 170, and which can be continuously updated essentially in real time, control system 180 may implement an optimised cleaning procedure for the FOUP 112 to be cleaned as well as for future containers, especially FOUPs, which require cleaning.

Claims

1 . Method for cleaning a container configured and adapted to hold wafers or reticles, comprising generating at least one digital twin simulation model configured and adapted to predict cleaning requirements and/or cleanliness states of the container, and implementing at least one cleaning procedure based on the at least one digital twin simulation model.

2. Method according to claim 1 , wherein the at least one digital twin simulation model comprises an airflow model, configured and adapted to predict a flow of gas, especially a purge gas, within the container.

3. Method according to any one of the preceding claims, which especially takes into account CAD models and structural information of the container..

4. Method according to any one of the preceding claims, wherein the at least one digital twin simulation model comprises a surface model of the container, which takes into account the material of which the container is made and its interaction with cleaning agents and/or AMC/VOC and or humidity, especially taking into account outgassing properties of a material the container is made of.

5. Method according to any one of the preceding claims, comprising calibration of the at least one digital twin simulation model based on a comparison of generated simulation data with real measurement data and/or cleaning statistics data.

6. Method according to any one of the preceding claims, comprising generating a digital twin of a container configured and adapted to hold wafers or reticles to be cleaned and/or of a cleaning system configured and adapted to perform cleaning of such a container.

7. Method according to any one of the preceding claims, comprising use of artificial intelligence Al and/or machine learning based ML based algorithms to generate the at least one digital twin simulation model.

8. Method according to any one of the preceding claims, comprising developing cleaning procedures based on historical cleaning procedures and data.

9. A cleaning system for cleaning a container configured and adapted to hold wafers or reticles, especially a Front Opening Unified Pod (FOUP), comprising: a cleaning unit (120) comprising a cleaning chamber (121 ) configured and adapted to receive a container (112) to be cleaned, a contamination detection and analysing unit (140) including one or more sensors configured and adapted to obtain contamination data of a container, especially data relating to particle levels, chemical residues, or other contaminants, a simulation unit (170) configured and adapted to generate at least one digital twin simulation model for predicting cleaning requirements and/or cleanliness states of the container (121) based on at least one of the contamination data obtained by the contamination detection and analysing unit (140) or container data obtained by at least one measurement sensor provided within the cleaning system or a container identification unit provided in the cleaning system or data provided by the cleaning unit (120), and a control unit (180) configured and adapted to control the operation of the cleaning system, especially to control at least one of the cleaning unit (120), especially a particle removal unit, a chemical treatment unit, a drying unit, a vacuum unit, and the contamination detection and analysing unit (140) based on the cleaning requirements and/or cleanliness states of the container (112) predicted by the digital twin simulation model.

10. System according to claim 9, wherein the simulation unit (170) is adapted to generate the at least one digital twin simulation model as a digital twin in realtime.

11 . System according to claim 9 or 10, wherein the simulation unit (170) is adapted to learn, especially to adapt the at least one digital twin simulation model based on simulation data and/or cleaning or cleanliness data taken from previous simulations and/or simulation data and/or cleaning or cleanliness data from other cleaning systems.