Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185339
Masakazu Hamasaki, Y. Hagio, K. Kasa, Yoshimitsu Kato, Manabu Takakuwa, Tsutomu Obata, Shunichi Nakao, Manabu Miyake, Katsuya Kato, Yosuke Takahata, A. Nakae
Intra-die overlay is becoming one of the key challenges in high accuracy overlay. While recent progress of overlay metrology has made it viable to monitor intra-die overlay signature by nondestructive methods, traditional intra-field correction does not work well enough to reduce intra-die overlay error of lithography process. In this paper, we demonstrated for the first time that the intra-die overlay correction does work by synchronizing scan speed to intra-die fingerprint and this method is actually applicable to treat both lot-to-lot and intra-wafer variation of intra-die overlay.
{"title":"Novel overlay correction by synchronizing scan speed to intra-die fingerprint on lithography scanner","authors":"Masakazu Hamasaki, Y. Hagio, K. Kasa, Yoshimitsu Kato, Manabu Takakuwa, Tsutomu Obata, Shunichi Nakao, Manabu Miyake, Katsuya Kato, Yosuke Takahata, A. Nakae","doi":"10.1109/ASMC49169.2020.9185339","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185339","url":null,"abstract":"Intra-die overlay is becoming one of the key challenges in high accuracy overlay. While recent progress of overlay metrology has made it viable to monitor intra-die overlay signature by nondestructive methods, traditional intra-field correction does not work well enough to reduce intra-die overlay error of lithography process. In this paper, we demonstrated for the first time that the intra-die overlay correction does work by synchronizing scan speed to intra-die fingerprint and this method is actually applicable to treat both lot-to-lot and intra-wafer variation of intra-die overlay.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"8 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77214630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185317
Viboth Houy, J. Lam, H. Ali
In semiconductor fabrication, diffusion process plays a critical role, ranging from Oxidation, Low Pressure Chemical Vapor Deposition (LPCVD), Thermal Processing, Plasma Processing, Atomic Layer Deposition (ALD) and Epitaxial Si. Diffusion Low-k application, one of the six diffusion process categories, is an Atomic Layer Deposition process (ALD) to create a spacer. The spacer provides various applications in the transistor fabrication process. Its low k value reduces capacitance between the gate and contact. However, the process is notorious for particle defects. This paper is intended to explore ways to improve particle performance which, in turn, optimizes its above mentioned functions. It covers a design of experiment (DOE) to manipulate gas flows in order to achieve its desired results. The paper, however, does not seek to introduce new hardware to the current furnace configurations.
{"title":"Particle Improvement for Low-K Process in Diffusion Furnace","authors":"Viboth Houy, J. Lam, H. Ali","doi":"10.1109/ASMC49169.2020.9185317","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185317","url":null,"abstract":"In semiconductor fabrication, diffusion process plays a critical role, ranging from Oxidation, Low Pressure Chemical Vapor Deposition (LPCVD), Thermal Processing, Plasma Processing, Atomic Layer Deposition (ALD) and Epitaxial Si. Diffusion Low-k application, one of the six diffusion process categories, is an Atomic Layer Deposition process (ALD) to create a spacer. The spacer provides various applications in the transistor fabrication process. Its low k value reduces capacitance between the gate and contact. However, the process is notorious for particle defects. This paper is intended to explore ways to improve particle performance which, in turn, optimizes its above mentioned functions. It covers a design of experiment (DOE) to manipulate gas flows in order to achieve its desired results. The paper, however, does not seek to introduce new hardware to the current furnace configurations.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"1 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77945205","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185357
Illhoe Hwang, H. Cho, S. Hong, Junhui Lee, SeokJoong Kim, Y. Jang
We present a reinforcement learning-based algorithm for route guidance and vehicle assignment of an overhead hoist transport system, a typical form of automated material handling system in semiconductor fabrication facilities (fabs). As the size of the fab increases, so does the number of vehicles required to operate in the fab. The algorithm most commonly used in industry, a mathematical optimization-based algorithm that constantly seeks the shortest routes, has been proven ineffective in dealing with fabs operating around 1,000 vehicles or more. In this paper, we introduce Q-learning, a reinforcement learning-based algorithm for route guidance and vehicle assignment. Q-learning dynamically reroutes the vehicles based on the congestion and traffic conditions. It also assigns vehicles to tasks based on the overall congestion of the track. We show that the proposed algorithm is considerably more effective than the existing algorithm in an actual fab-scale experiment. Moreover, we illustrate that the Q-learning-based algorithm is more effective in designing the track layouts.
{"title":"Q-learning-based route-guidance and vehicle assignment for OHT systems in semiconductor fabs","authors":"Illhoe Hwang, H. Cho, S. Hong, Junhui Lee, SeokJoong Kim, Y. Jang","doi":"10.1109/ASMC49169.2020.9185357","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185357","url":null,"abstract":"We present a reinforcement learning-based algorithm for route guidance and vehicle assignment of an overhead hoist transport system, a typical form of automated material handling system in semiconductor fabrication facilities (fabs). As the size of the fab increases, so does the number of vehicles required to operate in the fab. The algorithm most commonly used in industry, a mathematical optimization-based algorithm that constantly seeks the shortest routes, has been proven ineffective in dealing with fabs operating around 1,000 vehicles or more. In this paper, we introduce Q-learning, a reinforcement learning-based algorithm for route guidance and vehicle assignment. Q-learning dynamically reroutes the vehicles based on the congestion and traffic conditions. It also assigns vehicles to tasks based on the overall congestion of the track. We show that the proposed algorithm is considerably more effective than the existing algorithm in an actual fab-scale experiment. Moreover, we illustrate that the Q-learning-based algorithm is more effective in designing the track layouts.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"31 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76470434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185239
C. Edwards, M L N Swamy, Ravi Garg, Tim Karaniuk, C. Morgan, Debashis Panda
We demonstrate a new real-time inspection system developed to monitor tool health and detect defects during the epoxy dispense process. The system includes both hardware and software components. The hardware was designed to be low-cost and fit into a small footprint within the existing tools. Our software contains a tool setup/calibration utility and a user interface for recipe creation and real-time inspection. The software also provides extensive logging of key results including tabulated data, annotated images, and live inspection results on the user interface. The algorithm uses a mixture of advanced machine learning and computer vision algorithms to identify unwanted process variation. The new system has provided excellent results, an order of magnitude below the qualification targets, while ensuring the throughput time targets are not impacted.
{"title":"Real-Time Tool Health Monitoring and Defect Inspection during Epoxy Dispense Process","authors":"C. Edwards, M L N Swamy, Ravi Garg, Tim Karaniuk, C. Morgan, Debashis Panda","doi":"10.1109/ASMC49169.2020.9185239","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185239","url":null,"abstract":"We demonstrate a new real-time inspection system developed to monitor tool health and detect defects during the epoxy dispense process. The system includes both hardware and software components. The hardware was designed to be low-cost and fit into a small footprint within the existing tools. Our software contains a tool setup/calibration utility and a user interface for recipe creation and real-time inspection. The software also provides extensive logging of key results including tabulated data, annotated images, and live inspection results on the user interface. The algorithm uses a mixture of advanced machine learning and computer vision algorithms to identify unwanted process variation. The new system has provided excellent results, an order of magnitude below the qualification targets, while ensuring the throughput time targets are not impacted.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"120 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87830260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185389
Gang Liu, Rommel Relos, Bohumil Janik, Robert Davis, T. Myers, D. Allman, Jeff Hall, S. Vandeweghe, S. Menon, Ed Flanigan
Products in automotive applications demand polysilicon fuse One-time programmable (OTP) solutions with extremely low failure rates. Fundamental understanding of the programming mechanism and key design/programming factors are indispensable to achieving such a goal. This paper presents a real-time poly fuse voiding model supported by electrical waveforms, simulations and physical analysis data. Impacts of fuse design and programming condition changes are also examined.
{"title":"Polysilicon Fuse Electrical Voiding Mechanism AP/DFM: Advanced Patterning / Design for Manufacturability","authors":"Gang Liu, Rommel Relos, Bohumil Janik, Robert Davis, T. Myers, D. Allman, Jeff Hall, S. Vandeweghe, S. Menon, Ed Flanigan","doi":"10.1109/ASMC49169.2020.9185389","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185389","url":null,"abstract":"Products in automotive applications demand polysilicon fuse One-time programmable (OTP) solutions with extremely low failure rates. Fundamental understanding of the programming mechanism and key design/programming factors are indispensable to achieving such a goal. This paper presents a real-time poly fuse voiding model supported by electrical waveforms, simulations and physical analysis data. Impacts of fuse design and programming condition changes are also examined.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"15 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82554379","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185334
Xin Li, Scott Veirs, Tony Betts, John Dalziel, Ania Zemlerub, Yuee Feng, D. Saigal
The sub-fab, a sophisticated environment where vacuum and abatement systems are located, has evolved dramatically over the years. It became essential in supporting semiconductor chip manufacturing. Closely tied into the process tool safety system, the abatement and vacuum components have a direct influence on the fabrication uptime and yields. The already strong and further growing influence have led chip manufacturers to realize the significant value of adopting advanced monitoring and data analytics to optimize sub-fab operations. The value is proven by adoption of such systems for the main fabrication area over the last decade. This article presents an example of successful application of integrated sub-fab monitoring system at an R&D facility for both dry pumps and abatement systems. This implementation example successfully demonstrates excellent data visibility to all level users, quick data collection enabling significant reduction in troubleshooting time, initial reduction in unscheduled abatement down events, and ability to quickly obtain comprehensive historical data for abatement state comparison. The success of the monitoring implementation has led to planning of applying predictive health monitoring function to further increase of sub-fab equipment uptime.
{"title":"Integrated Sub-fab Monitoring System Improving DataVisibility and Abatement Uptime : Category: APC, EO, SM, DM","authors":"Xin Li, Scott Veirs, Tony Betts, John Dalziel, Ania Zemlerub, Yuee Feng, D. Saigal","doi":"10.1109/ASMC49169.2020.9185334","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185334","url":null,"abstract":"The sub-fab, a sophisticated environment where vacuum and abatement systems are located, has evolved dramatically over the years. It became essential in supporting semiconductor chip manufacturing. Closely tied into the process tool safety system, the abatement and vacuum components have a direct influence on the fabrication uptime and yields. The already strong and further growing influence have led chip manufacturers to realize the significant value of adopting advanced monitoring and data analytics to optimize sub-fab operations. The value is proven by adoption of such systems for the main fabrication area over the last decade. This article presents an example of successful application of integrated sub-fab monitoring system at an R&D facility for both dry pumps and abatement systems. This implementation example successfully demonstrates excellent data visibility to all level users, quick data collection enabling significant reduction in troubleshooting time, initial reduction in unscheduled abatement down events, and ability to quickly obtain comprehensive historical data for abatement state comparison. The success of the monitoring implementation has led to planning of applying predictive health monitoring function to further increase of sub-fab equipment uptime.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"9 1","pages":"1-3"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84084540","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185298
T. Ning, CH Huang, J. Jensen, V. Wong, H. Chan
Time delay between chamber measurements and optical emission spectroscopy (OES) data were estimated using the cross-spectral analysis in this paper. The time delay between control and key variables provides useful feedback in etching control processes. We found in our study that ramping the chamber pressure during the etch process leads to an increasing time delay at the first harmonic between the chamber pressure and bias voltage measurements and a decreasing time delay between the chamber pressure and a selected OES wave band.
{"title":"Estimation of Process Time Delay between Chamber Measurements and Optical Emission Spectroscopy : APC: Advanced Process Control","authors":"T. Ning, CH Huang, J. Jensen, V. Wong, H. Chan","doi":"10.1109/ASMC49169.2020.9185298","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185298","url":null,"abstract":"Time delay between chamber measurements and optical emission spectroscopy (OES) data were estimated using the cross-spectral analysis in this paper. The time delay between control and key variables provides useful feedback in etching control processes. We found in our study that ramping the chamber pressure during the etch process leads to an increasing time delay at the first harmonic between the chamber pressure and bias voltage measurements and a decreasing time delay between the chamber pressure and a selected OES wave band.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"90 1","pages":"1-5"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84348632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185299
M. Shoudy, H. Shobha, Huai Huang, S. Nguyen, Chao-Kun Hu
This paper will provide an in-depth study of the run path impact an in-situ thermal Hydrogen preclean performed on a Copper interconnect has on the selective Cobalt cap deposition process. A loss in selective Cobalt cap thickness was observed with a wafer order dependence which resulted in a degradation in Electromigration performance. With Electromigration lifetimes dropping due to interconnect scaling, it is important to maintain and improve the Electromigration reliability of devices as we move to smaller nodes. By altering the wafer run path through a multi-chamber process tool, we were able to recover the loss of Cobalt selective deposition and thickness.
{"title":"In-situ Preclean Run Path Impact on Selective Cobalt Cap Deposition and Electromigration","authors":"M. Shoudy, H. Shobha, Huai Huang, S. Nguyen, Chao-Kun Hu","doi":"10.1109/ASMC49169.2020.9185299","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185299","url":null,"abstract":"This paper will provide an in-depth study of the run path impact an in-situ thermal Hydrogen preclean performed on a Copper interconnect has on the selective Cobalt cap deposition process. A loss in selective Cobalt cap thickness was observed with a wafer order dependence which resulted in a degradation in Electromigration performance. With Electromigration lifetimes dropping due to interconnect scaling, it is important to maintain and improve the Electromigration reliability of devices as we move to smaller nodes. By altering the wafer run path through a multi-chamber process tool, we were able to recover the loss of Cobalt selective deposition and thickness.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"104 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87747725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185375
K. McLean, Calvin Ma, S. Roy, Fen Guan, H. Ding, Bart Green
Optical communication has recently seen a resurgence driven by the demand for high-speed networking. Silicon Photonics (SiPh) has gained recent interest as a low cost and high volume method for creating photonic integrated circuits (PIC). PICs create new challenges for manufacturability since the devices require optical probing in addition to RF probing. This paper discusses a low cost method for aligning an optical fiber to a vertical grating coupler for wafer optical probing using a pre-defined device layout. This method is suitable for high volume wafer manufacturing. The alignment takes 0.5-1.5s and is reliable across multiple products and designs.
{"title":"Practical considerations for high throughput wafer level tests of silicon-photonics integrated devices","authors":"K. McLean, Calvin Ma, S. Roy, Fen Guan, H. Ding, Bart Green","doi":"10.1109/ASMC49169.2020.9185375","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185375","url":null,"abstract":"Optical communication has recently seen a resurgence driven by the demand for high-speed networking. Silicon Photonics (SiPh) has gained recent interest as a low cost and high volume method for creating photonic integrated circuits (PIC). PICs create new challenges for manufacturability since the devices require optical probing in addition to RF probing. This paper discusses a low cost method for aligning an optical fiber to a vertical grating coupler for wafer optical probing using a pre-defined device layout. This method is suitable for high volume wafer manufacturing. The alignment takes 0.5-1.5s and is reliable across multiple products and designs.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"8 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88895908","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2020-08-01DOI: 10.1109/ASMC49169.2020.9185292
Janghwan Lee, Wei Xiong, Wonhyouk Jang
In this paper, we propose the “trace data analytics” for classifying fault conditions from multivariate time series sensor signals using well-known deep CNN models. In our approach, multiple sensor signals are converted into two dimensional representations using the proposed conversion methods to optimize the classification performance. Many studies on the prediction of manufacturing results using sensor signals have been conducted in the field of fault detection and classification for display and semiconductor manufacturing processes. It is challenging to apply machine learning to real-life manufacturing problems due to practical limitations, class imbalance and data insufficiency, which also make it difficult to produce a generalized model. To overcome these challenges, we propose using omni-supervised learning but with a new approach to knowledge distillation that ensembles predictions from multiple instantiations of a CNN model of synthetically generated data samples from a deep generative model. Our experiment results show that the fault classification accuracy improves substantially by applying trace data analytics to manufacturing data from display fabrication lines. The results also show that the quality of trained CNN models using the proposed knowledge distillation is maintained steadily and stably.
{"title":"Trace Data Analytics with Knowledge Distillation : DM: Big Data Management and Mining","authors":"Janghwan Lee, Wei Xiong, Wonhyouk Jang","doi":"10.1109/ASMC49169.2020.9185292","DOIUrl":"https://doi.org/10.1109/ASMC49169.2020.9185292","url":null,"abstract":"In this paper, we propose the “trace data analytics” for classifying fault conditions from multivariate time series sensor signals using well-known deep CNN models. In our approach, multiple sensor signals are converted into two dimensional representations using the proposed conversion methods to optimize the classification performance. Many studies on the prediction of manufacturing results using sensor signals have been conducted in the field of fault detection and classification for display and semiconductor manufacturing processes. It is challenging to apply machine learning to real-life manufacturing problems due to practical limitations, class imbalance and data insufficiency, which also make it difficult to produce a generalized model. To overcome these challenges, we propose using omni-supervised learning but with a new approach to knowledge distillation that ensembles predictions from multiple instantiations of a CNN model of synthetically generated data samples from a deep generative model. Our experiment results show that the fault classification accuracy improves substantially by applying trace data analytics to manufacturing data from display fabrication lines. The results also show that the quality of trained CNN models using the proposed knowledge distillation is maintained steadily and stably.","PeriodicalId":6771,"journal":{"name":"2020 31st Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC)","volume":"97 1","pages":"1-8"},"PeriodicalIF":0.0,"publicationDate":"2020-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89303012","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：