ENBIS Spring Meeting 2017

28 – 30 May 2017; Monastery of Schlägl in Upper Austria Abstract submission: 11 November 2016 – 5 March 2017

My abstracts

 

The following abstracts have been accepted for this event:

  • Reliability Assessment of Failure Free Stress Tests

    Authors: Horst Lewitschnig (Infineon Technologies Austria AG), Nikolaus Haselgruber (CIS Consulting in Industrial Statistics)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Reliability
    Keywords: Life data analysis, Reliability methods, Robust validation, Zero defect
    Submitted at 20-Feb-2017 05:30 by Horst Lewitschnig
    Accepted
    30-May-2017 11:15 Reliability Assessment of Failure Free Stress Tests
    Semiconductor devices for safety critical applications, like automotive, aviation, railways, ..., have to fulfil very high quality and reliability requirements. These requirements even increase with new applications and higher complexity, like autonomous driving for cars.

    When it comes to product qualification, such devices should not and typically do not fail; even if the stress duration is extended.

    Qualification tests are set up as zero failures sampling plans. This makes it difficult to estimate parameters of any underlying lifetime distribution, because such methods require failures.

    We review alternative methods, how reliability as a function of time can be estimated even without failures in the stress test:

    • One way is to use the Clopper-Pearson estimator for the failure probability after the stress test. For lower stress times, this value can be linearly scaled down.
    • The Clopper-Pearson estimator can be regarded from a Bayesian point of view as the likelihood function combined with the a priori distribution Beta(0,1). When the parameters of this a priori Beta distribution are changed, based on expert knowledge, then this could lead to a more accurate estimation of the failure portion at the end of the stress test.
    • The cumulative hazard function can be modelled by a Gamma-process.
    •The WeiBayes approach assumes a Weibull distribution with a given shape parameter.

    We present an alternative approach. We use expert judgement, how the failure behavior beyond the censoring limit would be. In this way, we’re looking for failure scenarios that could occur if we would extend the stress test further. By doing this, we can fit lifetime distributions to those scenarios and finally assess them for the best fitting distribution as well as for a certain confidence level.

    Basically we add more information to the stress test, which gives a more accurate picture about the devices’ reliability. If the stress test is really extended and failures occur, our assumptions are replaced step by step with real data; then the classical parameter estimators can be employed.

    The work has been performed in the project eRamp (Grant Agreement N°621270), co-funded by grants from Austria, Germany, Slovakia and the ENIAC Joint Undertaking.
  • Multi Failure Mode Reliability in Adults

    Authors: Chris McCollin (Nottingham Trent University), Shirley Coleman (ISRU Necastle upon Tyne)
    Primary area of focus / application: Reliability
    Secondary area of focus / application: Modelling
    Keywords: Failure modes, Langevin equation, Mokken scaling, Lévy process
    Submitted at 20-Feb-2017 12:23 by Chris McCollin
    Accepted (view paper)
    29-May-2017 16:55 Multi Failure Mode Reliability in Adults
    We consider sequential loss of ability to perform activities of daily living and incidents through life which affect emotional wellbeing. These former are rationalised using Mokken scaling and the latter are modelled with a Langevin equation approach to the Lévy process. The results show that the loss of abilities follow a common pattern and that life can be modelled as a series of cycles where an event related parameter comes into play. This wear parameter is affected differently at different phases of life due to the age of the recipient and external events. Examples are supplied for a number of cases.
  • Modeling the Variance of the Durations of Maintenance Activities in Semiconductor Fabs

    Authors: Diamanta Benson-Karhi (The Open University of Israel)
    Primary area of focus / application: Reliability
    Keywords: Availability, Variability of availability, Preventive Maintenance (PM), Imperfect PM, Optimization
    Submitted at 21-Feb-2017 15:07 by Diamanta Benson-Karhi
    Accepted
    30-May-2017 12:55 Modeling the Variance of the Durations of Maintenance Activities in Semiconductor Fabs
    The reduction of equipment variability of availability can significantly reduce cycle times, improve yields, reduce time-to-market and lower overall operational costs. Despite extensive efforts to reduce variability of availability, however, the execution of preventive maintenance (PM) is still one of the main sources of equipment variability of availability in the industry. In this paper, a model for PM duration variance is proposed. The model is based on the parts of the PM and the standard PM flow, and it provides a framework for modeling and segmenting the PM duration variance in the semiconductor industry while highlighting inefficiencies and improvement opportunities.
  • Modeling the Lifespans of Lithium-Ion Cells by Using D-Optimal Designs

    Authors: Ernst Stadlober (University of Technology Graz, Institute of Statistics), Gerhard Gößler (Virtual Vehicle Research Center Graz), Martin Cifrain (Virtual Vehicle Research Center Graz)
    Primary area of focus / application: Design and analysis of experiments
    Secondary area of focus / application: Reliability
    Keywords: Lithium-ion cells, Cell-aging experiments, Reliability, Design of Experiments, Response surfaces
    Submitted at 21-Feb-2017 18:00 by Ernst Stadlober
    Accepted
    29-May-2017 16:15 Modeling the Lifespans of Lithium-Ion Cells by Using D-Optimal Designs
    Lithium-ion batteries seem to meet the demands of electric mobility better than other battery types, because of a high energy density, hardly any memory effect, and only a slow loss of charge when not in use. The reliability of a battery cell is definitely one of its key performance indices. It is measured as the battery’s lifespan which depends on a large number of influential factors and there exists no comprehensive mechanistic understanding of the relationship between these factors and the process of cell aging. Due to the complex nature of the process under investigation, there is an obvious need for the use of statistical methods. The design as well as the analysis of the experiment can therefore only be conducted on a sound statistical basis.
    As a proper indicator of cell aging we define as life time of the cells the time needed to reach 70% of their initial capacity. It is also necessary to establish a method for extrapolating this life time, because a significant number of cells do not reach end-of-life within the experiment. Simple exponential models lead to stable extrapolations with respect to the characteristics of the experimental settings.
    The experiment is based on an initial design developed at the beginning of the project. This design includes seven factors (e.g. temperature, state-of-charge or current rates). However, the direct implementation of this design is not possible due to physical restrictions. Hence, we follow a strategy for revision of the design which focuses on a balance between constraints concerning available resources and the increase of the D-efficiency of the design. The response surface can be reasonably described at least by two multiple regression models, the simpler one with six variables and the more complicated one with ten variables. Both models are able to discriminate between harsh and moderate validation points and indicate a satisfying quality of prediction. For conducting further experiments one of the suggested models may be used as a starting point for design construction.
  • Applied and Interpretable Multivariate Process Control of Wafer Acceptance Tests

    Authors: Peter Scheibelhofer (AMS AG), Günter Hayderer (AMS AG)
    Primary area of focus / application: Process
    Secondary area of focus / application: Reliability
    Keywords: Statistical Process Control, Multivariate statistics, Robust statistics, Implementation, Semiconductor manufacturing
    Submitted at 22-Feb-2017 15:27 by Peter Scheibelhofer
    Accepted
    29-May-2017 09:45 Applied and Interpretable Multivariate Process Control of Wafer Acceptance Tests
    In order to assess the functionality and reliability of chips on a wafer after its fabrication, the wafer acceptance test (WAT) is used as a crucial quality check. The various chip functionalities are described by a large number of electrical parameters. Based on their outcome it is decided if the chips on a wafer fulfill the design requirements. Classical SPC charts are used to monitor each WAT parameter separately. However, these charts are univariate and do not reflect the fact that WAT parameters are interrelated, i.e. do not behave independently of one another. This makes the overall WAT data inherently multivariate. With classical monitoring approaches anomalies in the inherent relationship structure as determined by physics cannot be detected. However, measurements where parameter relationships deviate from the reference behavior can cause discrepancies to the design requirements of a chip and can affect its functionalities. Furthermore, such deviations can lead to significant yield loss when at the same time all variables are within their associated univariate SPC limits. In order to adequately monitor WAT measurements, multivariate control charts have to be implemented. We propose the use of a robust Hotelling’s T^2 statistics to monitor the univariate as well as the multivariate behavior of new WAT
    measurements in a single control chart. The Mason-Young-Trac decomposition of robust T^2 values offers the possibility to interpret out-of-control observations, determining the single or multiple variables responsible for an abnormality. This way, univariate as well as multivariate deviations from the reference can be detected in a simple and practically favorable way. In order for the T^2 statistics to be useful and applicable an appropriate software environment was implemented in R and TIBCO Spotfire. The implementation is able to connect advanced statistical modeling with an easy-to-use interface for process engineers.
  • The Use of Statistical Prediction Methods for Process Control of Additive Manufacturing

    Authors: Eva Scheideler (OWL University oAS, Lemgo), Andrea Ahlemeyer-Stubbe (Data Mining and More)
    Primary area of focus / application: Process
    Secondary area of focus / application: Modelling
    Keywords: Additive manufacturing, Process control, Predictive modeling, Meta model
    Submitted at 23-Feb-2017 08:47 by Eva Scheideler
    Accepted
    29-May-2017 16:35 The Use of Statistical Prediction Methods for Process Control of Additive Manufacturing
    Additive manufacturing (AM) is increasingly used to design and produce new products. This is possible due to further developments in the AM processes and materials. A key technological barrier that prevents manufacturers from adopting this technology to further productions is a lack of quality assurance of AM built parts. The quality of an additive manufactured part is influenced by more than 50 parameters, which makes process controlling challenging. Several research projects were done to solve this problem. The types of solutions can be grouped in two categories: Backward control solutions mainly used for powder bed fusion processes. They are using real-time monitoring of the melt pool as feedback control for the laser power. The second category are feedforward control solutions used for laser cladding.
    We want to develop a third category that will be used for powder bed fusion and non powder bed fusion related processes. It is based on combining meta-model thinking and predictive modeling.
    This talk illustrates challenges and opportunities of applying statistical prediction methods and self-learning aspects to predict potential discontinuities. As a result we want to enable real-time regulation of laser power or other controllable parameters to avoid discontinuities of the additive manufactured parts. The talk will present the project set up and the actual work in progress.