On Regression Methods for Virtual Metrology in Semiconductor Manufacturing

Virtual metrology (VM) aims to predict metrology values using sensor data from production equipment and physical metrology values of preceding samples. VM is a promising technology for the semiconductor manufacturing industry as it can reduce the frequency of in-line metrology operations and provide supportive information for other operations such as fault detection, predictive maintenance and run-to-run control. The prediction models for VM can be from a large variety of linear and nonlinear regression methods and the selection of a proper regression method for a specific VM problem is not straightforward, especially when the candidate predictor set is of high dimension, correlated and noisy. Using process data from a benchmark semiconductor manufacturing process, this paper evaluates the performance of four typical regression methods for VM: multiple linear regression (MLR), least absolute shrinkage and selection operator (LASSO), neural networks (NN) and Gaussian process regression (GPR). It is observed that GPR performs the best among the four methods and that, remarkably, the performance of linear regression approaches that of GPR as the subset of selected input variables is increased. The observed competitiveness of high-dimensional linear regression models, which does not hold true in general, is explained in the context of extreme learning machines and functional link neural networks.

Formation of Novel Di-Nuclear Lanthanide Luminescent Sm(III), Eu(III) and Tb(III) Triple Stranded Helicates from a C2 Symmetrically 2,6-Dicarboxylic Amide based 1,3-Xylene Ligand in CH3CN

The synthesis of the C2-symmetrical ligand 1 consisting of two naphthalene units connected to two pyridine-2,6-dicarboxamide moieties linked by a xylene spacer and the formation of LnIII-based (Ln1/4 Sm, Eu, Tb, and Lu) dimetallic helicates [Ln2 · 13] in MeCN by means of a metal-directed synthesis is described. By analyzing the metal-induced changes in the absorption and the fluorescence of 1, the formation of the helicates, and the presence of a second species [Ln2 · 12] was confirmed by nonlinear- regression analysis. While significant changes were observed in the photophysical properties of 1, the most dramatic changes were observed in the metal-centred lanthanide emissions, upon excitation of the naphthalene antennae. From the changes in the lanthanide emission, we were able to demonstrate that these helicates were formed in high yields (ca. 90% after the addition of 0.6 equiv. of LnIII), with high binding constants, which matched well with that determined from the changes in the absorption spectra. The formation of the LuIII helicate, [ Lu2 · 13 ] , was also investigated for comparison purposes, as we were unable to obtain accurate binding constants from the changes in the fluorescence emission upon formation of [Sm2 · 13], [Eu2 · 13], and [Tb2 · 13].

A Robust Curve-Fitting Procedure for the Analysis of Plasmid DNA Strand Break Data from Gel Electrophoresis

A robust method for fitting to the results of gel electrophoresis assays of damage to plasmid DNA caused by radiation is presented. This method makes use of nonlinear regression to fit analytically derived dose response curves to observations of the supercoiled, open circular and linear plasmid forms simultaneously, allowing for more accurate results than fitting to individual forms. Comparisons with a commonly used analysis method show that while there is a relatively small benefit between the methods for data sets with small errors, the parameters generated by this method remain much more closely distributed around the true value in the face of increasing measurement uncertainties. This allows for parameters to be specified with greater confidence, reflected in a reduction of errors on fitted parameters. On test data sets, fitted uncertainties were reduced by 30%, similar to the improvement that would be offered by moving from triplicate to fivefold repeats (assuming standard errors). This method has been implemented in a popular spreadsheet package and made available online to improve its accessibility. (C) 2011 by Radiation Research Society

Modeling Multivariate Interest Rates Using Time-Varying Copulas and Reducible Nonlinear Stochastic Differential Equations

We propose a new approach for modeling nonlinear multivariate interest rate processes based on time-varying copulas and reducible stochastic differential equations (SDEs). In the modeling of the marginal processes, we consider a class of nonlinear SDEs that are reducible to Ornstein--Uhlenbeck (OU) process or Cox, Ingersoll, and Ross (1985) (CIR) process. The reducibility is achieved via a nonlinear transformation function. The main advantage of this approach is that these SDEs can account for nonlinear features, observed in short-term interest rate series, while at the same time leading to exact discretization and closed-form likelihood functions. Although a rich set of specifications may be entertained, our exposition focuses on a couple of nonlinear constant elasticity volatility (CEV) processes, denoted as OU-CEV and CIR-CEV, respectively. These two processes encompass a number of existing models that have closed-form likelihood functions. The transition density, the conditional distribution function, and the steady-state density function are derived in closed form as well as the conditional and unconditional moments for both processes. In order to obtain a more flexible functional form over time, we allow the transformation function to be time varying. Results from our study of U.S. and UK short-term interest rates suggest that the new models outperform existing parametric models with closed-form likelihood functions. We also find the time-varying effects in the transformation functions statistically significant. To examine the joint behavior of interest rate series, we propose flexible nonlinear multivariate models by joining univariate nonlinear processes via appropriate copulas. We study the conditional dependence structure of the two rates using Patton (2006a) time-varying symmetrized Joe--Clayton copula. We find evidence of asymmetric dependence between the two rates, and that the level of dependence is positively related to the level of the two rates. (JEL: C13, C32, G12) Copyright The Author 2010. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org, Oxford University Press.

Forward and backward least angle regression for nonlinear system identification

A forward and backward least angle regression (LAR) algorithm is proposed to construct the nonlinear autoregressive model with exogenous inputs (NARX) that is widely used to describe a large class of nonlinear dynamic systems. The main objective of this paper is to improve model sparsity and generalization performance of the original forward LAR algorithm. This is achieved by introducing a replacement scheme using an additional backward LAR stage. The backward stage replaces insignificant model terms selected by forward LAR with more significant ones, leading to an improved model in terms of the model compactness and performance. A numerical example to construct four types of NARX models, namely polynomials, radial basis function (RBF) networks, neuro fuzzy and wavelet networks, is presented to illustrate the effectiveness of the proposed technique in comparison with some popular methods.

A two-stage algorithm for identification of nonlinear dynamic systems

This paper investigates the two-stage stepwise identification for a class of nonlinear dynamic systems that can be described by linear-in-the-parameters models, and the model has to be built from a very large pool of basis functions or model terms. The main objective is to improve the compactness of the model that is obtained by the forward stepwise methods, while retaining the computational efficiency. The proposed algorithm first generates an initial model using a forward stepwise procedure. The significance of each selected term is then reviewed at the second stage and all insignificant ones are replaced, resulting in an optimised compact model with significantly improved performance. The main contribution of this paper is that these two stages are performed within a well-defined regression context, leading to significantly reduced computational complexity. The efficiency of the algorithm is confirmed by the computational complexity analysis, and its effectiveness is demonstrated by the simulation results.

Analytical description of stochastic field-Line wandering in magnetic turbulence

A nonperturbative nonlinear statistical approach is presented to describe turbulent magnetic systems embedded in a uniform mean magnetic field. A general formula in the form of an ordinary differential equation for magnetic field-line wandering (random walk) is derived. By considering the solution of this equation for different limits several new results are obtained. As an example, it is demonstrated that the stochastic wandering of magnetic field-lines in a two-component turbulence model leads to superdiffusive transport, contrary to an existing diffusive picture. The validity of quasilinear theory for field-line wandering is discussed, with respect to different turbulence geometry models, and previous diffusive results are shown to be deduced in appropriate limits.

Improved fault diagnosis in multivariate systems using regression-based reconstruction

Subspace monitoring has recently been proposed as a condition monitoring tool that requires considerably fewer variables to be analysed compared to dynamic principal component analysis (PCA). This paper analyses subspace monitoring in identifying and isolating fault conditions, which reveals that the existing work suffers from inherent limitations if complex fault senarios arise. Based on the assumption that the fault signature is deterministic while the monitored variables are stochastic, the paper introduces a regression-based reconstruction technique to overcome these limitations. The utility of the proposed fault identification and isolation method is shown using a simulation example and the analysis of experimental data from an industrial reactive distillation unit.

Stochastic Sznajd Model in open community

We extend the Sznajd Model for opinion formation by introducing persuasion probabilities for opinions. Moreover, we couple the system to an environment which mimics the application of the opinion. This results in a feedback, representing single-state opinion transitions in opposite to the two-state opinion transitions for persuading other people. We call this model opinion formation in an open community (OFOC). It can be seen as "stochastic extension of the Sznajd model for an open community, because it allows for "special choice of parameters to recover the original Sznajd model. We demonstrate the effect of feedback in the OFOC model by applying it to a scenario in which, e.g., opinion B is worse then opinion A but easier explained to other people. Casually formulated we analyzed the question, how much better one has to be, in order to persuade other people, provided the opinion is worse. Our results reveal a linear relation between the transition probability for opinion B and the influence of the environment on B.

Relative role of deterministic and stochastic determinants of soil animal community: a spatially explicit analysis of oribatid mites

1. Ecologists are debating the relative role of deterministic and stochastic determinants of community structure. Although the high diversity and strong spatial structure of soil animal assemblages could provide ecologists with an ideal ecological scenario, surprisingly little information is available on these assemblages.
2. We studied species-rich soil oribatid mite assemblages from a Mediterranean beech forest and a grassland. We applied multivariate regression approaches and analysed spatial autocorrelation at multiple spatial scales using Moran's eigenvectors. Results were used to partition community variance in terms of the amount of variation uniquely accounted for by environmental correlates (e.g. organic matter) and geographical position. Estimated neutral diversity and immigration parameters were also applied to a soil animal group for the first time to simulate patterns of community dissimilarity expected under neutrality, thereby testing neutral predictions.
3. After accounting for spatial autocorrelation, the correlation between community structure and key environmental parameters disappeared: about 40% of community variation consisted of spatial patterns independent of measured environmental variables such as organic matter. Environmentally independent spatial patterns encompassed the entire range of scales accounted for by the sampling design (from tens of cm to 100 m). This spatial variation could be due to either unmeasured but spatially structured variables or stochastic drift mediated by dispersal. Observed levels of community dissimilarity were significantly different from those predicted by neutral models.
4. Oribatid mite assemblages are dominated by processes involving both deterministic and stochastic components and operating at multiple scales. Spatial patterns independent of the measured environmental variables are a prominent feature of the targeted assemblages, but patterns of community dissimilarity do not match neutral predictions. This suggests that either niche-mediated competition or environmental filtering or both are contributing to the core structure of the community. This study indicates new lines of investigation for understanding the mechanisms that determine the signature of the deterministic component of animal community assembly.

Supervised Aggregative Feature Extraction for Big Data Time Series Regression

In many applications, and especially those where batch processes are involved, a target scalar output of interest is often dependent on one or more time series of data. With the exponential growth in data logging in modern industries such time series are increasingly available for statistical modeling in soft sensing applications. In order to exploit time series data for predictive modelling, it is necessary to summarise the information they contain as a set of features to use as model regressors. Typically this is done in an unsupervised fashion using simple techniques such as computing statistical moments, principal components or wavelet decompositions, often leading to significant information loss and hence suboptimal predictive models. In this paper, a functional learning paradigm is exploited in a supervised fashion to derive continuous, smooth estimates of time series data (yielding aggregated local information), while simultaneously estimating a continuous shape function yielding optimal predictions. The proposed Supervised Aggregative Feature Extraction (SAFE) methodology can be extended to support nonlinear predictive models by embedding the functional learning framework in a Reproducing Kernel Hilbert Spaces setting. SAFE has a number of attractive features including closed form solution and the ability to explicitly incorporate first and second order derivative information. Using simulation studies and a practical semiconductor manufacturing case study we highlight the strengths of the new methodology with respect to standard unsupervised feature extraction approaches.

Modelling the nonlinear behaviour and fracture process of AS4/PEKK thermoplastic composite under shear loading

The accurate determination of non-linear shear behaviour and fracture toughness of continuous carbon-fibre/polymer composites remains a considerable challenge. These measurements are often necessary to generate material parameters for advanced computational damage models. In particular, there is a dearth of detailed shear fracture toughness characterisation for thermoplastic composites which are increasingly generating renewed interest within the aerospace and automotive sectors. In this work, carbon fibre (AS4)/ thermoplastic Polyetherketoneketone (PEKK) composite V-notched cross-ply specimens were manufactured to investigate their non-linear response under pure shear loading. Both monotonic and cyclic loading were applied to study the shear modulus degradation and progressive failure. For the first time in the reported literature, we use the essential work of fracture approach to measure the shear fracture toughness of continuous fibre reinforced composite laminates. Excellent geometric similarity in the load-displacement curves was observed for ligament-scaled specimens. The laminate fracture toughness was determined by linear regression, of the specific work of fracture values, to zero ligament thickness, and verified with computational models. The matrix intralaminar fracture toughness (ply level fracture toughness), associated with shear loading was determined by the area method. This paper also details the numerical implementation of a new three-dimensional phenomenological model for carbon fibre thermoplastic composites using the measured values, which is able to accurately represent the full non-linear mechanical response and fracture process. The constitutive model includes a new non-linear shear profile, shear modulus degradation and load reversal. It is combined with a smeared crack model for representing ply-level damage initiation and propagation. The model is shown to accurately predict the constitutive response in terms of permanent plastic strain, degraded modulus as well as load reversal. Predictions are also shown to compare favourably with the evolution of damage leading to final fracture.