Patents/ Research

We introduce the Generative Dense Chain Mapping, a training-free method that maps any discrete-time, possibly multivariate and variable-length, time series into a fixed-structure Hidden Markov Model, which we call a Generative Dense Chain (GDC). We show that this mapping is one-to-one, and thus we can construct a metric distance over time series by using a metric distance over GDC's. In empirical tests with semiconductor, ECG, and battery data this distance metric distance preforms comparably with similarity measures such as dynamic time warping. Additionally, we identify GDC's as having a number of desirable properties for time series analysis, such as accurate long-horizon forecasting, fully general forecasting pdfs, and explainability of forecasting predictions.  

An exemplary anomaly detection system is disclosed for a feature-based assessment of a semiconductor fabrication equipment or process, as well as other manufacturing equipment and processes, that employs Hidden Markov Model-based segmentation error correction of time-series sensor data in the assessment. Notably, the feature-based assessment and segmentation error correction have been observed to provide a high detection rate of defects in a fabricated device and associated fabricated techniques and with a low false alarm rate. 

This paper proposes a novel method for Virtual Metrology (VM) in plasma etch processes based on analysis of all time and wavelength samples of Optical Emission Spectroscopy (OES) signals. The new method flattens each OES signal into a single vector, after which Singular Value Decomposition (SVD) is performed on the matrix formed by vectors of flattened OES signals in the training dataset. Low rank SVD projections of flattened and standardized OES recordings served as inputs for Ridge Regression, Artificial Neural Network, and Random Forest based VM models. A VM study is then conducted on a dataset gathered from a major 300 mm wafer fabrication facility, showing that the use of newly proposed SVD-based OES features consistently outperformed benchmark VM model features. Additional analysis of feature importance performed based on the analytically tractable Ridge Regression VM model form demonstrated distinct time-frequency patterns of OES signal portions that were highly informative for prediction of relevant Critical Dimensions, clearly justifying the need to use the entire OES signals for VM. 

In this paper, we develop and demonstrate a methodology for Virtual Metrology (VM) of Critical Dimensions (CDs) in etch processes which uses physically interpretable features extracted based on a fully automated analysis of sensor trace recordings in their entirety rather than just on preselected portions of the raw trace data. This novel methodology was evaluated on a dataset consisting of sensor readings and corresponding physical CD metrology from an etch process performed in a 300 mm wafer fab on 114,774 wafers. It was observed that VM models consistently performed better when they were built using the novel, automatically generated features than when they relied on the benchmark features. Benchmark features were generated from preselected and expert-defined signal windows, and were utilized by fab engineers for process control and fault detection and diagnostics on the relevant tools. Furthermore, it was observed that fusing the benchmark and new, automatically generated features resulted in the best VM performance. Finally, analysis of contributions of various features to the overall VM model performance clearly showed that incorporating features from automatically identified, short-lived, transient portions of raw sensor signals consistently brought benefits to VM model performance.