Forward prediction based on wafer sort data — A case study

Abstract

This paper studies the potential of using wafer probe tests to predict the outcome of future tests. The study is carried out using test data based on an SoC design for the automotive market. Given a set of known failing parts, there are two possible approaches to learn. First a single binary classification model can be learned to model all failing parts. We show that this approach can be effective if the failing parts are compatible in learning. Second, an individual outlier model can be learned for each failing part. We show that this approach is suitable for learning failing parts such as customer returns, where each may have a unique failing behavior. We also show that with Principal Component Analysis (PCA), a learning model can be visualized in two or three dimensional PC space, which facilitates an engineer to manually select or adjust the model.

Screening customer returns with multivariate test analysis

“…A forward prediction methodology was derived in [10] that studied the potential of using parametric wafer probe test measurements to predict parts that fail at final test and on the customer's side. When predicting customer returns, the authors used PCA on a set of important tests that best describe the customer return's failing behavior and an outlier model was built to screen out returns.…”

“…This work employs two learning techniques previously suggested: (1) applying PCA to correlated tests proposed in [1] and (2) applying PCA in conjunction with SVM [2][3] one-class outlier analysis proposed in [10]. The goal is to develop a methodology to apply these learning techniques before and after examples of customer returns become available.…”

“…The parametric data of all parts that passed wafer sort testing are transformed using PCA. In the PCA space, an outlier model is built using the one-class ν-SVM algorithm [10]. In this algorithm, the value ν represents the upper bound on the fraction of outliers identified by the model (The actual number of outliers can be much smaller) [2].…”

“…. , t 10 . Suppose a future return can only be captured based on the model built on t 2 , t 4 , t 7 but the return used in the learning can be screened out with any combinations of three to five tests.…”

An experiment of burn-in time reduction based on parametric test analysis

“…The authors identified redundant tests belonging to one insertion and suggested that more expensive tests could be replaced with models built from those in less expensive insertions. In another example, the authors in [11] analyzed parametric wafer sort data from a high quality SoC and showed the potential for building models from the test data which were capable of predicting devices likely-to-fail at final package testing. Similarly, multivariate test analysis was used in [10,12] to predict parts that would fail in the field, i.e.…”