Outlier Detection to Assure Good Quality for Metrology Bias Acceptance Test

Abstract

The quality of measurement system in semiconductor manufacturing industry has been inevitably important to assure good quality of IC products being manufactured. Metrology tool's bias acceptance test is a regular test to decide whether a tool needs re-calibration. The current practice in manufacturing industries follows the renowned "Measurement Systems Analysis (MSA) Reference Manual" by Chrysler, Ford, and General Motors. However, most practitioners ignore outlier detection on the repeated measurement data, largely due to the fact that the MSA Reference Manual does not cover the outlier detection for bias acceptance test although ISO 5725 requires the outlier detection and replacement with correct values. This article discusses the impact of the outliers if existed on the bias acceptance test and the statistical methods of outlier detection. If the outliers are not detected and replaced with correct values, the bias acceptance test could fail to reject a metrology tool with unacceptable bias. This article introduces the statistical definition of the outlier, basics of statistical criteria for outliers and possible causes for the existence of outliers. Additional methods to detect outliers besides those recommended by ISO 5725 are introduced to assure good quality for metrology bias acceptance tests.