Catching Typos with Test Data: Using JMP to Identify Serial Number Errors (2025-US-30MP-2542)

In a high-mix, low-volume semiconductor manufacturing facility, devices are tested multiple times and at multiple temperatures throughout the process and tracked using manually entered serial numbers. While essential for traceability, manual data entry introduces the risk of transcription errors – leading to inconsistencies in device histories and potential misinterpretation of test results. This presentation showcases a real-world application of JMP to flag serial number entry errors by analyzing device behavior patterns across test stages.

Using electrical test data with thousands of measurements per device, we demonstrate a structured approach for identifying potential mismatches. We begin with data preparation, including z-score standardization, to bring measurements across temperature conditions onto a common scale. Next, Predictor Screening is used to isolate the most informative measurements, followed by multivariate analysis to reduce the parameter list to only the most representative predictors. This set of variables forms a behavioral fingerprint for each device. We then apply variable clustering to observe when serial numbers group tightly –  and when they don’t.

By visualizing and clustering devices based on this fingerprint, we identify cases where two different serial numbers exhibit near-identical electrical behavior –  suggesting a likely entry error. JMP’s built-in formulas, Predictor Screening, multivariate analysis, and clustering tools are used to guide exploration and support evidence-based conclusions.

This approach has been used to identify and resolve data integrity issues. This presentation targets engineers, analysts, and quality professionals seeking to apply analytics to real-world manufacturing challenges – without losing connection to the physical meaning of the data.