Schmoo Data Analysis Using Machine Language Algorithms
Rekha K Bangalore, Adeosun luwatosin Oluwatosin, Kelvin K. Lam · 2018
Manufacturing testing uses various methodologies to validate the product specifications. Some of the production specifications require extensive characterization using two or more parameters. The most common parameters are testing voltage, frequency and temperature across different margins to understand the operating region of the DUT (Device under Test) and apply the correct guard band to the tests to minimize customer defects. Schmoo Plots is one of the validation methodology analysis is a common way of characterizing a device over a range of parameters Voltage, frequency, and temperature. Schmoo plots are collected over multiple corners. Traditionally, the data extracted from testing is manually reviewed to identify marginality and corner cases. Visual review of the schmoo logs is time consuming. Task becomes tedious with increase in the sample space (no of units to get confidence In the testing). In addition manual reviews tend to miss corner cases and the defect shows up very late during product testing. Our approach is to use machine language algorithms to design a Schmoo analyzer model to detect Schmoo plots automatically. The Schmoo Analyzer makes efficient use of Schmoo data logs, by identifying consistently passing automatic test equipment (ATE) test patterns, Schmoo wall, schmoo floor. This helps in faster turn-around time for, Module test development, reject validation and eliminates the need for manual analysis. Based on our testing we are able to obtain an efficiency of 96% with a trained model with product data and efforts continue to improve the predictability that will result testing efficiency.