Speaker: Henry Qiao , General Electric, Quality Manager
Speaker: Henry Qiao , Quality Manager, General Electric
Topic: Estimating Process Capability with JMP® Scripted Statistical Models in New Product Validation
Topics: A JMP ® Scripted Modeling Approach to Estimate Process Capability in the Product Validation Process
summary:
In the process of new product verification of FPGA (field programmable gate array), characteristic tests of multi-factor crossover tests are often used to estimate long-term process capability. Since the estimation of process capability is based on a specific normal distribution, it cannot be directly used for the data of multi-factor crossover tests. This article aims to introduce a new model method to estimate the process capability of FPGA. This method first establishes a statistical model to obtain the worst value of the process output, estimates the size of the process variation, and then calculates the size of the process capability based on the worst value. Therefore, the process capability index obtained by this method is not based on a single normal distribution, but the worst process capability obtained based on multiple normal distribution data. This article also introduces how to use JMP scripts to automatically optimize the entire process of statistical model and standardized data analysis. This method is not only applicable to the product verification process of FPGA, but can actually be used for similar data analysis in any semiconductor industry.
Abstract：
In the product validation process of a Field Programmable Gate Array (FPGA) device, cross-factored characterization testing should be prescribed to estimate the effect of long-term process drifting. Process capability is usually estimated on the basis of one normal distribution. But the calculation cannot be applied to the cross-factored experiment data directly. This paper introduces a novel modeling approach to estimate the process capability of FPGAs. A statistical model is built to predict the worst-case values and estimate the inherent process variation, then calculate process capability indices. Therefore, the process capability indices are not obtained on the basis of one normal distribution. They actually show the worst process capability based on the cross-factored experiment data. An automated tool coded with JMP scripting language was also developed to optimize the model automatically, and to standardize the data process. This method can be applied not only to the data in the FPGA product validation process, but to all similar data in the semiconductor industry.
