A More Efficient Way To Calculate Device Specs Of Thousands Of Tests For Improved Quality And Yield

Today’s devices are required to pass thousands of parametric tests prior to being shipped to customers. A key challenge test engineers face, in addition to optimizing the number of tests they run on the device, is how to quickly and accurately define the true specification limits that should be used to determine if the device is “good”.

Device specification limits that are too wide may allow “outliers” to sneak through, resulting in poor quality parts being shipped into the supply chain. Conversely, device specification limits defined to be too narrow will reduce product yield and profitability. However, in the case where specification limits are too narrow, it may not be possible to widen them if the limits used were driven by the product datasheet. Complicating matters further, there is typically no one-size-fits-all approach when it comes to specifying spec limits as they may need to vary depending on which type of test is being run. This is not only true during the NPI phase of a product but also during the HVM phase once more variability is gathered, allowing the test limits to be continually reassessed and recalibrated as necessary. Therefore, to mitigate any issues with yield and quality, product and test engineers have the time-consuming task of re-evaluating the device’s performance and manually recalculating the limits for each individual parametric test over the manufacturing lifespan of the product.

What is needed is a more efficient method to quickly analyze the data collected for each parametric test and then statistically and automatically calculate new limits per test based on the process capability index, or Cpk. Cpk is widely used in process improvement efforts and measures the proportion of natural tolerances between the center of the process and the original predetermined upper and lower specification limits based on actual device data.

Figure 1 below shows a typical parametric trend plot of a given test on devices of a specific lot as measured against the original specification limits.

Figure 1.  Typical parametric data as measured against the original specification

Figure 2 below shows similar parametric test results but now using more accurate, statistically calculated upper and lower test limits that were automatically generated. The limits shown in this example have been tightened compared to the original specification limits, thus enabling outlier devices to be flagged and automatically binned out as bad parts.

Figure 2.  Statistically calculated specification limits

Engineers can also collectively view the results of one or more tests simultaneously and have the corresponding histograms displayed as shown in Figure 3. Note that the Cpk is also derived and is used to statistically calculate the actual upper and lower test limits per each individual test from the test data previously captured. This flexibility of estimating specification limits for parametric tests based on statistical means is highly beneficial to engineers as it allows them to optimize test specification limits in an automated and swift manner.

Figure 3.  Measured test data per parametric test

Optimizing test limits for every new product is a very time consuming task. To be able to use actual device performance data to quickly generate quality test limits based on statistical measurements is a significant improvement in efficiency for product and test engineers as well as a critical safeguard to ensure high quality and maximizing profitability. In addition, the ability to simulate the effects on yield using previously captured data by varying the values of the limits provides the engineers with the necessary information and insight to make informed tradeoffs between quality and yield when establishing their limits.

To learn more about how semiconductor companies are leveraging these types of automated technologies to aid in parametric testing of their devices, visit us at www.optimalplus.com.