June 28, 2010

The philosophical foundation for Vision was established in 1998. Since that time, we at Radiant have been working to enable all of the capabilities we envisioned in 1998. Each major revision, for instance Version 4, adds one or more of the features we originally intended for Vision. One of the features recently added in Version 4 is the ability to mine data from the DataSets you have created and filled with measurements. Under normal conditions, the Vision Editor allows you to make multiple, inter-related measurements of your samples and plot the results directly during the execution of a single test. The new data mining feature allows you to export measured data or filter plot results already in a DataSet to a central location on your disk. You can then construct a test in the Editor to import these special files into a Test Definition and treat them as if they are new measurements to be plotted by Filter tasks from the Vision Editor.

As an example, suppose that you have measured the capacitors from two different batches or, in the case of thin film devices, from two different wafers. A single DataSet may contain the two records, one for each batch or wafer that was sorted, or the records may exist in two separate DataSets. Each record contains the measurements of the capacitors in its batch or on its wafer. The data mining feature of Vision Version 4 will allow you to combine the results from the two sources without having to take the data again and without exporting the data as ASCII files to another plotting tool.

A second example is where a single sample is tested, annealed and tested a second time. It is impossible to run a Test Definition that will measure the device before and after the anneal step since the sample must be taken off the test fixture to accomplish the anneal step. Instead, the before and after tests can be run as separate Test Definitions and the results of the two records combined later as a third record in the DataSet.

How It Works

There are three steps to the data mining function.

First, the targeted data is exported in the "Vision" format from a measurement or filter record in the Archive. Each record in the Archive has an Export button which will allow you to export the data it contains in Vision format.

Second, a new Test Definition is built in the Editor and instructed to recall Vision-formatted files instead of making measurements. Each Task in the Editor has a small check box that, when enabled, forces the Task to look for Vision files on the disk or on the network.

Third, Filter Tasks are added to the Test Definition to collect, analyze and plot the recalled Vision files as you direct.

 

Demonstration of Data Mining for Wafer Sorting

Radiant places standard 50 micron by 50 micron test capacitors on each of its integrated capacitor wafers. The test capacitors are placed at the four cardinal locations on each wafer as well as in the center, yielding five test capacitors per wafer. For a recent piezoMEMs lot fabricated by Radiant, the five test caps were tested on two of the wafers. The test program for each wafer consisted of a Pause Task to allow moving the probes to the next capacitor followed by a Check Hysteresis Task to collect the hysteresis loop of that capacitor. Five pairs of these Pause/Hysteresis duos made up a test. A Hysteresis Filter was placed at the end of the Test Definition to collect the loops from all five capacitors from a single wafer into a single plot for comparison. The Test Definition was run independently on two wafers from the lot and each were archived in a single DataSet. The measurements for Wafer 2 and Wafer 3 respectively are shown below. The capacitors consist of 1-micron thick 4/20/80 PNZT with platinum electrodes, an ILD protection layer and Chrome/Gold metal interconnect.

The new data mining feature of Vision Version 4 provides the capability to transfer both data plots above to a single graph for comparison. Both of the plots originate from Hysteresis Filter Tasks. The first step is to open each of those Tasks in the Archive and export them to disk as VIS files.

The menu above is displayed immediately after double-clicking on the first Hysteresis Task in the test Record in the Archive of the dataset. Note the Export button in the lower left-hand corner of the menu. Clicking on the Export button yields the following menu:

Select the "Export Vision" file format and then click on the "Browse for File Name" button to store the VIS file anywhere on the system disk or on the network. The Hysteresis Task records for both wafers were exported in this fashion.

The next step is to create a new Test Definition in the Vision Editor. The key requirement is that a Task can only import a Vision file of its own type! Therefore, to recall and plot the data from the two exported Hysteresis Filters, the Test Definition should look like this:

The first two Tasks are Hysteresis Filters set to recall the two exported Vision files. The third task is a Collect/Plot Filter to collect the data from the two preceeding Hysteresis Filter Tasks and plot the data together. To force a Task to recall a Vision file of its same type, click on the check box labeled "Read Data from Vision File". The setup menu for the first Hysteresis Filter of this Test Definition is shown below for reference. The check box for recall is in the upper right-hand corner of the menu. All Tasks, whether they be a Measurement Task or a Filter Task, have such a check box.

The graph resulting from the execution of the given Test Definition shows good agreement between the capacitors on the two wafers with only one flyer. Using the "Configuration Dialog" available in Vision from the right mouse key, the loops from Wafer 2 were set to Blue while the loops from Wafer 3 were set to Red.

The combined plot shows that the Pmax values for both wafers have an extremely tight grouping. The coercive voltages for the two wafers are slightly different. The cause for this difference is unknown. If the devices on these two wafers are used in circuits that depend only on the final polarization at saturation, then capacitors from the two wafers will perform identically. If the capacitors are used in a manner that depends upon the value of their coercive voltages, capacitors from the two wafers will behave slightly differently.

Utility of the Vision Data File

The figure below shows an elaborate combination of the possible uses of the Vision Data File export/import option. A single Quiklook execution at 5.0 Volts and 10.0 ms is exported to a Vision Data File. Then 7.0-Volt and 9.0-Volt Hysteresis measurements and their associated Collect/Plot Filter data are expored to the VDF format from separate DataSets. A new DataSet is constucted and the Test Definition is configured to import the 5.0, 7.0 and 9.0-Volt data while making new 6.0 and 8.0-Volt measurements.New Collect/Plot Filter Tasks are introduced to gather data from the various sources and display the data together.

Conclusion

The data mining feature of Vision Version 4 provides a sophisticated analysis tool to the user. Vision now has the capability to recall and compare any historical measurements across an array of samples, lots or process conditions.

Copyright © Radiant Technologies, Inc. 2005, 2006, 2007, 2008
2835D Pan American Freeway NE Albuquerque, NM 87107 • Call: 505-842-8007, 800-289-7176 • email: radiant@ferrodevices.com

Web Design by Evangeline "Eve" Burge