1. Background
Part Average Analysis is a statistical method for latent defect diagnosis and avoidance. By comparison of product characteristics to “standard” values, devices with higher pre-damage risk are detected.

1.1 Off-line Analysis
A successful application of a self-acting PAA module in production/test lines requires a sound evaluation of historical data to determine significance of test parameters, size of sliding data frame etc. This is best done by a product specialist in coordination with the customer.

The PAA Module within the reduFin software application provides the ideal instrument for off-line PAA analyses of large amounts of test data and simulation of on-line operations.

The programme is built on the guidelines found under www.paa-web.de. For convenience, the main features of the method are listed in the sequel.

1.2 Some Mathematics
The PAA is dynamic in the sense that each test result is evaluated w.r.t. the local average and standard deviation computed over a sliding data window. The size of the latter is determined by

where RND := round up/down and N is the number of devices in the data set.

By default and away from start/end blocks (within Mlocal/2, resp. N-Mlocal/2), the DUT of interest is put in the middle of the observation window for the computation of local means and standard deviations:

The aim is to localise “outliers” such as

Selection Rules

Anomalies have to be ranked as not each one should lead to a reject of a DUT. Therefore, both the significance of an abnormal reading and the relevance of the test in which it occurs have to be preset. The following repeats the standard procedure:

Significance Sa of anomaly

Let .  Then, for

Relevance RT of observed Test T (to be set by Expert!)
RT = 9 - Clear indication of technical defect (Leakage Current, Voltage Drop)
RT = 3 - High probability to indicate technical defect (default value)
RT = 2 - Low probability to indicate technical defect
RT = 1 - Less relevance
Based on these definitions, the Risk Ri for the ith- DUT is defined as

where  is the number of anomalies, i.e., number of non-zero Sa readings for a DUT.
A DUT has to be selected for further inspection if .

The envisioned application of that classification is

2. The PAA Module in reduFin

After starting the PAA Module, a Settings Window appears:

There, the settings described in the preceding section can be chosen. In particular, Boolean tests are automatically excluded, for parametric tests the choice is up to the Expert.

After running an analysis, the results are depicted as

The various parts are:

• The PAA Summary indicates how many tests and devices (= log-files, i.e., no retest considerations) are concerned.

• The  PAA Test Grid lists all tests taking part in the actual PAA with colour code corresponding to the (global) Cpk ranges shown in the summary. A test’s trend plot with anomalies highlighted (in yellow) can be opened from here

• The  Relevance of a test has to be set by the expert in the Settings Window shown above. The # Anomalies counts the devices having PAA anomalies in this test. Corresponding histogram and weighted Cpk, Mean and Standard Deviation are seen below.

• Suspicious devices are listed in the rightmost window. The ones above the critical risk threshold  are coloured in red. A click on a device shows the test(s) concerned in the window below. Clicking on one of the tests leads to an excerpt of the test’s trend plot around the device with anomaly (highlighted by a yellow square):

Impressum

Copyright© optimiSE 2007