Diagnose Process Variations By Using Wafer Map and Highlighted Histogram Plot

Introduction

Cross wafer parametric variation can be a significant source of yield loss in IC manufcturing. Often cross-wafer problems are not seen as the data is lost when 'wafer averages' are taken in simple statistical analyses. This new wafer map feature in SPAYN is a useful tool that can assist process and integration engineers to find out process steps that cause the cross wafer yield loss.

Figure 1. Wafer Map display of statistically correlated
parameter values for LD_N parameter.

 

Figure 2. Setup window for highlighting
histogram window by selecting attributes.

Lot Distribution

As an example a SPICE Level 3 data set for a CMOS process is used. After loading the data into SPAYN, an attribute search is performed in order to get 107 sets of data collected from Lot1. Five wafers were sampled in this case. After a parameter search, invoke the histogram plot feature to display the parameters. LD for NMOS was selected as an example. The histogram plot displays the lot distribution of LD_N. A more useful plot would be a display of how the LD_N values are distributed access different wafers. By using the Highlighted by Attribute feature from the Set Up menu bar, we could generate a highlighted histogram plot by plotting the values of LD_N from different wafers by different colors. As indicated in Figure 3, the lightest grey color is used for wafer two. Notice that the values sampled from wafer two are distributed to the lower part of the distribution. Some are close to the nominal value, some are quite low. At this point, we want to know what are the process steps that cause such a low value. This is where the wafer map feature can be used.

Figure 3. Highlighted histogram display of parameter values.

 

Spatial Distribution On A Wafer

Perform Data Search -> Search Results-> Wafer Map... operations to invoke the wafer map. The die location information that was saved in a separate file is now loaded into SPAYN in the wafer map window. By using the SetUp feature on the wafer map window, one can select a particular wafer for analysis. For this analysis, wafer 2 was selected. Then display the LD_N values on the wafer map. This display gives the spatial distribution of LD_N values on a wafer map. By selecting different colors to represent different ranges of values, user can get an indication for the cause of low values. In our display of LD_N, we could easily see lower values are distributed toward the left side of the wafer which is a strong indication that there are some problems with some of the process steps. A yield or process engineer can use this data to identify which process steps have connections to this data skew.