Page 27 - ChipScale_Jan-Feb_2020-digital

The data demonstrates a strong correlation
between WFPI and chromatic confocal
microscopy. In the current WFPI system, the
data acquisition is fast (less than a second),
however, the data analysis can take up to 2s.
Most of the time used for data analysis is
spent acquiring the amplitude of the light.

Local wafer geometry
To measure local variations (NT
and roughness), a 2-inch blank silicon
wafer with specifications according
to SEMI standards [13] was used for
WFPI measurements. Two images were
acquired with the WFPI system: an
intensity image (Figure 8a) and a global
topography image. From the topography
image, a global depth map and a 3D map
(Figure 8b) were generated. A double
Gaussian high-pass filter was applied
on the global topography map of the
wafer to remove low-frequency spatial
resolution, which revealed roughness
with a spatial resolution equal to the

Figure 7: Comparison of reference mirror measurement using three instruments. All data is shown with a
false color image and a line scan through the center of the mirror sample. a) (top) WFPI snapshot (100ms); b)
(middle) Chromatic confocal at higher speed raster scanning #1 (2min [10]); and c) (lower) Chromatic confocal
microscopy #2 with lower raster scanning speed (11min [11]).
being capable of only measuring one other factors. In all data sets of the flat
point each time—needing a translation sample, one can clearly see there is some
stage when acquiring a surface profile warp in the single-digit micron range.
[9,10]. Several commercial reference However, this warp is still within the
mirrors (Table 2) with a known bow specifications of these samples. It is also
were used as target samples to compare clear that when a raster scanning system
the performance of WFPI against acquires data faster, the raster scanning
two commercial chromatic confocal noise increases. No raster scanning noise
microscopes [11,12]. The first set of data can be seen when using WFPI.
was done on a flat mirror, which means The last sample had bow of 98.01µm
there is no bow (i.e., radius of the bow is from the center low point to the edge
infinite, ∞) (see Figure 6). high point (radius of the bow is 3m, a
Some bow a nd wa r p a re to be smaller radius gives a higher bow and
expected even on a flat surface as a a large height difference between the
result of temperature changes, among lowest and highest points) (see Figure 7). Figure 8: Images of: a) intensity map, b) 3D depth
map and c) high-pass filtered NT and roughness.

25
Chip Scale Review January • February • 2020 [ChipScaleReview.com] 25