semi合集-English.pdf - 第6000页
SEMI P43-0304 © SEMI 2004 9 Figure 12 The 4 Default Orientati ons of a 90º Corner 8.2.3.2 corner are a gain — special case of clipped feature area gai n , in wh ich the region of interest contains one corner of a feature…
SEMI P43-0304 © SEMI 2004 8
(clipped) feature area loss. Note that the value of the
absolute (clipped) feature area deviation is always
positive. Same mandatory information as in (clipped)
feature area gain.
NOTE 17: This qualification parameter is recommended for
features for which it does matter where the gain and loss are
situated.
NOTE 18: This reasoning could be extended to Pattern area
gain /loss /difference /deviation if several features are present
in the region of interest.
NOTE 19: Edge roughness may influence feature area
deviation, unlike feature area difference, in which the effect
of edge roughness is filtered out by allowing the resulting area
gain and area loss to cancel out (see Figure 9).
NOTE 20: Normalized feature area, normalized feature area
gain, normalized feature area loss, normalized feature area
difference and normalized feature area deviation are defined
as the ratio between the actual value of the considered
parameter and the nominal feature area.
8.2.3 Specific Case of Corner Rounding (CR)
8.2.3.1 corner rounding — deviation of an actual
feature corner from the nominal one.
8.2.3.1.1 The definition above is qualitative, and can be
quantified in all practical cases by treating it as a
special case of feature area difference (see further in
Sections 8.2.3.2–7). Quantitative determination
requires as mandatory information:
• the designed angle (see Figure 10), DEFAULT
angle is 90 degrees.
• feature tone: dark vs. clear.
• corner type: outer vs. inner. Corner type is ambig-
uous without feature tone specified (see Figure 11).
• other features or feature corners within the
proximity range from the corner of interest,
DEFAULT is an isolated corner.
• orientation of the corner (determined by the 2
linear sections, e.g., 90^ 180, 160^ 210,…, (see
Figure 10). DEFAULT is all 4 orientations with
linear sections along X- and Y-axes (Figure 12) for
both dark and clear feature tone.
NOTE 21: The purpose of the default corners is to help the
user of this document to select representative features to
qualify, for example, the corner rounding fingerprint of a
mask making process.
Y
X
α
β
(α , β)
Figure 10
Nomenclature of Feature Corners by the Angles of
Their Linear Sections
(a)
(b)
Figure 11
Illustration of nomenclature used for feature
corners (elbow as example)
(a) dark feature; (b) clear feature;
bottom left in each sub-figure: inner corner;
top right in each sub-figure: outer corner.
SEMI P43-0304 © SEMI 2004 9
Figure 12
The 4 Default Orientations of a 90º Corner
8.2.3.2 corner area gain — special case of clipped
feature area gain, in which the region of interest
contains one corner of a feature.
8.2.3.3 corner area loss — special case of clipped
feature area loss, in which the region of interest
contains one corner of a feature.
NOTE 22: In non-corrected isolated cases normally there is
only a loss for outer corners and a gain for inner corners. In
other cases (i.e., corrected corners) there may be a loss and a
gain, which makes it necessary to also define the difference
and deviation, as done below.
8.2.3.4 corner area difference — corner area gain
minus corner area loss. As such it is a special case of
clipped feature area difference, in which the region of
interest contains one corner of a feature.
8.2.3.5 corner area deviation — the sum of corner
area gain and corner area loss. As such it is a special
case of clipped feature area deviation, in which the
region of interest contains one corner of a feature.
NOTE 23: As the corner shape is expected not to be decisive
for its printability (at least for an isolated corner), but rather
the balance between area gain and loss, it is recommended to
use area difference for corner qualification rather than area
deviation.
NOTE 24: Certain features with non-isolated corners, such as
line-ends and contacts, have specific definitions listed in
Sections 8.2.4 and beyond.
8.2.3.6 In present practice, corner rounding quantifica-
tion is done without comparison to the nominal corner.
However, this is only valid if the corner is isolated,
non-corrected, and its area gain is negligible. Typically
it is done based on the determination of a reference
corner which is obtained by the extrapolation of the
linear sections of the corner, if present (Figure 13).
Figure 13
Corner Rounding Determination by Extrapolation
of the Linear Sections of the Actual Feature Contour
8.2.3.7 equivalent corner rounding radius (ECRR) —
an equivalent, effective corner rounding radius
calculated from the area difference. It assumes that the
corner is a circular arc. The ECRR is calculated as
ECRR = sqrt( 4 * corner area difference / (π - 4) ), for a
90 degree corner.
8.2.3.7.1 The ECRR is defined only for negative corner
area differences, i.e., where the corner area loss is
larger than the corner area gain.
NOTE 25: This definition actually gives a 1D representation
for a 2D quality assessement, but it is found useful when
comparing mask quality to wafer printing results, which are
typically characterized by 1D measurements, such that a
dimensionless MEEF (mask error enhancement factor) can be
used. As with area based assessment, also this term
disregards the shape at the feature corner.
NOTE 26: Current methods to determine a corner rounding
radius based on fitting a circle to an actual corner have been
experimentally shown to deliver unreliable results and are
therefore strongly discouraged.
8.2.3.8 corner pull-back (CPB) — the distance
between the reference corner and the actual feature
contour
. This may be based on the minimum distance
(minimum CPB) or that determined along the bisectric
(bisectric CPB) (see Figure 14). The choice of CPB
technique is mandatory information.
NOTE 27: Edge roughness may have an important influence
on the corner pull-back, such that contour averaging may be
necessary to produce a meaningful result. The method of
contour averaging is mandatory information for corner pull-
back, if used.
SEMI P43-0304 © SEMI 2004 10
Figure 14
Corner Pull-Back: Bisectric (Full Line Arrow) vs.
Minimum (Dotted Arrow)
8.2.4 Specific Case of Line-End Shortening
8.2.4.1 line-end shortening — deviation of the actual
feature from the nominal feature at the nominal line-
end. This is still qualitative, and can be quantified in
general cases by overlaying the actual line contour to
the nominal line (see Section 8.4). Alternatively, a test
pattern such as Figure 17 may overcome the need to
overlay to the nominal case.
NOTE 28: “line-end” is also used for the darkfield case (for
spaces).
NOTE 29: In non-corrected isolated cases normally there is
only a loss (or shortening). In other cases (i.e., corrected line-
ends) there may be a loss and a gain (or extension), which
makes it necessary to also define the difference and deviation,
as done below.
NOTE 30: The minimum (= DEFAULT) region of interest
must include all feature area divergence until the feature can
be treated as one-dimensional (see Sections 5.3 and 5.4).
Figure 15
Quantification of line-end shortening by pull-back
(arrow) or area comparison. Top: shortening case,
below: extended case (e.g., caused by
overcompensation)
Figure 16
Quantification of line-end shortening by
equivalent line-end shortening (shaded area is equal
in the two figure halves)
Figure 17
Proposed test patterns to overcome the need to
overlay to the nominal case for the determination of
line-end shortening
8.2.4.2 line-end area gain — special case of clipped
feature area gain, in which the region of interest
contains a line-end.
8.2.4.3 line-end area loss — special case of clipped
feature area loss, in which the region of interest
contains a line-end.
8.2.4.4 line-end area difference — line-end area gain
minus line-end area loss. As such it becomes a special
case of clipped feature area difference, in which the
region of interest contains a line-end.
8.2.4.5 line-end area deviation — the sum of line-end
area gain and line-end area loss. As such it becomes a
special case of clipped feature area deviation, in which