Proximity effect correction calculations by the integral equation approximate solution method
-
Upload
celia-thomas -
Category
Documents
-
view
216 -
download
0
Transcript of Proximity effect correction calculations by the integral equation approximate solution method
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 1/6
Proximity effect correction calculations by the integral equation approximate solution
method
J. M. Pavkovich
Citation: Journal of Vacuum Science & Technology B 4, 159 (1986); doi: 10.1116/1.583369
View online: http://dx.doi.org/10.1116/1.583369
View Table of Contents: http://scitation.aip.org/content/avs/journal/jvstb/4/1?ver=pdfcov
Published by the AVS: Science & Technology of Materials, Interfaces, and Processing
Articles you may be interested in The representative figure method for the proximity effect correction [III] J. Vac. Sci. Technol. B 9, 3059 (1991); 10.1116/1.585369
Methods for proximity effect correction in electron lithography J. Vac. Sci. Technol. B 8, 1889 (1990); 10.1116/1.585179
Exact solution of the proximity effect equation by a splitting method
J. Vac. Sci. Technol. B 6, 432 (1988); 10.1116/1.583969
Approximate solution of Fredholm integral equations by the maximumentropy method J. Math. Phys. 27, 2903 (1986); 10.1063/1.527267
Solution of Linear Integral Equations Using Padé Approximants
J. Math. Phys. 4, 1506 (1963); 10.1063/1.1703931
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 132.239.95.91 On: Wed, 09 Apr 2014 06:30:39
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 2/6
Proximity effect correction calculations by the integral equation approximate
solution method
J.
M. Pavkovich
Varian Associates, Palo Alto, California 94303
(Received 17 June 1985; accepted 4 October 1985)
The
task
of successfully deal ing
with
the proximity effect problem involves
many
aspects, all of
which must be
dealt
with in
a reasonably satisfactory
manner. If one
chooses
to
correct
for
the
proximity effect by doing dose compensation one
major
task is
the
solution
of
the i.ntegral
equation whichdescribes
the
resultant exposure
in
terms ofthe incident flux of electrons.
One
well
known method
is
the
self-consistent
method. This paper
describes
a method which
provides
a
relatively accurate approximate solution
to the
integral equation which is easy to calculate and
which
provides
information on where
features
should be fractured to obtain good
dose
compensation.
Although the
relationship between
the incident
flux
and the resultant exposure
is
linear, the development process itself s not. This means that
the
usual integral equation should be
modified slightly so that
the
resultant exposure is defined in a
manner
which
more
closely
matches
the real problem. This paper will attempt to describe how information from the development
process
can be
used
to
define
the
exposure problem in
a manner which
provides
more
desirable
solutions.
I.
INTRODUCTION
In electron beam lithography
the
problem of proximity ef
fect compensation is well
known
and several techniques have
been suggested to solve it. These include shape correction,
dose compensation,z multilayer resist techniques,3 and
the
method
of equalization of
background
dose,4 or GHOST
method.
AU
of these
methods have theiradvantages
and dis
advantages. This paper describes a technique primarily use
ful with
the
dose compensation
method although it can
be
used in conjunction with other methods. This paper
de
scribes a
method
of solving
the
equation for the incident dose
in
terms
of
the
desired energy deposition in
the
resist.
The
solution is an approximate solution, but appears to be accu
rate
to within 2% or 3% of
the
correct solution.
U.
THE INTEGRAL EQUATION AND ITS SOLUTiON
The energy deposition in a resist layer on top of a silicon
substrate subjected to a patterning e-beam
can
be described
by
the
following equation:
E x,y)
= KJ x - x ,y - y )i(x ,y )dx dy , 1)
where
K
is a
constant
relating
the
energy deposition
in the
resist
to the incident current
density,
f(x,y)
is
the incident
current
density, f x,y) is the function which describes
the
distribution of energy which would be deposited
by an
inci
dent delta function
of
current. I t is normalized so
that
f f x,Y)dX dy = 1
2)
t will
be
assumed that
f x,y)
is adequately described by
a
double
Gaussian-type
function
which
will
be written
in
the
following form:
f x,y)
= exp - - --.:..-=--
(1) (
x
2
+ 2 )
1+ 7J 7T{ } f3
}
17(1)
( X 2+ y 2 )
- 1 + 7J rrf3 xp {3
.
3)
In this equation /3 = width parameter of forward scattered
energy,
{3
b
= width parameter
of
backward
scattered ener
gy, and
J = ratio
of total energy
in backward portion to the
total
energy in
the forward
portion. t
should be noted that
K,
3
f,/3
b and 1] are also all functions ofz the vertical coordi
nate
in
the
resist. For
the moment
however, we will assume
that
a
z coordinate can be chosen which adequately charac
terizes
the
average
energy
deposition vertically.
The
backward
scattering
parameter
{J b
can be
of
the
order
of 1 or 2
J1
and is
the
source of
the proximity
problem.
The
backward scattered energy is quite broad and slowly varying
spatially
and thus
generates
a background
level of energy
on
which the more sharply
defined energy deposited
by the
for
ward scattering rests.
The forward scattering paramet er f f is small compared to
the minimum
feature size.
This must
be so, otherwise
there
would
be no
hope
ofeven
doing lithography
since
one
feature
edge would directly affect another. However, {3 does affect
the
sharpness of
the
energy profile
at the
beam edge. More
over,
the
steep
portion of the
energy deposition profile is
determined
not
only by
the
broadening due
to
the
forward
scattering, but by the profile of the incident
beam
itself. Since
both
are
usually
represented
by
Gaussian-type
functions,
and the convolution
of
Gaussian functions is still Gaussian
it
is appropriate to approximate the forward scattering por
tion of
the
distribution function by a
delta
function. The
forward scattering,
although
still
important can
be consid
ered to contribute
to simply a less
sharply
defined
incident
beam. Also it is more appropriate to consider the beam edge
broadening problemseparately from
the
proximity problem
and
define
the proximity problem
as
the problem caused by
only
the broad
background scattering.
Thus
the distribution
function for
the
proximity problem
can
be written as
159 J.
Vac. ScI.
Technol
B .. (1),
Jan/Fab
1986
0134-211)(/86/010159-05 01.00 @
1986
American
Vacuum SOCiety
159
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 132.239.95.91 On: Wed, 09 Apr 2014 06:30:39
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 3/6
160
J. M. Pavkovich: Proximity effect correction calculations
f(x,y) = _ 1 _
8
(x,y) + _ YJ-S(x,y),
1
+ YJ
1 + YJ
(4)
where
S(x,y) = exp - ,
X2
+
y2
17 3 f
(5)
and will be referred to simply as the spreading function. By
writing things in this manner, we can separate the proximity
problem, which is really global in nature, from the precise
shape
ofthe
beam edge, which is really a
rather
local prob
lem. With this simplification, Eq. 1) can be written in the
form
E
(x,y)
=
K [_1_ i(x,y) +
_TJ_
1+TJ 1+TJ
X
II
(x
-x ,y - y )i(x ,y )dx d y l (6)
I f we further consider only one dimension, the above equa
tion becomes simply
E (x)
=
K
[_1_
i(x)
+
_TJ_
f
(x
- X )z (X )dX ],
(7)
1+1] l+TJ
This equation describes
the
resulting exposure from an inci
dent beam,
For the moment consider the three limiting cases depicted
in
Fig.
1.
Case I is a very narrow isolated line. If the line is
very narrow, there will
be
almost no contribution to the
background and thus the resultant exposure will consist only
of the first term in Eq. 7). Case II is a narrow gap between
two infinitely large regions. In this case, the background is at
its maximum value over the whole region. Thus even in the
gap there is essentially a full contribution from the second
term in Eq.
7).
Case
III
is the edge of an infinitely large
region.
At
the edge, the value
of
the integral in the second
term of Eq. 7) is 0.5.
Thus
the resultant exposure has the
step caused by the first term sitting
on
top of he contribution
of he second term. Far outside the region the contribution of
the second term will drop
to
zero.
Far
inside, the contribu
tion will approach T I
(1 +TJ)·
The proximity problem arises because of the significant
change in the profile and level
of
the energy deposition when
the feature size approaches the value of b When the feature
size is large, all the edges have an energy profile similar to
that depicted in case III. Even though the background expo
sure level is present, it has a relatively constant value at all
edges.
I t
should also be noted that the method of equaliza
tion
of
background dose attempts to make the edge profiles
all look like those for case II.
I t
s important to realize that it
is not whether the edge profiles are good or bad in some
sense,
but
whether they are all similar
that
is important.
In the traditional method of dose compensation, the ob
jective is to choose the incident current so
that
the resultant
exposure is uniform in all exposed areas. No constraints are
placed on the exposure in unexposed areas. Equation
7)
can
not be solved for i(x) since one does not know how to choose
(x) outside
of
features
to
generate
an
acceptable non-nega
tive) solution for i(x). Thus to obtain the equation to be solved
for
i(x),
one must multiply Eq.
7)
by
P (x),
where
P
(x) repre
sents the patterns to
be
exposed and is equal to 1 inside a
J. Vac. Sci. Techno . S, Vol. 4, No.1, Jan/Feb 1986
CASE I
NARROW
LINE
w4 i
i3b
l
CASE II
NARROW GAP
Hli
n-,l, .
I I
w __
w4 i
ilb) I H I I l IHIIIIIIHII
UK
I --;-+:;) 0
I I
I I
I I
, I
I
l
, I
I ,
_w_
CASE III EDGE OF LARGE FEATURE
IHIIIIIIIIIIlIl
K
1 ( 7) .
2 1 +
r;
0
160
FIG.
1. Three
limiting cases illustrating
the
proximity effect. Case I is a very
narrow line. Case II is a very narrow gap between two large features. Case
III is
the
edge of
an
isolated large feature.
feature and 0 outside. The result is
EoP(x) = K [_1_ i(x)P(x) +
_TJ_P(x)
l+TJ 1+1]
X
I (x - X )i(X )dX ] '
(8)
where Eo is the desired value of exposure within a feature.
Since
P
(x) is 0 where i(x) is 0, and 1 where i(x) is not zero, Eq.
8)
can be written in the form
aP (x) = K [_1_ i(x) + _ 1 ] _
I+TJ
I+TJ
X
I (x)S (x - x )P (x )i(x )dx ].
(9)
This is a linear integral equation of the second kind with a
symmetric kernel. Techniques are available to solve such
equations,
but
they are not very useful for the problem
at
hand. Moreover, the kernel is a function ofthe pattern and
thus there is no general characteristic type transfer function
relating
output
to input as one sees with linear filters.
The situation, however, is not hopeless. The functions
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 132.239.95.91 On: Wed, 09 Apr 2014 06:30:39
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 4/6
161
J. M. Pavkovich: Proximi ty
effect
correction calculations
within the integral are all well behaved and never negative.
The form
of
the integral is like a convolution which for this
case tends to make the result more slowly varying than the
input functions. The spreading function S
(x
-
x')
is a local-
ized function and thus tends to smooth out the variations in
i(x).
Moreover,
it
is really the changes in
Pix)
which contri-
bute most to the variation
of
the value
of
the integraL Thus
one is led to consider moving
i(x )
outside of he integral; thus
I
(x)S (x - x')P (x')i{x')dx' --+P (x)i(x)
J
(x - x')P
(x')dx'.
10)
With this approximation, the integral can be evaluated and
the resulting equation solved for
fIx).
The resulting approxi-
mate solution is then given
by
i(x)
=
(EoIK)P(x) _
(1/1 +
11)
+ ( 11/
1
+
I])(p*S)
11)
where we have written (P *S) for the integral since it now
truly represents the convolution of the pat tern function
P
(xl
with the spreading function S (x).
This approximate solution has the correct limiting behav-
Ior. In the case
of
an isolated line, (P S is essentially zero
and the dose should be increased by a factor
of
(1
+
1J).
In
the case
of
a small gap between two large features,
(P
>I<
S
is
close to 1 and thus there is no need to increase the incident
dose. Since this approximate solution was not derived ana-
lytically and since it is impossible to solve the integral equa-
tion analytically, it is somewhat difficult to assess its accura-
cy.
In
order to get some idea of the quality of this
approximate solution, a program was written which calcu-
lated both the approximate solution as well as determined
the correct solution numerically for arbitrary cases of lines
and spaces. Also
it
was possible to consider slight modifica-
tions to the approximate solution.
The kinds of modifications considered were adding terms
of the form [V (P >I<
S
F or
[ V
(P
*S 1where
' . a . a k a
v = l - J - .
ax Jy az
Such terms
will
be zero when
(P S)
is constant and are coor-
dinate system independent. This effort produced two results.
First, the edge of a large region really represents one of the
most difficult cases since it represents the fastest transition
from a region where (P S
is
1 to a region where it is O
Secondly, the results of various test problems indicated that
the above approximate solution could be improved slightly
by adding a term
of
the form
_ I1_)2[V(P*SlF
1+1]
12)
to Eq. 11). Thus the recommended approximate solution is
given by
r 1
i(x) = EoIK p x l--------
. [(l/1 +
1])
+
(1]/1
+ 1J)(P*S)
+
c: 1JY[V(P>I<
S
W} .
13)
J.
Vac. ScI. Techno .
e,
Vol. 4, No.1, Jan/Feb 1986
161
TABI.E 1. Comparison of
the approximate
solution
and
the numerical solu-
tion for the case
of an
isolated edge for increasing values
of
11.
Maximum
Maximum
Edge
dose
exposure
exposure
T
error
error
error
(%)
(%) (%)
0.75
2.6
2.1
0.75
1.00
3.5
2.8
0.37
1.25
4.5
3.6 0.20
LSO
6.2
4.2
0.86
III. COMPARISON WITH NUMERICAL SOLUTIONS
For
the proximity problems in which we are interested,
values
of 11
range up to around 1.0
or
perhaps a little more.
At a value of 1.0, halfof he deposited energy
is
going into the
background. It should be expected that as the value
of
'I]
increases, the accuracy
of
the approximate solution should
decrease.
In
this work, three measures were used to compare
the approximate soluti.on with the numerical solution. They
were, first, the greatest error between the correct and ap-
proximate solution, second, the greatest error between the
resulting exposures, and third, the difference between the
resultant exposure at the edge. The results are tabulated in
Table I for various values
of
1]. As can be seen, the results are
reasonably good, even for values
of 11
greater than
1.
Also
plotted in Fig. 2 is a plot of the correct numerical solution at
the edge of a large region for 1] =
1
IV.
CALCULATION OF THE APPROXIMATE
SOLUTION
The calculation
of
the approximate solution is quite
straightforward. One first overlays the pattern with a square
grid
of
mesh points. The size
of
the mesh is not related to the
feature size, but rather is related to
f
b the backward scatter-
ing width parameter. We wish to calculate the valueof P *
S )
at each mesh point. t is important to think
of
the mesh
points as sample points at which the value of (P
*
S is calcu-
lated rather than an average over some area. The value
of
(P
*S) at any arbitrary point can then be found by interpola-
tion. Since the spreading function S (x) is extremely wen be-
haved, (P
*S)
is also extremely wen behaved, This means one
can use a very coarse mesh and a high order interpolation
INCiDENT
DOSE
DEPOSITED
ENERGY
1.414
'I =
1
0.293
FIG. 2. Numerical solution of he integral equation illustrating the incident
dose and the resulting deposited energy for the edge of a large isolated
feature.
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 132.239.95.91 On: Wed, 09 Apr 2014 06:30:39
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 5/6
162
J. M.
Pavkovich: Proximity
effect
correction calculations
formula to find the value of(P
*S
at an arbitrary point,
Thus
how one chooses the mesh size is a compromise between the
number of mesh points at which one calculates
(P ..
S and the
type of interpolation one wishes to use. t is likely that the
optimum approach is to use a very coarse mesh initially
and
then generate a finer grid
of
points by interpolation so that
simple interpolation can be used to evaluate
(P
*S at arbi
trary
points.
The calculation of
(P
*S) at the grid points is very simple
for rectangles. The formula for the value
of
P
*S at an arbi
trary point is given by the following equation:
P *S) = [erf X 2 p ~ X _
rf
X l p ~ X
]
X ~ [ e r f Y ~ ~ Y ) - e r f Y ~ ~ Y ) l
14)
where erf(x) is the
error
function and is defined by
erf(x) = - e u
2
du
LX
[iTo
15)
and the coordinates
of
the rectangle are defined by
x
X
•
Y
and
Y2
Notice that the calculation is the product of two
terms where each term is dependent on only one coordinate.
This means
that
one can calculate a factor to be associated
with a row of grid points which depends only on they coordi
nates of a rectangle. A factor can also be calculated for a
column which depends only on the
x
coordinates of a rectan
gle. The contribution at each grid point is then simply the
row factor times the column factor.
After calculating (P
*S
at all the grid points, the values at
each grid point can be converted to values
of
the approxi
mate solution and thus the desired values of incident current
intensity.
The
more difficult
part of
applying this method
is
then to
determine how to actually do the exposure since even vector
scan machines cannot vary the exposure continuously. Typi
cally, each rectangle which is exposed can have its own level
of ntensity, but the intensity is constant across the rectangle.
This means that one must determine the appropriate average
intensity to use in a rectangle and whether or not to fracture
any particula r rectangle into two or more pieces to achieve
the best appropriate exposure.
How
such decisions are made
depend on defining rules of thumb to build into a program
which define how great the deviations from the continuous
solution can be. This, in
tum,
depends on understanding the
development characteristics ofthe resists
5
and
how great the
latitudes are in the development process.
V. WHAT IS THE CORRECT PROBLEM?
Thus far it has been assumed that the goal
of
the exposure
process was to achieve uniform deposited energy within each
feature. This in fact may
not
be the best goal.
In
a region
of
high background, the edges will continue to develop
at
a
faster rate than where the background is low. t may there
fore be worthwhile, in fact, to require that the total resultant
exposure decrease as the background level increases. Thus
one would replace the left-hand side of Eq, 9) by a term of
the form
J. Vae. Sci. Techno . e, Vol.
4,
No,
1,
Jan/Feb 1986
162
16)
where
k
=
a factor which relates how the deposited energy
should be reduced as the background level increases.
The
effect of introducing this factor in the equation is simply to
increase the effective value of J and to modify the definition
ofEo. The same approximate solution techniques can still be
used although the constants change slightly.
Another
choice might be to define the problem to be
solved in a manner which tried to keep the center
of
the
forward scattering portion
of
the energy deposited at a con
stant level. This approach would be useful with a resist
which
had
a very high contrast. The important point is that
defining the problem to be solved depends to some degree on
the resist characteristics.
VI. CONCLUSIONS
An approximate method of solving the integral equation
relating the incident current density to the deposited energy
has been described. This solution technique appears to give
quite good results for the values of parameters typically en
countered in practice. This technique should be useful
if
one
wishes to do proximity correction using dose compensation.
However, the dose compensation technique requires good
understanding
of
the resist development characteristics in
order to determine a set of development parameters which
provide sufficient latitude. Also i t should be clear from this
work that one should use parameters which characterize the
deposited energy and background level when studying resist
development. Incident current density is
not
an appropriate
parameter.
The
GHOST
technique of proximity correction is clearly
the method
of
choice in a raster scan machine. Although
approximations are made in that method in terms
of
achiev
ing background dose equalization, it is likely
that
the lati
tudes required in the resist development are smaller than
with dose compensation. However, in a vector scan machine,
dose compensation may provide faster throughput than the
GHOST method. Whether
or
not this is true depends largely
on
the characteristics of the resist since it is the resist charac
teristics which determine how
much
fracturing is necessary.
The technique described ignored the effect
of
forward
scattering. Forward scattering can dearly be a problem,
even the dominant problem for very narrow lines and thick
resists. Moreover, ifbias
6
is used, the lines which
are
exposed
become even narrower, which compounds the seriousness
of
forward scattering. The problem of forward scattering
should be treated separately from the backward scattering
problem since it is a local problem and occurs because the
region related to the development of a line edge directly
overlaps the region associated with the development of the
edge of a neighboring feature.
Thus
one should compensate
for forward scattering by performing some form
of
shape
adjustment. This in
turn
could depend somewhat
on
the lev
el of background present. Again, it is true that any shape
adjustment requires understanding the resist development
process and how the energy density and background level
affect
that
process.
Redistribution subject to AVS license or copyright; see http://scitation.aip.org/termsconditions. Download to IP: 132.239.95.91 On: Wed, 09 Apr 2014 06:30:39
8/19/2019 Proximity effect correction calculations by the integral equation approximate solution method
http://slidepdf.com/reader/full/proximity-effect-correction-calculations-by-the-integral-equation-approximate 6/6
163
J.
M.
Pavkovlch: Proximity
effect correction
calculations
IN.
D. Wittels
and C.
I.
Youngman in
Proceedings
of
he 8th International
Conference on Electron
and
Ion eam Science and Technology
edited by
Robert
Bakish (The
Electrochemical
Society,
New York
1978),
p.
361.
2Mihir
Parikh J.
Appl.
Phys.
SO 6,
4:>.71
(1979).
3J
B. Kruger
P.
Rissman
and
M. S. Chang J.
Vac. Sci. TechnoL 19, 1320
(1981).
J. Vac Sci Technol S Vol
4
No.1 Jan/Feb 1986
163
4G. Owen
and
P. Rissman J.
Appl. Phys.
54, 6, 3573 (1983).
5A.
S. Chen
A.
R. Neureuther andJ.M.
Pavkovich
J.
Vac. Sci. Techno . B
3, 148 (1985).
6M.
G. Rosenfeld, A. R. Neureuther,
and
C. H. Ting, J. Vac. Sci. Technol.
19, 1242 (1981).
Redistribution subject to AVS license or copyright; see http://scitation aip org/termsconditions Download to IP: 132 239 95 91 On: Wed 09 Apr 2014 06:30:39