Professional Documents
Culture Documents
SPIEDigitalLibrary.org/conference-proceedings-of-spie
Event: Photonics West '96, 1996, San Jose, CA, United States
ABSTRACT
Dammann gratings are binary phase diffractive optical elements that may be used to split an
incoming beam of light into a one- or two- dimensional array of beams. The 1-D gratings can be designed
for different numbers of diffraction orders (N). For a given value ofN there are 21)possible 1-D solutions.
Each solution is distinguished by a particular sign combination of the phases of the diffraction orders and is
obtained by optimizing the N adjustable length elements in the grating design. The number of iterations
required and the likelihood of converging on a solution is strongly dependent upon the starting point of the
iterative process. In an effort to produce an efficient way of designing the gratings for larger N, a
methodology to select the starting point was developed. The relationship between efficiency and sign
combination of the phases was studied to determine which phase combinations were most likely to produce
higher efficiency solutions.
Keywords: Diffraction gratings, binary phase, array generation, beam splatter, multiple imaging
2. INTRODUCTION
Diffractive optics as a field includes the calculation, fabrication and application of diffractive
optical elements. In this project a binary phase diffractive optical element (DOE) has been designed to
function as an array generator. This array generator is an optical system that splits an incoming beam of light
into one- or two- dimensional array ofbeamlets. Applications of such a device are optical computing', image
processing2, fiber optic star coup1ers, coherent summation of incoherent laser beams4 and parallel laser
processing, to name a few.
Many different approaches, for constructing array generators have been proposed such as pulse
width modulation (PWM), pulse position modulation (PPM)5 and use of lenslet arrays6 as phase gratings.
One of the oldest and most successful approaches is the use of binary diffraction gratings proposed by
Dammann7. This approach is interesting because it is based on the use of phase gratings as compared to
amplitude gratings, which allows a higher efficiency because of zero absorption losses. Also, since these
gratings are composed of binary features it is easier to fabricate them using the well established fabrication
techniques such as photolithography and etching.
This study covers the investigation of various design parameters of Dammann gratings, such as the
significance of the phase combination of a particular solution, the corresponding efficiency and the smallest
feature size. The surface profile required for the Dammann gratings is determined using a non-linear
constrained optimization technique in which an optimum solution to a series of non-linear equations is
sought. A design for an array of 1 x (2N+l) beams involves N equations each with N unknown variables.
The optimization process, the choice of both a starting point and a cost function and their influence on the
computational complexity were also given due consideration. The purpose of this approach was to produce
an efficient way of designing these diffractive gratings for larger N and to develop a methodology of
selecting which phase combinations are most likely to produce efficient optimized solutions. An attempt was
made to efficiently solve for all the possible solutions for a particular N, ranging from N=l to N=12.
3. THEORY
3.1 Design
The pattern for Dammann gratings, for two dimensional design, can be described mathematically by
a binary transmission function G(x,y)8, which can be expressed as a product of two one dimensional
functions
G(x, y) =
(x)y() (1)
The grating transmission function is completely characterized by the N transition points where the
phase change from 0 to 71 or vice versa, occurs. For simplicity g(x) is considered to be symmetric9 about the
origin. Hence a Dammann grating with N transition points will have 2N+1 diffraction orders (-N ., -1, 0,
+1 , +N). The basic period ofthe transmission fimction is -0.5 x 0.5.
I [ Xn+l+Xnhl
N Ix- 2
g
x
(x) =
n=0 (l)'rec . (2)
Xn+1_Xn
[ [ jj
The amplitudes of the diffraction orders are obtained by the Fourier transform of the grating
transmission function. Applying the boundary conditions, x0 = 0 and XN+1 = 0.5, the amplitudes are given
by
A0 —
4n_1)n+lxn + (1)N (3)
2N n+1.1
Am (—i)
mr n=1 sm22rmxn) (4)
For a given phase of the zeroth diffraction order and for a given value of N there are 2N possible
phase combinations. If the grating transmission function is shifted in the -x direction by half a period, it
results in a solution that is essentially the same as the solution of the original grating7. Hence, the possible
number of essentially different solutions is reduced by a factor of two resulting in 2' essentially different
solutions.
For ease of identification each phase combination was represented by an integer, referred to as the
sign number (s). To obtain the actual sign vector associated with a particular phase combination, the integer
was represented by its binary conversion with all the zeros replaced by -1.
Consider a case of N=4, and a sign number s=7
binary conversion of s=7 is {0 1 1 1]
sign vector associated with the phase combinations will be [-1 +1 +1 +11
In the sign vector the least significant bit represents the phase associated with highest diffraction order.
0.6 X
x
0
0.55
X x
0.5 0
0.45 X
X x
0
3 0.4 0
0
0.35
0o X
0
0.3
X
0.25
0.2
0.15
u 2 4 6 8 10 12 14
s, sign no.
Fig 1. Efficiency versus sign number for N=5. 'x' for original cost function, 'o' for
alternate cost function.
ForN=7
0.8
0.7 X X x X
XX
0
X XxX 1
0.6 0
C.)
X 'x 0
XX
0
OX
x X0
00
0.5x°o ox0X x 0X xo0x x x x o
XX X
O
x0° 0 xXOX
X XOX x
XX X X
04 X X
X X X
O 0X 0 0 X
X o X
0.3 0
X 0 X
X 0 X
0.2
D 10 20 30 40
0—. 50 60 70
s, sign no.
Fig 2. Efficiency versus sign number for N=7. 'o' for original cost function, 'x' for
alternate cost function.
The practical design for a Dammann grating begins by the selection of a cost function
C(x1 2 XN ) that will be mized the N densional space with respect to the ansition
points. Various non-linear optimization techniques, such as the Newton Raphson method'° and simulated
annealing'1, have been used to solve this problem since no analytical solution is known.
The non-linear optimization technique used, was based on Sequential Quadratic Programming
(SQP)'2" methods. SQP methods, proved to perform very well in terms of efficiency , accuracy and the
percentage of successful solutions as compared to other methods for the particular problem solved here. The
reason could be attributed to the fact that the cost function used was also quadratic in nature. In SQP
methods, at each major iteration an approximation is made of the Hessian of the Lagrangian function using a
quasi-Newton method. This is used to generate a quadratic programming (QP) sub problem whose
solution is used to form a search direction for a line search procedure.
The cost function used was based on the mean square deviation ofthe diffraction orders from the
zeroth order.
A common goal in the design of Dammann gratings is to maximize the efficiency and to minimize
the maximum relative deviation of the intensity from the average power in the desired beams. In an effort to
achieve higher efficiency designs some other cost functions were investigated as possible alternatives. One
of the cost functions selected was to deliberately force the undesired higher orders towards zero. The
mathematical form is as shown
N1 - Am)\2 N+n 2
C(x) = + (6)
m=l 5mA0 m=N Am
This cost function introduces n more variables for the design of N diffraction orders. Hence, the problem
becomes more complex and requires a longer computation time than minimizing the mean square deviation
of amplitudes from the zero order. Forcing the higher orders to zero typically results in an increase in the
amplitudes of even higher orders thus lowering the overall efficiency (fig.l).
Since the zeroth diffraction order appears to be most sensitive to any fabrication defects it was
considered to force the zeroth diffraction order to zero and minimize the mean square deviation of the
amplitudes from the first diffraction order rather than the zeroth order.
C(x) = + A02
m2 (5mAi — Am) (7)
Though the starting point used for the main cost function no longer proved to be appropriate for the new
problem, for some sign combinations this cost function generated designs of overall higher efficiency. This
cost function would lead to a design generating an even number of spots with the central two orders more
widely spaced than the rest of them. Fig (2) shows a comparison between the efficiencies for this cost
function and the original cost function.
The choice of a starting estimate is very crucial for any iterative optimization technique. It has an
influence on the number of iterations taken to converge to an optimum solution and also whether the
problem converges to a reasonable result or not.
Unless there is an algorithm to precompute the positions of the minima of the problem, the starting
value is selected randomly. This is not a very efficient way of solving the complex optimization problem that
is being discussed here. In addition little effort has been reported in solving for all the possible 2'
solutions for a specific N. Hence an attempt has been made to develop a strategy, based on the choice of the
starting point, to solve all the possible solutions for a given N, in a systematic way. For a given N the first
solution, represented by s=O, the starting vector was generated by dividing the entire period (x=O to x=O.5)
into (N+l) equal parts. For all the subsequent sign combinations s=1 (2N1>.l a group of starting points
were used. These starting points were:
a) Solution ofthe previous sign combination.
b) Starting point for the previous solved sign combination.
c) Dividing the entire period into (N+l) equal parts.
Every possible solution was solved using all the three starting points in an attempt to solve as many
cases as possible. More than fifty percent ofthe solutions were able to converge using this technique. All the
solutions obtained from this first technique were then used as starting points for the more stubborn cases that
did not solve using the first technique. Using the combination of these two techniques, it was possible to
solve a majority ofthe cases (fig. 3). Most ofthe cases that did not solve at the first step were represented by
an odd integer (sign number).
,(,(, S $ S
0
0
—, 0
.E
0
0
0
80
C 0
o 75
c 70L x
C.)
0 65
x
.) x
60L
x
55
6 8 10 12
Fig 3 . Percentage of solutions solved versus N. x' after the first approach, o'
after the second approach
An attempt was also made to study how the starting point moves in the N dimensional space as it
approaches the optimized solution. It was observed that if the choice of the starting point leads to a solution
in a relatively small number of iterations, the distance of the starting point from the fmal solution steadily
reduces as it moves in the N dimensional space. If the starting point is not good, the distance from the fmal
solution oscillates for the first few hundred iterations before starting a downward trend and thus leading a
higher number of iterations, figs. 4(a), 4(b).
0
C)
0
C)
C)
No. of iterations
(a)
ForN=9, s=1
0
0
C)
0
0
C)
C)
No. of iterations
(b)
Fig 4. Distance of the point from the fmal solution in the N dimensional space
versus number of iterations. (a) For a good starting point, (b) for a not so
good starting point.
The performance of diffraction gratings can be characterized by the diffraction efficiency and the
array intensity uniformity The diffraction efficiency is defmed as the fraction of the incoming light that is
2
diffracted into the orders of concern. Since intensity and amplitude are related as Tm A m ' the
mathematical form ofthe efficiency looks like
2 N 2
= A0 +
2mi Am (8)
The maximum achievable efficiency for a one-dimensional grating is not strongly correlated with N and has
values between 60% and 80%. For a specific N, each of the possible 2' solutions is characterized by a
different efficiency.
It was observed from the studies of variation of the efficiency with the sign combination of the
beam phases that the lowest efficiency represents the case with alternating phases (sign combination +1 -1
+1 ..). It was also observed in each case that the corresponding sign number((N—i)
2 —
\2
O.5)x — provided
the lowest efficiency or one of the lowest efficiencies. This value of s also represents a local minimum in
efficiency with steadily falling values of efficiencies as this special value of s is approached. Another
observation was that the efficiencies steadily fall as the extremes of the sign numbers are reached, i.e.
S -3 0 or s —* 2(N—1) This is shown in figs. 5,6. There also seemed to be a general trend that for
higher efficiency solutions the number of iterations required to converge has a higher mean and a higher
standard deviation.
6. FABRICATION DEFECTS
Fabrication defects have a bearing on the performance of the gratings. During processing, because
of over- (or under-) exposure and development it is possible that the spatial features become modified. This
is possible due to undercutting also, in the case of wet etching processes. When the features do not line up or
in the case of overetching, the unetched features are a little smaller and the etched features are a little bigger
than desired. This means that the transition from 0 to t will no longer occur at the computed transition points
thus leading to some diffracted intensity variations.
When the features do not line up exactly, the two dimensional grating function can no longer be
treated mathematically as a separable function in the x- and y- directions. It cannot be separated into two one
dimensional functions. To analyze the effects of undercutting on the diffraction order intensities, the grating
function was considered to be a sum of two separable functions. The two separable functions can be
represented as
g1(x, y) = g1(x)gi(y) (9)
g2(x, y) = g2x(x)g2y(y) (10)
The mathematical form of the first grating transmission function looks like
(x÷1 + +
(N — x2j)1 + (x÷1 x21)1
2 2
g1(x) = + recll X21 — X21 — 26 (11)
i=O recll X211 — X2 — 28
L ] L i
0.6 -x X*XXX -
0.5 xX .XXX... -
XX '%xXX
CXXXX%X xx XxX)Xx
XX X
C.)
. 0.4
Xx,(X% X X.xx* xX
xxxxx
C.)
xXx )X x
x *X)1( X)CXX
XXX... XXXX*xX *) x
0.3
X
X XX X XC
X XX
XX
0.2
: :
¶
0.1 0 100 200 300 400 500 600
s, sign no.
N=20
0.135
X
0.13 X
X
X
0.124 X X.
X X
X X
x
X X
0.115 X XX
C.) X
C) x X
C.) 0.11 X X
XX
C) X
X
o.io4 X
X
0.1
XX
0.095
x
05 10 15 20 25
s, sign no.
30 35 40 45 50
The mathematical expression for the second grating transmission function looks like
g2(x) =
(N_iy [X — (x22 + x2+1) i
_ 28 red
[ + (x212 + x21+1) i
L j L ]
The amplitudes obtained from a similar analysis as for the first grating function are
(N-i)
(N-i)
(N-i)
Hence for the case when the features are not exactly lined up the intensity of the zeroth order is given by
(N-I)
2
A0(x) — A02(x) = — x21) — 48} —
{2(x2+2 — x21) — 48}]
=A0+26 (18)
As is clear from equation (1), with an increase in the undercut 6 , the features become smaller and
the intensity of the zeroth diffraction order increases causing the relative intensity of the other orders to fall
correspondingly (fig. 7).
0.8
0.7 x x
xxx
0.6 xx
C 0.5 xx x
'-C x x
0.3
0.2
)0xx x x
,oc
x X
x X'X
Diffraction Orders
Fig 7. Amplitude versus diffraction orders as obtained for undercut
The etch depth of the features determines the phase shift imposed on the incident light beam. The
mathematical expression for the phase shift () as a function of etch depth (1), is given by
= k0nl (19)
Ifthe etch depth is such that instead of introducing a phase shift of it, it introduces a phase shift of 0 = t —
intensities of the diffraction orders are altered. It has been shown that the zeroth order intensity increases
appreciably8 and the intensities for the higher orders reduce. It has been observed that this effect
is more pronounced in gratings designed for larger N as compared to smaller N.
7. CONCLUSION
A methodology to select the starting points for the iterative process involved in the design of these
gratings has been developed. We have been able to successfully extend this methodology to higher values of
N. If the choice of a starting point is good the distance of the point from the fmal solution progressively
decreases.
It has been observed that for a particular value of N, as the extremes of the sign number is
approached, the efficiency steadily drops. The lowest efficiency or one of the lowest efficiencies occurs for
2
the sign number s=(2'-O.5) . For higher efficiency solutions the number of iterations required to
converge to the solution has a higher mean. These results show that there is a connection between the sign
number, describing the phases of the diffracted orders, and the performance properties such as efficiency of
the optimized Dammann grating. These results have been obtained for small values of N (1 to 12), but are
believed to also apply to larger values of N.
These results are helpful in developing design strategies for selecting Dammann design solutions
that are more likely to produce desired features such as high efficiency. This is particularly helpful for larger
N where both the computation time for each solution and the number of possible solutions increase
exponentially with N.
This research was performed in the Microfabrication Applications Laboratory at the University of
Illinois at Chicago. The authors would like to thank Tony Cocco and John C. Pincenti for their valuable
suggestions and support. Financial support from the Aileen S. Andrew Foundation is also gratefully
acknowledged.
9. REFERENCES