You are on page 1of 8

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL 7. NO.

9, SEPTEMBER 1989

1343

Measurement of Optical Waveguide Coupling Coefficients Using Multiple Waveguide Systems

Abstract-A method for simply and accurately determining the coupling coefficient between integrated optical waveguides is presented. Measurement of only a few features of the output power distribution from a linear array of identical guides is required. The result is most accurate for long samples, but approximates the coupling coefficient to well within a few percent even for samples shorter than one two-guide coupling length. Since waveguides in the array would be fabricated in the same way, perhaps on the same substrate, as the device structures of interest, the method ought to be a useful and convenient one for device designers.

I. INTRODUCTION RACTICAL designs of many integrated optical structures require knowledge of the coupling coefficient between neighboring parallel waveguides. This is usually a difficult quantity to estimate with the required accuracy; moreover, the fabricated waveguide geometries and their propagation constants are not always controllable to the required extent. In this paper we present a method for measuring the waveguide coupling coefficient with precision, using a simple test structure fabricated along with the devices of interest. It eliminates the need for more than a rough a priori estimate of the coupling coefficient and gives an accurate result which compensates for the fabrication process being used. A linear array of a large number of identical singlemode waveguides, with identical spacing between them, has a well-defined output distribution when light is incident into the center guide of the array. This distribution is governed by the way in which light is coupled from the center guide to the two adjacent guides, then again from those to their nearest neighbors, and so on, as shown in Fig. 1. The power output of guide +_n,where the center guide is labeled guide 0, is given by the squared Bessel where K is the coupling coefficient function .: K Z e ) I~ ( between adjacent guides, (Y is the single-guide attenuation coefficient, and z is the length of the sample [ 11. In weakcoupling theory, the waveguide output distribution is independent of loss for a uniform array, as each guide experiences the same attenuation over the sample. In terms of the two-guide coupling length L,, defined by L, =

Fig. 1. Structure of channel optical waveguide coupler array

a / 2 ~the argument is simply ~ K = 2 ( a / 2 L c ) 1 L , = l a , , Z where the length of the sample is expressed as some number 1 of these coupling lengths. Since the sample length is known, the form of the output distribution of light from an array of guides is thus a measure of the coupling coefficient. In Fig. 2 the outputs for several successively longer arrays are plotted as bar graphs, showing how the maximum output power appears in a waveguide successfully farther from the center. This reflects the fact that the Bessel function of order rz is close to zero for arguments smaller than rz, climbs to a positive peak near the point where its argument is equal to n , then falls off and oscillates about zero. This oscillation also produces alternating light and dark signals from guides near the center of the array. The

'

Manuscript received October 28, 1988; revised January 24, 1989. The authors are with AT&T Bell Laboratories, Holmdel, NJ 07733 IEEE Log Number 8928063.

'The above form of the output distribution function, along with the results herein which make use of it, applies only in cases where standard coupling-of-modes theory is valid. Where the geometry precludes the use of the simple formalism, the Bessel function must be replaced by a new expression for the guide output powers derived with cross-power terms taken properly into account [2].

0733-8724/89/0900-1343$01 .OO 0 1989 IEEE

1344

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7. NO. 9. SEPTEMBER 1989

Waveguide label

(a)
0.12

I
k

0.08

t 0.10

b n

5
U

0.06

0.08 0.06

Q
c c

Q c

n
0.04

fi .b z

0.04
0.02
n nn

.fi -

0.02

0.00

Waveguide label

(d) Fig. 2. Bar graphs indicating the peak output power of waveguides in successively longer arrays: (a) 0.5 two-guide coupling lengths, (b) 1 .O coupling length, (c) 3 . 0 coupling lengths, (d) 6 . 0 coupling lengths.

largest outputs appear in guides + 2 when the sample length is exactly equal to one coupling length Lc. 11. EXPERIMENT To demonstrate the method, arrays were fabricated using polymethyl methacrylate (PMMA) channel waveguides, formed by spinning onto a pattern of stripes etched into a thick p-glass film on a Si wafer substrate. The structure is an inverted rib and is shown schematically in Fig. 3 . The dimensions were chosen within a simple effectiveindex model [3] to achieve single-mode guiding at 1.06pm wavelength and a two-guide coupling length under 1 cm. The arrays were about 5 mm wide (more than 600 guides in the array), and the substrate wafers were cleaved to give array lengths up to 6 cm. Light from a Nd : YAG laser was focused through a microscope objective onto the cleaved end of each sample, into one of the guides near the center of the array. The waveguide outputs were projected onto a screen and observed with a CPD video camera. The video image was digitized and a scan was taken of the intensity along a line through the array output spots to measure the relative peak heights. The sample was then moved across the input beam to excite successive guides along the array, in order

Fig. 3. Schematic drawing of the inverted rib waveguide structures used to demonstrate the coupling coefficient measurement. The guides were approximately 2 pm high by 4 pm wide in the example; the separation between waveguides in the array was 8 pm.

to obtain a set of output scans large enough for meaningful statistics to be gathered. In order to measure the coupling coefficient from an output observation such as the typical one shown in Fig. 4(a) for the PMMA waveguide array, the pattern of spots must be matched to the theoretical result, a set of squared Bessel function values for the unknown argument 2 ~ 2 , with positive integer orders set equal to the waveguide label magnitudes. The pattern is symmetric about guide 0, and with a scan of the intensity distribution as in Fig. 4(b) one can identify the output peaks with integer labels

GABRIEL A N D WHITAKER: MEASUREMENT O F COUPLING COEFFICIENTS USING MULTIPLE WAVEGUIDE SYSTEMS

1345

IO
I I I I I I I I I I I L

50

100

DISTANCE (arb. units)

(b)

b n
c

0.06

9 .

.2

0.04

E 5

0.02

n nnI _."..

r
I

n
. .

- - - - - - _ _ - I

o
.

-..-------o N _ o ,

Waveguide label

(C) Fig. 4. Typical waveguide array output power distribution. (a) Photograph of the pattern of output spots projected onto a screen. (b) Scan of a video image of the output distribution, taken along a line through the centers of the spots. (c) Bar graph of peak waveguide outputs computed for the mean value of the coupling coefficient for 20 data scans.

starting with f 1. One way to determine the coupling coefficient from this data would be to calculate the squares of the Bessel functions for all orders n , using various trial arguments, and plot bar graphs as in Fig. 4(c). The bar heights could then be compared with the corresponding measured peak output powers. The coupling coefficient estimate would be chosen as the one for which the argument ~ K gives the minimum mean-squared error. It is Z easier, however, and just as accurate to determine K using the properties of Bessel functions along with the measurement of just three of the guide outputs.
1 1 FIRSTAPPROXIMATION THE COUPLING 1. FOR COEFFICIENT

power gives a first level of approximation for K or I , through the relationship of the peak value of a Bessel function to its argument. To obtain a simple formula for this relationship, note that the Bessel function of large positive integer order, close to the point where the argument equals the order, is well approximated by an Airy function, using the formula [4]

J,(n

+ . $ r ~ ' / ~= 2'/3np1/3Ai(-21/3.$)+ o ( n - ' ) . )


(1)

Since the Airy function has its peak at the point where its argument is very nearly equal to - 1, solving for .$ gives the Bessel function argument at the peak:
2KZ

The maximum output appears in a given guide for only Z a narrow range of the argument ~ K = 7 r l . Therefore, simply identifying which guide n,, contains the largest output

np +

($)

'/3

1346

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7. NO. 9. SEPTEMBER 19x9

where n,, is the label of the guide where the maximum output occurs. Because of the assumption above, this approximation works best when np is large; however, as long as the sample length is greater than about 80 percent of L,, i.e., for n,, I 2 , the approximation predicts the argument at the peak to within 2 percent. For short samples where n,, = 0 or 1, a different approximation, described in Section V , should be used to extract 2 ~ 2 . As a typical example, for a PMMA waveguide array of length 2.85 cm, 20 scans of the output were taken, for excitation successively into adjacent guides. This produced 40 sets of output peaks, one for each half of the array in each case. The maximum output appeared in guide f 14 in 38 of the cases, and was in guide 15 in the other two. The argument x = ~ K Z as computed from the , approximation above, had mean X = 15.965 and standard deviation a, = 0.228 (1.4 percent of the mean) for the set of scans. IV. INTERPOLATION The range of 2 ~ over which a given waveguide n,, con2 tains the highest power of the array is bracketed by the points at which the power in guides n,, - 1 and n,, or n,, and np 1 are equal. As the distance I increases throughout this range, power is coupled from guide n,, - 1 to n,, 1 without much change in the output of guide n,,. We define the crossoverpoint for guide n,, as the argument for which guides n,, - 1 and n,, 1 have equal output power; this is also the point where the power in guide n,, reaches its exact peak. The first approximation above estimates the crossover point and makes explicit calculations of the predicted outputs for n,, - 1 and n,, 1 unnecessary. The quality of this estimate improves as n,, increases, as shown by a plot of the error for values of n,, between 1 and 30 in Fig. 5 . In most cases, however, the array length will be somewhere between the crossover points for adjacent guides. The next level of approximation is a correction to the estimate above based on the ratio

10

20

30

Waveguide label

Fig. 5 . Percent error in the Airy estimate of the crossover point as a function of the waveguide + n p in which the peak output appears.

as

where PflP the measured output power level from waveis guide n,,, etc. The slope of the ratio Rn,,(x) at the crossover point can be found using Bessels equation [4]:

and the fact that the crossover point is the point at which J & ( x ) = 0. The solution is

(7)
Then, substituting forx = 2KZ from ( 2 ) , the correction to the first approximation of Section I11 becomes

R , ( ~ K z )=

Jn,

+1(

~ K Z - Jn,

-1

( 2 ~ ~ 1 JA,, ( 2 2 1 K = -2-

Jnp(2Kz)

J&4.

(3) The second equality follows from the identity for Bessel functions [4]

J n - , ( x ) - J n + l ( x >= 2 J A ( x ) . (4) At the crossover point, R , ( ~ K z ) = 0. Elsewhere within the range for a given n,,, as long as n,, I2 , R , ( ~ K z )is monotonically increasing with ~ K Z it is also nearly linear ; for n,, = 2 and becomes more so as n,, increases. Hence a 2 correction when the argument 2 ~ is above or below a crossover point can be calculated from the slope of the function RnP 2 ~ 2along with a measurement of the output ( ) 1 and n,, - 1. In terms of power in waveguides n,,, n,, these quantities, the ratio above is computed from the data

Using this interpolation to improve the estimate in the example above, the argument becomes

222 = 16.184, a, = 0.151 (9) for this array. The standard deviation for the coupling coefficient from the data is now 0.93 percent of the mean value.
=

GABRIEL A N D WHITAKER: MEASUREMENT OF COUPLING COEFFlCIENTS USING MULTIPLE WAVEGUIDE SYSTEMS

1347

Finally, the complete general expression for the estimate of ~ K = nl from array output data is Z

2Z K
where

= T l

+ AX ~ ~
'/3

(10)

approximated very well by a second-order polynomial: the error is less than k 0 . 3 percent for np = 1 and less than k0.5percent for np = 0 provided that the sample is longer than about 0.15 two-guide coupling lengths L,. When the largest output power is in guide np = 0, the argument is

xAiY= np
and

)( :

2Kz = -0.0163

+ 2.1687

142 1
r
1

0.7264

[2]
(15)

and for np

2 Z = 0.0422 K

+ 4.1731

* [

6-G K

1.

Id?]

- 1.5961

[?].

(12)

VI. SENSITIVITY WAVEGUIDE TO IMPERFECTIONS To compare the theoretical output distribution with the collected data, the array outputs for the example above were computed from approximate formulas and recursion relations for Bessel functions [4], using the mean value of the argument for this array. A bar graph of the result is shown in Fig. 4(c) along with the photograph of the output light distribution and one of the line scans in Fig. 4(a) and (b). Not all the features of the output distribution match exactly; in addition, there was considerable variation between corresponding peak heights over the many scans taken. It is a major advantage of the array measurement that such large variations in the data do not contribute unacceptable error to the estimate of the coupling coefficient. Since the estimate of K requires only relative power levels, uniform loss does not degrade measurement accuracy. However, scattering centers or other imperfections which exist do in general distort the array output power distribution. This was indeed found to be the case in the experiments above. As an extreme example of a single imperfection, the change in output due to a broken waveguide, i.e., a localized point within the array at which power flow is interrupted, can be computed. Using the numerical model above, the discontinuity was simulated by launching a second excitation at the chosen break point, with matched amplitude but opposite phase, to exactly cancel the effect of the propagating wave in that guide. Since the outputs of only three guides on each side of the array contribute to the measurement of K , the effect of a break depends on its position relative to the path of power flow to these six outputs. The extent to which one break can corrupt the estimate of the coupling coefficient is illustrated by considering a representative array, with length equal to 3L,. The output power distribution was computed using the model for each of various positions throughout the array at which a single break was introduced. For an array with no imperfections, a single input into guide 0 would produce the largest out-

As above, np is the (positive integer) label of the waveguide in which the peak output power appears, and the values of Pn,, Pn,+l, and P n p - l are the measured output power levels in guide n and its nearest neighbors. In the worst case, i.e., for np = 2 and x far from the crossover point, the error in this approximation, computed from a numerical model of the array, is less than 2.5 percent of ; the true value of x = ~ K Z closer to the crossover point, and especially as np increases, the error decreases to well under 1 percent. In fact, as in the example presented here, most of the uncertainty in the estimate of the coupling coefficient will usually result from variability in the data, or from inaccuracy of the measurement of the sample length z , rather than from the error of the approximation. V. APPROXIMATION SHORTER FOR SAMPLES In order to obtain the best accuracy from the' array measurement of the coupling coefficient K , the number of coupling lengths L, in the sample should be made as large as possible. In many cases, however, experimental or material constraints may not allow the fabrication of long samples. Since the method above does not work well when the largest output power appears in the center guide or in guide 1 or - 1 , some empirical results from a numerical model of a linear array are provided in this section. Rather than using the expression

Rnp(2KZ)=

Jn,

+I(

~ K z - Jnp- I ( 2 ~ z ) )
Jnp(2Kz)

which has no useful definition for np = 0 and is not a monotonic function of ~ K for np = 1 , a better estimate Z of the argument can be calculated from ( P n p + l / P n p ) for np = 0 or 1 . The numerical model shows that this ratio is

1348

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7, NO. 9, SEPTEMBER 198)

0.15 0.3 0.45 0.6 0.75


Y
Q

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8
-3

8 8 4 4 4 4 4 2 2 8 8 8 8 O 8 8 8 8 8 8

8 8 8 8 4 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 -1

0 2 4 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8
0

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 1

8 8 8 8 8 7 7 7 7 8 8

8 8 8 8 8 8 8 8 8 8 8 4 4

8 8 8 8 8 8 7 4 1 1 1 8 8

8 8 8 8 8 8 8 8 8 8 8 ~ ~

8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8

8 8 8 8 8 8 8 8 8 8 8 8 8 7 7 7 7 7 7 7
8

8 8 8 8 8 8 8 8 8 8 8 8 7 8 8 8 8 8 8 9
9

8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8

0.9 1.05 1.2 1.35


1 .5

a a
d

c
d

U)

E =
Q
~

.C

1.65 1.8

c
1.95
Q

2.1 2.25 24 2.55 2.7 2.85 3 .O

4 4 4 8 8 2 2 2

8 8 8 8 8 8 8
3

8 8 8 8 8 8 8

7 7 8 8 8 8 8

8 7 7 7 7 7 8

8 8 8 8 7 7 7

-10 -9

-8

-7 -6

-5 -4

-2

10

Waveguide i n which break o c c u r s Fig. 6 . Values of np for an array of length 3L,, where one guide contains a break. For positive values of the waveguide label, n,, is taken on the same side of the array as the break: for the negative labels given, n,, is on the side opposite from that of the break. In all cases the input guide is guide 0.

put power in the guide pair np = 8, as shown in Fig. 7(a). For distortions of the output in which the brightest spot remains in guide 8, the measurement error is always less than the maximum A x from the interpolation of Section IV, which is in this case about 7 percent. It is useful therefore to determine the circumstances for which a single localized break causes the highest output power to appear in a different guide. Fig. 6 gives the np values computed for the positive-n side of the array, as a function of the position of a single break within the array, where the input guide is always taken to be guide 0. The position of the break in the indicated guide is given in units of the two-guide coupling length L,. In most cases the largest output appears in the correct guide. From symmetry, values for the negative-n side are found simply by changing the sign of the waveguide indices. For example, when guide 2 is broken 1.2 coupling lengths from the array input, then the highest output power levels will appear in guides + 7 and -2. When guide -4 is broken at 1.95 coupling lengths, then the brightest output spots will be in guides + 8 and - 1. Fig. 7(b)-(e) shows the output power distribution for some

selected table entries, which may be compared with the ideal case in Fig. 7(a). Data should be collected for these measurements by successively exciting adjacent waveguides and taking the average of the many pairs of estimates thus generated. During the course of such a measurement, exceptionally bright outputs in obviously wrong guides (such as 0-4 in the example above) could be ignored and reasonably good estimates of K would still be computed from the relative power levels in the correctly located guides ( 7 , 8, and 9 above). A grossly asymmetric output distribution would be a signal that serious imperfections are present. Systematic variations in K or geometry across the array would produce strong asymmetry for every successive input, making a good measurement impossible. In more typical cases when the sample is of good quality, however, the worst errors would be made when a scattering center along the main path of power flow causes np to shift by one. The redundancy provided by the presence of both sides of the array, together with the averaging process from exciting a series of adjacent input guides, minimizes the effect of any one scattering center on the quality of the estimate.

GABRIEL AND WHITAKER: MEASUREMENT O F COUPLING COEFFICIENTS USING MULTIPLE WAVEGUIDE SYSTEMS

I349

Waveguide label

Waveguide label

(a)
0.2 0.12

(b)

zl P
n
c c

n
0

0.1

n
0

w .-

z
0.0

E b

8
U

E b

0.1 0

5 P

B
n
0

0.08

0.06
0.04 0.02 0.00
Q I

2! -

E b

mI

m I

w I

Waveguide label

(e) Fig. 7. Output power distribution for an array of length 3L,.. (a) Output for an ideal array. (b) Output when a break is present in guide - 2 at 0.9 coupling lengths. (c) Output for break in guide + 2 at 2.1 coupling lengths. (d) Output for break in guide +5 at 2.25 coupling lengths. (e) Output for break in guide + 3 at 1.5 coupling lengths.

VII. CONCLUSION An accurate and simple method is proposed for measuring coupling coefficients between integrated optical waveguides, using a linear array of guides with excitation in the center. The approximation given in (2) is usually

sufficient to obtain the coupling coefficient to within a few percent, and involves only identification of the waveguide in which the largest output power is found. However, should greater accuracy be required, it can be gained from a measurement of only three relative output power levels

1350

JOURNAL OF LIGHTWAVE TECHNOLOGY. VOL 7. NO. 9, SEPTEMBER 1989

from adjacent waveguides in the array. The method is general, and applies to any array of coupled waveguides, although a modified equation for the array output distribution must be used when the assumptions of standard coupling-of-modes theory are not applicable. Moreover, the measurement is nondestructive and unambiguous, an improvement over commonly used methods based on successive cleaves. Even for the relatively crude inverted rib design used here, and without particular care taken to eliminate significant fabrication imperfections in the guides, the method works well. Results improve with the quality of the waveguides, limited only by the assumptions of the theory and the accuracy with which the length of the sample can be measured. Severe problems such as nonuniform loss and scattering, or slowly varying changes of dimension across the structure, ought to be easily identified by asymmetry in the output distribution. In more typical circumstances, the measurement ought to be tolerant of the output variations which appear, as long as a series of adjacent guides in the array are excited to generate a set of data for statistical computations. To obtain the most useful information with this method, the test array of waveguides should be fabricated using the same procedure as for the waveguide device structures of interest. The number of guides should be large, at least six times the expected number of two-guide coupling lengths in the sample. This is not an unreasonable requirement, since this array would take up only a small portion of a standard mask and would add no extra steps to the fabrication sequence for device structures on the same substrate. The test structure is simple enough to be useful as a design tool included on any photolithographic mask used in fabricating coupled-waveguide devices. The accuracy of models which predict the coupling coefficient or coupling length for a given structure could be checked with this method, and variations in coupling with changes in fabrication process parameters could also be examined in this way.

ACKNOWLEDGMENT Waveguide arrays were provided by Dr. W. R. Holland and Dr. J. D. LeGrange. Refractive index measurements were made by Dr. Holland. REFERENCES
[ I ] S . Somekh, E. Garmire, A. Yariv, H. L. Garvin, and R. G. Hunsperger, Channel optical waveguide directional couplers, Appl. P h p .
Lett., vol. 22, p. 46, 1973. [2] H . A . Haus, W. P. Huang, S . Kawakami, and N. A. Whitaker, J r . , Coupled-mode theory of optical waveguides, J . Lightwave Techno!., vol. LT-5, pp. 16-23, 1987. [3] R. M. Knox and P. P . Toulios, Integrated circuit for the millimeter through optical frequency range, in Proc. MRI Symp. Submillinterer Wuves, J . Fox, Ed. Brooklyn, NY: Polytechnic Press, 1970, pp. 497516. [4] M. Abramowitz and I. Stegun, Handbook of Mathematical Functions. New York: Dover, 1972.

*
M. Christina Gabriel (S81-M84) received the
B.S.E.E. degree from the University of Pittsburgh in 1978, and the S.M. and Sc.D. degrees in electrical engineering and computer science from the Massachusetts Institute of Technology in 1981 and 1985, respectively. From 1978 through 1985 she was an AT&T Bell Laboratories GRPW Fellow. Since 1985 she has been a Member of the Technical Staff in the Optical Computing Research Department at AT&T Bell Laboratories in Holmdel, NJ.

*
Norman A. Whitaker, Jr., (S81-M85) is a native of Hudson, NH. He received the B.S., M.S., and Ph.D. (1986) degrees from the Electrical Engineering Department at the Massachusetts Institute of Technology, Cambridge. He is currently a member of the Optical Computing Research Department at AT&T Bell Labs in Holmdel. NJ.

You might also like