492 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988
Nonlinear-optical effects in ion-exchangedsemiconductor-doped glass waveguides
C. N. Ironside, T. J. Cullen, B. S. Bhumbra, and J. Bell
Department of Electronics and Electrical Engineering, University of Glasgow, Glasgow G128QQ, UK
W. C. Banyai, N. Finlayson, C. T. Seaton, and G. I. Stegeman
Optical Sciences Center, University of Arizona, Tucson, Arizona 85721
Received August 28, 1987; accepted November 4, 1987We have measured nonlinear-optical effects in ion-exchanged channel waveguides fabricated in semiconductor-doped glass. The nonlinearity is manifested as optically induced saturable absorption displaying a resonantenhancement. The relaxation time of the nonlinearity was found to be tens of picoseconds. The relaxation timealso exhibited wavelength dependence. Waveguide transmission and pulse-probe experiments using a picosecondmode-locked dye laser were used to carry out these measurements.
1. INTRODUCTION
Integrated optics offers considerable advantages for all-opti-cal logic devices based on nonlinear-optical phenomena. In-tegrated optics is compatible with optical-fiber technology;more importantly, the light is confined in waveguides so thathigh intensities can be maintained over much longer dis-tances than those available with unconfined light, and there-fore interaction distances are limited by absorption ratherthan diffraction effects. Semiconductor-doped glass (SDG)has been shown in a number of studies to have potentiallyuseful nonlinear properties that include a large nonlinearitytogether with a fast response time.'-7 SDG consists of nano-crystallites of CdS,-.-Se. with diameters of 10-100 nm em-bedded in a glass matrix. The crystallites provide the non-linear-optical properties, while the glass host is easily modi-fied to permit waveguides to be formed, which makes SDGan ideal candidate for the fabrication of nonlinear integrat-ed-optical devices. Several all-optical device configurationshave been proposed. 8-'2 A Mach-Zehnder interferometeroperated as an all-optical gate in LiNbO3 has been demon-strated by Lattes et al.9 Nonlinear directional couplershave been demonstrated in stress-induced GaAs channelwaveguides by Kam Wa et al.10 and in dual-core fibers byFriberg et al." It is possible to fabricate these devices inSDG with currently available waveguide technology.13 Inthis paper we present a demonstration of nonlinear-opticaleffects in channel waveguides fabricated in SDG's.
In Section 2 details of waveguide fabrication are present-ed. The variation of transmission of the waveguides as afunction of incident optical power and wavelength is given inSection 3. An increase in transmission as a function ofpower is observed, which is strongly enhanced in the vicinityof the band gap of the crystallites. Time-resolved pulse-probe measurements of transmission are presented in Sec-tion 4. These show properties similar to those of the bulkglasses, and resonant enhancement is again exhibited. Wediscuss the results of the experiments in Section 5. Weconclude that our results (similar to those obtained for the
bulk glasses) constitute an important technological step for-ward for the SDG's in that the nonlinear-optical effects canbe induced in an integrated-optical channel waveguide form.
2. WAVEGUIDE FABRICATION
The channel waveguides used for this study were fabricatedin SDG by an ion-exchange process. After early attempts atNa+/Ag+ ion exchange, in which unsatisfactory waveguideswere produced, we established that low-loss waveguides canbe manufactured by Na+/K+ ion exchange.' 3 The wave-guides were produced in soda-lime SDG, which was speciallyprepared by Schott Ltd. (the standard Schott color-filterglasses are made from low-sodium borosilicate melts, whichare difficult to ion exchange). Standard photolithographictechniques were employed to mask off all but a 4-Am by 6-mm stripe upon the surface of the substrate with evaporatedaluminum. Immersion in KNO3 at 340°C for 1 h producedsingle TE-TM-mode channel waveguides. The ends of thesubstrate were polished in order to permit the use of end-firecoupling as a means of coupling light into the waveguide.
3. TRANSMISSION STUDIES
We measured the optical transmission of ion-exchangedwaveguides as a function of incident optical fluence andwavelength by using a simple direct-detection method, asshown in Fig. 1. A mode-locked cavity-dumped dye lasersynchronously pumped by a frequency-doubled Nd:YAGlaser was used as the light source. The laser repetition ratewas 3.8 MHz, the autocorrelated pulse width was 6 psec, andthe maximum energy per pulse was 2.6 nJ. A Pockels cellwith a bias voltage set by a wideband (800-Hz) programma-ble power supply was used to provide 256 increments ofattenuation over 3 decades of optical power. Silicon photo-detectors calibrated against a Coherent 212 powermeterwere used to detect the average power entering and leavingthe waveguide. The detectors were checked for linearity
Vol. 5, No. 2/February 1988/J. Opt. Soc. Am. B 493
L INPUTI DETECTOR
Fig. 1. Experimental setup for measuring fluence dependence oftransmission of a K+/Na+ ion-exchanged waveguide in SDG.
i.0Ld-
U
C-z
F_~0.8
HU)0.6
H
W .
F_
CDz
- i5
H
6!ccI-
" 570 580 590 600 610
WAVELENGTH (nm)Fig. 2. The linear transmission of the channel waveguide as afunction of wavelength.
14.00
12.00
'-10.00
°_ 8.00
U):2 6.00
< 4.00
I-
2.00
Ann L-
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00
FLUENCE (mJ/cm 2 )
Fig. 3. Waveguide transmission as a function of input fluence atnine wavelengths. The transmission increases linearly at lowfluences and saturates at high fluence.
over the range of powers used. At each discrete level ofattenuation set by the Pockels cell the detector output was
sampled 128 times and averaged to reduce noise. M60 mi-croscope objectives were used to couple light into and out ofthe waveguides. A Jarrell-Ash 82-000 Series spectrometerwas used to measure the operating wavelength.
A plot of low-fluence transmission versus wavelength is
shown for the 6-mm-long channel waveguide in Fig. 2. The
input coupling efficiency was measured to be 37% at 610 nm
and assumed to remain constant when the wavelength waschanged. Assuming an effective waveguide cross-sectionalarea of 16 ,4m2, the incident power density at 1-mW averagepower was approximately 100 MW/cm 2. We were workingat very high intensities in these experiments; the intensityvaried between 10 MW/cm 2 and 1 GW/cm2 . We were alsoworking in a transient regime in terms of the induced carrierpopulations; that is, our laser pulse widths were shorter thantypical relaxation times quoted for darkened (solarized)SDG.' In this regime the relevant quantity is fluence, orenergy density. The energy densities used in our experi-ments were in the range 0.06 to 6 mJ/cm 2 .
The change in transmittance as a function of fluence andwavelength is plotted in Fig. 3. The transmittance clearlyincreases with increasing fluence at all wavelengths. Atshort wavelengths the transmission changes linearly withfluence, whereas the transmission changes begin to saturate,
14.00
12.00
zI 10.00
° 8.00
>_ 6.00Uf)z< 4.00
-. 0
2.00
0.00 576.00 580.00 584.00 588.00
WAVELENGTH (nm)Fig. 4. Transmission versus wavelength in a SDG channel wave-guide. A blue shift in the transmission edge occurs as the inputfluence is increased.
10.00
8.00
E0
O-1 577nm6.00
z' 578nm
_ -0nm581 nm
n 4.00 582nmi)~~~~~~~ ~583nmo 584nmU) 585nmm<- 2 _
0.00 I I . .0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00
FLUENCE (mJ/cm 2 )
Fig. 5. Absorption as a function of fluence for different excitationwavelengths below the gap. The saturation intensity decreases asthe detuning from resonance increases.
Ironside et al.
I
4
2
.,\
.vv
494 J. Opt. Soc. Am. B/Vol. 5, No. 2/February 1988
for comparable fluence levels, at the longer wavelengths.The experimental data are plotted in terms of change intransmittance versus wavelength for several values offluence in Fig. 4. The net transmission changes correspondto a blue shift in the absorption edge, implying that thechange in absorption is electronic rather than thermal inorigin. The corresponding absorption changes, calculatedas a weighted average over the length of the waveguide, canbe seen in Fig. 5. Although data are unavailable for photonenergies well above the band gap, it is clear that resonantlyenhanced saturable absorption has been observed in thisexperiment; that is, the saturation of the absorption coeffi-cient exhibits a strong wavelength dependence. Saturationof the absorption is evident at all wavelengths, with themaximum fractional change in absorption decreasing as thedetuning from resonance increases. It is evident from Fig. 5that the absorption cannot be completely saturated with thefluences that we were using.
4. TIME-RESOLVED PULSE-PROBEEXPERIMENTS
In this section we present the results of a pulse-probe mea-surement of the temporal behavior of the nonlinearity in theguided-wave environment. The previously discussed non-linear absorption measurements were performed with a sin-gle polarization; however, the pulse-probe polarizations arecrossed. We have neglected any relation between the effec-tive third-order-susceptibility tensor elements of the ran-domly oriented microcrystallites for parallel and crossedpolarizations. However, we make the assumption that therelaxation times of the change in absorption in the twoconfigurations are identical since the relaxation mechanismsrespond to the excited carrier density.
Pulse PropagationThe propagation of high-intensity pulses in these wave-guides was studied by using an autocorrelation technique, asany pulse distortion would have to be taken into account ifone were interpreting the data from the pulse-probe experi-ments. Autocorrelations of the input and output pulses of awaveguide were measured. Apart from there being morenoise on the output-pulse autocorrelation, there was no ap-parent substantial distortion. As a result, we were able touse the pulse-probe technique to measure the relaxationtime of the nonlinearity.
Pulse-Probe MeasurementsFigure 6 shows the experimental arrangement for a pulse-probe experiment that measures the time dependence ofabsorption recovery in the waveguide. In the usual pulse-probe experiments the pulse and probe are propagated atdifferent angles, thus permitting spatial separation. This isnot possible in a waveguide configuration, and therefore inthis experiment the pulse is polarized perpendicular to theprobe, so that an analyzer can be used for discriminationbetween them. Such an approach is not completely success-ful because there is some depolarization of the pulse duringpropagation through the waveguide (even at low powers),which means some leakage of pulse light into the probepolarization. Further isolation of the probe from the pulseis achieved by chopping the probe light at low frequencies
(near 100 Hz) and synchronously processing the signal with alock-in amplifier.
In Figure 7 the probe signal is shown as a function of delaybetween probe and pulse for wavelengths close to the bandedge of the SDG. Saturation of the absorption is clearly
DYELASERPULSE
A-SEMICONDUCTOR
DOPED GLASSWAVEGUIDE
Fig. 6. Experimental arrangement for pulse-probe measurementsof the time recovery of saturable absorption in ion-exchanged SDGwaveguides. The weak probe (1% of the pulse), which is polarizedorthogonally to the pulse, is chopped to permit synchronous detec-tion.
80 X=581 nm X=586nm
Z 40 -J Z40
0(n 80U X=583nm X=588 nmU)
U) 40-z
LUJ 80~ X \584 nm X=590 m
z0 H
40 80 1200
TI M E (psec.)40 80 120
Fig. 7. Time and wavelength dependence of absorption in SDGwaveguide (band gap at approximately 590 nm). A large saturatingpulse of 5-psec duration and a fluence of 1.5 mJ/cm2 is introducedinto the waveguide. The plots show the changes in transmission ofa weaker probe (0.015 mJ/cm 2) as a function of time delay and theoperating wavelength. The probe is polarized TE, and the pulseTM.
Ironside et al.
Vol. 5, No. 2/Pebruary 1988/J. Opt. Soc. Am. B 495
evident; the probe transmission is increased by approxi-mately 80% at a wavelength of 581 nm. The full widths athalf-maxima of the transmission curves increase from 12psec at 581 nm to 25-30 psec at 590 nm. Assuming that theturn-on time of the nonlinearity remains constant at thesewavelengths, we conclude that the relaxation time is wave-length dependent, increasing with decreasing photon ener-gy-
5. DISCUSSION
A number of studies of nonlinear-optical phenomena in bulksamples and in fiber SDG have been undertakenl-6; thesehave established that the resonant nonlinearity in SDG isrelatively large, with an intensity-dependent refractive-in-dex coefficient n2 of 10-15-10-'3 m2/W and a fast relaxationtime of approximately 10-11 sec.
The mechanism for the nonlinearity is best explained byusing a band-filling or dynamic Moss-Burstein model.'This model predicts a blue shift in the absorption edge,which is electronic in origin. Our results of a blue-shiftedabsorption edge are consistent with the band-filling mecha-nism. Any thermal contribution (red shift) that may maskthe maximum blue shift that can be observed is assumed tobe small. Our pulse separation was only 260 nsec-thermalrelaxations are assumed to be longer-and a significant por-tion of the energy was absorbed at shorter wavelengths, yetthe maximum decrease in absorption was observed at theshortest wavelength used.
The samples used in our experiments were assumed to bephotodarkened since they had previously been subjected tomoderate-to-high fluences. Roussignol et al. have concludedthat a fast response occurs only with photodarkened glass-es.' This conclusion is consistent with the observation ofYao et al.
5 of a luminescence lifetime of approximately 10psec in bulk SDG. Our results support this conclusion butalso reveal a wavelength dependence of the recovery time forthe nonlinear absorption. We are currently investigatingthe relaxation dynamics to determine the relative influencesof trapping levels, electron-hole recombination, excited-car-rier absorption, and nonradiative processes on the measuredlifetimes.
The nonlinear behavior of the semiconductor crystallitescan be crudely modeled as an inhomogeneously broadened,saturable two-level system:
a(w) = [1 + ao( &) (1)
where a(co) is the absorption observed at input intensityI(w), ao(co) is the linear absorption, and I(w) is the satura-tion intensity. The predictions of the two-level model werecompared with experimental results. The model predicts aresonant X3ff of 8 X 10-8 esu, corresponding to an intensity-dependent refractive-index coefficient n2 5 x 10-14 m2/W,which is in order-of-magnitude agreement with values mea-sured by other authors. 4 It is clear, however; that the ex-perimental data are not well modeled by the two-level sys-tem, except at low fluences. The practical consequences arethat the maximum achievable refractive-index change willbe somewhat less than that predicted by the two-level sys-tem and the accompanying absorption will be higher. Weare currently analyzing these results with more-detailed
semiconductor theory in order to gain better insight into thebehavior of the nonlinearity. The role of detuning becomesof critical importance, for only by detuning can an optimumtrade-off among refractive-index change, absorption, relax-ation time, and saturation intensity be achieved.
The significance of these measurements for integratedoptics is that they demonstrate some simple facts aboutusing darkened SDG's in a guided-wave format. The satu-ration intensities required in waveguides that are millime-ters in length are extremely large. The absorption can neverbe completely bleached, however, even at these intensities.Much smaller saturation intensities could be expected forundarkened samples, at the expense of a much longer relax-ation time.
6. CONCLUSIONS
The nonlinear-optical effects measured here are crucial tothe modeling of all-optical logic device operation in devicesmade from photodarkened SDG ion-exchanged waveguides.We have fabricated ion-exchanged channel waveguides inSDG and confirmed that the ion-exchange process does notaffect the nonlinear-optical properties significantly. Wehave carried out dc and time-resolved measurements of reso-nant nonlinear absorption in these waveguides. Our esti-mate of X3 is similar to that calculated by other authors forbulk SDG, and we have established that the relaxation timeof the nonlinearity is between 12 and 30 psec, depending onthe detuning from resonance.
ACKNOWLEDGMENTS
This research was sponsored in the UK by the Science andEngineering Research Council and in the United States bythe National Science Foundation (ECS-8501249), the U.S.Army Research Office (DAAG29-85-K-0173), and the U.S.Office of Naval Research (N00014-86-K-0719).
REFERENCES
1. P. Roussignol, D. Ricard, J. Lukasik, and C. Flytzanis, J. Opt.Soc. Am. B 4, 5-12 (1987).
2. B. J. Ainslie, H. P. Girdlestone, and D. Cotter, Electron. Lett.23, 405-406 (1987).G. R. Olbright and N. Peyghambarian, Appl. Phys. Lett. 48,1184-1186 (1986).
4. R. K. Jain and R. C. Lind, J. Opt. Soc. Am. 73, 647-653 (1983).5. S. S. Yao, C. Karaguleff, A. Gabel, R. Fortenberry, C. T. Seaton,
and G. I. Stegeman, Appl. Phys. Lett. 46, 801-802 (1985).6. G. Bret and F. Gires, Appl. Phys. Lett. 4, 175-176 (1964).7. C. N. Ironside, T. J. Cullen, J. Bell, B. S. Bhumbra, W. C.
Banyai, N. Finlayson, C. T. Seaton, and G. I. Stegeman, post-deadline paper presented at the 4th European Conference onIntegrated Optics, Glasgow, Scotland, May 1987.
8. G. I. Stegemian and C. T. Seaton, J. Appl. Phys. 58, R57-R78(1985).
9. A. Lattes, H. Haus, F. Leonberger, and E. Ippen, IEEE J. Quan-tum Electron. QE-19, 1718-1723 (1983).
10. P. L. Kam Wa, J. E. Sitch, N. J. Mason, J. S. Roberts, and P. N.Robson, Electron. Lett. 21, 26-27 (1985).
11. S. R. Friberg, Y. Silberberg, M. K. Oliver, M. J. Andrejco, M. A.Saifi, and P. W. Smith, Appl. Phys. Lett. 51, 1135-1137 (1987).
12. L. Thylen, N. Finlayson, C. T. Seaton, and G. I. Stegeman, Appl.Phys. Lett. (to be published).
13. T. J. Cullen, C. N. Ironside, C. T. Seaton, and G. I. Stegeman,Appl. Phys. Lett. 49, 1403-1405 (1986).
