Mid-to far-infrared femtosecond pulses

Transcription

1 Invited Paper Mid-to far-infrared femtosecond pulses H. M. van Driel, A. Hach and G. Mak University of Toronto, Department of Physics and Ontario Laser and Lightwave Research Centre Toronto, Ontario, M5S 1A7, Canada ABSTRACT The emergence of new, efficient, nonlinear crystals and the development of passively mode-locked, solid state lasers, such as the Kerr-lens mode-locked Ti:sapphire laser, has rekindled interest in optical parametric oscillators (OPO' s) while opening up new regions of the spectrum to high repetition rate, high average power femtosecond pulses. With synchronous pumping techniques it is now possible to generate pulses as short as 40 fs for (nominally) 1< < 4.im, at repetition rates of 108 Hz, and with average powers measured in 100's of mw. Tuning can be achieved in critically or non-critically phase matched KTP or LBO via crystal angle, temperature or pump wavelength tuning each of which has its merits in terms of tuning range, ease of use, noise properties, etc. Harmonic generation of the OPO or the Ti:sapphire laser beams permit the visible and near UV region to be accessed. Through difference frequency mixing of the two output beams of the OPO in crystals such as CdSe, AgGaSe2, etc. it is possible to generate femtosecond pulses in the mid-infrared so that the full wavelength range accessed by the Ti:sapphire laser, and up-conversion or down-conversion processes can extend from 200 nm to beyond 20.tm. 1. INTRODUCTION The development of the first dye laser and optical parametric oscillator (OPO) occurred within two years of each other in the 1960's. However, whereas the dye laser has had a large impact on spectroscopy and the development of ultrashort pulses for the investigation of ultrafast phenomena, the OPO has not been widely utilized. Although the bandwidth and tuning range of OPO's are much larger than dye lasers, which operate primarily in the visible and near-infrared, the frequency downconversion process is based on nonlinear optical effects in crystals, requiring high intensity. In the past the high intensity was typically obtained (along with high energy) in low repetition rate picosecond and nanosecond pulses. The intensity fluctuations in the pump source were enhanced in the parametric process and the crystals typically had to be used near their damage threshold. The use of shorter pulses with lower energy but higher intensity and synchronous pumping techniques pointed the way to overcoming some of these difficulties, as Burneika et al.1 have demonstrated initially with Ba2NaNb5O15. Lauberau2 in an excellent article has reviewed the development of a variety of picosecond, tunable parametric sources until Since then there have been two major developments which have allowed the OPO to be viewed in a more favorable light and to use it as a reliable generator of low noise, widely tunable pulses as short as 40 fs. The first major advance is the development of high quality, high nonlinearity crystals such as BBO(-BaB2O4), LBO (LiB3O5), and KTP (KTiOPO4) all of which also have high optical damage thresholds. They have already seen extensive use in a variety of picosecond and nanosecond parametric oscillators and generators. The other major reason for advances in short pulse OPO's is the advent of wide band-width solid-state lasers such as the Ti:sapphire laser3 which make use of non-spectrally selective mode-locking techniques. Such lasers offer tunable, 100 fs pulses from 720 to 1000 nm in low noise beams with average powers of greater than 1W and repetition rates of 108 Hz. These high average and high peak power lasers can be used to synchronously pump OPO's and produce as short or 50 / SPIE Vol /94/$6.00

2 shorter pulses in other regions of the spectrum. In particular with KTP the range from 1-4 p.m can be covered. When these sources are combined with sum- and difference-frequency mixing techniques, femtosecond pulses in the wavelength region from 200 nm to morethan 20 p.m can be produced. In this article we will review some of our work and that of others to exploit the characteristics offered by the new crystals and the novel pump sources to develop such sources. 2. BASICS OF PARAMETRIC PROCESSES The general principles of optical parametric conversion have been discussed in several excellent review articles4 and here we will simply outline some of the underlying physics for these processes and then consider how the ideas which apply to long pulse parametric devices have to be modified to allow short pulses to be generated. In general, parametric processes involving three beams of frequency oi, o2 and 0)3 satisfy the condition 0)3 0)1+0)2. (1) Note that a down-conversion process permits a wide spectrum of pair frequencies to be generated. The amplitude of each beam evolves according to a nonlinear polarization density source term. For example, in the case of collinear beam propagation with specific polarization states one can describe the nonlinear part of the polarization density for each beam by i = c02de 2E 3 = e02de 1E' 3 pu3 =e02de1e2 (2) Typical values of d for specific crystal axes in the conventional contracted notation are given in table I. It can easily be shown that with this form of the polarization density, negligible pump depletion, and the slowly varying envelope approximation, the wave equation for the (i=1,2) down-converted beam reduces to the following equation for its field envelope )1 = ike EW3 exp(iakz) (3) where z is the direction of propagation, j = (2,1), ic =od/nc, n1 is the refractive index of beam "i" and Ak = k3 k2 k1 is the difference in the propagation wavenumbers; a1 describes the linear loss term for field "i". To obtain optimum conversion in a given length, L, of crystal, one requires AkL <it. Phase matching (Ak =0) can be achieved in a variety of ways, but for bulk crystals it is usually obtained by employing the birefringence properties of crystals and allowing the three beams to propagate as ordinary (o) or extraordinary (e) rays. In type I phase matching the pump beam propagates as either an o- or an e-ray and the two other beams propagate as the other type of ray and with orthogonal polarization (e.g., o * e -i-c, or e * o +o). In type II processes one 'has o e +o, or e > e +0. Equation 3 does not have a general analytical solution. However, for many cases one can assume perfect phase matching, little scattering or absorptive loss and negligible pump depletion. With an input beam of magnitude E ' (0) Eq. 3 yields SPIE Vol / 51

3 E ' (L) = Et 1 (O)cosh(fL) E '2(L) = /CO2fll E 1 (O)sinh(FL) O)1fl2 (4) after the beams have propagated a distance L. The coefficient F is given by r2 = 2o1o21d12 13 (5) e0n1n2n3c and can be identified as the gain coefficient for both down-converted fields. Note that the gain is highest for degeneracy (both down-converted frequencies the same) which, in the absence of any other spectrally selective effects, will dictate together with phase matching which frequencies will preferentially be amplified. For typical values of pump beam intensity (13 > 100 MWcm2) and a d- coefficient typical of KTP, the gain coefficient still is << 1 cm1). Therefore for significant amplification one typically requires an intense input beam or feedback, such as that provided by an oscillator. In the down-conversion process the power change in each of the beams satisfies the Manley-Rowe relations for power flow namely, APw3 = _ Apw2 \1 (6) 0)3 2 O)i. so that the down-conversion process converts pump power into both the output beams. For a fixed geometry, the power change in either of the converted beams varies linearly with the power in the pump beamabove threshold. Therefore, the percentage fluctuation in amplitude of either of the downconverted beams is essentially the same as that of the pump beam. The derivation outlined above assumes a 1-D geometry and monochromatic plane wave beams. In reality, one must usually consider Gaussian pulses so that there will be spatial, temporal and spectral width effects.59 Dealing first with the spatial aspects, since the gain in the parametric system for the two down-converted beams is related to the intensity of the pump beam it can be anticipated that some focusing of the pump beam is desired. However, extreme focusing is not. First of all, a focused beam, with its angular spread, may not be able to satisfy the phase matching condition for all components of the beam. Second, tight focusing also leads to a strong variation of intensity in the direction of propagation and a non-optimum, spatially varying gain coefficient. As several authors have shown, maximum conversion is achieved if the crystal thickness is comparable to the pseudo-collimated beam length (Rayleigh length) of a Gaussian beam, with the amount of focusing limited by the angular acceptance of the crystal for phase-matching. A detailed analysis shows that the product AOL where L is the length of the crystal and A9 is the angular convergence/divergence of the beam, must be less than a material dependent value, typically near 10 mr-cm. Another important parameter for the choice of crystal and geometry relates to the sensitivity of the phase-matching condition to temperature change through variation of the refractive index with temperature. One then speaks of a ATL figure of merit, a typical value being in the range C-cm. As with the angular aperture, this parameter is dependent on the wavelength. The choice of phase-matching geometry is therefore dictated by the dispersion of the refractive indices, optimization of the effective "d" constant, acceptance angles of beams, etc. The phase matching condition also connects thebandwidth generation of a crystal to its length. Typical values for the length-bandwidth products (AL) of a crystal are 1-10 nm-cm, and for the femtosecond generation processes discussed below is the main source of bandwidth in the down-converted pules at 52 / SPIE Vol. 2041

4 low intensity. Table 1 : Optical Properties of Important Nonlinear Crystals for mid-jr Nonlinear Crystal KTP LBO BBO Point Group mm2 mm2 3m Refractive Indices (at 0.8 pm) n=1.75o2 n= n= n= n0= n=1.844o n= (at 1.6 pm) n=l.7286 n= ne=l.53ol n= n= n0= n= n=l.5948 Optical Axis (x,y,z)=(a,b,c) (x,y,z)=(a,c,b) Transparency (Rm) Phase-matching CPM CPM CPM only (o e+o) (o e+o) (e *o+o) NCPM NCPM (narrow tuning (o * e + e) range) Nonlinearity (pmjv)(shg at pm) d d d d d d22=-d21=-d Damage Threshold (CWIcm2) Chemical nonhygroscopic slightly hygroscopic nonhygroscopic (melting point C) (834) (925) (934) Data taken from Laser Focus Magazine, Sept., 1990; CPM=critical phase-matching, NCPM=noncritical phase-matching The requirement of phase matching implicitly contains the possibility of achieving tunability (and the selectivity not given by Eq. 1). For a given input wavelength tuning can be achieved by variation of the temperature or by changing the angle of the crystal (critical phase matching), the latter leading to greater ease of operation and range of tuning. Alternatively, one can vary the pump wavelength. However, the phase matching condition is more constrained, e.g. by the characteristics of the pump laser, and usually yields a reduced tuning range. SPIE Vol /53

5 The birefringence of the crystal in critical phase matching leads to a difference in direction of the themselves will have directions that differ by up to a few degrees. This leads to a spatial walk-off of the beams, smearing them in a transverse direction, and reducing the intensity and conversion efficiency. Hence, one requires that LOw (7) where 0w 5 the maximum walk-off angle between the three beams and w is the beam waist. The walkoff effect may be mitigated by choosing a non-collinear phase matching geometry, with the different wavevectors pointing in slightly different directions, and then reducing the walk-off angle to zero for at least one pair of beams. By operating in a non-critical phase matching geometry (input beam propagating at 9O relative to the crystal optic axis) the angular acceptance of the crystal is higher allowing tighter focusing to be used; there is also no spatial walk-off between the different beams. The fact that the "d" value is usually highest for non-critical geometries, together with the reduced walkoff usually results in large parametric gain in the crystal. If one wishes to preserve or reduce the pulse-width of an ultrashort pulse, one must also be concerned with temporal walk-off between the three interacting pulses. Temporal walk-off leads to a temporal smearing of the pulses, and reduces the intensity and conversion efficiency, just as spatial walk-off does. This effect can be shown to be related to the group velocity dispersion (GVD) of the material at the three different wavelengths. The temporal walk-off between the "i"th and "j"th beams is given by tg [?L() X()1. (8) If one wishes to maintain the incident pulse width, (tp), of the pump beam in each of the downconverted beams then the crystal should be thin enough that the spreading of the pulse due to GVD is less than the pulse width or tg (tp). (9) It should be noted that unlike sum frequency conversion processes such as second harmonic generation where the high bandwidth of short optical pulses can inhibit frequency conversion of the entire incident pulse, since it is not possible to obtain phase matching for all components, this is not the case for down-conversion processes because any change in o due to a change in o can be compensated by an opposite shift in (02 in the down conversion process. Indeed, one usually has the opposite problem in down-conversion, namely poor spectral selectivity, and one must look for ways to limit pulse bandwidths in order to speak of transform-limited pulses. The various restrictions put on the parametric interaction by the need to minimize beam spatial and temporal walkoff for short pulse generation, forces one to consider non-collinear phase matching geometries (including a selection between type I or II), and a certain range of pump wavelengths. For example, the interaction geometry which would maximize, e.g., the gain coefficient r may not preserve the short pulse width. In general then, one must be prepared to deal in various trade-offs in designing femtosecond parametric systems and this makes them fundamentally different from nanosecond or even picosecond pulse OPO's or lasers. With these general comments in mind we now consider the applicability of the three materials of Table 1 for femtosecond OPO's. In general BBO is not suitable for OPO applications at the nearinfrared wavelengths offered by the Ti:sapphire laser; tuning ranges and parametric gains are small. However KTP and LBO can both be used. We first consider KTP pumped by a 765 nm source. Plotted in Fig. la is the tuning curve of the signal and idler for type II critical phase matching based on beam propagation in the x-z plane (o 4 e + o). Also indicated is the Poynting vector walk-off angle of the signal versus crystal internal angle. A large tuning range from p.m including the idler, can be 54 / SPIE Vol. 2041

6 covered with this phase-matching configuration. Plotted in Fig. 2 are the temporal walkoffs between signal and idler beams relative to the 765nm pump as a result of GVD. The differences in group velocities dictates that crystals of only mm lengths can be used if 100 fs and shorter pulses are to be generated. The wide tuning characteristics have prompted several groups610 to use this material in the critical phase-matching geometry for femtosecond OPO's. The walk-off angles of.2 limits the mode sizes to about 20 m for crystals lengths of a few millimeters. KTP can also be used in a noncntically phase-matched geometry (as demonstrated, e.g. by Nebel et al.11) although, as alluded to above there are serious deficiencies in the tuning range. 3 v1 a, I, a, C,, 2 a) '-4 1 I Internal Crystal Angle 0 (degrees) C!, b) I I I I I I I I Crystal Temperature (degrees C) Fig. 1 a) Tuning curves and walk-off angle for KTP as a function of internal crystal angle and b) tuning curve for LBO as a function of temperature. Pump wavelength in both cases is 765 nm. SPIE Vol /55

7 LBO cannot be phase-matched for OPO operation at room temperature when Ti:sapphire pumping is considered with wavelengths near or below 800nm. When the crystal is heated however it can be noncritically phase-matched (type I: o * e + e) to achieve tuning over a limited range. Plotted in Fig. lb is the tuning curve of the signal and idler for LBO versus crystal temperature for a pump wavelength of 765 nm. The tuning range is as large as jtm if the crystal temperature is about 200 C. This has been exploited by Kafka et J12 to construct a femtosecond OPO based on this material. Note that because the signal and idler have the same polarization, they are not distinct. An unusual feature is that at degeneracy the relative temporal walk-offs are very close to zero; this makes femtosecond OPO's pumped at this wavelength quite interesting. Unfortunately, this is not a general trend for LBO. The dispersion between the signal and the pump beam is higher than with KTP pumped at the same wavelength, but because the signal and idler have the same polarization, the dispersion between them is much smaller. 0 ci 0 V3 : 0 '-4. 0 'U signal a' r_. 0 1 i _ I - I _ I I i i I I I i. -_ - I I. i I i i i C-, N I Internal Crystal Angle 9 (degrees) Fig. 2. Inverse group velocity dispersion characteristics for KTP 3. FEMTOSECOND PARAMETRIC OSCILLATOR Some success in achieving short pulse operation at high repetition rates via synchronous modelocking of an OPO was achieved by Piskarkas et al.13 and Edelstein et al6., making use of BBO and KTP crystals. The former system was pumped by a mode-locked Nd:glass laser and produced pulses in the picosecond range while the latter made use of the high intracavity power in a colliding-pulse modelocked laser and yielded 100 fs pulses in the near IR. In both cases the average output power was low. The development of a synchronously pumped hybridly mode-locked dye laser allowed us to use external pumping for a KTP OPO, enabling average powers up to 30 mw to be attained.8 Substantially higher powers could be achieved when the dye laser was replaced by a self-mode-locked Ti:sapphire laser. We now describe this system9 in detail. Figure 3 presents a schematic diagram of the Ti-OPO. The cavity arrangement is similar to that used for the dye-pumped OPO. The singly resonant ring-cavity consists of a plane high-reflector (HR), a 2% output coupler, and a pair of curved HR's with 20 cm radius of curvature. The mirrors have high reflectivity in the nm range. The 1.5 mm thick KTP crystal is positioned at the intracavity focus of the short radius mirrors. It has been cut for type II non-collinear phase-matching, and is anti- 56 / SPIE Vol. 2041

8 reflection coated for 1.3 p.m on both sides. The pump beam is focused onto the crystal noncollinearly with the resonant signal beam; a typical non-collinear angle is The pump laser is a Kerr lens mode-locked Ti:sapphire laser (Coherent Mira), which is configured to produce = 765 nm, 110 fs pulses at a 76 MHz repetition rate with an average output power of 800 mw. M2 P1 PZT HR 110 fs 76 MHz 800 mw A=765 nm S M3 62fs 76 MHz 175 mw ,am Fig. 3 A diagram of the externally pumped femtosecond optical parametric oscillator, showing the KTP nonlinear crystal (KTP), the output coupler (OC) and the back high-reflectors (HR) mounted on a piezoelectic transducer (PZT). The external cavity GVD compensator consists of two SF4 prisms (P1 & P2). The compressed beam is vertically displaced for output coupling. The signal beam (S) is shown with a solid line, the idler beam (I) with a dashed line and the pump beam (P) with a dotted line. Pumping of the OPO by the Ti:sapphire laser at 765 nm offers significant advantages over pumping with the dye laser at 645 nm. Besides the higher average power and shorter pulses, which are important in the nonlinear generation process, the group velocity mismatch between the e-polarized signal beam (at.i = 1.3 p.m) and the 0-polarized pump beam (X3 =765 nm) is reduced, by at least an order of magnitude. The better group velocity match between the pump and the signal beams and also the factor of two improvement in the pump-idler beam temporal walkoff implies that a shorter signal pulse width is possible. The reduced group velocity walk-off also improves the parametric gain of the system. However, this advantage is partly offset by the fact that Ti-OPO is operating away from the degeneracy point for which the threshold is slightly higher. Also, because of the longer pump wavelength the effective nonlinear coefficient "d" is slightly smaller. The threshold for Ti-OPO has been measured to be about 180 mw, in good agreement with calculations which take into account the Gaussian beam profile, GVD, and Poynting vector walk-off in the non-collinear phase-matching geometry. The Ti-OPO tuning range is similar to that of the OPO-dye and is also limited by the high reflectivity bandwidth of our cavity mirrors but extended tuning is possible with multiple mirror sets. To date we have covered the range p.m with signal and idler beams. Tuning the output wavelength (a rotation of up to 60 of the KTP crystal) is very simple and requires only minor adjustment of the cavity length to optimize the signal power. Average output powers as high as 180 mw are obtained in the signal beam, and by the SPIE Vol /57

9 Manley-Rowe relations, approximately the same power exists in the idler beam. The total conversion into both beams is therefore close to 50%! Optimization for highest output power (up to 200 mw in the signal beam) requires more extensive cavity realignment at each wavelength. It is also possible to tune the output of the OPO by adjusting the pump wavelength. Higher output powers should be possible with higher pump powers although crystal damage may represent a limitation. We estimate that the average intracavity circulating power is 10 W and with focused spot sizes at the KTP crystal of about 40 im it is possible to have focused peak power densities greater than 40 GWcm2! It is possible to make this OPO operate in both polarizations near degeneracy, i.e., to make it work in the signal (which we have defined as being e-polarized) or the idler wave. This does not imply that both polarizations have the same threshold. Besides the difference in pulse widths, there is polarization selection in the noncollinear geometry as well as polarization selection through the spectral selection (i.e. the finite bandwidth of the mirror set). Thus, the polarization can be selected in two ways: tune the opo until only one polarization can lase, and second, physically "flip" the KTP crystal orientation to allow higher gain for the opposite polarization. This noncollinear geometry is important for polarization selection and for realizing certain gain characteristics. For a given crystal orientation, the effective noncollinear angle between the pump and the resonated wave (signal or idler) is polarization dependent. Each crystal orientation favors the opposite polarization than the other - which chooses one wavelength or the other also! The tunability of the illustrated system is limited by the high reflectivity bandwidth of the mirror set. With other mirror sets we have achieved tunability of the signal beam from p.m and the idler beam from p.m. A wider range of tunability can be obtained with other mirror sets. The issue of what combination of crystal length and beam focusing to use to optimize OPO power is complicated and must be considered together with desired pulse width. Is there an optimum crystal length for a given pulse width? For Ti:sapphire laser pumping, our gain calculations show that decreasing the crystal length to 1 mm, at constant confocal parameter such that the beam waist decreases to 16 p.m, actually increases the gain by 17%. The reason why this is true is that the optimum parametric focusing scaling rules do not normally include GVD, which expects that the gain increases with L. But, with GVD, the temporal separation of the pulses increases with L. To first order we can say that in a regime where GVD is important, these two effects cancel and that the gain will be roughly constant at optimum focusing (constant confocal parameter). Clearly, for Ti:sapphire pumping, even when the signal-pump GVD is small, the idler-pump GVD is not. Thus, for crystal lengths larger than 0.5 mm the above statement will be true. Quantitatively, for = 756 nm and = 1.3 p.m, the peak gain is for L=O.75 mm with 15% drop in gain for 0.48 mm L 1.45 mm, which confirms that the dependence of threshold on crystal length is relatively weak. Within this range of crystal lengths there may be more important considerations than the threshold power: e.g. difficulty of focusing harder and the influence of self-phase-modulation (SPM) on pulse width, etc. We have quantitatively characterized the pulses before and after pulse compression by measuring the pulse widths with an interferometric, autocorrelation technique, and also measuring the spectral content with a 0.5 m spectrometer. The time-bandwidth product of the Ti:sapphire laser is 0.48 (with a spectral width of 30 nm) whereas for the highest intensity of the signal beam the time-bandwidth product is 1.63 (spectral width of 60 nm) indicating significant departure from the Fourier transform limit of about 0.3. Figure 4 shows the interferometric and intensity autocorrelation for approximately 1 80 mw in the signal beam. The existence of chirp in the signal pulse leads to a loss of coherence and gives rise to interference lobes in the wings of the autocorrelation trace. Fig. 4b shows the corresponding collinear intensity autocorrelation trace for the same pulse, with the solid curve indicating the best sech2 fit yields a 149 fs pulse width for the uncompressed pulse. The OPO signal pulse spectrum as shown in Fig. 5 is asymmetric an.d broadens significantly with increasing output power. At the peak OPO output power the spectra width is quite sensitive to the cavity length, and the noise in the spectra seems to be larger than the total RMS noise (.5%). This is due to a lack of either frequency stability in the pump laser or optical cavity length fluctuations. Because of the "hard" gain 58 / SPIE Vol. 2041

10 aperture of the pump laser the OPO signal pulse round-trip time will track the inverse frequency of the pump laser by spectrally tuning itself along the dispersion curve defined by the mirrors and KTP crystal. The dispersion at 1.3 p.m (used in our experiments), 2.6 fs/loo nm, is about a factor of 20 smaller than the experiments at 0.8 pm6, which implies a very high sensitivity to both frequency and cavity length fluctuations than seen before (partially explaining the noisy spectra). This is a disadvantage of infrared OPO's if spectral noise performance is important. However we are currently considering ways to minimize this. 8.0 Cl) r 2.0. Cl) a) 'I Cl) Time Delay (fs) Fig. 4 Interferometric and intensity autocorrelation traces for the signal beam at 1.3 rim. (a) and (b) correspond to approximately 180 mw average pulses before compression, (c) and (d) to the same pulses after compression. We have identified the chirping of the pulse with self-phase modulation of the high intensity beams at the KTP crystal. Because of the strong chirp of pulses in the cavity we have used an external prism pair to compress the pulses as illustrated in Fig. 4. With the equilateral SF4 prisms oriented at Brewster's angle we were able to obtain compressed signal beam pulses as short as 62 fs while still maintaining more than 95% of the beam energy. The time-bandwidth product of the compressed signal pulses (but without the correction for the autocorrelator crystal dispersion) is 0.68, which is 1.4 times that of the Ti:sapphire pump pulse. The fact that the beats exist even at lower intracavity power and after SPIE Vol /59

11 pulse compression demonstrates that these pulses are still not "transform-limited" and there is still pulse energy in the wings. Furthermore, such pulses may still contain nonlinear chirp or the positive GVD in the OPO cavity may not be completely compensated. As Diels et a!. have previously discussed14, only intracavity prisms can truly give transform-limited pulses with smooth symmetric spectra; it is impossible to modify the spectra with only linear optical elements outside the cavity. Recently, a Ti: sapphire pump OPO has been demonstrated with intracavity prisms'0. The output pulses were transform-limited and the spectra was smooth and symmetric, but the prisms reduced the output power by a factor of three by increasing the cavity loss and reducing the temporal overlap of the 57 fs circulating pulses with the 125 fs pump pulses. In comparison, extracavity prisms have the advantage of higher output power while still being able to achieve comparable pulse widths, because higher power creates greater self-phase modulation. The trade-off is non-bandwidth-limited output pulses. C/Ti. C') Q) II Wavelength (nm) Fig. 5 Spectrum of signal pulse at 1.3 p.m for average power of a) 185 mw and b) 30 mw. The spectrum of the (signal) pulse is always asymmetric. This reflects the complex nature of the parametric interaction. The relative height of the long and short wavelength peaks alters with cavity detuning with the long wavelength peak highest when the cavity length is slightly shorter than ideal. This also is the situation required for optimum power, and reflects the group velocity dispersion induced temporal walkoff of the pulses relative to each other. The asymmetry also reflects the fact that the pulse shape that develops is asymmetric with a steeper rising than falling edge, mainly due to pump depletion. This leads to a particular chirp characteristic in the observed spectra. Finally, it is interesting to note that, because of the 10 times higher intracavity intensity and the change in phase-matching angle for =756 nm, we see collinear, unphase - matched intracavity second 60 / SPIE Vol. 2041

12 harmonic generation (at 655 nm) of the JR signal beam and noncollinear, unphase - matched sum frequency mixing (at 480 nm) of the signal and the pump, with power of approximately one milliwatt and 50 iw, respectively. The second harmonic generation process is e+e.e, exploiting the large d33 nonlinear tensor element. The sum frequency generation process is e+o o (signal + pump SFG). The position of the residual pump beam, sum frequency beam and second harmonic beam are consistent with the noncollinear geometry. It is possible to frequency double the intracavity beam or mix it with the residual pump beam to develop tunable femtosecond sources in the visible as indeed Ellingson and Tang16 have now shown near 600 nm using an additional subcavity and a BBO crystal. For long term stability we have actively stabilized the cavity length with one of the cavity mirrors mounted on a PZT stage which is part of a feedback circuit based on sampling the cavity output. With the low noise of the Ti:sapphire laser, < 1% rms amplitude noise on the OPO output has been achieved over hours. The diffraction-limited Ti:sapphire beam results in a smooth profile, with no evidence of higher order modes. Recently Nebel et al. have shown that KTP can also be used to produce fs pulses in a noncritical phase matching geometry, although the range of tuning is not as high as in the critically phase-matched geometry. Also the group at Spectra Physics12 have demonstrated an OPO with pulses as short as 40 fs in noncritcally phase-matched, temperature-tuned LBO but details of the system are unknown. 4. EXTENSIONS OF WAVELENGTH RANGE The use of a high average power, stable laser such as the Kerr-lens mode-locked Ti:sapphire laser together with the Ti-OPO naturally allows one to seriously contemplate the possibilities of generating femtosecond pulses over a wide spectral range.15 Nebel and Beigang16 have demonstrated efficient external conversion of a continuously mode-locked picosecond Ti:sapphire laser using second, third and fourth harmonic generating schemes. Wavelengths as short as 205 nm have been achieved with average powers 1 0 mw. A similar scheme should make it possible to generate femtosecond pulses from the Ti:sapphire laser as illustrated in Fig. 6. Intermediate wavelengths not accessible from harmonic generation can be produced with a visible parametric oscillator based on one of the nonlinear crsytals indicated above. Alternatively, one can use the infrared, KTP-based optical parametric oscillator, with external or internal doubling to fill in the range below that covered by the Ti:sapphire laser itself as Ellingson and Tang17 have done. One can then envision having femtosecond laser pulses tunable from 200 nm to more than 4.0 ppm. At the peak OPO output power the spectra width is quite sensitive to the cavity length, and the noise in the spectra seems to be larger than the total RMS noise (5%). This is due to a lack of either frequency stability in the pump laser or optical cavity length fluctuations. Because of the "hard" gain aperture of the pump laser the OPO signal pulse round-trip time will track the inverse frequency of the pump laser by spectrally tuning itself along the dispersion curve defined by the mirrors and KTP crystal. The dispersion at 1.3.tm (used in our experiments), 2.6 fs/100 nm, is about a factor of 20 smaller than the experiments at 0.8 im6, which implies a very high sensitivity to both frequency and cavity length fluctuations than seen before (partially explaining the noisy spectra). This is a disadvantage of infrared OPO's if spectral noise performance is important. However we are currently considering ways to minimize this. To extend the wavelength into the infrared region one can consider the possibility to using difference frequency mixing of signal and idler beams to generate pulses even further out in the infrared and possibly as far as 22 im using the difference frequency mixing scheme illustrated in Fig. 7. Several authors2'1822 have demonstrated this approach for nanosecond and even picosecond pulses in materials such as GaSe, CdSe, AgGaS2 and AgGaSe2. Some of the properties for four of these materials are listed in Table 2. These crystals, however, are very soft and susceptible to damage even when used with picosecond pulses. We are currently working15 to see if the expected higher damage threshold for SPIE Vol /61

13 O.45-O.7m 4.5p.m /1.m Fig. 6. Scheme to produce femtosecond pulses from pm. OPTICAL PARAMETRiC OSCI LLATOR SIGNAL (w1) STAL "Wi -Wi' IDLER (w2) Fig. 7. Scheme to produce femtosecond pulses in the far-infrared using difference frequency mixing of the signal and idler beams from a femtosecond pulseopo. femtosecond pulse operation will occur. With appropriate dispersion compensation under ideal circumstances it should be possible to generate only one cycle of radiation near 20 p.m! The output power will be very low since the conversion efficiency for difference frequency scales as the square of the difference frequency. Using a 1mm thick piece of AgGaSe2 we have been able to generate <50 tw of average power in a single pass geometry using difference frequency mixing, but details of the pulse characteristics are still being measured. This is similar to that reported by Dykaar et al. 23 who used difference frequency mixing of the two beams of a two colour self-mode-locked Ti:sapphire laser. Their technique represents an alternative approach to generating femtosecond pulses in the infrared, although starting with shorter wavelengths will limit the tuning possibilities of such a source and the minimum wavelength they can generate will be more than 3.5 jim, limited by the tuning range of their Ti:sapphire laser. 62 / SPIE Vol. 2041

14 Table 2: Properties of Some Far-infrared Nonlinear Optical Crystals Crystal GaSe AgGaSe CdSe AgGaS Symmetry 4m2 4m2 6mm 4m2 Nonlinearity (pm/v) d d36 65 d d Transparency (tim) Data takenfrom Reference 4 5. FUTURE DIRECTIONS The last few years have seen significant advances in the generation of short light pulses in the infrared region of the spectrum in the femtosecond time domain. In future one can expect to see tailored engineering of infrared parametric oscillators for different purposes taking advantage of the variety of nonlinear crystals, pumping geometries, tuning techniques etc.. At the same time, given possible applications in nonlinear physics and chemistr.y, efforts will occur to amplify the oscillator sources or use cavity-dumping schemes. As with the variety of routes available to obtaining femtosecond pulses via different routes in the visible region of the spectrum, there are a variety of routes available to achieve high energy pulses in the infrared. Already, Joosen, and co-workers24 have constructed an infrared parametric generator/amplifier based on a two stage system with type-i phase-matched BBO crystals. The 10 Hz system which is pumped by an amplified, dispersion-compensated CPM laser produces 150 fs pulses and up to 100 pj per pulse; tunability is between 775 and 3000 nm. Mother significant trend involves the use of diode-laser pumping technology. Ebrahimzadeh et al25 have experimented with a doubly-resonant OPO based on an LBO crystal pumped with a Q-switched mode-locked diode pumped Nd:YLF source (pulsewidth of 33 ps). They obtained a tuning range of jim, with microjoules of energy per pulse. The crystal was used in a type-i, non-critically phase-matched geometry and could be temperature tuned. McCarthy and Hanna26 have produced a singly-resonant, synchronously pumped OPO based on pumping KTP by a continuously mode-locked, diode-pumped Nd:YLF laser. The system, shown in Fig. 3, produces up to 42 mw of average power in 1.5 p5 pulses tunable from 1.00 to 1.1.tm. Certainly the tunability of these low average power sources is somewhat limited at the present time but one can anticipate continuing development of diode-pumped systems because of the lower cost, smaller size, and higher efficiency compared to main-frame based systems. With the suitability of diode-pumping (at 680nm) for the seif-modelocked LiCAF laser, it may even be possible to see all-solid state OPO's with the LiCAF as the pump source. The future looks very good for the development of diode-pumped systems. 6. CONCLUSIONS The development of new nonlinear crystals with better figure of merits for nonlinear downconversion and the availability of high average power, femtosecond pulse sources such as the Ti:sapphire laser, and more recently, the LiCAF and LiSCAF lasers, are causing a revolution in femtosecond pulse generation via parametric processes and the future will continue to see many new developments in this area. Diode-pumped solid state lasers will also see increasing use as OPO pumps in the future. To obtain femtosecond pulses in the infrared a variety of crystals and frequency mixing configurations can be explored. 7. ACKNOWLEDGMENTS We gratefully acknowledge the Natural Sciences and Engineering Research Council of Canada, the SPIE Vol / 63

15 Premier of Ontario's High Technology Fund, and Bell-Northern Research that have generously supported our research. 8. REFERENCES 1. K. Burneika, M. Ignatavicius, V. Kabelka, A. Piskarskas, and A. Stabinis, IEEE J. Quan. Elec. 8, 574 (1972). 2. A. Lauberau, "Optical Nonlinearities with Ultrashort Pulses", in Ultrashort Light Pulses and Applications, Ed. by W. Kaiser (Springer-Verlag, New York, 1988), p D.E. Spence, P.N. Kean and W. Sibbett, Opt. Lett. 16, 42 (1991); D.K. Negus, L. Spinelli, N. Goldblatt and G. Feugnet, in Digest of Topical Meeting on Advanced Solid State Lasers, (Optical Society of America, Washington, D.C., 1990); M. Pichë, N. McCarthy and F. Salin, in Digest of Optical Society ofamerica Meeting (Optical Society of America, Washington, D.C., 1990), paper MB8. 4. R.L. Byer in Quantum Electronics, vol. 1, part B, ch.9, Ed. by R. Rabin, C.L. Tang (Academic New York), 1975.; C.L. Tang in Quantum Electronics, vol 1, part B, ch.6, Ed. by R. Rabin, C.L. Tang (Academic New York), 1975; R. L. Byer and R. L. Herbst, Parametric Oscillation and Mixing, in Nonlinear Infrared Generation, Topics in Applied Physics, Ed. by Y.-R. Shen (Springer-Verlag, New York, 1977), p M.F. Becker, D.J. Kuizenga, D.W. Phillion and A.E. Siegman, J. Appl. Phys., 45, 3996(1974). 6. D.C. Edelstein, E.S. Wachman, and C.L. Tang, Appl. Phys. Lett. 54, 1728 (1989); E.S. Wachman, D.C. Edelstein, and C.L. Tang, Opt. Lett. 15, 136 (1990); E.S. Wachman, W.S. Pelouch, and C. L. Tang, J. Appl. Phys. 70, 1893 (1991). 7. E.C. Cheung and J.M. Liu, J. Opt. Soc. Am., 8, 1491 (1991). 8. G. Mak, Q. Fu and H.M. van Driel, Appl. Phys. Lett. 60, 542 (1992). 9. Q. Fu, G. Mak and H.M. van Driel, Opt. Lett., 1006 (1992); G. Mak, Q. Fu and H.M. van Driel, Ultrafast Phenomena VIII (Spring Verlag, New York, 1992); Q. Fu, G. Mak and H.M. van Driel, Conference on Lasers and Electro-Optics, Baltimore, Md., paper CWD1, (1992). 10. W.S. Pelouch, P.E. Powers and C. L. Tang, Opt. Lett., 17, 1070 (1992) A. Nebel, C. Fallnich and R. Beigang, Conf. on Lasers and Electro-optics, Baltimore Md., 1993, Paper JW J.D. Kafka, M.L. Watts, J.W. Pieterse, Conf. on Lasers and Electro-optics, Baltimore Md., 1993, Paper CPD A. Piskarskas, V. Smil'gynvicyus A. Umbrasas, A. Fix and R. Wallenstein, Opt. Commun. 77, 335 (1990). 14. J.-C. M. Diels, J.J. Fontaine, I. McMichael and F. Simoni, Appl. Optics, 24, 1270(1985). 15. H.M. van Driel and G. Mak, Can. J. Phys. 71, 47 (1993). 16. A. Nebel and R. Beigang, Opt. Lett. 16, 1729 (1991). 17. R.J. Ellingson and C.L. Tang, Opt. Lett. 18, 438 (1993). 18. A.J. Campillo, R.C. Hyer, S.L. Shapiro, Opt. Lett. 4, 325 (1979). 19. T. Elsaesser, Seilmeier and W. Kaiser, Opt. Commun. 44, 293 (1983); T. Elsaesser, A. Seilmeier, W. Kaiser, P. Koidl and G. Brandt, App. Phys. Lett. 44, 383(1984); T. Elsaesser, H. Lobentanzer and A. Seilmeier, Opt. Commun. 52, 355 (1985). 20. R.C. Eckardt, Y.X. Fan, R.L. Byer, C.L. Marquardt, M.E. Storm and L. Esterowitz, Appl. Phys. Lett. 49, 608 (1986). 21. R.L. Byer, M.M. Choy, R.L. Herbst, D.S. Chemla and R.S. Feigelson, Appl. Phys. Lett. 24, 65 (1974). 22. D.C. Hanna, B. Kuther-Davies, R.C. Smith and R. Wyatt, Appl. Phys. Lett. 25, 142 (1974). 23. D.R. Dykaar and S.B. Durack, Conf. on Lasers and Electro-optics, Baltimore Md.,1993, Paper CFG W. Joosen, P. Agostini, G. Petite, J.P. Chambaret and A. Antonetti, Opt. Lett. 17, 133 (1992). 25. M. Ebrahimzadeh, G.J. Hall and A. I. Ferguson, Appl. Phys. Lett., 60, 1421 (1992). 26. M. J. McCarthyand D. C. Hanna, Opt. Lett., 17,402 (1992). 64 / SPIE Vol. 2041