1. Introduction

In recent years there has been a high level of interest in the properties and applications of fast semiconductor saturable absorbers operating at λ ∼ 1.5μm. These have been used for fabrication of passively modelocked lasers [1–3] and for the wavelength conversion of optical data streams through cross-absorption modulation [4]. To enable these effects to be utilised at the highest repetition rates, very fast absorption recovery times are required, in materials that can be conveniently integrated with optical gain elements. Recovery times of the order of 5ps have been obtained in reverse biased electroabsorption modulators [4], and in shorter wavelength quantum well (QW) material, [5] but a full study of absorption recovery in 1.5μm emitting quantum wells is presently lacking. To obtain recovery times significantly shorter than this, several techniques have been used. These include the introduction of excess defects into the material have been used to provide recovery times τ_a well below 5ps, including ion-implantation and non-stoichiometric crystal growth [6, 7]. Similarly short values of τ_a have been achieved by the use of intersubband transitions [8]. At λ ∼ 1.3μm, very short recovery times have been demonstrated, reducing to sub-ps values under high reverse bias (∼ 10 V) [9]. However, these methods present the problem that they do not permit the integration of saturable absorbers with QW gain sections without the use of epitaxial regrowth steps.

In this work we present experimental evidence for values of τ_a below 5ps in reverse biased AlInGaAs multiple quantum wells (MQWs) designed to emit at λ ∼ 1.5μm. This materials system is widely used for 1.5μm semiconductor lasers due to its superior performance at high temperatures when compared to equivalent structures using InGaAsP [10, 11]. This effect is observed within the gain peak of the material when forward biased, and so presents an ideal system for the integration of gain elements with fast saturable absorbers. Recent studies of MLLs fabricated from this material [3, 12, 13] have shown that very short pulse durations can be obtained; this can be understood in terms of a shorter τ_a.

2. Sample structure

The samples used in the experiment were based on commercial AlInGaAs 1550 nm laser diode (LD) wafers. The active region consisted of 5 MQWs (6-nm-thick quantum wells separated by 10-nm-thick barriers), where both wells and barriers are of Al-quaternary material. Compressive strain (1.2% in the QWs and 0.5% in the barriers) ensured that the radiative transition was predominantly TE polarized. Monolithic ridge waveguide-based lasers were fabricated using state-of-the-art electron-beam lithography and dry etching techniques. A range of different device structures, in terms of reflector design and cavity length, were used in during the work. The photocurrent measurements reported in Section 3 were carried out by injecting light into the reverse biased gain section of a simple Fabry-Pérot mode-locked laser (sample A). In Section 4, dealing with the wavelength conversion of short optical pulses, two different samples were used. Sample B was a DBR laser based on a 140μm long numerically optimized surface-etched grating [14], where the light was incident on the grating end of the device with an antireflection coated facet. A variable DC reverse bias was applied to the grating section of the device, which had an effective length at transparency of ∼ 35μm and a calculated reflectivity (R) of 37%. The high attenuation of the unbiased gain section avoided unwanted reflections from the far facet. Sample C had a 10° angled front facet (R ∼2%), a 50μm long cavity and a deeply etched facet-like termination of a ridge waveguide (etch depth ∼ 3μm with R calculated as ∼30%), reflecting the pulse back along the waveguide, to which a variable reverse bias could be applied. All three measured samples were fabricated from the same epitaxial wafer.

3. Photocurrent measurements of recovery times

To provide a first measurement of the absorption recovery time, measurements were made using a pump-probe technique with detection based on the photocurrent induced in the sample. Conventional pump-probe techniques have limitations in the measurement of absorption recovery times in waveguide samples, since to obtain an accurately measured pump-probe trace, a measurable quantity of probe light must be incident on the detector even at long delay times when the absorption has fully recovered. If high probe intensities are to be avoided, then very short waveguides must be used; as well as presenting difficulties in fabrication and handling, this means that the measurement cannot be carried out on existing laser samples. In order to avoid these difficulties, we used a photocurrent detected technique [15]. This has the advantage that it can be performed on samples of any waveguide length, meaning that investigations of the carrier dynamics under reverse bias need not be confined to specially fabricated samples.

To carry out the experiment, the output of a 10GHz, 2.5ps mode-locked fiber laser was divided into two equal beams using a 3dB fiber coupler. One beam was delayed relative to the other using a variable fiber delay line, and the beams were recombined using another 3dB coupler; the polarization of the two beams could be independently adjusted to ensure that both were TE polarized. The two beams were then coupled into the waveguide of sample A using a lensed optical fiber. A schematic diagram of this experimental setup is shown in Fig. 1(a). The power coupled into the sample from each beam is estimated as ∼ 2mW.

 

Fig. 1 (a) Schematic diagram of the photocurrent detected pump-probe setup used to measure τ_a (b) Symbols show the measured absorption recovery times τ_a, as a function of the applied bias. The error bars represent the standard deviation of a number of measurements under the same conditions. The blue line shows the values of τ_a calculated using Eq. (1). (c) Typical trace of photocurrent versus delay line position. (d) The absorption recovery rate γ_a plotted on a log scale versus applied bias. The blue line shows the rate calculated using Eq. (1).

Download Full Size | PDF

The absorption recovery time was obtained by measuring the photocurrent as a function of delay line position, and fitting the results to a single exponential function. By using slow electronics, the detected current is the sum of the photocurrents due to the two arms of the setup. If the difference in optical path length between the two beams is small, then the absorption saturation arising from the first pulse will not have recovered when the second pulse arrives, and so the induced photocurrent will be reduced. A typical trace of detected photocurrent against delay line position is shown in Fig. 1(c); its symmetrical shape about the zero delay point arises because the two pulses have the same power and polarization.

Measurements of τ_a were carried out at different levels of reverse bias. The results of this are shown in Fig. 1(b). A monotonic decrease in τ_a can be seen, as the reverse bias is increased. This is attributed to an increase in the rate of thermionic emission of carriers from the quantum wells. The same data expressed in terms of the absorption recovery rate ${\gamma }_a={\tau }_a^{-1}$ is plotted on a logarithmic scale in Fig. 1(d). It can be clearly seen that these lie close to a straight line, indicating that even at the highest bias investigated, thermionic emission of carriers dominates over any tunnelling process [9]. Based on the assumption that the absorption recovery is predominantly due to thermionic emission, it is possible to calculate the expected recovery times as a function of reverse bias, V using the expression [16]:

4. Recovery times measured through wavelength conversion

The very short values of τ_a measured in this material reveal its potential to be used in a number of applications where high speed cross- and self- absorption saturation can be used along with the associated modulation. These applications include the wavelength conversion of optical data streams, and associated functionality such as multicasting, conversion of data to RZ format and the retiming of RZ datastreams. It is a particularly interesting material with regard to these applications, since it can be conveniently integrated with optical gain elements operating at the same wavelength.

As a first step towards an investigation of these applications, we performed measurements of wavelength conversion through cross-absorption modulation. These measurements were carried out in a reflection geometry; a schematic diagram of the setup used is shown in Fig. 2(a). A stream of 10GHz, 2.5ps pulses was combined with a slightly different wavelength CW probe beam using an optical coupler and focused into the sample waveguide using a lensed fiber. A circulator was used to separate out the light reflected from the sample, which was then amplified, and filtered so that only the probe wavelength remained. This was then visualized using the optical input on a sampling oscilloscope with a nominal bandwidth of 65GHz.

 

Fig. 2 (a) Schematic diagram of the setup used for the wavelength conversion measurements. (b) (Top trace) λ ∼ 1550nm input pulses used for the experiment. This trace is not to scale and has been offset in the vertical direction for clarity. (Other traces) λ ∼ 1570nm output traces obtained from sample A under bias conditions from 0 to 3.5V, in 0.5V steps (from top to bottom) (c) The extinction ration and peak output powers obtained from sample B as a function of applied bias.

Download Full Size | PDF

The wavelength converted pulses obtained from sample B under a variety of applied biases are shown in Fig. 2(b), together with the incident pulses. The reduction in τ_a with increasing bias can be clearly seen. It should be noted that the apparent pulse duration of the incident and wavelength converted pulses at high bias is limited by the bandwidth of the oscilloscope. In this case, the mode-locked fiber laser was set to produce pulses at ∼1550nm and +5dBm, while the CW probe beam was at ∼1570nm, −10dBm, resulting in the conversion of the incident pulses to a longer wavelength. Figure 2(c) shows the extinction ratio, r and the peak power obtained from these traces. The extinction ratio of the wavelength converted pulses, showed a steady increase up to a maximum of 51%. This occurred at a bias of 2.5V, after which it decreased again. In general, r is related to the interplay between saturable and unsaturable losses. In the reflection configuration used for these measurements, the measured r will be adversely affected by additional reflections, such as from the facet of the device, raising the baseline of the detected 1567nm signal. It is expected that much higher values of r could be achieved by using an optimized transmission geometry for wavelength conversion in the material.

Figure 3(a) shows results obtained from similar experiments, using sample C. In this case, the relative spectral positions of the input pulses and the probe beam were reversed, using ∼1570nm incident pulses, and CW probe beams around 1545–49nm. The measured reflectivity spectrum of this sample (Fig. 3(c)) shows clear evidence for a residual Fabry-Pérot component to the spectrum, owing to the finite reflectivity of the angled front facet (R∼2%). This was suppressed in the previous measurements using sample B, due to the high absorption of the unbiased gain section of the device. The probe wavelength used at a given bias was chosen to lie close to a Fabry-Pérot peak, giving the highest extinction ratio for the wavelength converted pulses. At 0.5V and above, recovery between successive pulses was sufficient to allow the trailing edge of the wavelength converted pulses was fitted to a single exponential function, giving an estimate for the value of τ_a, which is plotted as the black curve in Fig. 3(a). The values obtained from the photocurrent detected pump-probe measurements reported above are plotted in blue. A close agreement between the two methods can be observed at high biases, where the recovery time τ_a is a small fraction of the repetition rate of the pulses. In addition, we have also plotted the values of τ_a reported by Nishimura et al. from their measurements of wavelength conversion in an electroabsorption modulator based on InGaAsP MQWs [17]. Higher bias voltages are required to obtain a given value of τ_a in the InGaAsP device, but without further device details it is not possible to give a full comparison in terms of electric field.

 

Fig. 3 (a) The black curve shows values of the absorption recovery time τ_a obtained by fitting the wavelength converted signal to a single exponential function. The symbols denote different wavelengths used for the CW beam: ■ 1546.5nm, ● 1547.2nm, ▲ 1548.4nm, ♦ 1549.4nm. The blue open symbols and line show the values of τ_a measured from using the photocurrent detected pump-probe technique. The grey line shows values of τ_a reported by Nishimura et al. for InGaAsP MQWs [17]. The top axis shows electric field values calculated for the devices investigated in this work, and does not apply to the the Nishimura et al. data. (b) The measured FWHM of the wavelength converted pulses. Symbols have the same meaning as those in panel (a). (c) Reflection spectra of sample C measured under applied biases from 0V to 4V, at 1V intervals, showing the residual Fabry-Pérot resonances. The Stark shift of the absorption edge with increasing voltage can also be clearly seen.

Download Full Size | PDF

Figure 3(b) shows the measured full-width half maxima of the wavelength converted pulses. Due to the limited bandwidth of the oscilloscope, the input pulse FWHM was measured as 7.5ps versus an autocorrelator measurement which gave 2.5ps. At a bias of 4.0V, the measured output pulse FWHM under these circumstances is 9.8ps. It is anticipated that the actual width is smaller than this value. The use of an improved device architecture could potentially reduce these values further, for example by following the wavelength conversion process with a delayed interferometer [4].

5. Discussion and conclusions

The short recovery times measured here provide insight into recent results in 1.5μm emitting MLLs. Recent evidence suggests that by using AlInGaAs quaternary material for the gain and saturable absorption, the output pulse durations can be much shorter than those previously obtained from phosphorus containing QW-based material [3]. These properties can be more easily understood if the absorption recovery time is lower in Al-quaternary material. Recent theoretical investigation of MLL dynamics [18] has indicated that for absorption recovery times below ∼ 10ps, the pulse duration is closely connected to τ_a. (Fig. 5(e) in [18]). These studies also indicate that for recovery times in the 10 – 25 ps range, the pulse duration is relatively unaffected by τ_a. We note that in [3] broad output pulses (FWHM ∼ 13ps) were obtained with the absorber section unbiased, and a consistent trend towards shorter pulses was observed as the absorber was biased towards 3.2V, the highest bias studied in that work. The results presented here show that over this range, τ_a is changing from > 30ps for the unbiased material to < 5ps. It should be noted that for a complete picture of the effects of absorption recovery on MLL dynamics it is necessary to include possible power dependence of τ_a, as has been reported by Dahdah et al. [19] for electro-absorption modulators. The pulse energies used in this work are in the 50–100fJ/pulse range. This is comparable with the circulating intracavity power of the devices reported in [3] only up to 10–20% above laser threshold. For higher output powers than this, the absolute values of the recovery time may be expected to be longer, but the trend of decreasing τ_a with higher bias will remain.

In conclusion, we have measured the absorption recovery time in AlInGaAs QWs, finding it is below 5ps for reverse biases above 3V. This short recovery time offers a potential explanation for recent reports of very short pulse durations from mode-locked lasers based on this material. Cross absorption modulation based wavelength conversion of a 10GHz pulse stream was carried out indicating the potential of this material for use in a number of applications requiring the integration of ultrafast absorption saturation with optical gain elements operating at the same wavelength.