A. Material Structure
In this work, 1.06-µm InGaAs/GaAs DQW laser structures, with asymmetric waveguide, were grown on n-GaAs(100) substrates by metalorganic chemical vapor deposition (MOCVD). The detailed epitaxial structure is shown in the Table I , similar to our previous study [22].
TABLE I.1.06-µm InGaAs DQW laser with asymmetric heterostructure layers
Material
|
Thickness (nm)
|
Doping (cm-3)
|
GaAs
|
250
|
p, ˃2e19
|
Al0.55Ga0.45As
|
300
|
p, 6e18
|
Al0.35Ga0.65As
|
300
|
p, 8e17
|
Al0.3Ga0.7As
|
300
|
p, 8e17
|
Al0.05Ga0.95As
|
200
|
p, 8e16
|
GaAs
|
8
|
barrier
|
In0.2Ga0.8As
|
7
|
quantum well
|
GaAs
|
8
|
barrier
|
In0.2Ga0.8As
|
7
|
quantum well
|
GaAs
|
8
|
barrier
|
In0.03Ga0.97As0.95P0.05
|
200
|
non-doped
|
In0.32Ga0.68As0.4P0.6
|
300
|
n, 8e16
|
In1-xGaxAs0.5P0.5
|
100
|
n, 8e17
|
In1-xGaxAs0.7P0.3
|
100
|
n, 8e17
|
In1-xGaxAs0.9P0.1
|
800
|
n, 8e17
|
GaAs
|
200
|
n, 2e18
|
B. Device structure and manufacturing process
The MLL fabrication process is similar to those reported in Refs [23, 24]. First, a 6 µm-wide and 1.3 µm-deep ridge waveguide was formed by standard lithography and wet etching. A 300-nm-thick SiO2 film was deposited as an electrical isolation layer, and a 3 µm-wide p-type electrode window was opened on the etched ridge less than 6 µm-wide (due to the ridge etch). Then, a 10 µm-wide electrical isolation region was patterned by another lithography step. Ti/Au layers were evaporated to form the p-side ohmic contact. After that, lift-off process was carried out to expose the electrical isolation region, and a consequent wet etching process was used to improve the electrical isolation effect. An isolation resistance of ~1kW was achieved between the two sections. Finally, the substrate was thinned, Ni/Ge/Au/Ni/Au layers were evaporated on the n-side, and then an ohmic contact layer was formed after 410 °C annealing. The processed wafer was cleaved into laser chips for device characterization. During the characterization, a thermo-electrical cooler (TEC) was used to control the operation temperature for the non-soldered MLL. In this study, the MLL was tested p-side up on the TEC, the length of the gain section (Lg) and the saturable absorber (SA) section (La) are 1576 µm and 404 µm, respectively. The gain section is forward driven by current (Ig) while the absorber section is reversely biased with Va. The side view of the two-section MLL with gain section and SA section is shown in Fig. 1. No coating on the cavity facet of MLLs was prepared.
C. Measurement setup
The reverse bias voltage is directly applied to the saturation absorption section by Keithley 2520 power supply, and the photocurrent value is automatically read.
The output light from the gain section facet was first coupled into a single mode fiber, then was split by a 10:90 fiber coupler, the 10 percent was guided into an optical spectrum analyzer (OSA, AQ6370C), and the 90 percent was further split into two equal parts: one was fed into a high-speed photodiode (PD) detector monitored by a 30-GHz real-time oscilloscope (DSO93004L); another was fed as the signal source for an electrical spectrum analyzer (ESA, N9030A). The details are shown in the Fig.2.
D. Experimental results and analysis
The measurement sensitivity is set constant for all measurements. For the tested laser in this study, the wet etched ridge width provides single spatial mode operation condition while operating in the mode locking regime. Figure 3 shows the light output power characteristics under different reverse bias voltages of the absorber section for the tested device in the continuous wave (CW) operation at room temperature (RT). The bias voltage (Va) applied to the absorber section varied from +1V to -3 V, with a 1V step, and Ig increased from 0 to 200 mA under each Va. The use of long absorber sections (up to 20.4 % of total length) resulted in increased absorption and was beneficial to mode locking and obtain ultrashort pulse. As the negative bias on the absorption region increases, the threshold current changes obviously differently in two intervals, one is the interval from +1 to -1V, where the threshold current increases slowly; the other is the interval from -1 to -3V, the threshold current increases rapidly with increased negative bias. This phenomenon should be caused by the strong absorption of the absorption section under a strong negative bias Va. The L-I and V-I curves show the good operating characteristics of the MLL.
Stable mode locking was achieved over a wide range of Va (from 0 to -3V). The RF spectra were measured at Ig=180 mA and Va=-0.9 V as shown in Fig. 4(a) using a resolution bandwidth (RBW) of 10 kHz.. The fundamental repetition rate with more than 42 dB signal to noise ratio (SNR) at room temperature is ~19.56 GHz, corresponding to the photon round-trip time in the 1.99 mm-long laser cavity. The second harmonic frequency at ~39.12 GHz is also marked in the figure with ~15 dB SNR, which indicate an efficient mode locking. The RF signal from the photodetector was also fed into a 30 GHz real-time oscilloscope (DSO93004L). Fig. 4(b) shows the pulse train observed on the oscilloscope. The time interval between two pulses is ~50.78 ps, corresponding well to a repetition rate at ~19.69GHz, which is very close to the fundamental repetition rate ~19.56 GHz. The response resolution of the real-time oscilloscope in picosecond and femtosecond level is not enough. Therefore, the measured time pulse train is actually an envelope of ultra-narrow pulses. It can also be seen from Fig. 4(b) that the distribution of the time-domain pulse train presents a more uniform and regular time distribution and the cycle time change is very small and barely noticeable, which shows that the time-domain pulse envelope is much closer to no chirp. It also proves that it is weakly affected by the chirp effect during the test. This provides a basis for calculating its ultra-narrow pulse width using the ideal sech2 shaped of the time-bandwidth product (TBP) later.
The full width at half maximum (FWHM) of the emission spectrum (Dl) is ~1.08 nm, and the optical spectrum of the laser is centered at 1065.35nm. The tested peak and fitted peak curves are showed in Fig. 5. The ideal Lorentz-shaped TBP of ~0.22 is a smaller value than the ideal sech2 shaped pulses TBP of ~0.315. Here, taking into account that it is not significantly affected by the chirp, and assuming an ideal Fourier transform-limited pulse, a pulse width (PW) of ~1.1 ps can be expected by the sech2 shaped. Under this condition, the average power from the MLL is ~71 mW, corresponding to a peak power ~3.3 W. To our knowledge, this is the narrowest pulse width and the highest peak output power obtained from an electrically pumped edge emitting GaAs-based two-section semiconductor DQW-MLL.It can be seen that pulses get broadened with increasing Ig, and narrowed with increasing Va.
Figure 6 shows the electroluminescence spectrum measured at different Va (a: Va=+1V ; b: Va=0V ; c: Va=-1V ; d: Va=-2V ; e: Va=-3V ). It can be found that the redshift of the lasing peak occurred is moved further away from the center of the spontaneous emission spectrum with an increasing reverse bias. As the reverse bias Va increases, the band-gap of the absorption region shrinks significantly due to the quantum confined Stark effect (QCSE). In the gain spectrum of the MLL, lasing tends to happen at long wavelength while mode gain reaches the threshold current [24].
The peak is shifted from 1063 to 1074 nm near the threshold current under different Va. The peaks redshift, and become narrower and higher with Ig under increasing bias Va in Fig6.(a-e) . Especially, it is also found that lasing peak will be getting obviously narrower as reverse bias changes from +1 to -3V. This should be caused by the more bandgap shrinkage under big reverse bias.
Gain can be determined by amplified spontaneous emission (ASE) spectrum measurement. The net modal gain (Gnet) of the two-section MLL was computed using the Hakki-Paoli method. Fig.7 shows the net modal gain spectra at different Ig when Va was set at +1, 0, -1, -2 and -3V, respectively. The net modal gain is extracted from the peak to valley ratio of sub-threshold Fabry-Perot oscillations, which was computed using the following relation [25].
where Γ is the confinement factor, gm is the material gain, αi is the internal loss related to defects and dopants in the waveguide and QW. S is the peak-to-valley ratio of the FP resonances which can be directly obtained from the ASE spectra, L is the whole cavity length (~1.99 mm), and R1, R2 are facet reflectivity (A value of 0.346 is used for both cleaved facets). The ASE spectra were recorded with a wavelength step size of 0.02 nm. From Fig. 7, it can be seen that, the net modal gain increases with increasing current, and reaches a value close to the mirror loss (am=(1/2L)ln(1/R1R2) =~5.33 cm-1), which is the threshold loss.
In the following, a comparison of ASE and gain spectra for InGaAs/GaAs DQW MLLs is implemented. Fig.7 plots the modal gain spectrum as a function of gain section currents, which is derived from the ASE spectra as shown in Fig.6, and it also shows the lasing spectrum near threshold. The laser lases exactly at wavelength of the net modal gain peak, and same results are observed at other Va, including an internal loss, ai=~(1.56±0.34)cm-1. An obvious redshift in the gain peak is also observed with increasing Va, which is attributed to carrier thermalisation within the DQW active layer. It can be seen that the distribution of net modal gain curve tends to be closer near the threshold current at a larger negative Va value. From formula (1), a larger negative Va causes the first item on the right side of the formula to become closer to zero, which means the value of S approaching to +¥.When S value is getting larger, the modes of the MLL becomes more stable, also exhibiting a higher power output.
Figure. 8 shows the peak net modal gain as a function of gain section injection current at +1, 0, -1, -2 and -3V bias voltage, respectively, all measured just at or below the threshold. With increasing reverse bias voltage, the peak net modal gain decreases at constant gain current, which is similarly reported It revealed that the dissociation time of the exciton (the width of exciton absorption peak) in the saturable absorber determines the pulse width of mode locked laser. The exciton binding energy of is higher for InP system and is lower for GaAs/AlGaAs than that for InGaAs/GaAs. The exciton absorption peak shift toward the longer wavelength as the reverse bias is applied (QCSE effect). On the other hand, the gain peak will shift slightly toward shorter wavelength due to band filling effect as the cavity loss increases with reverse bias to the SA section. As a result, the gain peak and absorption peak are separated (as Refs [26, 27]. also observed). Therefore, in this paper, the bandgap of SA was shifted to the shorter wavelength to match the gain and the absorption peaks. This helps to generate short pulse for mode-locked lasers.
for the QD lasers [28,29]. Peak gain saturation trend is observed under a high bias voltage at room temperature, while the temperature condition in the reference is at 80 ̊C. The slope of peak net modal gain versus current declines obviously with reverse bias voltage increasing, which may be caused by the increased the thermal effect of photocurrent in the absorber. This will be further discussed in the following part.
In Fig. 9, the net modal gain spectra at threshold for the same sample at different Va are plotted. The linear fitting is usually used to derive the internal loss (αi), which relies on the assumption that the internal differential efficiency keeps constant for lasers with different cavity length. The modal internal loss (αi) of the DQW laser is derived to be (1.56±0.34) cm-1, which is a higher value than the data reported previously [30]. The high internal loss should be caused by the significant shrinking of band gap of the saturated absorption region under negative Va. The Va-dependent net modal gain spectra from +1 to -3 V, are measured near threshold current. It can also be seen from Fig 9 that the gain peak is red-shifted and the gain bandwidth is narrowed. These factors lead to an increase in internal optical loss. This is consistent with the internal loss result of the test.
The net modal gain peak redshifts consistently about 10 nm with increasing negative Va, which may be mainly due to the absorption spectrum shift by the quantum well band-gap shrinking effect. Since the lasing start exactly at the location of the net modal gain peak of the MLL, it also implies a potential application for wavelength tunable or multi-wavelength laser [31]. Both of the polynomial fitting curves imply a very good adjustability. As expected, at all absorber biases, the peak value of net modal gain has nearly a similar value which is mainly determined by mirror loss.
Figure 10 shows the photocurrent (Is) at absorption region increases as Ig increases, and increases with the Va negatively increasing. Is comes from the saturated absorption region under reverse bias, and its current direction and magnitude are related to the applied bias. When Ig>140mA, Is saturates in the reverse bias voltage range of -3 to -2V. Comparing the effect of Ig on photocurrent, a higher Ig has much greater influence on the photocurrent of MLL, while Va has important effect on photocurrent saturation. For example, when Va is in the range of -3 to -2V, Is increases with Va negatively increasing, and then saturates. Especially, in Fig.10, the increment of Is versus gain current (Ig) is uneven under -2V bias. And the efficiency that gain current converts photon into photocurrent gets to the highest value from 120 to 140mA under -2V reverse bias, which indicates the most photons are absorbed by the absorption section. So mode-locking always happen at high photon absorption status of the absorber, such as Va= -0.9V while Ig=180mA, as shown in Fig. 4(a).