Large area photoconductive terahertz emitter for 1.55 μm excitation based on an InGaAs heterostructure

Terahertz (THz) time-domain spectroscopy (TDS) exploits the broad emission spectrum of sources which are excited with femtosecond near-infrared pulses. The possibility to register these pulses in amplitude and phase offers a wide range of applications such as studies of the complex dielectric function of semiconductors, highly correlated systems and organic molecules [1, 2], nondestructive materials investigations [3] and security applications [4]. The most widely used THz sources in TDS systems are photoconductive antennas [5, 6] based on GaAs, which are excited with Ti:sapphire lasers at wavelengths around 800 nm. In order to make use of the full power available from state-of-the-art laser systems large area emitters with active areas in the range of 0.1–1 mm² have been employed [7, 8]. The large area, combined with convenient operation at low bias voltages, is provided by scalable emitters [9]. Such emitters, excited with a standard Ti:sapphire laser oscillator, are capable of providing THz fields above 250 V cm⁻¹ [10]. For production of turnkey THz spectrometers at significantly lower cost it is highly attractive to replace Ti:sapphire lasers by fibre lasers operating at the telecommunication wavelength of 1.55 μm. This requires THz emitters based on semiconductors with an energy gap below 0.8 eV. Small-gap materials like InSb, InAs and GaSb have been employed since the early times of THz-TDS in the form of surface field emitters or photo-Dember emitters [11–15]. However, the efficiency of these emitters is limited by the unfavourable emission geometry, which also prevents direct coupling of such emitters to optical fibres. Recently lateral photo-Dember emitters have been built, which do not suffer from these limitations [16, 17]. The development of photoconductive antennas on low-gap semiconductors is challenging, since these materials exhibit low resistivities. The low resistivities are caused by the absence of a significant Schottky barrier and the presence of a large carrier concentration. Photoconductive antennas for excitation at 1.55 μm have been successfully demonstrated at materials engineered for high resistivity such as low-temperature (LT) grown InGaAs [18], ion-irradiated InGaAs [19, 20] and InGaAs/InAlAs heterostructures [21, 22]. Due to the large area and the electrode geometry, the resistivity requirements for scalable emitters are much higher than for standard photoconductive emitters. Recently a scalable emitter based on InGaAs with ErAs nanoparticles has been demonstrated and THz fields of 0.7 V cm⁻¹ have been achieved [23]. In this paper we present a scalable emitter based on a LT-InGaAs/AlGaAs heterostructure capable of emitting THz fields of 2.6 V cm⁻¹. In particular we study the saturation behaviour of these emitters and the optimum near-infrared spot size. The results indicate that this emitter is well suited for TDS systems operated with amplified fibre lasers.

The scalable emitter design provides a large active area and at the same time allows one to achieve high acceleration fields with low bias voltages. GaAs based scalable emitters excited with 800 nm radiation are very efficient as demonstrated by near-infrared-to-THz conversion efficiencies of up to 2 × 10⁻³ [24, 25]. The emitters for this study feature electrodes in the form of two interdigitated combs of metallization on the photoconductive material. This electrode pattern causes the electric field in two neighbouring gaps to point into opposite directions. However, without covering every second gap, the THz fields of two neighbouring gaps would cancel out each other in the far field. To prevent this, every second gap is covered with gold, which is electrically insulated from the electrodes by a 1 μm thick layer of Si₃N₄ (see figure 1(c)). The electrodes and the gaps exhibit a width of 5 μm and 7.5 μm, respectively. The microstructured pattern covers an area of 300 μm × 300 μm. The emitter chip is mounted with a thermally conductive adhesive on a 1-inch aluminum holder, which also serves as heat sink (see figure 1(a)).

Zoom In Zoom Out Reset image size

Before describing the specific properties of the LT-InGaAs/InAlAs heterostructure we briefly discuss the material requirements for pulsed THz generation. The emission mechanism, i.e. the acceleration of photogenerated carriers in a bias field, is most efficient when a large temporal change in photocurrent can be induced upon excitation with short near-infrared pulses. This is typically achieved in materials with small effective mass and large carrier mobility. Even though the ultrafast carrier dynamics can differ significantly from the equilibrium condition characterized by the mobility [26], the mobility can still serve as a performance indicator when different photoconductive materials are compared. In order to apply strong bias fields, high breakdown fields as well as high resistivities are required. Due to the electrode geometry, much higher resistivities are required for scalable emitters as compared to photoconductive antennas. According to these arguments it is clear that semi-insulating GaAs (SI-GaAs) is an ideal material for emitters excited with 800 nm radiation. We note that, unlike for continuous wave (CW)-photomixers and photoconductive detectors, short carrier lifetimes are not required for pulsed photoconductive THz emitters. It has been shown that emitters based on low-temperature grown GaAs (LT-GaAs) are actually less efficient in comparison with SI-GaAs-based emitters [10, 27–29].

For InGaAs the challenge is to prepare materials with a high enough resistivity. The low resistivity along with the small band gap prevents the formation of a significant Schottky barrier, hence only insufficient bias voltages can be applied to a metal–semiconductor–metal structure. One path to increase the resistivity is to trap carriers in deep impurity states. To this end low-temperature growth and Fe⁺-implantation has been employed in order to induce trapping centres [18, 19]. However, unlike LT-GaAs, LT-InGaAs exhibits a fairly low resistivity because the trapping centres in LT-InGaAs are energetically located in the vicinity of the conduction band edge, while they form mid-gap traps in LT-GaAs. This gives rise to a high n-type carrier concentration in LT-InGaAs. The carrier concentration can be reduced by doping LT-InGaAs with Be acceptors; however, it is very difficult to adjust the Be doping level for perfect compensation. Therefore additional means to trap carriers without deteriorating the carrier mobility are desired. One option is to introduce LT-InAlAs layers, which offer additional trapping centres. These traps are energetically located below the conduction band of InGaAs (cf figure 1(d)). To ensure that the photogenerated carriers from the InGaAs layers can reach the trapping centres in the InAlAs layers via tunnelling, the layer thickness should not exceed a few nanometres. Separating the carrier trapping centres spatially from the region where the photogenerated carriers are generated is highly attractive, since the carrier acceleration is reduced only slightly [22]. The photoconductive material employed in our emitters consists of a stack of 100 periods of 8 nm thick In_0.52Al_0.48As alternating with 12 nm of beryllium doped In_0.53Ga_0.47As. The layer structure is grown lattice matched by low-temperature molecular beam epitaxy on semi-insulating InP:Fe (see figures 1(c) and (d)). The growth temperature was 200 ° C; further details on the growth conditions can be found in [21]. We note that due to the large band gap of 1.46 eV no photogeneration takes place in the InAlAs layers. Furthermore these layers exhibit a high resistivity and therefore basically do not contribute to the dark current.

Hall measurements on the multilayer material revealed a resistivity of 576 Ω cm, a residual n-type carrier concentration of 2.7 × 10¹³ cm⁻³, and a mobility of 406 cm² V⁻¹ s⁻¹. This mobility is more than one order of magnitude larger than the mobility of pure LT-InGaAs with similar resistivity [18]. Taking into account the emitter geometry and the resistivity of the material, a simple estimate yields an expected value of 16 kΩ for the device resistance. After deposition of the first metallization the resistance of the emitters was approximately 8 kΩ, which indicates the good contact quality of the metallization. Due to each of the following processing steps, namely the deposition of the Si₃N₄ and the second metallization, the resistance of the emitters decreased by a factor of two, resulting in a final device resistance of about 2 kΩ. The decreased resistance after the Si₃N₄ deposition shows that Si₃N₄ does not serve as a passivation layer on InGaAs. In contrast, significant resistance increases have been observed for GaAs due to the passivation effect of Si₃N₄ [30].

The emitters exhibit pronounced superlinear I/U characteristics (cf figure 2(a)). We attribute the superlinear behaviour mainly to a highly negative temperature coefficient. Increasing the bias voltage results in a higher device temperature which decreases the resistance of the THz emitter thus leading to the superlinear dependence. For a total current of roughly 20 mA the current increases with time until the device is destroyed by overheating; the actual device temperature was not measured. The dark current is about ten times higher than the photocurrent, hence the dark current is responsible for the majority of the heating. Higher bias voltages can be reached by applying a pulsed dc voltage instead of a rectangular ac voltage. The superlinearity of the bias dependence of the current was decreased for the 10% duty cycle.

Zoom In Zoom Out Reset image size

In our setup (cf figure 2(b)) we employ an amplified fibre laser system with a wavelength of 1.55 μm and pulse duration of 100 fs. The repetition rate of the system is 78 MHz and the average output power of the laser is approximately 200 mW. A beam splitter (BS) is used to divide the beam into two parts. The first part is used for pumping the THz emitter, the second part is frequency doubled for electro-optic detection of the THz pulse [31]. The pump beam is focused onto the THz emitter by a lens (L1) with a focal length of 150 mm. This lens is mounted on a stage, which allows one to change the spot size of the pump laser on the emitter position. The generated THz emission is collimated and refocused by two off-axis parabolic mirrors with a reflected focal length of 100 mm. The electro-optic detection is performed with a 200 μm thick 〈110〉-oriented ZnTe crystal. The probe beam is passed through a hole in the second parabolic mirror and focused on the ZnTe crystal. Due to the THz path length of ∼330 mm in the unpurged setup, absorption lines of the water vapour can be observed in the THz spectra. The temporal delay of the probe beam is adjusted by a variable delay stage, which is operated in step-scan mode. After passing the ZnTe crystal, a λ/4-plate and a polarization sensitive beam splitter (PBS), the probe beam is detected by a balanced detector. The detector signal is recorded with a lock-in amplifier (integration time 300 ms), which employs the 6 kHz modulation frequency of the bias voltage as a reference. Assuming perfect phase matching in the ZnTe crystal we calculate the THz field according to [31]. In figure 3 a typical THz transient and the corresponding amplitude spectrum are depicted. The spectrum reaches 4 THz, which is the limit of the electro-optic detection with a ZnTe crystal. The signal-to-noise ratio for the electric fields around 1 THz is above 100.

Zoom In Zoom Out Reset image size

A constant excitation power of 75 mW was applied for the experiments at varied bias fields. For measurements with bias fields up to 16 kV cm⁻¹ the bias was applied as a rectangular shaped ac voltage. To increase the bias field above 16 kV cm⁻¹ the bias was instead applied as a pulsed dc voltage with duty cycles of 10–50%. With this we could increase the bias field beyond 30 kV cm⁻¹ without destroying the emitter. The observed spectrum is basically independent of the bias field (cf figure 4(a)). We note that the reduced spectral resolution of the spectra displayed in figure 4(a) compared with that in figure 3(a) is caused by the shorter temporal length of the recorded THz traces. The dependence of the THz field amplitude on the bias field remained linear for all bias fields (see inset figure 4(a)); however, for the 10% duty cycle the signal-to-noise ratio was lower (see figure 4(b)).

Zoom In Zoom Out Reset image size

For the highest bias field a maximum THz field amplitude of 2.6 V cm⁻¹ was observed. The operation at a bias field of 32 kV cm⁻¹ with a low duty cycle confirms that the main problem for higher bias voltages is the heating of the emitter chip due to a high total current, and not the breakdown of the photoconductive material. Etching away the InGaAs/InAlAs heterostructure in every second gap would lead to a resistance two times higher; furthermore the second metallization would not be needed any more [23, 32]. A second way to increase the resistance of the emitter is by decreasing the active area.

The saturation of the THz emission for high laser fluences is the main limitation of THz emitters based on photoconductive antennas. The pump laser has to be focused tightly to the centre of the antenna structure with a spot size of the order of 10 μm (depending on the antenna structure). Due to screening effects in the photoconductive material the THz amplitude saturates for higher laser powers [33, 34]. It has been shown that the screening is predominantly caused by the emitted THz field and that space–charge screening of the bias field plays only a minor role [7, 35]. For InGaAs based stripline emitters with a 25 μm gap Dietz et al [22] observed a clear saturation already at 10 mW average laser power. To investigate the saturation characteristics of our emitter, we performed measurements at different pump powers for two different spot sizes, namely 50 μm full width at half maximum (FWHM) and 100 μm FWHM. The pump power was varied in a range from 3 to 75 mW, while the bias field was kept constant at 11 kV cm⁻¹. For the spot size of 100 μm the dependence of the THz amplitude on the pump power was linear (see figure 5, linear fit). Decreasing the spot size to 50 μm resulted in a sublinear increase of the THz amplitude.

Zoom In Zoom Out Reset image size

The saturation behaviour for the smaller spot size is well described by the function

$$\begin{eqnarray} &&A\propto \frac{{P}_{\mathrm{pump}}/{P}_{\mathrm{sat}}}{1+{P}_{\mathrm{pump}}/{P}_{\mathrm{sat}}}. \end{eqnarray} \tag{ 1 }$$

Here A is the amplitude of the THz signal, P_pump is the pump power and P_sat is the saturation power where the emitted amplitude is reduced by 50% [33]. The best fit to the experimental data is obtained for P_sat = 130 mW. Taking into account a nearly linear dependence for the 50 μm spot size below 20 mW one can estimate that the saturation of the 100 μm spot would begin above pump powers of about 80 mW, since the illuminated area is four times larger for the 100 μm spot. For a pump power of 300 mW the amplitude would be approximately 30% smaller than the amplitude expected without saturation. Since the photocurrent is in the range of 10% of the dark current, heating due to the laser power can be neglected. For even higher laser powers one could increase the spot size and the emitter area accordingly.

In figure 6(a) the amplitude of the THz signal is plotted as a function of the pump beam spot size for different pump powers. For small pump powers up to 30 mW the THz amplitude stays constant for spot sizes of less than 100 μm. For spot sizes significantly larger than 100 μm part of the pump beam extends beyond the active area. Therefore the THz amplitude decreases for larger spot sizes for all pump powers. Interestingly a different behaviour was observed for a SI-GaAs-based scalable emitter [36]. In particular, for an emitter based on SI-GaAs (active area 3 × 3 mm²) the THz emission decreased strongly as the spot size was reduced to values below 300 μm. This was attributed to suppressed emission of THz waves with wavelength larger than the spot size. This interpretation was corroborated by the observed strong red shift in the emission spectrum of the SI-GaAs emitter as the spot size was increased [36]. In the experiment with the LT-InGaAs emitter, however, we do not observe a similar spectral shift (cf figure 6(b)). For the LT-InGaAs material the trapping processes strongly determine the carrier dynamics and in particular lead to a fast deceleration of carriers. For SI-GaAs, on the other hand, the carrier lifetime is long and the carrier relaxation dynamics depends on the side-valley transfer and the screening dynamics. The first mechanism is related to the bias field, while the latter is influenced by the near-infrared fluence. Consequently, the THz emission of a SI-GaAs emitter is blue-shifted with increasing bias field as well as with increasing pump fluence, while this effect is not observed for emitters based on GaAs with a short carrier lifetime [9, 37]. The fact that the emission spectrum of our LT-InGaAs emitters does not depend on the bias field, excitation fluence or excitation spot size points to the dominant role of the trapping processes in the carrier dynamics of this material. The stability of the emitted spectrum with respect to external parameters is highly attractive for spectroscopic applications.

Zoom In Zoom Out Reset image size

In this section we compare our emitter to other THz emitters suitable for excitation at 1.55 μm. We cannot compare the emitted THz field strength of our emitter to literature values for InGaAs based photoconductive antennas, since in the reported experiments on emitter antennas the detection is also performed with photoconductive antennas and therefore no quantitative values for the THz field are available. Nevertheless one can state that photoconductive antennas should be more efficient in the range of excitation fluences, where no saturation occurs. This is due to the fact that in case of the antenna all excitation photons are exploited as they are focused to the photoconductive gap, while in case of the scalable emitter a large part is reflected by the metallization. On the other hand, for excitation fluences where strong saturation is observed, the THz emission from scalable emitters will exceed the emission of photoconductive antennas. From our data and the study of Dietz et al [22] we expect that the transition between these two regimes occurs at pump pulse energies of 0.1–1 nJ (∼10–100 mW average power for a 78 MHz system).

Comparing the maximum THz field achieved by different scalable emitters we find that the 2.6 V cm⁻¹ from our InGaAs/InAlAs heterostructure is about three times larger than the maximum fields reported for scalable emitters based on InGaAs with ErAs nanoparticles [23]. For lateral photo-Dember emitters THz fields of 4 V cm⁻¹ have been achieved [4], but with a THz bandwidth (∼2 THz) roughly two times lower as compared to our scalable emitter. The question whether lateral photo-Dember emitters or scalable photoconductive emitters are better suited for a TDS system depends on the particular application. Photo-Dember emitters are attractive for simple compact emitter heads as they do not require electrical contacting. Photoconductive emitters, on the other hand, allow for fast modulation of the THz emission via the bias voltage.

Compared to the THz fields achieved with scalable emitters based on SI-GaAs excited with 800 nm radiation the achieved fields of all InGaAs based emitters are still very low. This can be mainly attributed to trapping centres, which are required to reach suitable resistivities, but which deteriorate the carrier acceleration and in turn the THz emission. In fact it has been recently shown that photoconductive antennas based on InGaAs/InAlAs heterostructures with trapping centres only in the InAlAs layers exhibit a sixfold increase in the emitted field as compared to InGaAs/InAlAs heterostructures with trapping centres in all layers [22].

Finally we note that the electro-optic detection employed in the experiments, needed here for quantitative assessment of the produced THz fields, may be replaced by the more compact detection with photoconductive antennas. Then no frequency doubler is required and fibre coupling of the detector head is possible. For easy alignment one could use the same device for both emitter and detector. For improving the signal-to-noise ratio in the detection part, antennas with low dark current, which can be achieved by mesa-etching of the photoconductive material, are highly attractive [38].

The combination of a scalable large area emitter structure with a low-temperature grown InGaAs/InAlAs heterostructure provides a broadband THz emitter adapted for fibre lasers operating at 1.55 μm. The resistivity of the material is suitable for emitters with an active area of 300 × 300 μm², which is large enough to excite the emitter with amplified fibre lasers in the 100 mW range without significant saturation. Compared to photoconductive antennas on this photoconductive material, large area emitters allow one to use the full laser power delivered by femtosecond fibre laser systems. To increase the resistance of the devices, and therefore increase the maximum bias voltage, one could actively cool the device, for example with a Peltier element. An additional advantage is the easy alignment of scalable emitters. A key feature of the THz emitter based on the LT-InGaAs/InAlAs heterostructure is the independence of the emitter spectrum of the parameters spot size, pump power, and bias voltage. This feature, which is attributed to the carrier dynamics being strongly influenced by the trapping centres, is highly attractive for stable THz emission in TDS systems.

We thank H Hilliges and C Neisser for cleanroom processing.