1. Introduction

THz technology continues to advance from the original Austin switch ([1], 1984) to support a wide variety of applications in spectroscopy, science, inspection, communications, etc. The vast majority of these applications use common photoconductive antennas (PCAs) to transmit and receive THz radiation through an optical path formed using bulk-optical THz elements. However, to decrease cost and increase performance and functionality, it is essential to develop a next generation of THz system-on-chip (TSoC) [2–6] comprising multiple THz components and functions. In analogy with MMICs operating at microwave frequencies, such TSoCs would include active devices (photoconductors, photomixers, amplifiers, mixers, etc.), passives (filters, matching elements, capacitors, chokes, etc.) and, as is the focus of this work, low-loss and non-dispersive waveguides and transmission lines.

Experiments that demonstrated sub-picosecond pulse transmission were first completed in the 1980s [1,7]. In [7] a coplanar waveguide (CPW) fabricated on a silicon-on-sapphire (SOS) substrate was used to a transmit a subpicosecond pulse over 8 mm after which it broadened to 2.6 ps. Later in the 1990s a coplanar strip (CPS) transmission line was fabricated on a thin silicon dioxide/silicon nitride membrane which supported subpicosecond pulse transmission over 4 mm [8]. After this initial work a number of other waveguide technologies [9–12] were investigated which achieve low loss and minimal pulse distortion over relatively long distances (>24 cm). However, they are rather bulky mechanically, which limits their suitability for direct connection to active devices or precise fabrication of complex TSoC. As a result, these are typically excited by capturing a portion of the far-field radiated from a standard photoconductive antenna (PCA). Previously we utilized a different approach to achieve low loss and low dispersion THz waveguides by near-field coupling to a metallic slit waveguide [13]. However, this waveguide is still far from ideal for TSoC integration. In 2009 CPS transmission lines were investigated on low-permittivity substrates for spectroscopic applications where thin-film active regions were connected to a plastic substrate using pre-process bonding [14]. In [14] discrete active regions were bonded before the lithographically-defined transmission lines were deposited (thus requiring mask alignment), limiting the flexibility for complex system fabrication where many discrete components may be required. Post-process bonding is desirable because the active regions can be placed onto an existing conducting circuit, which is similar to placing low-frequency surface-mount components onto a printed-circuit board (PCB). In [15] a CPS transmission line on a 400 µm thick quartz substrate was excited by post-process LTG-GaAs bonding, however only short distances were investigated (≈ 1 mm).

This work is motivated by the development of a practical TSoC platform or workbench for engineering circuits for THz frequencies, circuits that may ultimately include a wide variety of active and passive components designed largely by scaling MMIC counterparts. To meet these objectives we require 1) compatibility with standard surface-mount fabrication techniques, 2) THz transmission lines with low loss and low dispersion and 3) ability to define and fabricate RF-engineered features (stubs, impedance transformers, multi-section filters, etc.) with high precision. After considerable investigation, we conclude that only approach that meets all three requirements is one similar to that reported in 1994 in [8]. Low-loss and low dispersion result from fabrication of the CPS on a thin membrane, which results in relatively weak dielectric loading, which minimizes group velocity dispersion (GVD) and Cherenkov radiation [16]. Precision is insured using standard photolithography to define circuit features.

In this paper we resurrect and extend the approach presented in [8] to demonstrate a simple example of this TSoC platform; a system consisting of a transmitter, waveguide and receiver. Several things have changed since the original 1994 work. We use thin (1 µm) silicon nitride membranes which are now commercially available over at least 1 cm × 1 cm areas, rather than the previously demonstrated oxide-clad membranes. In [8] the pulse was generated by photoconductive switching and detected by electro-optic sampling. While electro-optic sampling performs well, the optical elements required are not compatible with our TSoC objectives. Rather, we use small (40 µm × 20 µm) discrete thin-film photoconductive LTG-GaAs epi-layers as both transmitter and photoconductive receiver.

This paper combines several contributions in demonstrating the transmission of THz-bandwidth pulses and potential of this TSoC platform. Specific contributions include 1) extending prior work (mainly [8,17]) to fabricate a non-dispersive CPS transmission line on a thin silicon nitride membrane, 2) detection of pulses containing frequencies up to 1.5 THz after propagating for 10 mm, the furthest a pulse with this bandwidth has been transmitted on a CPS, 3) demonstration, for the first time, of post-processing placement of thin-film photoconductive LTG-GaAs devices on the membrane-based CPS, 4) investigation of contact efficacy between the CPS and placed devices, 5) demonstration of a simple radiatively-coupled DC block to isolate the photoconductive receiver from relatively high DC bias voltage applied at the transmitter and 6), numerical simulation of various CPS configurations to show a path forward to lower loss and increased functionality. Taken together, these contributions demonstrate a novel technology platform that allows precise and simple fabrication of THz-bandwidth circuits. Work continues to investigate various other applications of this platform to produce membrane-based TSoCs.

2. Design

The structure investigated in this paper is a CPS transmission line on a thin (1 µm) silicon nitride membrane. Silicon nitride was selected because thin-film membranes on silicon frames are commercially available. The dielectric loss for silicon nitride is reasonable at THz frequencies (thin-film loss tangent, tan δ_ε ≈ 0.00526 [18]). Note that the majority of the field is located in the surrounding air, thus the overall dielectric loss is minor. Figure 1(a) illustrates renderings of the overall structure and enlarged images of the transmitter and receiver regions are shown in Figs. 1(b)-1(c), respectively. Figure 1(d) is a microscope image of the fabricated receiver region. The cross-section of the CPS transmission line is shown in Fig. 1(e). Electrical contacts are patterned onto a commercial silicon/silicon nitride frame which is typically used for X-ray microscopy [19]. For this experiment we used a 10 mm × 10 mm silicon nitride window on a 15 mm × 15 mm silicon frame. We were most concerned with the proof-of-concept design (not heavily optimized) so we selected waveguide dimensions (S = W = 10 µm) which would have a moderate attenuation at high frequencies while keeping the total structure dimensions (S + 2W) relatively small compared to the spatial pulse length (<300 µm) to minimize radiation losses during excitation.

 

Fig. 1 Illustration of the THz platform. a) Overall structure on membrane. b) Rendering of transmitter LTG-GaAs connection. c) Rendering of receiver LTG-GaAs connection. d) Microscope image of receiver LTG-GaAs connection. e) Cross-section of CPS transmission line.

Download Full Size | PDF

Our selected circuit configuration resembles a standard “sliding contact” transmitter except that the transmitter is located at the end of the transmission line. This excitation method is used to maximize the transmitted power towards the receiver and minimize the resonances which can occur with biasing features. The receiver is similar to a standard PCA which is placed in close proximity to the DC biasing lines. This configuration was selected because it resembles a microwave bias tee and minimizes signal distortion which occurs from resonant cavities, this was confirmed by prior simulations. The transmitter and receiver bias lines are fed from opposing sides of the silicon frame primarily such that they fit onto a probe station.

The thin-film photoconductive LTG-GaAs layer originated from a 940 nm LTG-GaAs/900 nm AlAs/650 µm semi-insulating (SI)-GaAs structure. The LTG-GaAs layer was grown at 260^°C via molecular beam epitaxy (MBE). After growth the substrate was annealed at 450^°C for 10 minutes. We found that these anneal conditions give subpicosecond carrier lifetimes that are in agreement with [20] (1/e carrier lifetime of ≈0.23 ps). Note that the extremely thick sacrificial AlAs layer was not required for this experiment, however the material was available. A much thinner AlAs layer (≈ 50 nm [21]) should be used to reduce substrate deterioration by oxidization.

To fabricate the LTG-GaAs photoconductive devices we used a processing method similar to that described in [17]. To summarize the procedure we first used standard photolithography, gold deposition (5 nm Ti and 100 nm Au), and lift-off. Afterwards photoresist was used to protect the active regions via mask alignment and development. Then the entire substrate was submerged in a GaAs etching solution (citric acid and hydrogen peroxide) to create an array of mesa devices. Next an etch resistant wax was melted onto the surface, then the structure was submerged in Hydrofluoric (HF) acid which etches the sacrificial AlAs layer and releases the LTG-GaAs epilayer. The structure, still bonded to the wax, was reinserted into the GaAs etching solution until all the mesa devices were cleanly separated. The wax and structures were placed onto a Mylar membrane, then the wax was removed with Trichloroethylene (TCE). After this procedure we obtained thousands of 40 µm × 20 µm × 0.4 µm LTG-GaAs photoconductive devices which can be used as either transmitters or receivers.

3. Simulations

The transmission line was simulated in both the time and frequency domain using commercial software [22]. Time-domain simulations are used to characterize transmission-line excitation. Frequency-domain simulations are used for transmission-line design and insight. Due to the extreme dimensional ratio -a 1 µm thick membrane and a 10 mm length (1:10000) -long simulation times are required to characterize the structure in its entirety. To gain the most insight, evaluation the complex propagation constant, γ = α + iβ is needed, where α is the attenuation coefficient and β is the phase constant. To obtain γ a 2D simulation of the transmission-line cross section is required. After γ is obtained via simulation the pulse transmission can be characterized by application of the Fourier transform:

The following material properties are used in the simulations. For silicon nitride: the relative permittivity ε_r = 7.6, relative permeability µ_r = 1, electrical conductivity σ = 0 S/m, and loss tangent tan δ_e = 0.00526 [18]. For the gold contacts we assumed the best case scenario with negligible surface roughness and selected the standard bulk conductivity σ_Au =4.1 × 10⁷ S/m. Adjustment of the transmission-line separation, S, and width, W, can greatly impact γ. To illustrate this we simulated the transmission line with various S and W values (see Fig. 2). Figure 2(a) plots the attenuation coefficient, α, which can be significant and will limit the bandwidth of the system. Note that the majority of the attenuation originates from conductor loss (confirmed by comparing simulations with perfect and gold conductors). Figure 2(b) plots the phase constant, β. The difference between the light line (vacuum/lossless propagation) and the other traces indicates the presence dielectric loading. For convenience we selected to set S equal to W, however when optimizing system performance they are unlikely to be equal, Appendix A illustrates the impact of varying the W/S ratio for a fixed cross section (S + 2W).

 

Fig. 2 Frequency-domain simulations results for a CPS transmission line on a 1 µm silicon nitride membrane with S = W = 10 µm, 20 µm, 30 µm, 40 µm, and 50 µm. a) The attenuation coefficient. b) The phase constant.

Download Full Size | PDF

Thus far the simulation suggests that we should select a large CPS separation and width to achieve the best performance (i.e. S = W = 50 µm). However, as the cross-section of the transmission-line increases (i.e. S + 2W = 150 µm) more radiation will occur at the higher frequencies during excitation. Figure 3 plots the time-domain simulation results located 950 µm from the source, V(z = 950 µm, t), for various transmission-line widths and separations. From Fig. 3(b) it is clear that larger transmission-line cross-sections experience more loss at higher frequencies which is contrary to the data shown in Fig. 2(a). This occurs because the transmission line looks like a larger antenna during excitation (S + 2W is larger); however, once the pulse is coupled to the transmission line it will propagate with lower loss. This is the reason that we have selected to use S = W = 10 µm for the experiment (minimize radiation during excitation) although the transmission line attenuation is larger [Fig. 2(a)]. The time shift for the various peaks in Fig. 3(a) is the result of a variation in the effective dielectric constant. For smaller cross-sections the field is more localized to the dielectric thus the group velocity is reduced. The crossing traces in Fig. 3(b) originates from the complex nature of radiation during excitation and coupling to the transmission line. Future work will investigate optimized tapering structures which minimize the radiation during excitation (i.e. S = W = 5 µm) then transform to a low-loss configuration (i.e. S = W = 50 µm) for the majority of the propagation distance. Also, we have focused on membrane thickness of 1 µm, which reduce dielectric loading adequately for the frequencies considered. Substantially thinner (e.g. down to 100 nm) membranes are available providing another dimension for investigation.

 

Fig. 3 Time-domain simulation results for Gaussian pulses plotted 950 µm from the source for S = W = 10 µm, 20 µm, 30 µm, 40 µm, and 50 µm. These pulses are plotted 950 µm away from the source to ensure any ringing has sufficiently decayed.

Download Full Size | PDF

Once the complex propagation constant is known, the output pulse for an arbitrary input pulse can be obtained by numeric application of Eqs. (1) and (2). In this work we assume the excitation pulse is of the following form [23]:

We have selected to use common values for the following analysis [23]: τ_c = 0.5 ps, τ_s = 0.03 ps, and τ_p = 0.054 ps. Figure 4(a) plots the input and output pulses in the time domain. The input pulse is given by Eq. (3) and the output pulse is obtained by application of Eqs. (1) and (2) using the simulated complex propagation constant. Figure 4(b) plots the spectral response. It is apparent from Fig. 4(a) that the pulse both broadens and becomes attenuated after propagating 10 mm. Note that the peak amplitude of the received pulse is 0.066× the input amplitude. While the attenuation is large, if the pulse can be resolved with a reasonable signal-to-noise ratio (via a lock-in amplifier) then the transmission line can still be used to investigate circuit elements. However for longer interconnects it would be desirable to taper the transmission line to a lower loss configuration, for example, if S = W = 50 µm (not pictured) which would change the peak reduction to 0.367× the input amplitude (10 mm) -a potentially substantial improvement.

 

Fig. 4 Input and calculated output pulses via Fourier Transform (using simulated γ(ω)) for S = W = 10 µm.

Download Full Size | PDF

4. Experiment and results

The experimental setup used (Fig. 5) is similar to a standard THz-time-domain spectroscopy (TDS) measurement setup. A femtosecond laser (90 fs pulse width, 780 nm wavelength, 80 MHz repetition rate) is passed through a beam splitter; one beam is directed to the transmitter through a optical chopper (1100 Hz), the other to the receiver via a mechanical delay line. Note that we used optical chopping to remove the inductively coupled signal which is detected when using electrical chopping. The transmitter consists of a thin-film LTG-GaAs layer which has a 25 VDC bias applied. The receiver contacts are connected to a lock-in amplifier which is referenced to the optical chopper in the transmitter optical path. The received pulse is plotted by sweeping the physical path length difference between transmitter and receiver. Providing that the receiver has a minimal carrier lifetime (τ_c in the subpicosecond range) it becomes possible to resolve THz-bandwidth signals, which is why LTG-GaAs (τ_c ≈ 0.5 ps) is used.

 

Fig. 5 Experimental setup for testing the transmission line. For the FIB-bonded experiment the optical power delivered to the transmitter and receiver is 6.2 mW and 7.4 mW, respectively. For the VDW-bonded experiment the optical power delivered to the transmitter and receiver is 1.6 mW and 2.0 mW, respectively.

Download Full Size | PDF

Two different techniques were explored for the physical connection between the thin-film LTG-GaAs transmitter and receiver and their respective contacts on the silicon nitride membrane. Initially we bonded the transmitter and receiver similarly using the Van der Waals (VDW) technique [Fig. 6(b)] [17, 21]. In short, we placed the thin-film LTG-GaAs transmitter and receiver with their contacts touching the CPS conductors on the silicon nitride membrane, then using a micro-manipulator they were moved to their precise location (as illustrated in Fig. 1). A drop of deionized (DI) water was then placed over each, dried, then the thin-film LTG-GaAs layer was bonded to the silicon nitride membrane and gold contacts. Although there is no bonding material (i.e. Indium), after fabrication and testing this VDW-bonding technique [Fig. 6(b)] was found to function reasonably well. We were able to measure a strong photoresponse for both devices. However, with an applied DC bias, we noticed that as the optical power increased, the photocurrent would gradually become unstable, we suspect, due to heating, thus we limited the optical power to the VDW-bonded transmitter. To provide better thermal dissipation we decided to “weld” the transmitter contacts to the membrane contacts. The thin-film LTG-GaAs transmitter was removed, then a new thin-film LTG-GaAs device was placed upside-down, repositioned, and re-bonded via the VDW method except without any gold-on-gold contact. The exact same device was not reused because we were unsure if original device had become thermally damaged during testing; however, we did use a device from the same batch which performs near identically. Afterwards we directly welded the thin-film LTG-GaAs contacts onto the membrane contacts using Tungsten deposition in a Focused Ion Beam (FIB) [Fig. 6(a)]. This helped to reduce the photocurrent noise however it increased the dark current which we suspect is the result of Gallium implantation which occurred during Tungsten deposition via the FIB. With this modification higher optical powers could be used with the FIB-bonded configuration compared to the VDW-bonded configuration.

 

Fig. 6 Illustration of the LTG-GaAs bonding techniques. Note that the gold contact pads are both 12.5 µm × 10 µm and are separated by 5 µm, the feature resolution has a radius of curvature of 2.5 µm. a) Van der Waals bonding with FIB deposited tungsten. b) Van der Waals bonding without tungsten.

Download Full Size | PDF

Figure 7(a) plots the received temporal pulse where the inset is focused on the normalized leading edge of the pulse. Figure 7(b) plots the spectral response (magnitude) which is obtained by applying the Fast-Fourier Transform (FFT) to the received temporal pulse. The data in Fig. 7(b) is normalized to the maximum of the FIB configuration spectrum, and the VDW configuration spectrum is shifted down by 10 dB for clarity. Note that the phase is not plotted, as is common practice in literature. For the FIB-bonded method it is evident from the spectral response that the received pulse contains frequency components which extend to 1.5 THz. For the VDW-bonded method frequencies extend up to approximately 1.0 THz. This is explained by a difference in optical power. As previously mentioned, when the optical power for the VDW-bonded transmitter increases, the photocurrent becomes noisy, likely due to heating. To mitigate this issue for the VDW-bonded method a lower optical power is used: 1.3 mW (transmitter) and 2.0 mW (receiver). For the FIB-bonded method an average optical power of 6.2 mW (transmitter) and 7.4 mW (receiver) is used. For both cases the optical beam is focused to a waist diameter of approximately 7 µm on the transmitter and receiver. Since the noise issue was not noted in past work we suspect it is occurring because we used a very thin LTG-GaAs layer (0.4 µm). We selected to use this thickness simply because the material was available to us. Past work used the following LTG-GaAs layer thicknesses: 0.8 µm in [14] and 2.0 µm in [15,17]. Compared with the work in [14] (pre-process bonding), these post-processing bonding techniques have lower bandwidth and dynamic range; however, we did demonstrate comparable results when transmitting over a longer distance using a higher loss configuration (S = W = 10 µm compared to S = W= 20 µm in [14]). Given the large advantage in terms of fabrication (i.e. mask alignment is not necessary), post-process bonding becomes a very attractive option.

 

Fig. 7 Experimental result plotting the received THz-bandwidth pulse.

Download Full Size | PDF

Future work will explore the two related issues, device bonding and heat dissipation. Device bonding with the FIB is undesirable due to a high processing cost and potential Gallium implantation effects, however it provides better heat dissipation. VDW bonding is desirable because of the lower cost, however it appears to not dissipate heat as effectively. Another option to improve heat dissipation is to use a thicker metalization on the thin-film LTG-GaAs devices and the CPS transmission line. Alternatively, and preferably, use of well known flip-chip bonding with Indium as a bonding material may be a useful method to provide a better bond, low contact resistance and good heat dissipation.

In Fig. 7(a) the received pulse has a relatively flat tail following the main pulse which is desirable for high resolution system. The two peaks following the main pulse in Fig. 7(a) correspond to two resonant cavities, this is seen as ringing below 0.25 THz in Fig. 7(b) (two peaks at 59 GHz and 117 GHz). The first cavity occurs between the receiver and the back of the Silicon frame (along the air/Silicon interface). Quasi-statically the resonance is calculated as Δ f = 3 × 10⁸/(2 × 2.54× 117 GHz) = 0.505 mm which is close to the frame thickness (0.500 mm). The second cavity occurs between the receiver and the end of its contact pads (1.00 mm long). As approximation (ignoring the silicon nitride because it is very thin) this can be calculated quasi-statically as Δf = 3 × 10⁸/(2 × 2.54× 59 GHz) = 1.04 mm which is close the distance between the receiver and the end of its contact pads. The first ringing effect can be minimized by placing the silicon frame onto another thicker silicon mount which has a rough end-face to minimize reflections. The second ringing effect can be minimized by extending and smoothly flaring the receiver contact pads.

Appendix B illustrates the importance of the thin silicon nitride substrate by plotting the result which uses a glass coverslip as the substrate and the VDW-bonding method.

5. Conclusion

We have demonstrated a simple system-on-chip for the generation and detection of THz-bandwidth pulses after propagating 10 mm on a coplanar stripline defined on a thin silicon nitride membrane. Thin-film LTG-GaAs photoconductive devices are positioned and bonded using two separate bonding methods. Received pulses exhibit low distortion due to the minimal dielectric loading of the thin substrate. A pathway forward to lower propagation loss and increased system functionality is articulated. The combination of thin membranes, photolithographic feature definition and compatibility with surface-mount techniques provides an attractive platform for the construction of complex TSoCs. Challenges in device bonding and heat dissipation are under ongoing investigation as we explore implementation of more complex circuits.

Appendix A - Sweeping the CPS width and separation ratio

As previously mentioned the optimal W/S ratio for the CPS transmission line is not likely obtained when S = W. The concept is illustrated here. Figure 8 plots the simulated attenuation coefficient for a few different frequencies when the total CPS cross-section (S + 2W) is held at 90 µm. From Fig. 8 it is clear that the lowest attenuation occurs near W/S = 0.75 (S = 36 µm andW = 27 µm). Note when S = W the attenuation coefficient is not significantly different than the optimal value.

 

Fig. 8 Simulated attenuation coefficient for a number of S/W ratios at different frequencies.

Download Full Size | PDF

Appendix B - Experiment on a thick glass coverslip

Before constructing the CPS transmission line on the thin silicon nitride membrane we constructed the same waveguide structure on a 18 mm × 18 mm × 0.18 mm glass cover-slip for fabrication practice. It was expected to have significant pulse distortion due to heavy dielectric loading. Figure 9(a) illustrates the received pulse when using glass cover-slip. Figure 9(b) plots the spectral response. Comparison between Fig. 7 and Fig. 9 illustrates that the usage of a thin membrane can have a profound impact on the performance of the device (as expected).

 

Fig. 9 Detected pulse on glass cover-slip using the VDW-bonding method.

Download Full Size | PDF

Funding

Natural Sciences and Engineering Research Council (NSERC) of Canada.

Acknowledgments

We would like to thank Dr. Thomas Tiedje for the meaningful discussions. We would like to acknowledge the Waterloo Institute of Nanotechnology for fabricating the LTG-GaAs.

References

1. D. Auston, Picosecond Photoconductors: Physical Properties and Applications (Academic, 1984).

2. L. Samoska, “An overview of solid-state integrated circuit amplifiers in the submillimeter-wave and THz regime,” IEEE Trans. Terahertz Sci. Technol. 1, 9–24 (2011). [CrossRef]  

3. S. Koenig, D. Lopez-Diaz, J. Antes, F. Boes, R. Henneberger, A. Leuther, A. Tessmann, R. Schmogrow, D. Hillerkuss, R. Palmer, T. Zwick, C. Koos, W. Freude, O. Ambacher, J. Leuthold, and I. Kallfass, “Wirelesssub-THz communication system with high data rate,” Nat. Photonics 7, 977–981 (2013). [CrossRef]  

4. T. Nagatsuma, G. Ducournau, and C. Renaud, “Advances in terahertz communications accelerated by photonics,” Nat. Photonics 10, 371–379 (2016). [CrossRef]  

5. P. Siegel, R. Smith, M. Graidis, and S. Martin, “2.5-THz GaAs monolithic membrane-diode mixer,” IEEE Trans. Microw. Theory Tech. 47, 596–604 (1999). [CrossRef]  

6. H. M. Cheema and A. Shamim, “The last barrier: on-chip antennas,” IEEE Microw. Mag. 14, 79–91 (2013). [CrossRef]  

7. M. B. Ketchen, D. Grischkowsky, T. C. Chen, C.-C. Chi, I. N. Duling, N. J. Halas, J.-M. Halbout, J. A. Kash, and G. P. Li, “Generation of subpicosecond electrical pulses on coplanar transmission lines,” Appl. Phys. Lett. 48, 751–753 (1986). [CrossRef]  

8. H. Cheng, J. Whitaker, T. Weller, and L. Katehi, “Terahertz-bandwidth pulse propagation on a coplanar stripline fabricated on a thin membrane,” IEEE Microw. Guid. Wave Lett. 4, 89–91 (1994). [CrossRef]  

9. R. Mendis and D. Grischkowsky, “Undistorted guided-wave propagation of subpicosecond terahertz pulses,” Opt. Express 26, 846–848 (2001).

10. K. Wang and D. Mittleman, “Metal wires for terahertz wave guiding,” Nature 432, 376–379 (2004). [CrossRef]   [PubMed]  

11. T. Jeon, J. Zhang, and D. Grischkowsky, “THz sommerfeld wave propagation on a single metal wire,” Appl. Phys. Lett. 86, 161904 (2005). [CrossRef]  

12. M. Wächter, M. Nagel, and H. Kurz, “Metallic slit waveguide for dispersion-free low-loss terahertz signal transmission,” Appl. Phys. Lett. 90, 061111 (2007). [CrossRef]  

13. R. Smith, A. Jooshesh, J. Zhang, and T. Darcie, “Photoconductive generation and detection of THz-bandwidth pulses using near-field coupling to a free-space metallic slit waveguide,” Opt. Express 25, 26492 (2017). [CrossRef]   [PubMed]  

14. S. Yanagi, M. Onuma, J. Kitagawa, and Y. Kadoya, “Propagation of terahertz pulses on coplanar strip-lines on low permittivity substrates and a spectroscopy application,” Appl. Phys. Express 1, 012009 (2008). [CrossRef]  

15. L. Desplanque, J. F. Lampin, and F. Mollot, “Generation and detection of terahertz pulses using post-process bonding of low-temperature-grown GaAs and AlGaAs,” Appl. Phys. Lett. 84, 2049–2051 (2004). [CrossRef]  

16. D. Grischkowsky, “Optoelectronic characterization of transmission lines and waveguides by terahertz time-domain spectroscopy,” IEEE J. Sel. Top. Quantum Electron. 6, 1122–1135 (2000). [CrossRef]  

17. R. D. V. Ríos, S. Bikorimana, M. A. Ummy, R. Dorsinville, and S.-W. Seo, “A bow-tie photoconductive antenna using a low-temperature-grown GaAs thin-film on a silicon substrate for terahertz wave generation and detection,” J. Opt . 17, 125802 (2015). [CrossRef]  

18. G. Cataldo, J. A. Beall, H.-M. Cho, B. McAndrew, M. D. Niemack, and E. J. Wollack, “Infrared dielectric properties of low-stress silicon nitride,” Opt. Express 37, 4200 (2012).

19. Norcada Inc., https://www.norcada.com/.

20. I. Gregory, C. Baker, W. Tribe, I. Bradley, M. Evans, E. Linfield, A. Davies, and M. Missous, “Optimization of photomixers and antennas for continuous-wave terahertz emission,” IEEE J. Quantum Electron. 41, 717–728 (2005). [CrossRef]  

21. E. Yablonovitch, D. M. Hwang, T. J. Gmitter, L. T. Florez, and J. P. Harbison, “Van der waals bonding of GaAs epitaxial liftoff films onto arbitrary substrates,” Appl. Phys. Lett. 56, 2419–2421 (1990). [CrossRef]  

22. ANSYS HFSS V19.1, https://www.ansys.com/products/electronics/ansys-hfss.

23. Y.-S. Lee, Principles of Terahertz Science and Technology (Springer, 2009).