RECENT advances in the transistor technologies such as Si

440 IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 7, NO. 4, JULY 2017 Submillimeter-Wave Waveguide-to-Microstrip Transitions for Wide Circuits/Wafers Jungsik Kim, Wonseok Choe, and Jinho Jeong, Member, IEEE Abstract Microstrip transitions for wide circuits/wafers are proposed using stub arrays (SAs) and indented waveguides (IWs). Full-wave analysis is performed to show that the proposed techniques can effectively suppress the resonances excited in the slit region which is needed to accommodate wide substrates. H- band waveguide-to-microstrip transitions consisting of two directors, dipole antenna, and balun were fabricated including the proposed techniques, accommodating a 1220-µm-wide substrate which is about three times the width of WR-03 standard waveguide (430 µm). The slit region is 1300 µm wideand90µm high. The comparison of measurement results between four transitions proves that the proposed SAs and IWs can successfully eliminate in-band resonances and greatly enhance the bandwidth of the transitions. The transition with both SAs and IW exhibits the best performance, a back-to-back loss of 2.3 db across 234 314 GHz, where the variation in the insertion loss was less than ±0.5 db. This insertion loss includes the losses in the microstrip line and 3-cm-long waveguide section. The return loss is better than 13.7 db in this frequency range. Bandwidth for 10-dB return loss is as large as 97 GHz from 227 to 324 GHz. Index Terms Dipole transition, submillimeter-wave, terahertz, waveguide-to-microstrip transitions. I. INTRODUCTION RECENT advances in the transistor technologies such as Si metal-oxide semiconductor field effect transistors, GaAs high electron mobility transistors, and InP heterojunction bipolar transistors (HBTs) lead to the successful development of submillimeter-wave and terahertz monolithic integrated circuits (TMICs) for imaging, spectroscopy, and wireless communications [1] [6]. TMICs are generally mounted or packaged in rectangular waveguides using waveguide-to-microstrip transitions for module and system applications. For example, H- band (220 325 GHz) medium power amplifier (PA) module has been successfully developed by wire-bonding PA TMIC to the waveguide-to-microstrip transition using E-plane probes fabricated on 50-µm-thick quartz substrate [7], where a single transition loss was as low as 1.2 db at 220 320 GHz. Monolithic transitions or on-chip transitions are common approaches to minimize performance degradation during Manuscript received November 2, 2016; revised January 26, 2017; accepted April 20, 2017. Date of publication May 16, 2017; date of current version June 29, 2017. This work was supported by a grant to the Terahertz Electronic Device Research Laboratory funded by the Defense Acquisition Program Administration, and by the Agency for Defense Development (UD150043 RD). (Corresponding author: Jinho Jeong.) The authors are with the Department of Electronic Engineering, Sogang University, Seoul 04107, South Korea (e-mail: k1j1s3@sogang.ac.kr; choe0039@ sogang.ac.kr; jjeong@sogang.ac.kr). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TTHZ.2017.2701151 assembling of TMICs in the waveguides, since they can avoid parasitic effects due to bond-wires [8]. Submillimeter-wave monolithic transitions have been reported in the form of E-plane probe and dipole transitions [8] [12]. In [11], low-loss E-plane probe transition integrated on 50-µm-thick InP substrate was reported at 310 350 GHz, showing a back-to-back loss about 4 db including the approximately 450-µm-long coplanar waveguide. On-chip dipole transitions were also reported in [12] showing a single transition loss about 1 db at 340 380 GHz. Both E-plane probe and dipole transitions entail the air-filled channels or slits on the sidewalls of the waveguides which can generate high order modes or unwanted resonance modes of electromagnetic fields [8], [12]. In order to remove these modes out of band, the size of the channels and slits should be minimized, which limits the width of E-plane probe and dipole transitions and in turn the size of THz ICs. Unfortunately, standard rectangular waveguides are relatively small in submillimeterwave range compared with the circuit size. To overcome these problems, nonrectangular circuits were realized by using additional processes like laser dicing and deep dry-etching techniques, where the substrate around the transitions was removed to reduce the width of the transitions while keeping the circuit size [9], [14]. In this study, we propose new submillimeter-wave waveguideto-microstrip transitions using dipole antenna which can accommodate wide circuits/wafers without exciting unwanted waveguide modes. The techniques to remove in-band resonances in the slit region are proposed for rectangular substrates, so that there is no need of additional processes. The validity of the proposed techniques is proved by the simulation and measurement of the transitions designed on quartz substrates. The proposed idea can be easily applied to InP substrate. In Section II, we investigate the resonance problems in the slit region caused by wide circuit substrates. We also propose the techniques to remove in-band resonances and restore the bandwidth performance of the transitions; stub arrays (SAs) and indented waveguides (IWs). The measurement results of designed four H-band transitions based on the proposed techniques are discussed and compared in Section III, where the best design is also presented in terms of insertion/return loss and bandwidth. II. TRANSITIONS FOR WIDE SUBSTRATES Fig. 1(a) shows three-dimmensional and cross-sectional views of H-band waveguide-to-microstrip transition using dipole antenna, which accommodates 1220-µm-wide substrate. 2156-342X 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

KIM et al.: SUBMILLIMETER-WAVE WAVEGUIDE-TO-MICROSTRIP TRANSITIONS FOR WIDE CIRCUITS/WAFERS 441 Fig. 1. Waveguide-to-microstrip transition using dipole antenna for wide circuits. Length unit is µm. (a) 3-D and cross-sectional views. (b) Substrate layout with optimized dimension. We selected this width for the transition, considering the wide width of high PA ICs. The standard H-band waveguide (WR- 03) is 430 µm wide. The waveguide jig is split into two metallic blocks, top and bottom waveguides. The slit region (air gap) is formed along the E-plane of the bottom waveguide to accommodate the wide-width transition substrate. Electric fields in the waveguide are captured by the dipole antenna with directors. Two directors are used to improve the loss and bandwidth performance of the dipole transitions as shown in Fig. 1(b) [15]. Then, the fields are transformed to microstrip mode through coplanar strip-to-microstrip balun consisting of 40 o radial stub [15], [16]. In this study, a 50-µm-thick quartz substrate with dielectric constant of 3.78 is used for fast and low-cost fabrication. Front-side and back-side metal patterns are fabricated in a 300-nm-thick gold layer on top of a 20-nm-thick titanium on the quartz substrate. Dimensions of the dipole antenna, directors, and balun were optimized as shown in Fig. 1(b) by a three-dimensional full-wave structure simulator, Ansoft HFSS. The height and width of the slit region are determined to be 90 and 1300 µm, respectively, to accommodate the 50-µm-thick and 1220 µm-wide substrate in consideration of the tolerance in waveguide machining and substrate dicing processes. Unfortunately, this slit region, which deforms the standard waveguide, disturbs the electromagnetic propagation in the waveguide, and excites unwanted waveguide modes, even though its height is Fig. 2. Simulated S-parameters of the transitions. (a) Comparison between 400-µm-wide (dashed lines) and 1220-µm-wide (solid lines) transitions. (b) Simulated S 21 as a function of substrate width (W s ). quite low. Actually, it can store an electromagnetic energy at certain in-band frequencies like resonators, which seriously degrades transition performance at those frequencies and seriously limits the bandwidth of the transitions. Fig. 2(a) shows the simulated S-parameters versus frequency of the transition with 1220-µm-wide substrate shown in Fig. 1(a). It also contains the simulation results of the transition with 400-µm-wide substrate and the same layout to Fig. 1(a), which fits into the standard waveguide and thus does not require the slit region. We selected a little bit smaller substrate (400 µm wide) than the waveguide (430 µm wide), considering the tolerances in the machining and dicing process. The substrate resides on the metallic pedestal as shown in Fig. 1(a), which prevents the direct wave transmission from input to output. The transition with 400-µm-wide substrate shows very flat insertion loss ( 20log S 21 ) of 1.3 db from 235 to 330 GHz, as shown in Fig. 2(a). Metallic and dielectric losses of the materials were included in the simulations. The losses of the dipole antenna including directors, balun, and microstrip line were simulated to be 0.35, 0.2, and 0.2 db, respectively. Therefore, the coupling

442 IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 7, NO. 4, JULY 2017 Fig. 3. Simulated electric field intensity in the slit region (z =450 µm) at 306 GHz. (a) 400-µm-wide transition. (b) 1220-µm-wide transition [see Fig. 1(b)]. efficiency of the dipole and balun can be estimated to be 88.1%. Low-frequency roll-off in S 21 is caused by the dipole antenna and balun. On the contrary, the transition with 1220-µm-wide substrate exhibits sharp drops in S 21 at 256, 294, and 306 GHz which badly limits the bandwidth of the transition. Recalling the fact that the 400-µm-wide transition does not exhibit these drops, we can infer that they are caused by the slit region to accommodate wide substrate. Fig. 2(b) shows that the simulated S 21 as the substrate width (W s ) increases. Note that slit width is 80 µm wider than that of the substrate. The resonances occur in-band if the substrate width exceeds that of the waveguide. The wider the substrate or the slit, the more the in-band resonance frequencies. This simulation result implies that the transition performance becomes seriously degraded as the substrate or slit width increases. It was also found from the simulation that the resonance frequencies can be affected by the substrate length (L s ). That is, the longer the slit, the lower the resonance frequencies. Fig. 3 shows the distribution of electric field intensity in the plane of the slit region (z = 450 µm) at one of the resonance frequencies (306 GHz) of 1220-µm-wide transition. The 400- µm-wide transition shows relatively uniform and low-intensity electric fields in the slit region as shown in Fig. 3(a). On the contrary, the wide transition has electric field highly concentrated in the slit region as shown in Fig. 3(b), which disturbs the propagation of electromagnetic energy. These resonances badly reduce S 21 bandwidth to 35 GHz from 94 GHz of the 400-µm-wide transition. Note that S 21 bandwidth in this paper is calculated as frequency range within which S 21 exhibits the ripple less than ±0.5 db. From these simulation results, it can be stated that the resonances in the transmission characteristics of the transition are attributed to the rectangular cavity formed in the slit region. Therefore, these resonances can be moved out of the band or suppressed by perturbing the field of the cavity resonator. In order to remove in-band resonances and restore the bandwidth, we propose two techniques: SAs, which add metallic strips on the substrate in the slit region and IW, which deforms the cavity structure. Fig. 4(a) depicts the substrate with SAs at four corners. The microstrip SAs disturb the electric field distribution and prohibits the excitation of the resonance in the slit region [17]. The quarter-wave open-stubs in [17] were used to reduce Fig. 4. Proposed techniques to eliminate in-band resonances in the slit region. Length unit is µm. (a) SAs with alternating stub lengths. (b) Simulated S 21 depending on stub length (L c ). (c) IW used as top waveguide. the field in the slit region by enforcing short circuit along the waveguide wall. The field at certain frequencies can be excited at these kinds of the periodic metallic structures [18], which can perturb and suppress the resonance field of the cavity in the slit region. To improve the operation bandwidth of the SAs, we introduced alternating arrays of long and short stubs. Fig. 4(b) shows the simulated S 21 according to the stub length (L c ). Compared with the single length of stubs, the SAs combining short and long stub lengths achieve the best bandwidth performance. It is also found from the simulation that they can operate over wide range of the substrate width. In addition to the SAs, we propose the IW as shown in Fig. 4(c), where the top waveguide is turned upside down to better show the indentation. Unlike straight waveguide as shown in Fig. 1(a), the IW has cuts in the middle of the top block (the slit region of the microstrip line section), which can also disturb the field distribution in the slit region and suppress the resonances. Note that the active circuits will be placed in the middle

KIM et al.: SUBMILLIMETER-WAVE WAVEGUIDE-TO-MICROSTRIP TRANSITIONS FOR WIDE CIRCUITS/WAFERS 443 Fig. 5. Simulated electric field intensity in the slit region (at z = 450 µm) at 306 GHz. (a) Transition with SAs. (b) Transition with indented waveguide. Fig. 7. Photograph of the transition. (a) Close-up view of the transition with SAs (top waveguide removed). (b) Standard and indented waveguides used as top waveguide. Fig. 6. Simulated S-parameters of the proposed transitions. (SA: stub array, IW: indented waveguide). between input and output transitions, so that the proposed indentation on the top waveguide will provide enough room for dc bias circuitries and wire-bonding. One of the sidewalls of the indented part in the top waveguide can be opened for outer circuit interconnection such as dc bias supply. The simulation shows that this opening does not degrade the performance of the transitions. Dimensions of the SAs and IW were optimized from the simulation and are displayed in Fig. 4. Fig. 5 shows the electric field distributions of the transition with SAs [see Fig. 5(a)] and IW [see Fig. 5(b)] at the same plane and frequency to those in Fig. 3. They show nearly uniform field distribution in the slit region compared with Fig. 3(b), and the resonance is effectively eliminated. It was verified from the simulation that the optimized SA and IW for the 1220-µm-wide circuit can also eliminate the resonances for the circuit width from 920 to 1320 µm. The dimension can be modified to remove the resonances when the circuit width deviates from this range. Fig. 6 shows the simulated S 21 and S 11 of three transitions: one with SAs, another with the indented waveguide, and the other with both SAs and IW. Note that the sharp drops in S 21 shown in Fig. 2(a) are disappeared in all three transitions, dramatically extending the bandwidth. The transition with SA shows a little bit less bandwidth than that with IW, which is caused by the limited bandwidth of the SAs. The loss in the SAs increases the insertion loss of the transition. The transition with the IW exhibits the lowest insertion loss. The best bandwidth of insertion and return loss ( 20log S 11 ) were accomplished by combining SAs and IW techniques. III. MEASUREMENT RESULTS To verify the proposed idea, we designed and fabricated four transitions; one without SAs or IW, one with SAs only, one with IW only, and one with both SAs and IW. They all have the same size and layout of the substrate. The substrate is packaged in the waveguide jig of 3 3 3cm 3.Fig.7(a)showsthe close-up view of the transition with SAs sitting on the slit region (1480 1300 µm 2 ). The substrate was well-aligned with the waveguide blocks using a microscope, minimizing the performance degradation by the alignment errors in the substrate installation. The alignment error can be maximally 40 µm in consideration of the machining and dicing tolerances. We found from the simulation that this amount of error can slightly degrade the performance, but not seriously; S 11 just becomes higher than 10 db at certain frequencies. We believe from Fig. 7(a) that the error will not be greater than 20 µm, which does not affect the performance. The fabricated standard and indented top waveguides are compared in Fig. 7(b). The waveguide blocks were made of gold-plated aluminum. S-parameters of the transitions were measured using a vector network analyzer with WR-03 frequency extenders of which the ports are waveguides. The calibration was performed at these waveguide ports using through-reflect-line calibration kits. Then, the fabricated transition was placed between the calibrated WR-03 waveguide ports, so that the measured S-parameters represent the performance of the transition itself.

444 IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 7, NO. 4, JULY 2017 TABLE I COMPARISON OF MEASURED RESULTS OF H-BAND WAVEGUIDE-TO-MICROSTRIP TRANSITIONS FOR WIDE SUBSTRATE (1220 µm) Transition with no SA or IW SA only IW only both SA and IW S 21 BW (GHz) 1 45.0 78.0 79.0 80.0 S 21 (db) 2 2.1 2.4 2.1 2.3 Worst S 11 (db) 3 13.2 10.6 12.1 13.7 1 Bandwidth of S 21 with ±0.5 db ripple. 2 Average S 21 within S 21 BW. 3 Worst S 11 within S 21 BW. of 2.2 db of the 3-cm-long waveguide transition. Insertion loss per single transition is calculated to be 0.8 db, in consideration of the simulated loss of microstrip line and the measured loss of waveguide section. In addition, it also exhibits the best return loss better than 13.7 db within S 21 bandwidth. Fig. 8. Measured S-parameter of the back-to-back transitions. (a) S 21. (b) S 11. Fig. 8 shows the measured S-parameters of the back-to-back connected transitions. The traditional transition without SAs or IW exhibits very limited S 21 bandwidth of 45 GHz because of the resonance around 304 GHz as expected from the simulation. This resonance was effectively eliminated by SAs and IW as shown in Fig. 8(a), which leads to dramatic increase of S 21 bandwidth up to 80 GHz. Unlike the simulation results shown in Fig. 6, the performance improvement by employing both SAs and IW was not clear in the upper-end of the band, compared with the IW only case. The simulation shows that this can be caused by the machining errors in the waveguide manufacture which is found to be able to affect the performance of the transition especially at the frequency around 300 GHz. The ripples in S-parameters were attributed to the mismatches or misalignments between the waveguide flanges of the measurement instrument and 3-cm-long transition, which can be occurred by the waveguide machining tolerances. Table I summarizes and compares the measured performance of the transitions. The IW successfully removes the resonance without degrading insertion loss. On the other hand, the SAs degraded the insertion loss by 0.3 db. The transition with both SAs and IW shows the best bandwidth of insertion loss of 80 GHz (234 314 GHz), where the average insertion loss is 2.3 db with less than ±0.5-dB ripple. This is very close to the simulated loss IV. CONCLUSION We investigated the resonance problems in the waveguide-tomicrostrip transition using dipole antenna, which are caused by the slit in the sidewall of the waveguide used to mount wide circuits. We also proposed and designed the SAs and indented waveguides capable of suppressing in-band resonances and restoring the bandwidth performance. The fabricated H- band waveguide-to-microstrip transition with both SAs and IW exhibits low insertion loss over broadband with good return loss. The resonances in the slit region were moved out of band. The proposed techniques do not require additional processes such as laser dicing or deep etching techniques to form nonrectangular substrates. Therefore, they can be used for various broadband THz passive circuits with wider substrates than the standard waveguides. In addition, they can also be easily extended to monolithic on-chip transitions in submillimeter-wave wide-width ICs. REFERENCES [1] S. Moghadami et al., A 210 GHz fully-integrated OOK transceiver for short-range wireless chip-to-chip communication in 40 nm CMOS technology, IEEE Trans. THz Sci. Technol., vol. 5, no. 5, pp. 737 741, Sep. 2015. [2] I. Kallfass et al., All active MMIC-based wireless communication at 220 GHz, IEEE Trans. THz Sci. Technol., vol. 1, no. 2, pp. 477 487, Nov. 2011. [3] J. Kim, S. Jeon, M. Kim, M. Urteaga, and J. Jeong, H-band power amplifier integrated circuits using 250-nm InP HBT technology, IEEE Trans. Terahertz Sci. Technol., vol. 5, no. 2, pp. 215 222, Mar. 2015. [4] K. Eriksson, S. E. Gunnarsson, V. Vassilev, and H. Zirath, Design and characterization of H-band (220 325 GHz) amplifiers in a 250-nm InP DHBT technology, IEEE Trans. THz Sci. Technol., vol. 4, no. 1, pp. 56 64, Sep. 2013. [5] V. Radisic et al., Power amplification at 0.65 THz using InP HEMTs, IEEE Trans. Microw. Theory Techn., vol. 60, no. 3, pp. 724 729, Mar. 2012. [6] J. Kim and J. Jeong, Submillimeter-wave InP HBT power amplifier using impedance-transforming two-way Balun, Microw. Opt. Technol. Lett., vol. 57, no. 8, pp. 1831 1834, Aug. 2015. [7] A. Tessmann et al., Abroadband220 320GHzmediumpoweramplifier module, in Proc. IEEE Compound Semicond. Integr. Circuit Symp., 2010, pp. 1 4.

KIM et al.: SUBMILLIMETER-WAVE WAVEGUIDE-TO-MICROSTRIP TRANSITIONS FOR WIDE CIRCUITS/WAFERS 445 [8] W. R. Deal et al., THz monolithic integrated circuits using InP high electron mobility transistors, IEEE Trans. Terahertz Technol., vol. 1, no. 1, pp. 25 32, Sep. 2011. [9] A. Tessmann et al., Metamorphic HEMT MMICs and modules operating between 300 and 500 GHz, IEEE J. Solid-State Circuits, vol. 46, no. 10, pp. 2193 2202, Oct. 2011. [10] V. Radisic et al., 220-GHz solid-state power amplifier modules, IEEE J. Solid-State Circuits, vol. 47, no. 10, pp. 2291 2297, Oct. 2012. [11] L. Samoska et al., A submillimeter wave HEMT amplifier module with integrated waveguide transitions operating above 300 GHz, IEEE Trans. Microw. Theory Techn., vol. 56, no. 6, pp. 1380 1388, Jun. 2008. [12] K. Leong et al., A 340 380 GHz integrated CB-CPW-to-waveguide transition for sub millimeter-wave MMIC Packaging, IEEE Microw. Wireless Compon. Lett., vol. 19, no. 6, pp. 413 415, Jun. 2009. [13] W. R. Deal et al., Demonstration of a 0.48 THz amplifier module using InP HEMT transistors, IEEE Microw. Wireless Compon. Lett., vol. 20, no. 5, pp. 289 291, May 2010. [14] V. Radisic et al., Power amplification at 0.65 THz using InP HEMTs, IEEE Trans. Microw. Theory Tech., vol. 60, no. 3, pp. 724 729, Mar. 2012. [15] N. Kaneda, Y. Qian, and T. Itoh, A broad-band microstrip-to-waveguide transition using quasi-yagi antenna, IEEE Trans. Microw. Theory Techn., vol. 47, no. 12, pp. 2562 2567, Dec. 1999. [16] S. Koziel, S. Ogurtsow, W. Zieniutycz, and A. Bekasiewicz, Design of a planar UWB dipole antenna with an integrated Balun using surrogate based optimization, IEEE Antennas Wireless Propag. Lett., vol. 14, pp. 366 369, Feb. 2015. [17] J. Jeong, Y. Kwon, S. Lee, C. Cheon, and E. A. Sovero, 1.6- and 3.3-W Power-Amplifier Modules at 24 GHz Using Waveguide-based power combining structures, IEEE Trans. Microw. Theory Techn.,vol.48, no. 12, pp. 2700 2708, Dec. 2000. [18] L. Chen, Y. Wei, X. Zang, Y. Zhu, and S. Zhuang, Excitation of dark multipolar plasmonic resonances at terahertz frequencies, Sci. Rep., vol. 6, 2016, Art. no. 22027. Jungsik Kim received the B.S. degree in wireless communications engineering from Kangwoon University, Seoul, South Korea, in 2011. He is currently working toward the Ph.D. degree in electronic engineering at Sogang University, Seoul. His research interests include monolithic microwave integrated circuits, THz integrated circuits, and wireless power transfers. Wonseok Choe received the B.S. and M.S. degrees in electronic engineering from Sogang University, Seoul, South Korea, in 2012 and 2014, respectively, where he is currently working toward the Ph.D. degree in electronic engineering. His research interests include monolithic microwave integrated circuits and THz integrated circuits. Jinho Jeong (S 00 M 05) received the B.S., M.S., and Ph.D. degrees in electrical engineering from Seoul National University, Seoul, South Korea, in 1997, 1999, and 2004, respectively. From 2004 to 2007, he was in the University of California at San Diego, La Jolla, CA, USA, as a Postdoctoral Scholar, where he was involved with the design of highefficiency and high-linearity RF power amplifiers. In 2007, he joined the Department of Electronics and Communications Engineering, Kwangwoon University, Seoul. Since 2010, he has been with the Department of Electronic Engineering, Sogang University, Seoul. His research interests include monolithic microwave integrated circuits, THz integrated circuits, high-efficiency/high-linearity power amplifiers and oscillators, and wireless power transfers.