A Proton Irradiated CMOS On-Chip Vivaldi Antenna for 300 GHz Band Slat Array Implementation

As the CMOS transceiver reaches the sub-millimeter wave operating frequency, its circuit area cannot keep up with the shrinkage of the <inline-formula> <tex-math notation="LaTeX">$0.5 \lambda_0 \times 0.5 \lambda_0$ </tex-math></inline-formula> area limit for the typical 2-dimensional (2D) tile-based phased array topology. This article proposes an end-fire on-chip Vivaldi antenna on a standard 65-nm CMOS process for the 300 GHz band operation. The Vivaldi architecture was chosen for its broadband and end-fire radiation characteristics. End-fire antenna is required for slat array topology, which enables 2 D array implementation for transceivers with circuit area above <inline-formula> <tex-math notation="LaTeX">$0.5 \lambda_0 \times 0.5 \lambda_0$ </tex-math></inline-formula>. The antenna length was shortened to maximize beamwidth and reduce area. Additionally, comb-shaped slots were added to suppress side lobes and back radiation caused by the short length. To prevent higher mode surface waves from distorting the antenna radiation pattern and reducing efficiency, the substrate was thinned to <inline-formula> <tex-math notation="LaTeX">$50 \mu \mathrm{m}$ </tex-math></inline-formula>. A dual-layer proton irradiation process increases the substrate resistivity to <inline-formula> <tex-math notation="LaTeX">$1 \mathrm{k} \Omega$ </tex-math></inline-formula>-cm, allowing high-efficiency on-chip antenna implementation on low-cost CMOS processes. The manufactured on-chip Vivaldi antenna has an area of <inline-formula> <tex-math notation="LaTeX">$0.45 \lambda_0 \times 0.45 \lambda_0$ </tex-math></inline-formula>, with measurement results showing 6 dBi gain with 1 dB flatness from 220 GHz to 320 GHz (37% bandwidth) and 76° E-plane beamwidth at 270 GHz with 87% efficiency. A <inline-formula> <tex-math notation="LaTeX">$1 \times 4$ </tex-math></inline-formula> slat array implementation using the proposed on-chip Vivaldi antenna has been demonstrated, with measurement results showing a 56° beam steering range across the E-plane.


I. INTRODUCTION
I N RECENT years, research on 300 GHz band (220 GHz to 320 GHz) transceivers (TRX) has uncovered a lot of potential for applications such as imaging [1], [2], [3], radar [4], [5], [6], [7], [8], and ultra-high speed communication [9], [10], [11], [12], [13], [14].The large available bandwidth can be utilized to improve radar spatial resolution or achieve the Tb/s data rates required for future 6G wireless communication systems [15].The sub-millimeter wavelength reduces antenna size to the point where on-chip antenna (OCA) implementation is possible, eliminating additional losses, parasitics, and variations introduced by chip-to-PCB interconnects [16], [17].However, the 300 GHz band has a significant path loss [18] and cannot penetrate through obstacles.Adding a lens on top of an OCA [3], [4], [5], [6], [19] compensates for the path loss with the high lens gain.However, the narrow beamwidth and the static radiation pattern necessitate external mechanical parts for beam steering to detect or avoid obstacles.Such mechanical parts are undesirable as they increase system size and complexity.A phased array antenna implementation satisfies the high gain and electronically controlled beam steering requirement [20].The sub-millimeter-sized antenna enables the implementation of many elements within the same area footprint typically occupied by a single sub-10 GHz antenna.Furthermore, the feasibility of TRX implementation using c 2024 The Authors.This work is licensed under a Creative Commons Attribution 4.0 License.
For more information, see https://creativecommons.org/licenses/by/4.0/Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.a standard CMOS process has been demonstrated [8], [9], [10], [11], [12], [13], promising low cost, high yield, and large production capacity needed to manufacture the required high number of elements.
The implementation of a phased array below 60 GHz is typically done in tile configuration [21], [22], [23] using broadside antenna due to the ease of assembly, simplicity, and lower cost [24].However, the TRX chip area survey in Fig. 1 shows that as the frequency increases above 60 GHz, the TRX chip area stays roughly the same [25], [26], [27], [28], [29], [30], [31], [32], while the 0.5λ 0 × 0.5λ 0 tile area limit keeps shrinking until it becomes smaller than the TRX area (λ 0 is the wavelength in free space).The TRX shape can be modified to only maintain less than 0.5λ 0 length at one side [8], [28], forming a tile array depicted in Fig. 2(a).However, the array scalability is decreased from M×N into 2×N, which reduces array density and area efficiency.The slat array implementation using endfire OCAs, shown in Fig. 2(b), eliminates the TRX area limitation by introducing a third dimension to expand the array, restoring the array scalability to M×N.However, endfire OCA performance on sub-terahertz frequency suffers from the surface wave effect and resistive loss induced by the high relative permittivity ( r = 11.9) and low resistivity (around 3 to 15 •cm) of the silicon substrate used in the standard CMOS process [33], [34].Furthermore, simple substrate isolation methods commonly implemented in broadside OCAs, such as bottom metal shielding [35], [36], [37], [38], [39], [40] and artificial magnetic conductors (AMC) [41], cannot be applied to the end-fire antenna.Therefore, most end-fire OCAs were either implemented on a separate lowloss substrate like quartz [42] and glass [43] or utilized a special etching technique to remove the silicon substrate below the antenna [44], [45], resulting in more complex design fabrication and integration.A fully in-process endfire OCA implementation has been demonstrated in the SiGe BiCMOS process [46].However, the silicon substrate in the SiGe BiCMOS process has a higher substrate resistivity (around 50 •cm) compared to the standard CMOS process, which significantly reduces degradations due to the substrate resistive loss.
This article proposes an end-fire 300 GHz band Vivaldi OCA fabricated on the standard 65-nm CMOS process for the slat array implementation.The Vivaldi architecture was chosen for being a wideband end-fire antenna that can cover 220 GHz to 320 GHz frequency.Because singleelement gain becomes less important for large-scale arrays, the Vivaldi OCA was optimized for wide 3-dB beamwidth and high efficiency, leading to shorter antenna length, smaller area, and lower gain in exchange for wider beamwidth than the typical gain-optimized Vivaldi antenna.Combshaped slots were added to prevent back radiation from happening due to the short length.To improve efficiency, the substrate was thinned to 50 μm to prevent the excitation of higher mode surface waves, and the dual-layer proton irradiation was applied to raise the substrate resistivity.The dual-layer proton irradiation was developed as a postprocess and can be performed separately after the chip is manufactured [47], [48], allowing a high-resistivity substrate implementation while keeping the low-cost and high-volume production capacity of the standard CMOS process.Further details on antenna structure, working principle, and parametric analysis are described in Section II.A 1×4 slat array with the Vivaldi OCA as the element was designed as a proof of concept, with design details described in Section III.Section IV describes the verification of single-element and array design through measurement.Section V concludes the article.

II. ON-CHIP VIVALDI ANTENNA A. ANTENNA STRUCTURE AND OPERATING PRINCIPLE
Fig. 3 shows the overall implementation of the proposed Vivaldi OCA.The chip size, the antenna placement on the chip, and the chip placement on the PCB were made to closely replicate a single 300 GHz band TRX implementation.Fig. 4 shows the cross-section of the standard 65-nm CMOS process used to implement the Vivaldi OCA, consisting of nine copper metal layers enclosed with 8.75 μm silicon dioxide inter-metal dielectric (IMD) and 2.5 μm passivation layer formed above the silicon substrate with permittivity and resistivity of 11.9 and 3 •cm, respectively.
The Vivaldi OCA was implemented at the top metal (M9), with the detailed structure shown in Fig. 5.The antenna   radiates through the traveling-wave mechanism formed by the exponential-tapered slot [49].Theoretically, the feeding structure limits the upper-frequency cutoff, while the aperture width W M determines the lower frequency cutoff [50], [51].The taper length L A is proportional to the gain and inversely proportional to the beamwidth [51], [52].Therefore, optimizing for maximum beamwidth leads to lower gain and shorter L A , with area reduction as an added benefit.However, there  is a limit on how short the L A is before the antenna deviates from the typical traveling-wave antenna behavior [51].This anomalous behavior can be attributed to the increasing nonradiated energy flowing backward through the outer side of the flare as the L A becomes shorter and a high permittivity substrate is used.This excess energy can interact with structures outside the antenna and cause problems such as unwanted side lobes, resonance, and circuit interference.The antenna characteristics also become more sensitive to structural change outside the antenna, as illustrated by the significant changes of the radiation pattern in Fig. 7(a) and reflection coefficient in Fig. 7(b) due to additional reflector placed at the back of the antenna (Fig. 6(a) and 6(b)).Therefore, comb-shaped slots were added to the outer side of both antenna flares to increase the path impedance at those locations, forcing more energy to radiate forward and preventing further flow to the back of the antenna, as shown in Fig. 6(c) and 6(d).Consequently, the antenna characteristics become more resistant to structure variations outside the antenna, as shown in Fig. 7.The slot length L S was set approximately by 0.25λ 0 / √ r [53], with λ 0 denoting the wavelength at the center frequency.The slots should cover the whole flare's outer sides to maximize the suppression of the energy backflow.
Fig. 8(a) shows the slot line radial stub feeding structure used to convert the grounded co-planar waveguide (GCPW) transmission line to the slot line mode of the Vivaldi OCA.This architecture was chosen because it provides the largest bandwidth compared to other planar CPW-to-slot line transitions [54].The back-to-back structure simulation in Fig. 8(b) shows that the coverage of this feeding structure exceeded the target design frequency (220 GHz to 320 GHz), ensuring that the feeding structure does not limit the antenna bandwidth.
At the 300 GHz band, the 300 μm thick silicon substrate used in the 65-nm CMOS process enables the excitation of higher transverse electric (TE) and transverse magnetic (TM) modes within the substrate, reducing radiation efficiency and distorting the radiation pattern.The relationship between substrate thickness (d) and the cutoff frequency (f c ) of the TE and TM modes n for both grounded (g) and ungrounded (ug) substrates are given as follows [33]: where c, r , and μ r are the speed of light, substrate dielectric permittivity, and substrate dielectric permeability, respectively.Therefore, the substrate thickness must be reduced until the cutoff frequency of TE 1 and TM 1 mode is higher than the maximum operating frequency of the antenna.
Because the Vivaldi OCA metal structure behaves like a large ground cover, (1) was used to determine the suitable substrate thickness.The substrate thickness value of 50 μm was chosen to ensure that the higher order modes above 320 GHz are suppressed while the chip maintains enough structural integrity for further post-processing.dissipation and eddy current, which degrades antenna efficiency.Fig. 10 shows how increasing the substrate resistivity improves the antenna efficiency and gain, with the maximum improvement achieved at 1 k •cm.Therefore, dual-layer proton irradiation was used to increase the substrate resistivity above 1 k •cm [55], with process details shown in Fig. 11.A 100 μm thick nickel mask was used to localize the high-resistivity region formation around the OCA and protect the active circuit from radiation damage, eliminating the need for active device remodeling.A dual-layer profile was used instead of the typical single profile [47], [48] to reduce the total proton fluence requirement, shortening  the irradiation time [55].For this implementation, the duallayer profile consists of a main irradiation at 60 μm depth with 2×10 14 cm -2 fluence, and an interface irradiation at 10 μm depth with 2×10 14 cm -2 fluence.Fig. 12 shows the substrate resistivity after irradiation measured with the spread resistance profiling (SRP) method.The resistivity after 1 minute of 260 • C annealing was also measured to ensure the formed high-resistivity region can withstand hightemperature post-processing.

B. DESIGN PROCESS AND PARAMETRIC ANALYSIS
The Vivaldi OCA design process began with the initialization of all design parameters.First, the substrate thickness and resistivity were set to 50 μm and 1k •cm, in accordance with the discussion in the previous subsection.Then, the feeding structure was designed and optimized on the initialized substrate, resulting in optimized design parameters in Fig. 8, where the value of W T = 5 μm was determined.Finally, the rest of the OCA was built and initialized as follows (all units in μm, except R): W M = 450, W F = 25, L A = 300, L B = 200, L E = 150, L S = 80, S = 5, D S = 10, and R = 3. W F and L B were kept constant.The optimization and parametric analysis were performed within the 200 GHz to 340 GHz frequency range, with a target operating frequency of 220 GHz to 320 GHz.
The first parametric analysis was performed on the combshaped slots parameters L S , S, and D S .A back reflector similar to Fig. 6(d) was added to see the impact of the combshaped slots better.Fig. 13 shows how the slots suppress gain dips at 220 GHz and 312 GHz induced by the energy reflected from the reflector.Shorter L S provides better dip suppression at a higher frequency, while longer L S has better  low-frequency dip suppression.Therefore, L S was adjusted to around 80 μm to achieve balanced dip suppression performance centered around 270 GHz, which also shows good agreement with the slot design equation in [53].Fig. 14 shows the optimum value for both D S and S is between 5 μm to 10 μm, or around L S /10.The deviation of D S value from L S /10 has a minor effect, while the S value must be kept larger than 5 μm to prevent degradation at upperfrequency cutoff.After the optimization of comb-shaped slots, the reflector will not affect the antenna characteristics and can be removed during subsequent optimizations.
The rest of the design parameter, L A , W M , R, and L E , was investigated and optimized through parametric analysis.The results in Fig. 15 confirm that shorter L A leads to the wider E-plane and H-plane 3-dB beamwidth.However, there is a point where efficiency starts to degrade because the taper is too short to perform the traveling-wave radiation mechanism properly.Therefore, L A should be set at the minimum value where it still operates above 80% efficiency, which was 300 μm in this case.The impact of W M shown in Fig. 16 indicates that the overall reflection coefficient increases as W M decreases.This fits with the Vivaldi antenna design equation in [50], [51], where smaller W M moves the lower cutoff frequency closer to the operating frequency, causing the reflection coefficient to increase.To keep the antenna area small, W M was reduced to the minimum size where the overall reflection coefficient is still below −10 dB, which is around 450 μm.The parametric analysis results of R shown in Fig. 17 are mainly a trade-off between gain and reflection coefficient, where the gain and the reflection coefficient are   inversely proportional to R. Therefore, R was also reduced to the point where the overall reflection coefficient is still below −10 dB, which is around 3. Finally, the analysis of the L E variation summarized in Fig. 18 shows how it behaves as a dielectric load [46] that can be tuned to boost the gain or to adjust the reflection coefficient of the antenna.In this design, L E is tuned to achieve a balance between both gain and reflection coefficient, resulting in L E value set to 150 μ m.
The post-optimization design parameters for the OCA main structure can be found in Fig. 5.For the chip implementation, the area around the OCA was filled with metal dummies to fulfill the metal density requirement.Fig. 19 shows the final simulation results of the non-irradiated and irradiated OCA gain, reflection coefficient, and efficiency after dummy placement.After irradiation, the designed onchip Vivaldi antenna can achieve 6 dBi realized gain and around 80% efficiency across 220 GHz to 320 GHz.The reflection coefficient stays below −10 dB, and the gain never decreases below 3 dB of the peak gain across the operating frequency, indicating more than 37% impedance and 3-dB gain bandwidth.Fig. 20 shows the Vivaldi OCA simulated radiation pattern from 220 GHz to 320 GHz, with detailed E and H plane 3-dB beamwidth values shown in Fig. 21.From 220 GHz to 280 GHz, the 3-dB beamwidth for the E-plane and H-plane is constant at around 76 • and 100 • , respectively.As the operating frequency increases to 320 GHz, the beamwidth of both E-plane and H-plane gradually decreases to 58 • and 92 • , respectively.

III. 1X4 ON-CHIP SLAT ARRAY
To demonstrate the application of the proposed Vivaldi OCA as a slat array element, a fully-active 1×4 slat array prototype was designed with antenna configuration shown in Fig. 22.The phase control required to perform beam steering was achieved using an on-chip 300 GHz band TRX on each element, with TRX circuit details can be found in [9].The inter-element pitch D A was limited to 650 μm (0.58λ 0 at 270 GHz) due to the TRX area constraint.Fig. 23 shows how placing the previously optimized single element Vivaldi OCA, with W M = 450 μm, in array configuration causes the reflection coefficient to degrade above the −10 dB limit.Therefore, the W M value was reduced to 400 μm to increase inter-element distance, which reduce the active reflection coefficient to below −10 dB across the operating frequency and reduce the mutual coupling.This indicates different optimum parameters between stand-alone and array implementation, requiring additional re-optimization.Fig. 24 shows the simulated mutual coupling between ports and the impact of inter-element phase Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.the 37% impedance bandwidth.The low mutual coupling of around −20 dB keeps the change of the reflection coefficient to the minimum when θ is changed.
Fig. 25 shows the simulated gains and efficiencies of the array.The array maintained realized gain above 10 dBi and efficiency above 80% from 220 GHz to 320 GHz, maintaining the 37% 3-dB bandwidth of the single element Vivaldi OCA.

IV. MEASUREMENTS AND DISCUSSIONS A. ON-CHIP VIVALDI ANTENNA MEASUREMENT
Fig. 27 shows the die micrograph of the Vivaldi OCA manufactured using the TSMC 65-nm CMOS process.The antenna occupies 0.5 mm × 0.5 mm area on the chip.The fabricated chips were diced to 0.75 mm × 3 mm, and the substrates were thinned to 50 μm thick.The antenna gain and reflection coefficient were measured before and after irradiation.Another measurement was performed after 260 • C annealing for 1 minute to investigate the ability of the irradiated substrate to withstand high-temperature postprocessing (e.g., flip-chip, reflow soldering).Fig. 28 shows the measurement setup to measure the reflection coefficient, gain, and radiation pattern of the single-element Vivaldi OCA as a device-under-test (DUT).The setup consisted of two VDI WR-3.4 extenders connected to the Keysight PNA-X to perform S-parameter measurements from 220 GHz to 320 GHz frequency.The extender port-1 was connected to the DUT via a 325-GHz RF probe (Cascade Microtech i325-S-GSG), and the extender port-2 was connected to the VDI WR-3.4 diagonal horn antenna.The DUT was placed on the edge of the 3 mm thick MEGTRON6 board with a 300 μm distance between the antenna tip and the board edge to replicate the antenna implementation on the PCB.The minimum distance R between the horn and the DUT was determined using the far-field equation [50]: with D as the aperture size.Since the horn has a significantly larger aperture than the DUT, the horn aperture diameter of 5.6 mm was used to calculate the minimum distance.The distance of 7 cm was chosen because it meets the far-field requirement at 320 GHz with some margins.The extender with the horn antenna can be swept within ±60 • around the E-plane for radiation pattern measurements.The metal parts of the measurement setup, including the extender and the GSG probe near DUT, were covered with absorbers during the measurement session to reduce interference.The extender WR-3.4 ports were calibrated using the 2-port thru-reflectline (TRL) calibration.Then, after attaching the GSG probe to port-1 to move the measurement plane to the probe tip, an additional 1-port short-open-load (SOL) calibration was performed.
The measured DUT reflection coefficient includes both the antenna and the pad.The pad S-parameters were characterized through separate measurements, which were used to de-embed the pad from the antenna.The simulated and measured reflection coefficients for all irradiation conditions are compared in Fig. 29.The measured reflection coefficient was below −10 dB within the measurement frequency range Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply. of 220 GHz to 320 GHz, or more than 37% impedance bandwidth for all sample groups.The reflection coefficients of the irradiated DUT were around 5 dB to 10 dB higher than the non-irradiated DUT because the increase in substrate resistivity reduced the substrate loss, resulting in more energy available to be transmitted or reflected.
The antenna gain can be extracted from the S 21 measurement using the following equation: where G horn is the gain of the horn antenna, L pad is the loss of the pad, L wg is the interconnect loss from the horn to the extender, and R is the distance between the horn and the DUT.All terms are in dB.The values of G horn , L pad , R, and L wg were known from the datasheet or separate measurements.The gain measurement and simulation results for all sample groups are compared in Fig. 30.The peak measured gain for the non-irradiated DUT was 2 dBi at 270 GHz with 2 dB variation across the measurement frequency range, which improved by 4 dB to 6 dBi peak gain at 270 GHz with 1dB variation after irradiation.The 3-dB gain bandwidth was measured to be at least 100 GHz (>37%) for all irradiation cases.No significant gain variation was observed between the irradiated and the annealed DUTs, indicating that the gain enhancement provided by the irradiation process can withstand the high-temperature post-processing.Fig. 31 shows the measured and simulated effect of irradiation on the E-plane co-polarized radiation pattern at  270 GHz.The irradiation caused similar gain improvement at pattern within ±48 • angle and slowly diminished above that limit.The measured E-plane 3-dB beamwidth was around 76 • at 270 GHz both before and after irradiation because the beamwidth was within the ±46 • limit.The E-plane radiation pattern of the irradiated Vivaldi OCA across the measurement frequency is shown in Fig. 32.The measured E-plane 3-dB beamwidths were around 76 • from 220 GHz to 280 GHz, which slowly decreases to 72 • at 300 GHz and 54 • at 320 GHz.
Overall, the measurement results generally match the simulation results, with a discrepancy of more than 5 dB in some parts of the reflection coefficient and cross-polarization values.The possible causes of these discrepancies include the difference between actual and simulated substrate resistivity value, the rough chip edge after the dicing process, the probe affecting the antenna characteristics, and the adhesive used to fix the DUT.Due to the limitation of the measurement setup, the antenna efficiency and H-plane 3-dB beamwidth were determined from the simulation after the model was verified with measurement results.The simulated efficiency at 270 GHz is 32% for the non-irradiated DUT and 87% after irradiation.The simulated 3-dB H-plane beamwidth is 96 • before and after irradiation.
Table 1 shows the comparison with other published silicon on-chip antennas operating at the 300 GHz band.The proposed antenna offers the widest bandwidth, the highest radiation efficiency, and a relatively large 3-dB beamwidth compared to previous works.This work is the only end-fire on-chip antenna implemented using a standard CMOS process with less than 0.5λ 0 ×0.5λ 0 dimension, making it suitable for slat array implementation.

B. 1X4 ON-CHIP SLAT ARRAY MEASUREMENT
Fig. 33 shows the die micrograph of the 1×4 slat array manufactured using the TSMC 65-nm CMOS process together with the TRX circuit [9] for phase control and beam steering demonstration.The array covers 0.65 mm × 2.6 mm area.The chip was thinned to 50 μm thick and attached to the PCB through a flip-chip process.The array part was irradiated with dual-layer proton irradiation, while the circuits were covered with a nickel mask to protect them from radiation damage.The DUT was measured with the setup shown in Fig. 34.The required phase shift for each element to perform beam steering was adjusted by an external computer through a serial peripheral interface (SPI) controller.The measured E-plane beam steering pattern for various phase shift ( θ ) input is presented in Fig. 35, which shows good agreement with simulation results at ±45 • phase shift input.The discrepancy at the −90 • phase shift might be caused by the inaccuracy of the phase shifter, as the pattern appears to have phase shift input above −90 • .Other possible causes of discrepancy include rough chip edge after the dicing process, minor phase shifter inaccuracy, and imperfect gain calibration of each TRX element.The beam steering range of 56 • across E-plane has been demonstrated.However, the 3-dB steering range limit cannot be verified due to the limited TRX phase shifter range.

V. CONCLUSION
This article presented the design process, optimization, and parametric analysis of a 300 GHz band on-chip Vivaldi Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.antenna for implementation on a standard 65nm CMOS process with a proton-irradiated substrate.The designed OCA was fabricated, irradiated, and measured to verify the proposed design.The measurement results show 3-4 dB gain improvement across the operating frequency after irradiation, achieving 6 dBi peak gain and 87% efficiency at 270 GHz.The Vivaldi OCA achieved 76 • E-plane beamwidth at 270 GHz and fully covers the 300 GHz band (220 GHz to 320 GHz), which corresponds to 37% impedance and 3-dB gain bandwidth.The implementation of the Vivaldi OCA in 1×4 slat array was performed, which demonstrates 56 • beam steering range.The wide bandwidth, large beamwidth, and high efficiency demonstrated by the Vivaldi OCA make it suitable for future on-chip slat array implementation to achieve full 2-dimensional array scaling.

FIGURE 1 .
FIGURE 1. Survey of TRX chip area in relation to its center frequency on CMOS and SiGe BiCMOS process.

FIGURE 4 .
FIGURE 4. Process cross-section of the standard 65-nm CMOS process used for antenna implementation.

FIGURE 6 .
FIGURE 6.Comparison of simulated Vivaldi OCA E-field at 320 GHz between smooth flare edge (a) without reflector, (b) with reflector, and comb-shaped slots flare edge (c) without reflector, (d) with reflector.

FIGURE 7 .
FIGURE 7. Comparison of simulated (a) E-plane realized gain pattern at 320 GHz and (b) reflection coefficient between smooth and comb-shaped flare edge, both with and without reflector.
3, 5, . . ., for TE n = 0, 2, 4, . . ., for TM(1) Fig.9(a)shows how reducing the substrate thickness to 50 μm eliminates distortion on the radiation pattern.Fig.9(b) also shows how the efficiency decreases as more TE modes get excited with increasing thickness.The 3 •cm resistivity of the substrate used in the 65-nm CMOS process causes additional loss due to thermal

FIGURE 9 .
FIGURE 9.The simulated effect of substrate thickness at 270 GHz on (a) E-plane realized gain pattern and (b) efficiency of the Vivaldi OCA.

FIGURE 10 .
FIGURE 10.The simulated effect of substrate resistivity at 270 GHz on (a) E-plane realized gain pattern and (b) efficiency of the Vivaldi OCA.

FIGURE 11 .
FIGURE 11.The (a) top view and (b) cross-section view of the dual-layer proton irradiation process.

FIGURE 12 .
FIGURE 12.The SRP measurement results of the substrate resistivity after irradiation and annealing process.

FIGURE 13 .
FIGURE 13.The realized gain simulation for different LS.

FIGURE 14 .
FIGURE 14.The realized gain simulation for various (a) DS and (b) S.

FIGURE 15 .
FIGURE 15.The simulation of efficiency, E-plane 3-dB beamwidth, and H-plane 3-dB beamwidth at 270 GHz for different LA.

FIGURE 16 .
FIGURE 16.The simulation of reflection coefficient for different WM.

FIGURE 17 .
FIGURE 17.The simulation of (a) realized gain and (b) reflection coefficient for different R.

FIGURE 18 .
FIGURE 18.The simulation of (a) realized gain and (b) reflection coefficient for different LE.

FIGURE 19 .
FIGURE 19.The post-optimization Vivaldi OCA simulation results of (a) realized gain, reflection coefficient, and (b) efficiency before and after irradiation.

FIGURE 20 .
FIGURE 20.The simulated radiation pattern of the irradiated post-optimizationVivaldi OCA from 220 GHz to 320 GHz.

FIGURE 21 .FIGURE 22 .
FIGURE 21.The simulated E-plane and H-plane 3-dB beamwidth of the irradiated post-optimization Vivaldi OCA.

FIGURE 24 .
FIGURE 24.The simulated (a) active reflection coefficient with different phase shift input, and (b) mutual coupling of the 1x4 on-chip slat array with WM = 400 µm.

FIGURE 25 .
FIGURE 25.The simulated realized gain and radiation efficiency of the 1x4 on-chip slat array with WM = 400 µm.

Fig. 26
shows the beam steering pattern from various θ inputs at 260 GHz.The 3-dB steering range was found at θ input of ±132 • , which corresponds to −40 • and 34 • steering angle, or 74 • steering range.The steering range is close to the value predicted from the E-plane 3-dB beamwidth of the Vivaldi OCA, which is 76 • .

FIGURE 26 .FIGURE 27 .
FIGURE 26.The simulated E-plane beam steering pattern of the 1x4 on-chip slat array at 260 GHz.

FIGURE 28 .
FIGURE 28.Measurement setup for reflection coefficient, gain, and radiation pattern.

FIGURE 29 .FIGURE 30 .
FIGURE 29.Simulated and measured reflection coefficient of the fabricated Vivaldi OCA.

FIGURE 31 .
FIGURE 31.Simulated and measured E-plane radiation pattern of the fabricated Vivaldi OCA at 270 GHz.

FIGURE 32 .
FIGURE 32.Simulated and measured E-plane radiation pattern of the irradiatedVivaldi OCA across the measurement frequency.

FIGURE 33 .
FIGURE 33.The micrograph of the 1x4 slat array and the photo of the PCB implementation.

FIGURE 34 .
FIGURE 34.The setup for the 1x4 slat array beam steering pattern measurement.

FIGURE 35 .
FIGURE 35.The measurement results of the E-plane beam steering pattern at 260 GHz.