An Interlaboratory Comparison of On-Wafer S-Parameter Measurements up to 1.1 THz

IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 1 An Interlaboratory Comparison of On-Wafer S-Parameter Measurements up to 1.1 THz Faisal Mubarak, Member, IEEE, Gia Ngoc Phung, Member, IEEE, Uwe Arz, Senior Member, IEEE, Kamel Haddadi, Member, IEEE, Isabelle Roch-Jeune, Guillaume Ducournau, Member, IEEE, Thomas Flisgen, Ralf Doerner, Member, IEEE, Djamel Allal, Divya Jayasankar, Graduate Student Member, IEEE, Jan Stake, Senior Member, IEEE, Robin Schmidt, Member, IEEE, Gavin Fisher, Nick Ridler, Fellow, IEEE, Xiaobang Shang, Senior Member, IEEE Abstract—This paper reports on an interlaboratory measurement comparison involving on-wafer S-parameter measurements from 10 GHz to 1.1 THz. Seven laboratories are involved, and each participant has measured an individual reference substrate fabricated from a high-resistivity silicon wafer in the same batch. Oneand two-port co-planar waveguide (CPW) structures are designed, simulated, and fabricated. The measurements from 10 GHz to 1.1 THz, extending across six frequency bands, are conducted using different equipment in terms of vendors and specifications (e.g., probe pitch size). Despite such differences, this interlaboratory study has shown a generally good agreement between results from different participants when uncertainties are considered. The comparison with simulated reference values demonstrates agreement within 0.08 for |S11| Received 13 September 2024; revised 2 January 2025; accepted 21 January 2025. Date of publication 30 January 2025. This work was supported, in part by the 23IND10 OnMicro project. The project (23IND10 OnMicro) has received funding from the European Partnership on Metrology, co-financed from the European Union’s Horizon Europe Research and Innovation Programme, and by the Participating States. The project was also supported in part by the Dutch Ministry of Economic Affairs and Climate, in part by the Frech Renatech network, and in part by the Contrat de Plan Etat Region (CPER) WAVETECH @HdF project and Haut de-France Regional council, IEMN UHD flagship project, the IEMN CHOP platform, and France 2030-PEPR FUNTERA and SYSTERA projects, funded by ANR (Agence Nationale de la Recherche) under Grants ANR-22-PEEL-0006 and 22-PEFT-0006, French RENATECH network, the Equipex + Nanofutur (Grant IA-21-ESRE-0012). (Corresponding author: Faisal Mubarak) Faisal Mubarak is with with the Electricity and Time Department of VSL, the National Metrology Institute of The Netherlands, Thijsseweg 11, 2629 JA Delft, The Netherlands (e-mail: [email protected]). Gia Ngoc Phung and Uwe Arz are with Physikalisch-Technische Bundesanstalt (PTB), Bundesallee 100, 38116 Braunschweig, Germany Ralf Doerner and Thomas Flisgen are with Ferdinand-Braun-Institut gGmbH, Leibniz-Institut für Höchstfrequenztechnik, Berlin, Germany. Thomas Flisgen is also with Brandenburgische Technische Universität CottbusSenftenberg, 03046 Cottbus, Germany Gavin Fisher is with Formfactor GmbH, Suss Strasse 1, Germany. Kamel Haddadi, Isabelle Roch-Jeune, and Guillaume Ducournau are with Univ. Lille, CNRS, UMR 8520 - IEMN - Institut d’Electronique de Microélectronique et de Nanotechnologie, - Lille, France Robin Schmidt is with Keysight Labs, Wingepark 51, Belgium. Djamel Allal is with Laboratoire national de métrologie et d’essais (LNE), 29 Avenue Roger Hennequin, 78197 Trappes Cedex, France. Divya Jayasankar and Jan Stake are with Terahertz and Millimetre Wave Laboratory, Department of Microtechnology and Nanoscience, Chalmers University of Technology, SE-41296 Gothenburg, Sweden. Divya Jayasankar is also with the Research Institutes of Sweden, SE-50462, Borås, Sweden. Nick Ridler and Xiaobang Shang are with National Physical Laboratory, Teddington, TW11 0LW, UK. Color versions of one or more figures in this article are available at https://doi.org/10.1109/TTHZ.2025.3537461. Digital Object Identifier 10.1109/TTHZ.2025.3537461 and 2 dB for |S21|measurements of matched devices up to 1.1 THz. The measurement comparison demonstrates the need for a standardized measurement approach and, with that, a potential to achieve accurate on-wafer CPW measurements up to THz frequencies, underpinning the development of integrated circuits for such high frequencies. Index Terms—Calibration, coplanar waveguides (CPW), onwafer, S-parameter measurements, comparison, Terahertz metrology. I. INTRODUCTION THE ability to sense terahertz waves in a chip-scale technology has the potential for transformative applications in sensing [1], imaging [2], and security [3]. Operational frequencies between 100 GHz and 300 GHz are expected to meet the large bandwidth requirements necessary for supporting individual user data rates of up to 100 Gbit/s envisioned for 6th generation (6G) networks [4]. All these applications require devices and integrated circuits operating at these high frequencies, with semiconductor foundries demonstrating active devices with operating frequencies exceeding 1 THz [5], [6]. The measurement infrastructure for on-wafer device characterization is critical for developing such novel devices and systems operating beyond 1 THz. Commercially available measurement equipment for Terahertz characterization is increasing steadily, with frequency extender heads for vector network analyzers (VNAs) [7], on-wafer calibration kits [8], and ground-signal-ground (GSG) probes [9] as essential components. The complexity and diversity in developing such Terahertz testbeds make it increasingly difficult to establish accurate on-wafer measurements up to the WR 1.0 band. This is also evident from the lack of traceability for on-wafer coplanar waveguide (CPW) measurements above 118 GHz, with PTB as the only NMI worldwide having CMC’s for on-wafer measurements [10]. The literature review reveals increasing efforts to develop measurement techniques for on-wafer S-parameter characterization in the millimetre-wave and sub-millimetre-wave range. Williams demonstrated sub-millimetre wave transistor characterization by developing a dedicated calibration kit for the desired technology [11], an approach widely employed in on-wafer device characterization. Recent works © 2025 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/ IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 2 have explored novel techniques and devices for WR 1.0 band characterization using planar goubau lines [12], [13]. A mTRL calibration up to 1 THz was demonstrated with inverted grounded-CPW devices in an InP HBT Process [14]. Likely, traceable CPW measurements on fused silica substrates [15] and membrane technology [16] have been demonstrated for frequencies up to 110 GHz, including a comprehensive uncertainty budget for multiline-TRL (mTRL) on-wafer measurements, dominated by uncertainties arising from the reference calibration standards [15]. While [7], [17] demonstrated waveguide measurement up to WR 1.0 band and investigated the impact of repeatability and cable movements on the combined measurement uncertainty. Furthermore, other studies [18], [19] have focused on investigating the impact of drift error due to the frequency extender heads for the WR 1.0 band. Consequently, [20] recently demonstrated an on-wafer capacitor characterization up to 1.0 THz supported with uncertainty estimates. Measurement comparisons are fundamental for the validation of such uncertainty budgets and provide increased confidence in corresponding measurements. Interlaboratory comparisons in on-wafer measurements have highlighted several considerations necessary for comparable onwafer RF device characterization [21]–[25] such as strong dependency on the probe topology, chuck material, environmental conditions, measurement hardware, and operator skills [21]. In this effort, this work gathers five National Metrology Institutes (NMIs), two European universities, one research institute, and one leading manufacturer of millimetrewave and THz test and measurement instrumentation to provide a comprehensive on-wafer ultra-broadband 10 GHz – 1.1 THz CPW measurement comparison. To the best of the authors’ knowledge, this is the first interlaboratory comparison conducted over a frequency range spanning up to 1.1 THz. Where initial results were made available in [26], this work offers valuable and detailed insights into the expected accuracy of on-wafer measurements and EM simulation tools at high frequencies. The manuscript is organized as follows. Section II presents the design and fabrication process of the reference substrates. A wide range of calibration and validation standards are designed and fabricated to support increased calibration and verification accuracy. In addition, a parametrization study is presented to determine electrical reference values and related uncertainties. Then, Section III proceeds with describing the simulation-based approach for reference values and uncertainties evaluation of the devices used during the interlaboratory comparison. Subsequently, measurement details crucial for organizing a comparison are provided in Section IV. Finally, Section V presents the measurement campaign results and related discussions. II. REFERENCE SUBSTRATE FABRICATION A. Device selection and design A set of calibration and validation CPW structures has been designed on high-resistivity silicon (HR Si) wafer to establish metrology-grade accuracy in planar S-parameter measurements. The selection of the utilized technology is based on its availability. The reference substrate consists of a low-frequency set (kit 1) including 35 devices operating from 10 GHz to 330 GHz and a high-frequency set (kit 2) with also 35 devices operating from 330 GHz to 1.1 THz, as listed in Table I. Two separate sets are designed to ensure there is one well-defined fundamental propagation mode to ensure accurate implementation of mTRL calibration. To ensure single mode of propagation, the following condition is ensured in the design of CPW devices: fmax <c0 10dpϵr,e f f . (1) Here, fmax is 330 GHz for kit-1 devices, and 1100 GHz for kit-2 devices. Furthermore, d is the ground-to-ground distance of the CPW devices in kit-1 and kit-2, and c0is the speed of light. Finally, ϵr,e f f is the effective dielectric permittivity of the substrate. Each set includes eight transmission lines with varying lengths ranging between 610 µm and 4280 µm for kit 1 and between 120 µm and 1320 µm for kit 2. Besides transmission lines, both kits include multiple offset shorts, opens, matched loads, and fixed-distanced two-port structures embedding two one-port devices. This variety of devices allows several calibration techniques to be implemented, i.e., TRL, mTRL, unknown thru (SOLR), and 16-term methods. Furthermore, identical access structures are included in the design of every device to allow for an accurate definition of the measurement reference plane and to realize consistent and accurate measurement results. B. Fabrication process Two wafers have been manufactured, each with eight identical dies, including kit 1 and kit 2 structures. The microfabrication process on a 3-inch (76.2 ±0.3 mm) high resistivity (>5000 Ω/cm) silicon (Si) wafer from Siltronix, with 275 µm (±15 µm) thickness, started with thermal oxidation of the wafer in a furnace to produce a 50 nm layer of stable silicon oxide shown in Figure 1. After deoxidation with Hydrogen Fluoride (HF), a 50 nm dry thermal oxide (TOX) was grown by APCVD (atmospheric pressure chemical vapour deposition) at 1100 ◦C. Then, a resist bilayer was spin-coated and patterned using e-beam lithography with a Nano Maker EBPG-5000 Plus (VistecT M ) for the definition of thin-layer resistors. Following exposure, the resist was developed to remove the exposed areas and create the desired pattern. Subsequently, a 23 nm Titanium (Ti) layer (measured mean value) was evaporated, and the lift-off process was performed to form the set of resistors. Afterward, spin-coating, alignment, and e-beam lithography of a new IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 3 TABLE I: Overview of CPW devices in kit 1 and kit 2 Structure Physical length (µm) Ground type Structures in each kit Purpose Thru kit 1: 500 kit 2: 120 NA 3 system calibration Lines kit 1: 610 - 4280 kit 2: 155 - 1320 NA 8 system calibration Short-Short kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 system calibration Mismatched line kit 1: 1200, 2940 kit 2: 342, 894 NA 2 Interlaboratory comparison Attenuator kit 1: 944 kit 2: 340 NA 1 Interlaboratory comparison Filter kit 1: 944 kit 2: 340 NA 1 Interlaboratory comparison Short-Short kit 1: 0, 1044, 1264, 1444 kit 2: 0, 380, 450, 520 GC & GO 4 & 4 Interlaboratory comparison Open-Open kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 Probe crosstalk Load-Load kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 Probe crosstalk Load-Open kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 Probe crosstalk Open-Short kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 Probe crosstalk Load-Short kit 1: 944 kit 2: 340 Ground Connected & Ground Open 1 & 1 Probe crosstalk Fig. 1: A 2D mapping of the silicon oxide thickness with an average of 53 nm and 51 nm for both wafers, respectively. resist bilayer were conducted to define the CPW structures. The resist was again developed to ensure clean patterning before metal deposition. Finally, a 25 nm adhesion Ti layer and a Gold (Au) layer were deposited and lifted off (resulting in a Ti/Au average thickness of 485 nm). Eight identical reference dies were fabricated on the same wafer. In Figure Figure 2, Scanning Electron Microscope (SEM) images of the reference substrate are shown. C. Parameterization and characterization Relevant parameters affecting the electrical properties of the reference structures are identified and measured to assess high-frequency behavior through 3D electromagnetic (EM) simulations, as outlined Section III. First, before the fabrication of the reference substrate, the sheet resistance was calibrated on a 0.25-inch gallium arsenide (GaAs) wafer to achieve an experimental value of 50.5 Ω. Then, after the fabrication of the reference kits, a mechanical profilometer was used to measure Ti and Ti/Au thicknesses corresponding to the resistive and conductive layers, respectively. The sheet resistance was measured using a non-contact resistivity measurement technique. A Tescan Mira XMU scanning electron microscope (SEM) was used to measure the critical dimensional parameters of CPW structures. For each of the eight dies, 15 parameters of CPW devices from kit 1 and kit 2 have been characterised. The images were processed offline to limit electron charging by the SEM, as shown in Figure 2. The uncertainty for each parameter across eight dies is determined, as shown in Table II for the thru structure. The values obtained for the CPW dimensions were taken into account in the uncertainty estimation. TABLE II: The nominal values and uncertainties of thru device parameters from kit 1 and kit 2. Parameter kit 1 value kit 2 value Uncertainty Width (µm) A, A’ 100 29.2 2-4 % D, D’ 54 17.4 2-4 % G, G’ 100 29.2 2-4 % C 25 9.0 2-4 % F 11 3.5 3-4 % I 25 9.0 2-4 % Length (µm)L 500 120 2 % Spacing (µm) B, B’ 15.96 6.3 3-5 % E, E’ 8 2.6 3-5 % H, H’ 15.96 6.3 3-5 % Substrate ϵr11.80 11.80 3 % TSi (µm)275 275 5.5 % Thickness (µm)Tgold 0.525 0.525 4 % III. REFERENCE SUBSTRATE EVALUATION This section outlines the method for evaluating the reference values and uncertainties corresponding to the IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 4 (a) A’ A B’ B C D’ D E’ E F G’ G H’ H I L A’ A B’ B C D’ D E’ E F G’ G H’ H I L Reference planes A’ A B’ B C D’ D E’ E F G’ G H’ H I L Reference planes (b) Fig. 2: SEM image of (a) reference substrate showing a section including four CPW structures from kit 1 and (b) the thru standard from kit 2 including various device parameters measured for uncertainty evaluation. devices used for the interlaboratory comparison. First, EM simulations in CST Studio Suite®are employed to evaluate the S-parameters corresponding to each CPW device from kit 1 and kit 2, as listed in Table I. However, the fabrication process tolerances, as shown in Table II, introduce variations in the S-parameters of the CPW devices. The dimensional and material parameter variations resulting from the fabrication process tolerances are propagated via EM simulations using the method proposed in [27] to quantify the S-parameter uncertainty. The second step involves evaluating the reference value and combined uncertainty for each device resulting after the calibration. The fabrication process uncertainty of every calibration device contributes to the combined uncertainty corresponding to the devices used in the comparison. For this, Monte-Carlo simulations are employed to evaluate the combined uncertainty of each comparison device. Measurement plane Reference plane Measurement plane Reference plane (a) Reference planeReference plane (b) Fig. 3: (a) Acquisition of EM simulation-based raw data for the calibration standards. (b) Acquisition of EM simulationbased reference data for the calibration standards at the reference plane position. TABLE III: A summary of CST Studio Suite®simulation settings. Parameter Value Background normal (vacuum) Boundaries open boundary Solver frequency domain Solver order 3rd order Mesh type hexahedral mesh Mesh resolution > 1·106cells Accuracy 1·10−4 Excitation waveguide ports A. Simulation approach To calibrate each CPW device and determine the combined uncertainty, EM simulations are employed to acquire the necessary data for all CPW devices embedded in kit1 and kit-2. First, the so-called uncorrected (RAW) data is acquired with EM simulation of all structures, including advanced probe models designed for kit-1 and kit-2 operation. Dedicated probe models are designed for kit-1 and kit-2 structure simulations. In Figure 3(a), for brevity, only the probe model for kit-1 is shown. The probe is positioned to emulate realistic measurement conditions to the device, as it would during an actual on-wafer measurement. Different combinations of parameter variations, resulting from the fabrication process tolerances shown in Table II, are then propagated through these EM simulations to determine a configuration that produces the maximum deviation from nominal S-parameter results. The S-parameter deviation from the nominal values of a kit-1 offset-short structure is shown in Figure 4 for assorted combinations of parameter variations. The configuration with the most significant offset in S-parameters, denoted with the combination(max), is subsequently used to acquire the uncorrected S-parameter and uncertainty of all devices in kit-1 and kit-2. In Table III, a summary of CST Studio Suite®simulation settings is shown. It was not possible IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 5 to measure all dimensions of every structure throughout the eight calibration dies manufactured for the comparison participants, and only 15 selected dimensions on each die were measured. Hence, due to the lack of information, the configuration with the most significant offset is used to estimate device uncertainty instead of the most probable combination. 50 100 150 200 250 300 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 "|!short| 50 100 150 200 250 300 Frequency (GHz) -30 -20 -10 0 10 20 30 "phase(!short) (degrees) 0r(substrate) Lengths Thickness(gold) Widths Combination (max) Fig. 4: S-parameter variations for kit-1 short structure estimated with EM simulations. The S-parameter results acquired for the various parameter settings are normalized to the S-parameter results acquired for nominal parameter values. B. Reference S-parameter &Uncertainty Using the uncorrected S-parameter results acquired in the previous section, the corrected S-parameter values for all devices are determined using the Multiline-ThruReflect-Line (mTRL) calibration method. For this, first, S-parameters of only the TRL calibration devices are acquired with EM simulation at the pre-defined reference plan position, as shown in Figure 3(b). Once the reference data of the calibration devices is acquired, the uncorrected data of all devices is corrected using the TRL calibration method. Simultaneously, the combined uncertainty of each device is estimated using the Monte-Carlo approach, as shown in Figure 5. In total, 1000 samples were used for each frequency point during the Monte-Carlo simulations, which were found sufficient during the Monte-Carlo verification experiment. The corrected simulation results, referred to as reference values, for all comparison devices, are shown in Figure 6. The blue lines depict the reference value for each device, and the corresponding uncertainty is identified with the blue-colored region. The reference value uncertainty accounts for the fabrication process uncertainty of every calibration device used for correcting the comparison device S-parameters and the probe crosstalk uncertainty detailed in Section V-A. The uncertainty of the EM simulation itself is not accounted for in this uncertainty analysis, which can typically range up to 5 % or even higher for 3D structures, as it depends on several factors, i.e., solver type, mesh quality, material properties, and frequency considerations [28]. The simulation accuracy most likely degrades with increasing frequencies due to the lack of material accuracy at higher frequencies. IV. MEASUREMENT DETAILS This section outlines the interlaboratory measurement comparison details for on-wafer S-parameter measurements up to 1.1 THz. The comparison covers an extensive frequency range, including six frequency bands collectively supported by all participants. The measurement comparison participants and details of the various measurement systems are provided in section IV-A. Detailed information on the comparison devices and the corresponding measurement parameters are provided in section IV-B. The most critical part of a measurement comparison is the reference value corresponding to each parameter being compared. Description of the reference value and uncertainty corresponding to the comparison devices are provided in section IV-C. The reference substrate is designed to support mTRL calibration to support the extensive frequency range of the comparison, with more information provided in section IV-D. Finally, section IV-E summarizes uncertainty sources involved in on-wafer measurements and details of the combined measurement uncertainty supported by PTB measurements. A. Participants Each participant is provided with an individual reference substrate due to the destructive nature of on-wafer measurements. Each reference substrate includes two kits, ThruREF Reflect-1REF Reflect-2REF LineREF wafer fabrication process tolerances Reference data (REF) TRL calibration Error terms and uncertainty Output: TRL error-terms DUTRAW ΔS11dut S11 ΔS11thru ΔS11Reflect-1 ΔS11Reflect-2 ΔS11line ThruRAW Reflect-1RAW Reflect-2RAW LineRAW Raw data Fig. 5: Monte-Carlo simulation setup used to propagate wafer fabrication tolerances for mTRL calibration. IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 6 50 100 150 200 250 300 0 0.1 0.2 |S11| (a). kit 1: 50 + load Coaxial WR6 & WR5 WR3 Reference Uncertainty PTB FBH Keysight NPL kit 1 400 500 600 700 800 900 1000 1100 0 0.1 0.2 (b). kit 2: 50 + load WR2 WR1.5 WR1 50 100 150 200 250 300 0.7 0.8 0.9 1 1.1 |S11| (c). kit 1: Open 400 500 600 700 800 900 1000 1100 0.7 0.8 0.9 1 1.1 (d). kit 2: Open 50 100 150 200 250 300 -20 -15 -10 |S21| (dB) (e). kit 1: Attenuator 400 500 600 700 800 900 1000 1100 -20 -15 -10 (f). kit 2: Attenuator 50 100 150 200 250 300 Frequency (GHz) -4 -3 -2 -1 0 |S21| (dB) (g). kit 1: Mismatch line 400 500 600 700 800 900 1000 1100 Frequency (GHz) -6 -4 -2 0(h). kit 2: Mismatched line Reference Uncertainty Chalmers Lille FormFactor kit 2 Fig. 6: Comparison device reference values determined by the median of the measurement results for kit 1 and kit 2. Data depicted with round markers correspond to kit 1 devices, and data using square markers belong to kit 2 devices. with kit-1 supporting measurements in the coaxial frequency range up to 110 GHz, the WR6, WR5, and WR3 waveguide bands. Whereas kit-2 supports measurements in the WR2, WR1.5, and WR1 waveguide bands. Section II provides detailed information on the reference substrate design and fabrication. The measurements for kit 1 range from 10 GHz to 330 GHz and are conducted by the National Physical Laboratory (NPL), Physikalisch-Technische Bundesanstalt (PTB), Ferdinand-Braun-Institut (FBH), and Keysight Technologies®(Keysight). Whereas measurements for kit 2 range from 330 GHz to 1.1 THz and are conducted by the University of Lille (Lille), Chalmers University of Technology (Chalmers), and Form Factor®. Table IV shows an overview of the measurement systems used by the participants. B. Comparison devices The selection of the comparison devices is such to provide coverage across the match and highly mismatched loading conditions for the reflection coefficient (S11) parameter. Therefore, two one-port devices are selected to compare the S11 parameter: a matched load and a highly mismatched open device. For the transmission (S21) parameter, the selection of the devices enables comparison across a broad transmission parameter dynamic range. Hence, a mismatched transmission line and an attenuator device are selected. Both the magnitude and phase parameters are considered in the comparison. All comparison devices have different electrical characteristics compared to calibration standards used for the mTRL calibration employed to correct the data and, therefore, are appropriate for verification and comparison. In Table V, an overview of the comparison devices and the corresponding parameter is provided. C. Comparison reference value The interpretation of the comparison reference value is preferably based on traceable measurement results acquired during the comparison using methods as outlined in [29]. However, as only PTB provided the uncertainty estimates corresponding to their measurements, it is not possible to determine the reference values for the comparison devices using participant measurement results. Hence, IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 7 TABLE IV: Details of the measurement systems used by the participants. Frequency band Laboratory VNA Probe (pitch) Frequency extender Chuck Probing system coaxial Keysight FBH Keysight PNA-X 5247B Anritsu MS4647B Infinity (100 µm) T220A (50 µm) N5295A (M4) Anritsu MA25400A Ceramic Ceramic Form Factor CMX300 Cascade PA200 WR6 WR5 Keysight FBH NPL PTB Keysight PNA-X N5292A Anritsu MS4647B Keysight PNA-X N5247B Anritsu VectorStar DMPI (50 µm) T220A (50 µm) GGB (75 µm) GGB (50 µm) VDI Mini Anritsu MA25400A VDI VDI Ceramic Ceramic Ceramic Ceramic Form Factor CMX300 Cascade PA200 MPI TS-150 MPI TS-150 WR3 Keysight PTB Keysight PNA-X N5292A Anritsu VectorStar DMPI (50 µm) GGB (50 µm) VDI Mini VDI Ceramic Ceramic Form Factor CMX300 MPI TS-150 WR2 LILLE FormFactor Rohde & Schwarz ZVA24 Keysight PNA T-Wave (25 µm) T-Wave (25 µm) Rohde & Schwarz Z500 VDI Absorber Ceramic Formfactor EPS200 FormFactor Summit 200 WR1.5 LILLE Rohde & Schwarz ZVA24 T-Wave (25 µm) VDI VNAX Absorber Formfactor EPS200 WR1 LILLE Chalmers FormFactor Rohde & Schwarz ZVA24 Keysight PNA-X N5242A Keysight PNA T-Wave (25 µm) T-Wave (25 µm) T-Wave (25 µm) Rohde & Schwarz ZC1100 VDI VDI Absorber Absorber Ceramic Formfactor EPS200 MPI TS-150 FormFactor Summit 200 100 200 300 400 500 600 700 800 900 1000 1100 Frequency (GHz) -70 -60 -50 -40 -30 -20 -10 0 probe crosstalk (dB) Coaxial WR6 & WR5 WR3 WR2 WR1.5 WR1 device length 944 7m 944 7m 944 7m 340 7m 340 7m 340 7m Infinity (1007m) DMPI (507m) DMPI (507m) GGB (507m) GGB (507m) GGB (757m) T-Wave (257m) T-Wave (257m) T-Wave (257m) T-Wave (257m) T-Wave (257m) Keysight (kit 1) Chalmers (kit 2) PTB (kit 1) Form Factor (kit 2) NPL (kit 1) Lille (kit 2) Fig. 7: Probe crosstalk measured with two-port devices comprising of two one-port structures as shown in Table I, i.e. Load-Open (LO), Open-Short (OS), and Load-Short (LS). Data depicted with markers are estimated using the set of structures without a common ground plane, and data shown with dotted lines are for structures with a common ground plane. TABLE V: An overview of the comparison devices and parameters. Device Parameter Unit Matched load (1-port) S11 Linear magnitude Open (1-port) S11 Linear magnitude Attenuator (2-port) S21 Log magnitude (dB) Mismatched line (2-port) S21 Log magnitude (dB) Mismatched line (2-port) S21 Phase (degrees) the reference value and the corresponding uncertainty for each comparison device shown in Table V are determined with EM simulations using methods outlined in section III. D. Calibration method The mTRL method is widely regarded as one of the most accurate on-wafer calibration techniques [30]. For instance, in the inter-laboratory study [31] performed by three laboratories in the 140 to 220 GHz frequency band with different calibration approaches (Short-Open-Load-Thru (SOLT), on-wafer mTRL and off-wafer TRL) reliable results could only be obtained by on-wafer mTRL calibration. On-wafer mTRL currently represents the best choice for calibrations of millimetre-wave and THz VNA measurement systems. Hence, the mTRL method is employed in the measurement comparison campaign. Participants were allowed to use a different combination of eight transmission lines for the mTRL calibration of their system, as shown in Table I. It is common practice to verify the quality of the calibration using the measured complex propagation constant or the measured complex effective relative dielectric constant of the transmission lines. Higher-order modes of propagation and probe cross-talk may be observable, allowing us to relate these irregularities to the discrepancies observed in calibration results. Unfortunately, we did not collect this IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 8 data during the interlaboratory comparison. In addition, the participants calibrated their measurements based on different implementations of the mTRL algorithm [32], [33], [34]. A more particular case is Keysight, whose results were calibrated based on an optimal mTRL with additional probe misplacement compensation as realized in [35]. All measurement results presented in this interlaboratory comparison correspond to a reference impedance defined by the characteristic impedance of the transmission line standards. Therefore, the characteristic impedance Z0was not computed, and no renormalization to 50 Ωwas applied to the corrected results. Consequently, any discrepancies related to the reference impedance will most likely arise from the variations observed between different calibration substrates. E. Measurement uncertainty Several types of uncertainty sources generally stem from external and internal disturbances in on-wafer measurements. Internal effects originate from the transmission lines, i.e., radiation and dispersion effects. To incorporate the internal disturbances from radiation or dispersion effects, the uncertainty estimation has been revised to implement the new improved CPW model [36]. External effects, on the other hand, stem from disturbances such as e.g. the influence of microwave probes, the impact of the neighbourhood, propagation of parasitic modes (e.g. surface waves and parallel plate modes) due to the measurement boundary condition and radiation effects from the probes. Due to the complexity of the nature of these effects, a series of studies has been performed to allow for a better understanding of these effects [37]–[39]. With the benefit of hindsight, the layout of both kits has been optimized to mitigate the influence of these uncertainty sources as much as possible. Thus, the remaining uncertainty sources stem from the VNA, cables, and frequency extenders which can still be estimated according to the guide for VNA-based measurement methods [40]. The uncertainties corresponding to the PTB measurements are calculated according to the comprehensive measurement model and the procedure described in [15], making use of the improved CPW model of [36]. The PTB uncertainty budget takes into account instrumentation errors such as VNA noise, linearity, drift, cable stability, contact repeatability, uncertainties from crosstalk between the measurement ports, and CPW calibration standard uncertainties. Due to differences in the measurement equipment used by the participants, these uncertainties are valid for the PTB measurements only, which were performed in D-band and J-band, respectively. Table VI shows an overview of the uncertainty sources supported by the PTB and simulation uncertainty analysis. Furthermore, participants did not provide details corresponding to the exact combination of transmission lines used for the implementation of their mTRL calibration. Hence, no fixed and optimum combination of transmission lines was set, and additional measurement errors may have been introduced if delays between each pair of transmission lines had not been well chosen by the participants. For example, PTB used all eight transmission lines to implement the mTRL calibration. V. MEASUREMENT RESULTS AND DISCUSSION This section presents the interlaboratory comparison measurement results for kit 1 and 2 devices for frequencies up to 1.1 THz. The measurement results submitted by all participants for the comparison devices are shown in Figure 6. Here, measurement results for kit 1 and kit 2 devices are denoted with round and square markers, respectively. The results are collected across three frequency bands with kit 1 devices measured from 10 GHz up to 330 GHz. While measurement results for kit 2 devices range from 330 GHz to 1.1 THz and are accumulated across three waveguide bands. In Table IV, participant details are provided for each frequency band. Comparison results are analyzed by calculating the difference between the measurement, and the simulation-based reference value for each participant, as shown in Figure 8 and Figure 9. First, the probe crosstalk measurements are evaluated, and then the measurement results for each frequency band are separately discussed. A. Results for probe crosstalk The probe crosstalk is estimated by taking the maximum value acquired for the |S21|and |S12|measurement results acquired for devices comprising of two one-port equidistant structures as listed in Table I. This includes all devices identified with ’probe crosstalk’, i.e., Load-Open (LO), OpenShort (OS), and Load-Short (LS). The devices are designed to maintain a constant distance between the probes; for kit-1, the two one-port devices are distanced by 944 µm, and for kit-2, the distance is 340 µm. Furthermore, each set of devices is manufactured two-fold: a set with both devices sharing a common ground and a set with both devices having an isolated ground. The selection of devices is such to offer a wide range of loading conditions to the probes and allows the compilation of crosstalk performance for a variety of different probe types, as shown in Table IV. We estimate probe crosstalk for the set of devices with a common ground plane and the set of devices without a common ground plane separately. This approach includes sensitivity to the load impedance, as eight devices are used for probe crosstalk analysis using the following: TABLE VI: An overview of uncertainty sources accounted for by the PTB measurements and simulated reference values. Uncertainty source PTB measurements Simulated values calibration standards yes yes system noise yes no system linearity yes no system drift yes no cable stability yes no cross-talk yes yes contact repeatability yes no IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 9 50 100 150 200 250 300 -0.1 -0.05 0 0.05 0.1 |S11meas| - |S11sim| (a). kit 1: 50 + load Coaxial WR6 & WR5 WR3 400 500 600 700 800 900 1000 1100 -0.1 -0.05 0 0.05 0.1 (b). kit 2: 50 + load WR2 WR1.5 WR1 50 100 150 200 250 300 Frequency (GHz) -0.2 -0.1 0 0.1 0.2 |S11meas| - |S11sim| (c). kit 1: Open Reference (normalized) Uncertainty PTB FBH Keysight NPL kit 1 400 500 600 700 800 900 1000 1100 Frequency (GHz) -0.2 -0.1 0 0.1 0.2 (d). kit 2: Open Reference (normalized) Uncertainty Chalmers Lille FormFactor kit 2 Fig. 8: Comparison results for reflection coefficient measurements of one-port load and open from kit 1 and kit 2. probe crosstalk =max (x=y∈1,2)¡|Sx y (SS)|,|Sx y (OO)|, |Sxy (LL)|,|Sxy (LO)|,|Sx y (OS)|,|Sx y (LS)|¢. (2) Here, subscript denotes the combination of devices fabricated, i.e., SS denotes Short-Short, and LO denotes LoadOpen pair. The probe crosstalk results are shown in Figure 7. First, the probe crosstalk values for the coaxial frequency range are evaluated. Here, only Keysight could provide measurement data, including crosstalk information. Here, superior crosstalk performance is visible when comparing results acquired at 10 GHz with those collected at 110 GHz, an intuitive attribute. The more interesting attribute is the difference between the crosstalk performance for structures with an isolated ground shown in markers and those devices with a common ground depicted with dotted lines in Figure 7. Up to 110 GHz, devices with isolated ground provide better crosstalk performance. In the WR5 frequency range, three participants submitted crosstalk data. Here, two participants (PTB and NPL) used GGB (by GGB Industries Inc.) probes, and Keysight used DMPI (by Dominion MicroProbes Inc.) probes. It is evident from Figure 7 that the DMPI probes show better crosstalk performance as compared with GGB probes. In the WR5 frequency band, the GGB probes exhibit crosstalk degradation proportional to the frequency, a finding confirmed by data provided by NPL and PTB. The WR3 frequency range shows comparable crosstalk performance for the DMPI and GGB probes, comparable to the values noted in the WR2 band. Also, no significant difference in crosstalk performance is evident for the devices with and without a shared ground. The crosstalk data for the WR2, WR1.5, and WR1 bands is acquired using T-Wave probes (by FormFactor Inc.), collectively submitted by three participants (Lille, Chalmers, and FormFactor). The WR2 band shows slightly larger crosstalk values as compared to those acquired for the WR1.5 band. It is interesting to see notable 5-10 dB lower crosstalk values in the WR1.5 band when using an absorber, and there is no clear reason for this, as similar probes are also used in the other frequency bands. Also, devices with common ground show a better crosstalk performance in the WR2 band, while no notable difference is evident for the WR1.5 and WR1 bands. The crosstalk measurement results acquired up to 1.1 THz, covering six frequency bands, do not show abrupt shifts when transitioning between the different frequency bands. It is evident from Figure 7 that crosstalk performance does not show noteworthy degradation proportional to the frequency, a key attribute of this study. Also, no significant difference in crosstalk performance is noted for devices with and without a common ground. Finally, the crosstalk values shown in Figure 7 do not indicate the crosstalk performance seen in typical commercial nanoscale technologies, i.e., CMOS and SiGe BiCMOS. The limitations in the available area in such technologies will most likely not allow usage of probe distances as in this interlaboratory comparison. B. Results for the coaxial frequency range Figure 6(a),(c),(e), and (g) show the measurement and reference values for the kit-1 comparison devices. Here, two participants, Keysight and FBH, provided their measurement results acquired with test beds with frequency extenders as the most notable difference. When evaluating the reflection coefficient parameters for the matched load and open device shown in Figure 8, a good agreement is observed as the differences between IEEE TRANSACTIONS ON TERAHERTZ SCIENCE AND TECHNOLOGY, VOL. 15, NO. 3, MAY 2025 16 Jan Stake (S’95–M’00–SM’06) was born in Uddevalla, Sweden, in 1971. He received an M.Sc. in electrical engineering and a Ph.D. in microwave electronics from Chalmers University of Technology in Gothenburg, Sweden, in 1994 and 1999, respectively. In 1997, he was a Research Assistant at the University of Virginia, Charlottesville, VA, USA. From 1999 to 2001, he was a Research Fellow with the Millimetre Wave Group at the Rutherford Appleton Laboratory, Didcot, UK. He then joined Saab Combitech Systems AB, Gothenburg, Sweden, as a Senior RF/microwave Engineer until 2003. From 2000 to 2006, he held different academic positions with the Chalmers University of Technology, and from 2003 to 2006, he was also the Head of the Nanofabrication Laboratory, Department of Microtechnology and Nanoscience (MC2). In 2006, he was appointed Professor and the Head of the Terahertz and Millimetre Wave Laboratory at the Chalmers University of Technology. He was a Visiting Professor with the Submillimeter Wave Advanced Technology (SWAT) Group at Caltech/JPL, Pasadena, CA, USA, in 2007 and at TU Delft, the Netherlands, in 2020. He received an appointment as a visiting research fellow at the National Physical Laboratory, UK, in the year 2023. He is also the co-founder of Wasa Millimeter Wave AB, Gothenburg, Sweden. His research interests include high-frequency semiconductor devices, terahertz electronics, submillimeter wave measurement techniques, and terahertz systems. Prof. Stake served as the Editor-in-Chief for the IEEE Transactions on Terahertz Science and Technology between 2016 and 2018 and as Topical Editor between 2012 and 2015. From 2019 to 2021, he was chairperson of the IEEE THz Science and Technology Best Paper Award committee. He is an elected member of the International Society of Infrared, Millimeter and Terahertz Waves (IRMMW-THz) board. Robin Schmidt graduated in 2018 as an Electrical Engineer from l’École Supérieur d’Électricité (Supélec). Now employed as a researcher by Keysight Technology Belgium and pursuing the Ph.D. at University of Leuven (KU Leuven), his focus is the development of calibration for onwafer S-parameters measurements in mm and sub-mm-wave frequency. Gavin Fisher is applications specialist for the Centre of Expertise at FormFactor working on RF application layers for the systems group providing application support and technical services to FormFactor’s customers as well as evaluating and developing new product. With two decades of experience working with Formfactor / Cascade probe systems and probes and a broad spectrum of applications such as high-frequency measurement, calibration and power device measurement, he educates and trains customers on best practices to achieve accurate measurement results. He has made several presentations at European Microwave Week, International Microwave Symposium, MOS-AK workshops, and Agilent/Keysight seminars. He holds an upper-second degree from Brunel University in Mechanical Engineering with electronic systems. Nick M. Ridler received the B.Sc. degree from King’s College London, U.K., in 1981. He is currently Head of Science for Electromagnetic and Electrochemical Technologies at the UK’s National Physical Laboratory (NPL), U.K. He is also an Honorary Professor at the Universities of Glasgow and Liverpool, U.K., and a Visiting Professor at the Universities of Kent, Leeds, and Strathclyde, U.K. He is Non-Executive Director of LA Techniques Ltd and a member of the Board of Directors of the European Microwave Association (EuMA). He has more than 40 years’ experience working in industrial, government and academic research establishments. His current interests include making accurate electromagnetic measurements at high frequencies (i.e., from approximately 1 kHz to 1 THz). He is an NPL Fellow, and Fellow of the Institute of Electrical and Electronics Engineers (IEEE), the Institution of Engineering and Technology (IET) and the Institute of Physics (IOP). He was General Chair of the 2021 European Microwave Week. Xiaobang Shang received the B.Eng. degree (First Class) in electronics and communication engineering in 2008 and the Ph.D. degree in microwave engineering in 2011 from the University of Birmingham, U.K. He became a Senior Scientist at the National Physical Laboratory (NPL), Teddington, U.K., in 2017 and was promoted to Principal Scientist in 2023. He has authored or coauthored more than 120 scientific articles on microwave measurements and microwave circuits. Dr. Shang is a member of IEEE MTT-S Technical Committee (TC)-3 and TC-21 and was an Associate Editor for the IEEE Microwave and Wireless Technology Letters (2020-2024). He was a recipient of several prestigious awards including the IEEE MTT-S Outstanding Young Engineer Award in 2025, the Roberto Sorrentino Prize in 2022, the IEEE Tatsuo Itoh Award in 2017, and the ARFTG Microwave Measurement Student Fellowship Award in 2009.