Researchers have developed a discontinuity-aware divided electromagnetic analysis method that significantly improves the simulation accuracy of high-frequency differential circuits by accounting for structural discontinuities at component interfaces. In high-frequency integrated circuit engineering, sub-terahertz frequencies such as the D-band spanning 110 to 170 GHz offer wide bandwidths that are essential for realizing ultra-high-speed wireless communications exceeding 100 Gb/s. However, interconnecting differential circuits in these systems requires power-matching networks, where compact transformer-based circuits are commonly deployed. Because transformer layouts omit ground planes while adjacent active devices such as common-source MOSFETs require ground planes connected via ground walls, sharp physical discontinuities inevitably emerge at their connection interfaces. These structural transitions introduce severe boundary mismatches, causing unintended electromagnetic field distortions and reflections that conventional modeling workflows fail to capture adequately.
To manage prohibitive computational burdens in practical design, engineers routinely adopt divided electromagnetic analysis, where individual circuit blocks are simulated separately and subsequently combined into a full-system response. While dividing large layouts substantially decreases simulation memory and runtimes, traditional divided approaches cannot directly evaluate boundary interactions at connection interfaces. Conventional de-embedding techniques extend only the uniform transmission structure identical to the calibration reference plane, inherently overlooking the abrupt transitions that occur when grounded and ungrounded blocks meet. Consequently, standard divided workflows fail to reflect interface discontinuities, leading to pronounced discrepancies in both even- and odd-mode circuit responses when compared against full-structure electromagnetic simulations.
Addressing this computational challenge, the research team established a divided framework that directly integrates the actual connected geometry into the de-embedding procedure. Rather than assuming idealized uniform boundaries, the proposed formulation extends the section of the circuit that is physically connected across the entire layout and performs four-port de-embedding directly on that extended region. By embedding the actual interface geometry into the extracted error network, the electromagnetic influence of the grounded-to-ungrounded boundary is preserved within the resulting characteristics. The approach utilizes mixed-mode transformation, symmetric thru-short-delay calibration, characteristic impedance estimation, and renormalization, ensuring that interface discontinuity effects are simultaneously evaluated in both differential and common modes while retaining the computational benefits of divided modeling.
Constructing the underlying four-port de-embedding network required eliminating parasitic effects introduced by lumped excitation ports and localized field disturbances. The researchers implemented a symmetric thru-short-delay calibration sequence based on differential thru and delay test structures simulated with the Cadence EMX Planar 3D Solver. Because the differential thru standard is geometrically symmetric with respect to its reference plane, identical error networks were assigned to both sides, enabling error extraction without requiring an auxiliary short standard. The modal decomposition treated mixed-mode scattering parameters by decomposing them into separate even- and odd-mode blocks. Evaluation across 1 to 300 GHz confirmed that mode conversion between differential and common modes remained below -70 dB for differential coupled lines with and without discontinuities, confirming that off-diagonal coupling could be safely neglected during the initial calibration stage.
The physical stability of the de-embedding standard also depended on maintaining a uniform ground reference potential close to zero volts near the simulation ports. Surface currents tend to spread widely over ground walls at lower frequencies due to skin-depth dynamics, which can distort the reference potential and disturb modal field patterns. By simulating surface current density distributions at 10 and 110 GHz across ground-wall widths of 3, 15, and 30 micrometers, the investigators determined that a 30-micrometer wall suppressed edge surface current densities below 0.001 A/m, providing a sufficiently stable ground reference. Furthermore, parametric sweeps evaluating de-embedding network lengths identified 30 micrometers as optimal for minimizing odd-mode discrepancies. In benchmark coupled-line tests across the D-band, applying the baseline four-port network reduced reflection errors from 4.3 to 0.085 dB for the even-mode parameter See11 and from 5.0 to 0.07 dB for the odd-mode parameter Soo11.
The researchers subsequently verified the discontinuity-aware methodology on a representative differential coupled line incorporating a grounded-to-ungrounded interface. Performance was compared directly against full-structure electromagnetic simulations and conventional de-embedding that only extended reference-plane structures. Across the D-band, the proposed approach reduced even-mode reflection errors from 0.95 to 0.24 dB for See11 and from 2.0 to 0.21 dB for See22 relative to the full electromagnetic benchmark. Similarly, odd-mode reflection errors decreased from 1.6 to 0.15 dB for Soo11 and from 1.8 to 0.15 dB for Soo22. The discontinuity-aware technique reproduced characteristic low-frequency peaks and the subtle curvature near 70 GHz observed in full-wave analysis, validating that extending the actual connected structure captures localized reflections and boundary distortions.
In this circuit, common-source MOSFET stages with dedicated ground planes were interconnected using compact, ungrounded transformer-based matching networks. For the MOSFET layout, de-embedding networks were extracted by extending differential coupled lines without ground planes to match the transformer layout, directly embedding the grounded-to-ungrounded interface into the transistor block. For the transformer blocks, de-embedding was carried out using transmission lines possessing identical cross-sectional dimensions to eliminate lumped-port parasitics. Comparing the resulting circuit responses from 80 to 170 GHz demonstrated that the proposed method achieved superior agreement with full electromagnetic simulation compared to both uncalibrated divided analysis and conventional de-embedding.
At the amplifier’s peak-gain operating frequency of 106 GHz, the error in the odd-mode forward gain Soo21 fell from 0.20 dB under conventional de-embedding down to 0.01 dB with the proposed technique. Circuit-wide reflection characteristics exhibited substantial precision gains across both operational modes. Specifically, even-mode reflection errors See11 and See22 decreased from 0.95 to 0.15 dB and from 0.15 to 0.068 dB, respectively, while odd-mode reflection errors Soo11 and Soo22 were reduced from 0.51 to 0.042 dB and from 0.44 to 0.024 dB. These numerical outcomes confirmed that incorporating the actual connected boundary into the de-embedding process effectively bridges the discrepancy between full-structure electromagnetic modeling and modular divided circuit simulations.
Despite substantial gains across gain and reflection parameters, a residual discrepancy remained in the even-mode transmission characteristic See21. Because even-mode return currents flow partially through the underlying ground plane, the grounded-to-ungrounded discontinuity exerts a stronger influence on even-mode propagation. Detailed modal conversion analyses of individual circuit blocks revealed that the input and output transformer matching networks introduced noticeable mode conversion. The input transformer utilized a 2:1 turns ratio that created structural asymmetry in its crossover section, while the output transformer incorporated a parallel matching capacitor that disturbed high-frequency symmetry. Although the grounded-to-ungrounded boundary itself generated negligible mode conversion below -70 dB, the physical asymmetry inherent to the transformer components violated the assumption of ideal modal independence, leaving residual mode conversion after mixed-mode decomposition.
The study establishes that discontinuity-aware divided electromagnetic analysis provides an accurate and computationally practical framework for modeling complex differential millimeter-wave and sub-terahertz circuits. The authors noted that the formulation assumes small mixed-mode conversion within individual sub-blocks and a stable local ground potential around excitation ports, meaning that severe layout asymmetries can degrade modal extraction accuracy. Enhancing algorithmic robustness against pronounced component asymmetries represents an avenue for future development. Nevertheless, by capturing structural discontinuities directly within the de-embedding procedure, the framework offers a viable modeling foundation for large-scale D-band transceivers and advanced phased-array architectures where full-wave electromagnetic simulation of entire layouts is computationally intractable.
Subject of Research: Electromagnetic analysis method for differential circuits with structural discontinuities
Article Title: High-accuracy electromagnetic analysis method for differential circuits with discontinuities
Article References: Mochida, K., Tanaka, S., Yoshida, T., Amakawa, S., & Fujishima, M. (2026). High-accuracy electromagnetic analysis method for differential circuits with discontinuities. Journal on Wireless Communications and Networking, 2026(1), Article 125. https://doi.org/10.1186/s13638-026-02659-w
Image Credits: AI Generated
DOI: 10.1186/s13638-026-02659-w
Keywords: electromagnetic analysis, divided EM analysis, differential circuits, structural discontinuities, de-embedding, D-band, sub-terahertz, transformer matching, MOSFET, CMOS, mixed-mode S-parameters, TSD calibration, modal decomposition, differential amplifier, phased-array, ground plane, planar 3D solver
Cite Scienmag News
Denise Maddox. (November 20, 2026). Discontinuity-Aware Divided Electromagnetic Analysis Improves Differential Circuit Accuracy. Scienmag. https://scienmag.com/discontinuity-aware-divided-electromagnetic-analysis-improves-differential-circuit-accuracy/
Denise Maddox. "Discontinuity-Aware Divided Electromagnetic Analysis Improves Differential Circuit Accuracy." Scienmag, 20 November 2026, https://scienmag.com/discontinuity-aware-divided-electromagnetic-analysis-improves-differential-circuit-accuracy/. Accessed 20 November 2026.
Denise Maddox. "Discontinuity-Aware Divided Electromagnetic Analysis Improves Differential Circuit Accuracy." Scienmag. November 20, 2026. https://scienmag.com/discontinuity-aware-divided-electromagnetic-analysis-improves-differential-circuit-accuracy/








