Three-dimensional integrated circuits use through-silicon vias (TSVs) to connect active layers vertically, shortening interconnect paths between stacked chips. For engineers evaluating these connections, selecting the right TSV equivalent circuit model is essential: the electrical behavior of a via changes with frequency, and a model suited to one operating region may not accurately represent another. In particular, signal-ground TSVs can operate in a slow-wave region where a conventional quasi-TEM model may give less accurate results.
A slow-wave circuit model provides a way to represent this low-frequency behavior using lumped electrical elements. The distinction matters when estimating transmission characteristics and signal integrity in 3D IC interconnects. The model discussed here was compared with electromagnetic simulation, illustrating why the assumed propagation mode should match the frequency range being analyzed.
Why TSV behavior depends on frequency
A signal-ground through-silicon via (GS-TSV) consists of two TSVs. Each via has a central copper cylinder surrounded by a silicon dioxide liner, which electrically separates the copper from the silicon substrate. The substrate and liner affect how electric and magnetic fields propagate around the conductors.
As operating frequency changes, the silicon’s electrical response changes as well. The analysis divides the frequency range into three regions: slow-wave, dielectric quasi-transverse electromagnetic (quasi-TEM), and skin-effect modes. These are not interchangeable descriptions. Using a model for the wrong region can lead to inaccurate transmission results and, in turn, misleading design guidance.
What defines the slow-wave mode?
In the modeled case, slow-wave operation occurs below the silicon relaxation frequency, given as 15 GHz. In this region, interfacial polarization at the silicon dioxide–silicon boundary can create a thin space-charge layer. This Maxwell–Wagner polarization effect makes the interface behave like a lossy ground plane.
The resulting signal has a very slow phase velocity compared with propagation in the quasi-TEM region. That difference affects the electrical representation needed for circuit analysis. A model that assumes the silicon allows electric and magnetic fields to penetrate as they would in a dielectric may therefore miss behavior associated with the interface and substrate at lower frequencies.
Above the relaxation frequency, and below the characteristic frequency for silicon skin effect, the silicon behaves more like a dielectric. Electric and magnetic fields can penetrate it, making the dielectric quasi-TEM model more appropriate. At still higher frequencies, the product of silicon conductivity and frequency becomes sufficiently large that penetration depth decreases and skin-effect behavior emerges. The boundaries depend on the relevant silicon properties and structure dimensions.
Building a slow-wave GS-TSV equivalent circuit
The proposed circuit represents the GS-TSV using electrical parameters derived from microwave theory. Its elements account for the via’s geometry and the electrical properties of the copper, oxide liner, and silicon substrate. The model includes:
- Equivalent inductance: related to the via height, copper radius, and spacing between the TSVs.
- Oxide capacitance: associated with the silicon dioxide liner, its dielectric constant, and the via dimensions.
- Silicon substrate conductance: representing conduction through the silicon between the TSVs.
- Frequency-dependent resistance: capturing resistance behavior that varies with frequency and the silicon skin-effect characteristic frequency.
Together, these elements provide a circuit-level representation of the physical interconnect. The equations in the model link the parameters to dimensions and material properties, so changing the via geometry or substrate characteristics changes the circuit representation. This is useful for system-level analysis, where engineers need a circuit model rather than a full three-dimensional field solution for every evaluation.
Validating the model against electromagnetic simulation
The circuit was implemented in Keysight ADS, and its results were compared with simulations in ANSYS HFSS. The comparison used insertion loss, an S-parameter measure of signal transmission through the structure. In the slow-wave region, the ADS results from the slow-wave equivalent circuit agreed with the HFSS simulation data better than results from the quasi-TEM equivalent circuit.
The relative accuracy changed as frequency increased. The quasi-TEM circuit became more accurate in the higher-frequency region, while the slow-wave circuit became less accurate as the structure moved toward quasi-TEM operation. This is an important qualification: the slow-wave model is not presented as a universal replacement for other TSV models. Its value is in representing the frequency region for which it was developed.
Practical implications for interconnect analysis
For engineers and system integrators working on 3D ICs, the first modeling decision should be to identify the operating region, not simply select a familiar circuit template. Before interpreting insertion-loss or other transmission results, consider whether the silicon is expected to exhibit slow-wave, quasi-TEM, or skin-effect behavior.
A practical workflow is to:
- Establish the frequency range. Compare operating frequencies with the silicon relaxation and skin-effect characteristic frequencies.
- Account for the actual structure. Include relevant via dimensions and silicon properties when determining the applicable model.
- Use a mode-appropriate circuit. Apply the slow-wave representation below the relaxation frequency and a quasi-TEM representation in the region where its assumptions are suitable.
- Check circuit results against field simulation where appropriate. The ADS-to-HFSS comparison demonstrates how electromagnetic simulation can assess whether a circuit model represents the intended operating region.
Accurate TSV modeling supports more reliable evaluation of vertical interconnects in stacked-chip designs. The central takeaway is straightforward: frequency determines which physical mode dominates, and that mode should determine the equivalent circuit. For GS-TSVs in the slow-wave region, a dedicated model can improve agreement with electromagnetic simulation compared with a quasi-TEM circuit. As frequency rises and behavior changes, engineers should transition to a model suited to the new region rather than extrapolating the slow-wave circuit beyond its useful range.