When is co-simulation necessary versus circuit-only simulation?

Co-simulation becomes necessary when the operating frequency is high enough that physical dimensions approach a significant fraction of the wavelength, making lumped-element models inaccurate. Below 1 GHz, circuit-only simulation with well-calibrated lumped models is usually adequate. At 1 to 10 GHz, co-simulation is recommended for critical paths (matching networks, package parasitics, interconnects longer than lambda/20). Above 10 GHz, co-simulation is essential for virtually all design elements. Specific scenarios requiring co-simulation include on-chip inductor design (substrate coupling and eddy currents degrade Q-factor in ways that simple lumped models miss), BGA package modeling (hundreds of solder balls with mutual coupling create resonances above 5 GHz), antenna-PA integration (load-pull effects and antenna impedance variation with scan angle), and filter-amplifier co-design (where the filter's port impedance varies with frequency in ways that affect amplifier stability).

What are the computational costs and limitations?

EM simulation time scales as O(N^2) to O(N^3) with the number of mesh elements, which grows with electrical size and geometric complexity. A simple microstrip line at 5 GHz (1000 mesh elements) solves in seconds, while a complete QFN package with bond wires at 28 GHz (500,000 elements) takes 4 to 8 hours on a 64-core workstation. Co-simulation adds the circuit simulation time (typically seconds to minutes for harmonic balance) plus data transfer overhead. The main limitation is the interface between EM and circuit domains: S-parameters are linear, so they cannot capture nonlinear EM effects (ferrite saturation, plasma breakdown, high-power multipaction). For these cases, time-domain co-simulation with tight coupling (the EM and circuit solvers exchange data at every time step) is needed but extremely expensive computationally. Practical workarounds include macro-modeling (fitting the EM response to a compact circuit model), adaptive frequency sampling (solving EM only at key frequencies), and model order reduction (reducing the EM model from millions to thousands of degrees of freedom).

Electronic Design Automation

Co-Simulation

Q: How does EM-circuit co-simulation work?

The workflow involves three steps. First, the passive RF structure (package, transmission line, inductor, antenna feed network) is modeled in a 3D EM solver with ports defined at the connection points to active circuits. The EM solver computes the multiport S-parameter matrix across the frequency range of interest using FEM (finite element method), FDTD (finite-difference time-domain), or MoM (method of moments). Second, the S-parameter data (Touchstone .snp file) or an equivalent circuit model (fitted rational function or lumped element network) is imported into the circuit simulator as a black-box n-port network. Third, the circuit simulator connects the n-port to transistor-level SPICE models and runs harmonic balance (for steady-state nonlinear analysis), transient (for time-domain behavior), or envelope simulation (for modulated signals). The key advantage is that the EM solver captures distributed effects, coupling, and radiation that lumped models miss, while the circuit simulator handles nonlinear active device behavior that EM solvers cannot.

/koh-sim-yoo-lay-shun/

Co-simulation couples EM field solvers (Ansys HFSS, CST, Keysight EMPro) with circuit simulators (ADS, Cadence Spectre, AWR) for accurate multi-domain RF design. The EM solver computes S-parameters for passive structures; the circuit simulator connects these to active device SPICE models. Captures parasitic coupling, substrate effects, and radiation that lumped models miss, reducing design iterations 30 to 50% and improving first-pass success from 60% to 85%.

Category: Electronic Design Automation

Iteration reduction: 30 to 50%

First-pass success: 60% → 85%

Understanding Co-Simulation

RF design exists at the boundary between two physical domains. Active devices (transistors, diodes) are described by compact SPICE models with lumped parameters (g_m, C_gs, r_ds), while passive structures (transmission lines, inductors, package interconnects, antennas) exhibit distributed electromagnetic behavior governed by Maxwell's equations. At low frequencies where all dimensions are much smaller than a wavelength, lumped circuit models approximate the distributed behavior adequately. But as frequencies increase, this approximation breaks down: a 5 mm bond wire is electrically invisible at 100 MHz (λ/600) but becomes a significant inductor at 5 GHz (λ/12) and a resonant element at 30 GHz (λ/2).

Co-simulation bridges these domains by letting each solver handle its strengths. The EM solver uses finite elements or finite differences to solve Maxwell's equations in the 3D geometry, capturing distributed transmission line effects, electromagnetic coupling between adjacent traces, substrate currents and eddy effects, and radiation from open structures. The result is a frequency-dependent S-parameter matrix that accurately represents the passive structure's behavior at all ports. The circuit simulator then uses these S-parameters as a linear n-port network alongside nonlinear device models, applying harmonic balance, transient, or envelope analysis to predict the complete system's performance under realistic operating conditions.

Co-Simulation Interface Equations

S-Parameter Port Interface:
b_i = ∑_j S_ij(f) × a_j (frequency domain, linear)

Rational Function Fit (macro-model):
S(s) = D + ∑_k R_k/(s - p_k) (poles p_k, residues R_k)

Mesh Density Rule:
Δx ≤ λ_min/10 ; elements ≈ (L/Δx)³

Where a_j, b_i = incident/reflected power waves, s = jω = complex frequency, λ_min = wavelength at maximum frequency. A 10 mm structure at 60 GHz (λ = 5 mm): Δx = 0.5 mm, ~8,000 elements. At 300 GHz: ≈1M elements.

Co-Simulation Tool Combinations

EM Solver	Method	Circuit Simulator	Interface	Use Case
Ansys HFSS	FEM (3D)	Keysight ADS	S-parameter / Dynamic Link	MMIC, package, antenna-PA
CST Studio	FDTD / FIT	CST Design Studio	Built-in co-simulation	SI/PI, EMC, antenna
Keysight EMPro	FEM / FDTD	Keysight ADS	Native integration	Package, board-level
Cadence EMX	MoM (2.5D)	Cadence Spectre	CDF / Virtuoso	On-chip inductor, MIM cap
Sonnet	MoM (planar)	NI AWR / ADS	Touchstone export	Microstrip, filter, coupler

Common Questions

Frequently Asked Questions

How does EM-circuit co-simulation work?

3D EM solver computes multiport S-parameters (FEM/FDTD/MoM) with ports at active circuit connections. S-parameter data (.snp Touchstone) or fitted rational function is imported into circuit simulator as n-port network. Circuit simulator connects it to transistor SPICE models and runs harmonic balance, transient, or envelope analysis. EM captures distributed effects; circuit handles nonlinearity.

When is co-simulation necessary?

Below 1 GHz: rarely needed (lumped models adequate). 1 to 10 GHz: recommended for critical paths (matching, packages, interconnects > λ/20). Above 10 GHz: essential for virtually all elements. Always needed for: on-chip inductors, BGA packages, antenna-PA integration, and filter-amplifier co-design where port impedance variation affects stability.

What are the computational costs?

EM scales O(N²) to O(N³) with mesh elements. Simple microstrip: seconds. QFN package at 28 GHz (500K elements): 4 to 8 hours on 64 cores. Main limitation: S-parameters are linear, so nonlinear EM effects (ferrite saturation, multipaction) need time-domain tight coupling. Workarounds: macro-modeling, adaptive sampling, model order reduction.

Related Terms

← Co Pol Pattern Co Simulation EMC →