Standards, Specifications, and Industry Practices Design Process and Best Practices Informational

What is the electromagnetic simulation workflow for a millimeter wave component or module?

Q: How long does a mmWave EM simulation take?

Simulation time depends on model complexity and solver: HFSS (FEM, frequency domain): 15 minutes to 2 hours per frequency point for a typical mmWave package (10×10 mm). A 200-point frequency sweep: 8-24 hours with adaptive interpolation (sweeping fewer points and interpolating). With 64 GB RAM and 16-core workstation. CST (time domain): 1-6 hours for a broadband solve (all frequencies in one run). Generally faster than HFSS for broadband problems. Similar hardware requirements. ADS Momentum (2.5D MoM): 5-30 minutes for a planar mmWave circuit (microstrip filter, coupled-line coupler). Much faster for planar structures but does not capture 3D effects (bond wires, via mode conversion). Tips for speed: use symmetry planes (halves the problem size), use surface impedance boundaries instead of meshing conductor thickness, and optimize the simulation box size (absorbing boundaries should be at lambda/4 from the structure).

Q: Do I need to include the PCB package in simulation?

At mmWave: yes. The package (QFN, BGA, fan-out wafer-level) contributes significant parasitics at 30+ GHz: bond wire or bump inductance (50-200 pH), package substrate transmission line loss (0.5-2 dB), cavity resonances within the package, and mold compound loading on antenna elements. Package co-simulation is essential: simulate the MMIC die separately, then the package separately, then combine using S-parameter co-simulation. Alternatively, simulate the complete die-in-package as a single 3D EM model (most accurate but most computationally expensive). For flip-chip assemblies: the bump array and under-fill dielectric must be modeled because they affect impedance and create parallel-plate modes between the die and the package substrate.

Q: What accuracy can I expect at 60 GHz?

With properly calibrated materials and mesh-converged simulation: S21 (insertion loss/gain): ±0.3-0.7 dB. S11 (return loss): ±2-5 dB (return loss is very sensitive to impedance mismatches at mmWave). Resonant frequency: ±0.5-1.5% (±300 MHz-900 MHz at 60 GHz). Radiation efficiency (antenna): ±5-10%. These numbers assume: (1) Measured substrate dielectric properties at the operating frequency. (2) Copper roughness modeled using Huray or equivalent. (3) Actual manufactured dimensions used in the model. (4) Proper meshing with convergence verification. Without measured material data: simulation-measurement discrepancy can be 2-5 dB, making the simulation unreliable for optimization.

The electromagnetic simulation workflow for a mmWave component (30-300 GHz) follows a structured process: (1) Geometry creation: import the 3D CAD model (from Solidworks, Inventor, or the PCB layout tool) into the EM solver. Simplify the geometry: remove features smaller than lambda/20 (at 60 GHz: lambda = 5 mm, so features < 0.25 mm may be simplified). Retain all features that affect the RF path: transmission line cross-sections, via arrays, bond wire profiles, and mold compound shapes. (2) Material assignment: apply frequency-dependent material properties. At mmWave, dielectric constant and loss tangent vary significantly with frequency. Use measured data (not datasheet values at 1 MHz). Surface roughness of copper traces must be modeled (Huray or Hammerstad models); at 60 GHz, surface roughness of 1 μm RMS adds 0.3-1.0 dB/cm of excess loss beyond smooth-conductor predictions. (3) Port definition: define wave ports or lumped ports at signal entry/exit planes. For differential signals: use balanced port definitions. For waveguide interfaces: use the appropriate waveguide port mode. De-embed ports to a consistent reference plane. (4) Meshing: apply adaptive mesh refinement with convergence criterion of delta-S < 0.01 (S-parameter change between mesh passes < 0.01). For mmWave: initial mesh seeding with 6-10 elements per wavelength in the dielectric. Critical regions (via barrels, bond wire landings, thin dielectric layers) need mesh refinement settings. (5) Solve: frequency sweep from DC to 2× the operating frequency (to capture harmonics and out-of-band behavior). Interpolating sweep (HFSS) or broadband solver (CST time-domain) for efficient frequency coverage. (6) Post-processing: extract S-parameters, radiation patterns, field distributions. Identify loss mechanisms and resonant modes.

Category: Standards, Specifications, and Industry Practices

Updated: April 2026

Product Tie-In: Design Tools, Test Equipment

Millimeter-Wave EM Simulation

Millimeter-wave simulation demands higher accuracy and more computational resources than lower-frequency work because the wavelengths are comparable to component dimensions, making every geometric detail electromagnetically significant.

Meshing Strategies

The mesh must resolve the smallest features and the shortest wavelength in the model. For a 60 GHz design on a 4-mil (100 μm) Rogers 3003 substrate: wavelength in the dielectric = c/(f*sqrt(epsilon_r)) = 5e9/(60e9*sqrt(3.0)) ≈ 2.9 mm. Mesh element size: 0.29-0.48 mm (lambda/6 to lambda/10). However: the copper thickness (18 μm) is much smaller than the mesh element size. Do not mesh the copper thickness explicitly; use the impedance boundary condition (surface impedance) for conductor loss. The substrate thickness (100 μm) requires at least 3-4 mesh elements through the thickness for accurate field representation. Via barrels (typically 100-200 μm diameter, 100-400 μm depth) need local mesh refinement with elements no larger than 50 μm. Bond wires (25 μm diameter, 200-500 μm length) need cylindrical mesh elements along their length. Total mesh size for a typical mmWave module (10×10 mm package with 20 signal vias, 5 bond wires, and an integrated antenna): 500,000-5,000,000 tetrahedral elements in HFSS, requiring 16-64 GB RAM and 30 minutes to 4 hours per frequency point on a modern workstation.

Material Modeling at mmWave

Material properties are critical at mmWave: (1) PCB substrate: Rogers 3003 (epsilon_r = 3.0, tan_delta = 0.0013 at 10 GHz). At 60 GHz: epsilon_r may decrease by 2-5% and tan_delta may increase by 50-100% due to molecular relaxation. Use the manufacturer's frequency-dependent data (Rogers publishes data to 40 GHz; extrapolate or measure beyond). (2) Copper roughness: standard PCB copper has RMS roughness of 1-5 μm (depending on treatment: RTF, VLP, HVLP). The Huray model uses a snowball model of copper grains to calculate excess loss: L_rough/L_smooth = 1 + (3/2) × (4*pi*A_sphere*N_sphere)/(pi*delta^2), where A_sphere and N_sphere are the snowball parameters and delta is the skin depth. At 60 GHz, delta = 0.27 μm for copper; a 2 μm RMS roughness increases conductor loss by approximately 30-50%. (3) Mold compound: epoxy mold compounds used in IC packages have epsilon_r = 3.5-4.5 and tan_delta = 0.01-0.03. These must be included in the simulation for packaged mmWave ICs. (4) Wire bond: model as a cylindrical conductor with exact 3D profile (height, length, shape). Bond wire inductance: 0.7-1.0 nH/mm at DC, increasing at mmWave due to proximity effects.

Validation and Correlation

mmWave simulation results must be validated against measurement: (1) Fabricate test structures on the same substrate/process as the product: TRL calibration kit for de-embedded S-parameter measurement, thru lines for loss characterization, resonators for material property extraction. (2) Measure using calibrated probe station (GGB Industries, FormFactor) with LRRM calibration for on-wafer probing, or test fixtures with precision connectors (1.85 mm or 1.0 mm for 60+ GHz). (3) Correlation criteria: S21 within ±0.5 dB by 60 GHz, S11 within ±3 dB, frequency of resonances within ±1%. Larger discrepancies indicate: incorrect material data, missing geometric features, or measurement de-embedding errors. (4) Iterate: update material properties in the model based on test structure measurements, re-simulate the product design with corrected properties, and use this calibrated model for optimization.

Common Questions

Frequently Asked Questions

How long does a mmWave EM simulation take?

Simulation time depends on model complexity and solver: HFSS (FEM, frequency domain): 15 minutes to 2 hours per frequency point for a typical mmWave package (10×10 mm). A 200-point frequency sweep: 8-24 hours with adaptive interpolation (sweeping fewer points and interpolating). With 64 GB RAM and 16-core workstation. CST (time domain): 1-6 hours for a broadband solve (all frequencies in one run). Generally faster than HFSS for broadband problems. Similar hardware requirements. ADS Momentum (2.5D MoM): 5-30 minutes for a planar mmWave circuit (microstrip filter, coupled-line coupler). Much faster for planar structures but does not capture 3D effects (bond wires, via mode conversion). Tips for speed: use symmetry planes (halves the problem size), use surface impedance boundaries instead of meshing conductor thickness, and optimize the simulation box size (absorbing boundaries should be at lambda/4 from the structure).

Do I need to include the PCB package in simulation?

At mmWave: yes. The package (QFN, BGA, fan-out wafer-level) contributes significant parasitics at 30+ GHz: bond wire or bump inductance (50-200 pH), package substrate transmission line loss (0.5-2 dB), cavity resonances within the package, and mold compound loading on antenna elements. Package co-simulation is essential: simulate the MMIC die separately, then the package separately, then combine using S-parameter co-simulation. Alternatively, simulate the complete die-in-package as a single 3D EM model (most accurate but most computationally expensive). For flip-chip assemblies: the bump array and under-fill dielectric must be modeled because they affect impedance and create parallel-plate modes between the die and the package substrate.

What accuracy can I expect at 60 GHz?

With properly calibrated materials and mesh-converged simulation: S21 (insertion loss/gain): ±0.3-0.7 dB. S11 (return loss): ±2-5 dB (return loss is very sensitive to impedance mismatches at mmWave). Resonant frequency: ±0.5-1.5% (±300 MHz-900 MHz at 60 GHz). Radiation efficiency (antenna): ±5-10%. These numbers assume: (1) Measured substrate dielectric properties at the operating frequency. (2) Copper roughness modeled using Huray or equivalent. (3) Actual manufactured dimensions used in the model. (4) Proper meshing with convergence verification. Without measured material data: simulation-measurement discrepancy can be 2-5 dB, making the simulation unreliable for optimization.

Keep Reading