The Field Always Wins

You made a schematic. You completed a design. You saw 28 dB isolation in simulation. The work has just begun.

A schematic is, by definition, an abstraction. You assumed the trace is a perfect wire. The substrate is a perfect insulator. The signal goes exactly where you intended.

But physics has its own nature. Its own constraints. Its own boundaries.

First Principle: Boundary Conditions

Design does not mean just topology. Design means: What boundaries did you create for the electromagnetic fields?

In RF, there is no such thing as a wire. There is only the field.

And the field does not care what you wanted in your schematic. The field cares about:

• What is the geometry?
• What is the metal thickness?
• What is the substrate’s permittivity?
• Is that permittivity a tensor or a scalar?
• Is the metal bulk or thin film?

These are boundary conditions. Your simulator assumes them. Physics eventually reveals incorrect assumptions.

Second Principle: Anisotropy Is Not Chaos — It Is Structure

You have heard of substrates like sapphire, alumina, quartz. You think, “OK, epsilon value is 9.8. Everything is set.”

That assumption is incomplete.

Because in crystalline substrates, epsilon is not a single number. It is a tensor. Meaning: the speed of signal propagation depends on which direction you route your trace.

Understand it like this:

If you make a square pad, the fields inside that pad will propagate in two different directions—at two different speeds. Those two speeds will create two different resonances. You cannot fix this with a single inductor or capacitor. Why? Because the problem is tensor anisotropy, not a lumped parasitic.

Solution: Round pads. Octagonal pads. And if possible, use an air bridge—lift the metal off the substrate. Air has epsilon = 1. That will eliminate your problem.

Third Principle: Kinetic Inductance — When Electrons Get Crowded

Inductance means magnetic field.

That is incomplete. There is another kind of inductance. Kinetic inductance.

Understand it like this: At 3.5 GHz, the skin depth in gold is approximately 1.2 microns. You made your metal thickness 1 micron. Now, where does the current distribute? Across the entire cross-section? No. Only on the surface. But the thickness is so small that electrons become crowded. They jostle. Their inertia stores energy. This creates extra inductance—but only at high frequency. At DC, you will not see it.

This is why your impedance measures 50 ohms on a DC meter, but 52 ohms at 3.5 GHz. Your simulator assumed a perfect conductor. That assumption was incorrect.

Rule of thumb: Keep metal thickness ≥ 2.5 microns (two skin depths) at 3.5 GHz. If you do not, you are not building a coupler. You may accidentally optimize for heat instead of signal.

Fourth Principle: Geometry Becomes Behavior

You put a 90° corner in your design. At 100 MHz, it causes no problem.At 3.5 GHz, that same corner begins behaving like an unintended reactive structure. Why? Because current crowds at the inside corner. Charge accumulates at the outside corner. A resonance forms.

The central junction of a Wilkinson divider—where three ports meet in one small area—is the worst offender. Why? Because fields propagate into both air and substrate simultaneously. Different permittivities. Different speeds. Different resonances.

The simulation shows: 28 dB isolation.
The measurement shows: 14 dB isolation.

The topology did not change. Your design was not wrong. You simply did not model the boundaries.

Solution: Minimize the physical area of any central junction. Use tapered junctions (length ≥ 1.5× line width). Use rounded Y-branches with 120° spacing. Mitered bends are less effective than you think in thin film—because metal thickness is negligible. Your focus should be on field shaping, not corner cutting.

Fifth Principle: Your Simulator Is an Assumption

Your simulator assumes:

• Bulk conductivity (ignoring grain boundary scattering)
• Perfectly smooth surfaces
• Ideal vertical sidewalls
• Isotropic substrate
• Kinetic inductance = 0

Your actual thin-film stack has:

• Sputtered films with grain structure
• Etch bias and sidewall angle (typically 10–20° from vertical)
• Possible substrate anisotropy
• Real surface roughness
• Metal thickness possibly below 2× skin depth

That gap—28 dB simulated versus 14 dB measured—is not noise. It is a structural gap.

Sixth Principle: Prototype Success Does Not Equal Production Reality
The prototype:

• Built slowly
• Handled by senior engineers
• Tuned manually
• Measured under ideal conditions

This can easily show you 28 dB isolation.

But production introduces:

• Process variation
• Thermal drift
• Operator differences
• Material lot changes
• Plating thickness fluctuations
• Dielectric inconsistency
• Packaging stress

The real issue: You did not estimate the physics gap between prototype and production.

Final Principle: The Field Always Wins

This is the ultimate point.

If you did not respect the boundaries:

• Return path geometry
• Current crowding
• Anisotropy
• Kinetic inductance
• Junction parasitics

The field responds only to the boundaries present.

You are not designing traces. You are shaping fields. And manufacturing changes those fields.

Ask different questions:

• Where are your boundaries?
• Which path does the return current take?
• Where does the field concentrate?
• What is the physical area of your junction?
• How does your metal thickness compare to skin depth?
• Is your substrate anisotropic or isotropic?

The object is not the circuit. The field is the circuit.

Physics is patient. It eventually reveals every shortcut.