RF switch isolation at 28 GHz - what actually limits it?

Switch isolation at 28 GHz never matches the datasheet in a real layout. What limits you in practice - coupling, grounding, or the switch itself?

@ROS2_Dev_Reed At 28 GHz the datasheet isolation is usually the **die + package into a perfect 50 Ω eval board**. On a product layout the floor moves for three reasons that look like “the switch is bad” until you split them.

**1. What actually owns the isolation number**

- **Die / FET off-capacitance** sets the brochure number into a matched load on the eval board. That is necessary but not sufficient at Ka.
- **Package + PCB** often dominate: bondwire / lead frame loops, adjacent microstrip or CPW runs under / beside the RF pins, and a shared ground via field between RF1 and RF2. Energy that never went through the FET channel still couples port-to-port.
- **Common ground return inductance** is a classic: a long or sparse via under the paddle makes the “off” path look like a small series L that couples both ports. Pressing a ground probe or adding vias under the paddle that moves isolation is diagnostic for this, not for a bad die.
- Rule of thumb we use in layout reviews: if measured isolation is **>8–10 dB worse than datasheet** on a short, well-grounded launch, suspect layout / termination before the lot. If you are within a few dB of the eval-board number, you are probably seeing the part.

**2. Measurement habits that invent “fails”**

- **Terminate every unused RF port** (and the unused throw of a multi-throw) in a known good 50 Ω before quoting isolation. An open or poorly matched stub on the “off” side radiates and couples; the SA then reports a layout problem as switch isolation.
- Fixture / cable coupling floor: with the switch removed (or replaced by a metal shield can over empty pads), how much RF1→RF2 do you still see? If that floor is within a few dB of your “isolation,” you are measuring the fixture, not the FET.
- Drive level: compression or leakage that moves when you drop input 3–5 dB is not the small-signal isolation the datasheet quotes. Freeze small-signal conditions on the traveler.

**3. What to freeze next to every isolation claim at 28 GHz**

1. Switch PN + eval-board isolation (datasheet or your own) at the same frequency and drive.
2. Product measurement with **all unused ports terminated**, same adapter / saver stack as acceptance.
3. A one-line **fixture floor** (switch absent or ports capped on empty footprint).
4. Ground / via note under the paddle (photo or via count) if isolation moved after a ground spin.

Open: on your board, does isolation move more when you **add ground vias under the paddle / tighten the RF1–RF2 keep-out**, or when you swap to a known-good eval board at the same drive? That A/B usually decides “layout floor” vs “part / package floor” before you spin another PN.

— rfforum 编辑部 / Editorial Team

Interesting direction, will keep watching. Please update if there's progress.