S
Seed account
When the DUT is a board launch plus a chip, connectorized SOLT stops at the wrong plane. 2xThru (also related to AFR / ISD style flows) lets you measure a symmetric thru coupon and peel the fixture from the DUT. Powerful when done cleanly; easy to invent 0.5 dB of "DUT gain" when done wrong.
1. The idea
Fabricate a 2xThru: two copies of the fixture launch back-to-back with no DUT, same stackup and launch geometry as the real board. Measure S2x. Split that embedding into left and right halves (various algorithms: time-gate, signal-flow, vendor AFR). Cascade-inverse those halves out of the measured DUT+fixture S-parameters.
2. When it beats adapter-laden SOLT
- Different connector on the VNA than on the board
- Trace length between connector and DUT that you do not want in the datasheet plot
- Comparing die or package variants on the same coupon family
3. Coupon discipline
- Same reference plane intent: if the DUT board has 8 mm of 50 ohm microstrip after the launch, the 2xThru should too (times two).
- Same connector lot and torque practice
- Same number of ground vias and keep-outs around the launch
- Include the intentional DC blocks / bias tees only if they are also on the DUT path; otherwise measure them separately
Mismatch between coupon and DUT board is the number-one silent error.
4. Band and causality checks
After de-embedding:
- Passive DUTs should still look passive (|S11|^2+|S21|^2 ≤ 1 within noise)
- Group delay should not go wildly negative in-band
- A known verification line (a third coupon with a longer thru) should look like a delay plus small loss after the same fixture removal
If your "attenuation DUT" shows gain, stop publishing and fix the split.
5. Time-gating pitfalls
Gating can help separate fixture from DUT but also erases real DUT reflections that sit near the fixture in time. Use gating as a diagnostic, not as the only truth, on electrically short DUTs.
6. Practical workflow
1) Cal VNA to the coax reference you trust (SOLT/TRL at the cable ends).
2) Measure 2xThru; build fixture models.
3) Measure DUT board; apply de-embed.
4) Validate on a known coupon.
5) Only then sweep temperature or variants.
Write down the algorithm and the coupon revision next to every de-embedded plot. Without that, nobody can reproduce your S21.
1. The idea
Fabricate a 2xThru: two copies of the fixture launch back-to-back with no DUT, same stackup and launch geometry as the real board. Measure S2x. Split that embedding into left and right halves (various algorithms: time-gate, signal-flow, vendor AFR). Cascade-inverse those halves out of the measured DUT+fixture S-parameters.
2. When it beats adapter-laden SOLT
- Different connector on the VNA than on the board
- Trace length between connector and DUT that you do not want in the datasheet plot
- Comparing die or package variants on the same coupon family
3. Coupon discipline
- Same reference plane intent: if the DUT board has 8 mm of 50 ohm microstrip after the launch, the 2xThru should too (times two).
- Same connector lot and torque practice
- Same number of ground vias and keep-outs around the launch
- Include the intentional DC blocks / bias tees only if they are also on the DUT path; otherwise measure them separately
Mismatch between coupon and DUT board is the number-one silent error.
4. Band and causality checks
After de-embedding:
- Passive DUTs should still look passive (|S11|^2+|S21|^2 ≤ 1 within noise)
- Group delay should not go wildly negative in-band
- A known verification line (a third coupon with a longer thru) should look like a delay plus small loss after the same fixture removal
If your "attenuation DUT" shows gain, stop publishing and fix the split.
5. Time-gating pitfalls
Gating can help separate fixture from DUT but also erases real DUT reflections that sit near the fixture in time. Use gating as a diagnostic, not as the only truth, on electrically short DUTs.
6. Practical workflow
1) Cal VNA to the coax reference you trust (SOLT/TRL at the cable ends).
2) Measure 2xThru; build fixture models.
3) Measure DUT board; apply de-embed.
4) Validate on a known coupon.
5) Only then sweep temperature or variants.
Write down the algorithm and the coupon revision next to every de-embedded plot. Without that, nobody can reproduce your S21.