On this page
A gold printed interconnect can have a small resistance compared with its test fixture. Subtracting a short-circuit reading may appear to isolate the interconnect, but that correction is valid only when the control reproduces the unwanted path without removing part of the quantity of interest. A useful de-embedding plan defines exactly which gold track, pad transitions, wires and contacts belong to the reported result. It also reports how uncertain the difference remains after subtraction, particularly when two large readings leave a small residual.
Key design decisions
- Draw the measurement boundary before choosing the short or control coupon; a convenient metal bridge is not automatically an equivalent fixture.
- Remove unwanted lead drops with appropriate sense placement before relying on a numerical correction.
- Retain raw and corrected readings together and compare the corrected resistance with its uncertainty, not with display resolution alone.
Name the electrical quantity being isolated
Decide whether the target is the printed gold trace alone, trace plus bond pad, or a complete wire-to-track interconnect. These are different quantities. A control that bypasses both the trace and one bond pad cannot isolate the trace unless the pad contribution is independently known. Place the boundaries on the drawing and annotate the corresponding physical voltage-sense points.
The word fixture can include current leads, probe contacts, switching contacts and short printed routes on an adapter. Only the portions included within the voltage measurement can contribute directly to the reported resistance. Keep shared test equipment outside the specimen definition unless the intended acceptance test deliberately includes it. An agreement on these boundaries is more valuable than a correction with additional decimal places.
Establish the finite control value independently
Characterize the substitute element between its defined endpoints using an independent suitable method or a traceable value with a relevant uncertainty. Record temperature, material state and any dependence on mounting. A resistance assigned by subtracting the same unknown fixture from itself does not provide independent information about the control.
Four-wire sensing and stable voltage-offset correction should already be established for each observation. They do not remove an adapter or printed segment lying inside the sense boundary. If the control element has a known value of 0.4 mΩ, its full assembly reading cannot be treated as a pure fixture estimate. The fixture estimate is the assembly reading minus that 0.4 mΩ. A well-characterized finite control can be more useful than an apparently better short whose own value is unknown.
Design a control that preserves the unwanted path
A useful control keeps the same unwanted electrical path while replacing the target segment with an independently characterized element. Call the complete control-assembly observation Rcontrol,meas and the element alone Rshort. The measured control assembly includes both the unwanted path and Rshort. A fixture-only estimate is therefore Rcontrol,meas minus Rshort, not the raw control reading.
Use before-and-after control readings to test fixture stability around the specimen sequence. A control measured only at the beginning of a long session cannot reveal later contact wear or temperature drift. When removing the specimen changes the contact state, include reseating variation in the uncertainty rather than treating the control as an exact constant. A stable instrument does not guarantee a stable mechanical fixture.
Propagate uncertainty through the difference
For the matched additive model, subtract the raw control-assembly reading from the specimen-assembly reading and add back the known control-element resistance. Alternatively, subtract a fixture-only estimate from which the element has already been removed. Name the convention explicitly so that a later spreadsheet does not subtract the finite element twice.
Separate uncertainty from a permissible product tolerance. A corrected result of a few milliohms can have an uncertainty of similar size even though the instrument displays many digits. Repeated readings characterize some variation but do not automatically include incorrect control geometry, temperature error or calibration effects. The uncertainty budget should identify which contributions are measured, specified or bounded by an explicit assumption.
Rtarget = Rcombined − Rcontrol,meas + Rshort; for independent inputs, u²(Rtarget) = u²(Rcombined) + u²(Rcontrol,meas) + u²(Rshort)
- Rcombined: measured specimen assembly containing target and unwanted additive path, in ohms
- Rcontrol,meas: measured control assembly including its finite substitute element, in ohms
- Rshort: independently characterized resistance of the substitute element at the relevant state, in ohms
- u: standard uncertainty in the same resistance unit
- For correlated inputs, propagate the full covariance matrix with sensitivity coefficients +1, −1 and +1; shared influences are not automatically independent or perfectly cancelling.
The unwanted path is additive and equivalent between observations. The formula does not remove non-ohmic effects or establish contact equivalence.
Compare a small residual with the measurement capability
Take hypothetical specimen and control-assembly readings of 14.0 mΩ and 10.0 mΩ, with a known control element of 0.4 mΩ. The target is 14.0 − 10.0 + 0.4 = 4.4 mΩ. Independent standard uncertainties of 0.5 mΩ, 0.5 mΩ and 0.1 mΩ give √(0.25 + 0.25 + 0.01) ≈ 0.714 mΩ, about 16.2% of the corrected result. Omitting the finite control would instead report 4.0 mΩ, about 9.1% below the model result.
Suppose the control assembly reads 10.0 mΩ at minute zero and 10.8 mΩ at minute ten, while its element remains at 0.4 mΩ. A specimen reading of 14.0 mΩ at minute four would use a linearly interpolated control of 10.32 mΩ, giving 4.08 mΩ. This assumes approximately linear fixture drift that substitution does not disturb. Two endpoint readings do not establish that assumption.
If only a no-excursion range between those endpoint control values is justified, the corresponding target range is 3.6–4.4 mΩ before measurement uncertainty. Without that no-excursion assumption, even this range is not a guaranteed bound. A changed corrected result therefore needs a time model before it can be interpreted as a changed gold interconnect.
Test transfer with two characterized control values
Where practical, compare two independently characterized substitute elements within the same fixture boundary. If their resistances are 0.4 and 1.4 mΩ and the unwanted contribution is unchanged, their complete assembly readings should differ by 1.0 mΩ within the relevant uncertainty. This difference check can expose a control-dependent fixture contribution before either control is used to resolve a smaller target.
Agreement supports only the tested transfer conditions. A different pad area, current-spreading region or seating condition can change the unwanted path when the gold specimen is installed. Keep repeated substitutions separate from repeated readings without remounting. If the two controls disagree with their known difference, resolve that mismatch instead of choosing whichever control produces the lowest corrected resistance.
Use unexpected corrections as diagnostic evidence
An implausible result is not automatically a bad gold conductor. It can reveal that the assumed measurement model is incomplete. Retain the combined reading, control reading and timing so another engineer can reconstruct the correction instead of seeing only the final residual.
| Observation | Potential measurement issue | Next discriminating action |
|---|---|---|
| Residual rises when the finite control value is restored | Original subtraction may have used the raw assembly as fixture-only resistance | Check the three named terms and their units |
| Residual changes with control selection | Contact geometry or characterized element values may differ | Compare two known controls within the same boundary |
| Before and after controls disagree | Fixture state is time dependent | State the interpolation or bounding assumption |
| Small corrected resistance is negative | Control mismatch or uncertainty may exceed the residual | Retain the signed result instead of clipping to zero |
| Two known controls fail their difference check | Substitution changes the unwanted contribution | Resolve transfer before interpreting a smaller target |
Keep the corrected value traceable to the physical test
The delivered record should contain the fixture drawing, specimen and control identities, raw current and voltage, temperature, order of measurements, subtraction formula and uncertainty assumptions. Include photographs of the sense boundaries where their placement is difficult to reproduce. Keep test-induced surface changes in the record when the same pad will later be bonded or otherwise attached.
For a gold interconnect acceptance requirement, state whether the corrected quantity or the complete assembled path is controlled. A very low corrected trace resistance does not establish an adequate wire bond, adhesion strength or operating current. De-embedding answers a specific electrical question; separate evidence is needed for those mechanical, thermal and process functions.
Send the interconnect test boundaries
Provide the gold interconnect and its control fixture together so the correction can be reviewed against the actual physical path.
- Target drawing with force contacts, sense points, bond pads and included or excluded adapter segments.
- Control-coupon geometry and independently characterized element resistance, temperature, uncertainty and transfer evidence.
- Raw specimen/control assembly readings, actual timestamps, bracketing controls, current conditions and repeated-substitution observations.
- Correction model, uncertainty budget, electrical acceptance limit and any later bonding or attachment operation.
The drawing-upload form loads as you reach this section.