Executive answer
Locate the current before assigning the defect
- A tester can observe a combined current, not an isolated specimen property.
- In a simple parallel-leakage model, conductances add; resistance values are not background-subtraction quantities.
- A changing background needs investigation and an uncertainty decision, not an automatic corrected pass.
Current guidance context
The buyer decision: reject the sample or investigate the measurement?
A supplier reports a high insulation resistance, but the receiving tester returns a lower value. It is tempting to call the membrane switch defective or to subtract an empty-fixture reading until the numbers agree. Neither action identifies where the measured current flows. Preserve the result first, then ask whether the two laboratories measured the same path under a sufficiently characterized setup.
Tektronix/Keithley's data-acquisition primer identifies fixture insulation in parallel with the device under test as one reason a high-resistance reading can be too low. Here, device under test is abbreviated DUT. The worksheet below translates that measurement issue into a supplier-review decision; it does not claim that the fixture caused any particular product failure.
Define what the reported insulation value includes
Identify the source and return nodes, other-conductor states, installed components and specimen condition. Then identify the fixture contacts, cables and switching route used to reach those nodes. Keep the intended specimen insulation path distinct from unintended paths in the measurement system. A shield or backer must have a recorded state, not an assumption carried over from a different drawing.
ASTM's public catalog marks F1689-05(2020), Determining the Insulation Resistance of a Membrane Switch, withdrawn in 2023 with no replacement; the listing was updated November 30, 2023. An old reference therefore needs an explicit contractual method review. This article uses that status fact only, not the full standard or a current conformance claim.
Keep insulation resistance separate from the resistance of a pressed key and from a dielectric-withstand result. A number without its test boundary cannot substitute for either record. Ask the laboratory to resolve method, bias, timing, handling, safety and specimen disposition under its authorized procedures; this guide supplies no universal voltage or production-test recipe.
Worked example: a 1 GΩ specimen can display as 500 MΩ
Hypothetical HYR editorial model—not a measurement: assume a stable, linear 1 GΩ DUT path and a separate 1 GΩ fixture path connected across the same ideal source. Both currents reach the measuring channel. Their parallel equivalent is 1/Rdisplay = 1/RDUT + 1/Rfixture, giving Rdisplay = 500 MΩ. One gigaohm is 1,000 megohms. The display is lower than either individual resistance.
Use an assumed 5 V only to check the arithmetic, not to select or recommend a test setting. Each 1 GΩ branch carries 5 nA, so the channel sees 10 nA and 5 V divided by 10 nA is 500 MΩ. The fixture has changed what the channel observes; the calculation has not identified the condition of a real switch.
If only the assumed fixture path changes to 10 GΩ, its current becomes 0.5 nA. The combined 5.5 nA corresponds to approximately 909 MΩ, although the modeled DUT is still 1 GΩ. These two displays could arise from the same ideal DUT with different backgrounds. The example excludes instrument error, transients, nonlinear leakage and changes caused by installing the specimen.
Why subtracting the background is a separate decision
Do not subtract resistance values. In the ideal same-voltage parallel model, branch currents or conductances add. Inferring DUT current by subtracting a blank current requires evidence that the blank represents the same independent background in the loaded setup. A different contact geometry, voltage distribution or connection state can invalidate that assumption. Arithmetic alone does not authorize a correction.
A second hypothetical worksheet shows the sensitivity. Hold the total current exactly at 10 nA and the assumed bias exactly at 5 V, but let the blank-current evidence span 3–7 nA. Subtraction would leave 7–3 nA for the DUT, respectively. Those endpoints correspond to about 714 MΩ and 1.667 GΩ. This is a range-propagation example for one background term, not a full uncertainty interval or statistical confidence statement.
If a buyer invents an 800 MΩ minimum solely for this example, that inferred range crosses the proposed limit. It cannot establish a pass from the midpoint. The actual laboratory must evaluate the validity of the model, the relevant uncertainty and the agreed decision rule. A zero or negative corrected current is also a reason to investigate the model and measurement, not report an infinite insulation resistance.
Request system evidence without turning a guide into wiring instructions
Tektronix/Keithley's low-current application note separates system offsets, settling and leakage, and distinguishes a driven guard from shielding. It shows how fixture or cable leakage can join the measured current. That is a reason to ask how the system was characterized, not a basis for copying its settings or connections into another tester.
Ask the laboratory what background record represents the measurement route and how stable it is over the relevant sequence. Request the original current-versus-time evidence and the rationale for the selected reading time. Preserve range, polarity, environment, fixture revision and any changes between blank and specimen records. A single final resistance cell should not erase an unsettled trace or a changed setup.
If guarding is part of the authorized method, document which paths it excludes. Do not remove a current that the requirement intends to include in the specimen result, then call the new measurement equivalent. Guard, measurement reference and protective earth are not interchangeable labels. Only the qualified laboratory should establish the actual instrument-specific safety and connection scheme.
Five ways the review can assign the wrong conclusion
These are plausible review errors, not reported HYR defects. Use them to locate the missing evidence.
- The low reading is assigned to the membrane switch without checking the complete measurement route: retain both system and specimen evidence.
- An empty fixture is checked under a different condition: its record may not describe the background present during the specimen reading.
- A variable blank is reduced to one convenient correction: preserve the range and let the laboratory evaluate its significance.
- A guard arrangement excludes an intended specimen path: the reported value now answers a different question.
- A final display replaces the raw time record: settling, connection changes or unstable readings are hidden from the acceptance reviewer.
What OEM engineering and sourcing should put in the RFQ
Send the controlled circuit and stack drawings, the exact insulation path and the proposed sample state. Ask the supplier and receiving laboratory to describe their measurement boundaries in compatible terms. Include the method owner, instrument/fixture identification, authorized conditions, raw records and any proposed correction. Keep a suspected specimen defect and a suspected system contribution as separate unresolved questions.
Agree on the decision before comparing suppliers: which evidence can support an uncorrected result, which correction needs validation, what uncertainty matters near the limit, and when a repeat or specimen investigation is required. Do not solve a disagreement by changing the requirement after seeing the numbers. Preserve the first result, subsequent setup changes and the responsible engineer's disposition.
HYR's secure RFQ can collect the membrane-switch drawings and evidence request for review. It does not replace the laboratory's measurement responsibility or the OEM's final functional and safety validation. Use the closed-circuit resistance guide for the pressed-state budget; this page concerns confidence in the insulation measurement itself.