Lilienfeld 1925 Test for Control Induced Modulation

# Test for Control-Induced Modulation: Execute the Frozen Protocol and Score the Result

Step 22 executes the protocol Step 21 wrote and nothing else. The bias list, the interleaving order, the repeat count, the comparison statistic, and the two thresholds were all fixed before this step began. Step 22's only job is to run that plan faithfully, compute the pre-registered statistic from the data it produces, and report whether the result clears both thresholds — not to find a more favorable way to look at the data once it exists.

A positive result here is still not proof of a solid-state field effect in the sense of a later, well-controlled transistor. It is a claim that, under this specific device, this specific protocol, and these specific pre-registered criteria, a change in terminal current correlated with control bias in a way that leakage and baseline drift do not explain. That claim must be exactly as strong as the evidence, no stronger.

## 1. Run the frozen protocol without modification

Load the Step 21 bias list, interleaving schedule, and repeat count exactly as registered. Execute the identical Step 19 terminal sweep at every scheduled point and the identical Step 20 three-terminal accounting alongside it. If execution reveals that a scheduled bias point is not safely reachable — an unexpected compliance event, a leakage reading above the Step 20 bound — stop, log the deviation with its cause, and decide whether to amend the frozen protocol before continuing, exactly as Step 21 required. Do not quietly skip the point and continue as if nothing happened, and do not substitute a different bias "close enough" to the scheduled one.

Record the actual order points were run in, even though it was randomized in advance, because a mis-executed schedule (a point run twice, a point run out of turn) must be caught before analysis rather than discovered afterward.

## 2. Compute the pre-registered statistic exactly as specified

Using only the comparison rule Step 21 fixed, compute at each nonzero bias point

$$ \Delta I_{12}(V_{13})=I_{12}(V_{13})-\tfrac{1}{2}\left[I_{12}(0^{-})+I_{12}(0^{+})\right], $$

averaged across the completed repeats, together with its standard error from the repeat-to-repeat spread. Compute the matching leakage ratio at the same bias point from the simultaneous three-terminal data,

$$ f_{leak}(V_{13})=\frac{\left|I_{13}(V_{13})\right|}{\left|\Delta I_{12}(V_{13})\right|}. $$

Do not introduce a new statistic, a different averaging window, or a different reference convention at this stage because it produces a cleaner-looking number. If the frozen statistic is genuinely inadequate, that is a finding about the protocol, to be corrected in a new Step 21 revision for the next device, not a reason to re-derive the answer for this one after the fact.

Executed bias schedule and the pre-registered modulation statistic A plot shows the terminal current measured at each executed bias point across repeats, overlaid with the computed paired-difference statistic at each nonzero bias and its standard error, alongside the leakage ratio computed from the same simultaneous three-terminal data. RUN THE FROZEN PLAN, THEN COMPUTE THE FROZEN STATISTIC No new statistic, no new averaging window, no re-derivation after the data is seen EXECUTED BIAS SCHEDULE ACROSS REPEATS executed sequence position, three interleaved repeats terminal current I12 offonoffonoffonoff Each point is the full Step 19 sweep result; order shown is the actual randomized run order, logged as executed. PRE-REGISTERED STATISTIC, COMPUTED AFTER THE FACT, NOT BEFORE ΔI12(V13) = I12(V13) − average of bracketing I12(0) averaged across the completed repeats standard error from repeat-to-repeat spread f_leak(V13) = |I13(V13)| / |ΔI12(V13)| from the same simultaneous three-terminal data not a new leakage assumption No substitute statistic is computed even if the frozen one gives an unflattering result. A genuinely inadequate statistic is a finding for the next protocol revision, not a reason to re-derive this one.

## 3. Score against both pre-registered thresholds, together

Apply the rule exactly as written in Step 21: the paired difference must exceed the baseline repeatability bound, and the leakage ratio must not exceed the leakage limit, at the bias point being evaluated,

$$ \left|\Delta I_{12}(V_{13})\right| > \delta I_{min}\quad\text{and}\quad f_{leak}(V_{13})\le f_{leak,max}. $$

Both conditions must hold simultaneously. A large $\Delta I_{12}$ that fails the leakage test is evidence of a conduction path through the control circuit, not of field-effect modulation. A $\Delta I_{12}$ that passes the leakage test but does not exceed $\delta I_{min}$ is statistically indistinguishable from baseline noise and must not be reported as an effect regardless of how suggestive its sign looks.

Require the same sign and a comparable magnitude at the matched-magnitude positive and negative bias points unless a stated physical reason predicts asymmetry. An effect that appears only at one polarity, with no such reason given, is a flag for an uncontrolled asymmetry such as a biased contact or an asymmetric leakage path, not a stronger result.

## 4. Treat polarity, repeat-to-repeat, and run-order consistency as part of the evidence

A genuine control-bias effect should behave coherently across the structure of the frozen protocol, not merely clear the two thresholds once. Check:

  • consistency of sign and approximate magnitude across the independent repeats;
  • consistency of sign when comparing positive and negative control bias at matched magnitude;
  • absence of correlation between the computed effect and the randomized run order, which would indicate the "effect" is actually tracking elapsed time or drift rather than bias.

A result that clears both thresholds in one repeat but not in the others, or that correlates with run order rather than with bias polarity, is a hold, not a pass, even though the raw numbers at the headline bias point looked acceptable.

Scoring rule and consistency checks for a candidate modulation result A two-axis scoring plot places the computed effect against the resolvability threshold and the leakage ratio against its limit, defining a single pass region where both conditions hold, next to a table of repeat-to-repeat and polarity consistency checks and the final disposition rule. BOTH THRESHOLDS, TOGETHER, EVERY TIME A result must clear resolvability and the leakage limit, and repeat coherently across the protocol SCORING PLANE |ΔI12(V13)|, relative to δImin f_leak(V13), relative to f_leak,max δImin f_leak,max PASS REGION resolved above baseline, leakage stays below limit unresolved leakage explains it CONSISTENCY CHECKS AND FINAL DISPOSITION Sign and magnitude consistent across all completed repeats Sign consistent between matched positive and negative control bias No correlation between the computed effect and randomized run order PASSboth thresholds cleared at every scheduled point, all three consistency checks satisfiedHOLDthresholds cleared at some points or repeats but not coherently across the protocolREJECTunresolved, leakage-explained, run-order correlated, or polarity inconsistent with no stated cause

## 5. Re-verify the baseline and leakage bound have not moved during execution

Running the full Step 21 protocol delivers more cumulative charge and bias exposure to the device than any single prior step. Before interpreting the result, re-run the Step 19 baseline sweep and the Step 20 leakage check one more time, under the identical zero-control condition used at the start, and compare against both the pre-run values and the published Steps 19–20 thresholds. If either has shifted beyond its stated repeatability bound, the device that produced the Step 22 data is not the device that was characterized, and any computed effect must be reported as confounded by specimen drift rather than scored as a clean pass.

## 6. State the result as a disposition, not as a headline number

Report pass, hold, or reject for the protocol as a whole, using the rule from §3 and §4 together, and report it alongside the actual numbers rather than instead of them: $\Delta I_{12}(V_{13})$ with its standard error, $f_{leak}(V_{13})$, the per-repeat breakdown, and the polarity comparison. A single favorable number extracted from an otherwise mixed record is not a pass. State explicitly which mechanisms — leakage, baseline drift, displacement current, run-order drift — were checked and found not to account for the result, rather than asserting a field effect by default because nothing else was tested.

## 7. Minimum Step 22 record

Retain:

  • device genealogy and the registered Step 15–21 geometry, coverage, material, isolation, baseline, leakage, and frozen protocol;
  • the as-executed bias schedule and run order, including any logged deviation from the Step 21 plan and its stated amendment;
  • complete per-point, per-repeat terminal sweep data and simultaneous three-terminal accounting;
  • the computed $\Delta I_{12}(V_{13})$ with standard error and $f_{leak}(V_{13})$ at every nonzero bias point, using the frozen statistic only;
  • the polarity comparison, the repeat-to-repeat comparison, and the run-order correlation check;
  • the post-run Step 19 baseline and Step 20 leakage re-measurement and their comparison against pre-run values;
  • temperature, humidity, illumination, elapsed time, and cable-motion controls through the full run;
  • the final pass, hold, or reject disposition for the protocol as a whole, stated with the supporting numbers rather than in place of them;
  • an explicit list of the alternative mechanisms checked and ruled out, or not ruled out, for any reported effect.

The historical geometry is grounded in [J. E. Lilienfeld, US Patent 1,745,175](https://patents.google.com/patent/US1745175A/en), whose claim is a modulation of the current between the two main terminals by a potential on the control member. The requirement to pre-register a comparison rule and score a result against thresholds fixed before the data existed, rather than against a rule chosen afterward, is the discipline this step enforces so that the device's known leakage and drift behavior — established with the practices in the [Keithley Low Level Measurements Handbook](https://www.tek.com/en/documents/product-article/keithley-low-level-measurements-handbook---8th-edition) — cannot be mistaken for the claimed effect.

## Test for Control-Induced Modulation’s Place in the Process Lineage

Step 21 fixed the plan; Step 22 executes it and scores the result against the rule written in advance, re-verifying that the baseline and leakage bound established in Steps 19 and 20 have not moved during the run. Only a result that clears both pre-registered thresholds, behaves consistently across repeats and polarity, and shows no run-order correlation constitutes evidence of control-induced modulation under this protocol. A hold or reject outcome returns the finding to the device genealogy for root-cause work rather than being reported as a negative result in isolation.

Take lilienfeld 1925 test for control induced modulation further

Ask the copilot about this term, or have our engineers assess it against your process.

Ask CFSGPT about this → Request a technical assessment → Open the CFS app