Concepts & glossary

Plain-English explanations for the SMU-specific terms you'll encounter in the GUI and the CSV output. Roughly ordered by how often you'll hit them in a typical session.

Source vs measure (SMU basics)

A source measure unit (SMU) can both source and measure on the same pair of leads. The Keithley 2400 family is bipolar: you pick whether voltage or current is sourced; the other quantity is measured. ResistaMet GUI exposes that as the Voltage Source and Current Source modes; resistance-mode internally sources a known current and measures the resulting voltage and reports R.

Compliance

The maximum the measured quantity is allowed to reach before the instrument clamps. When you source 1 V into a short circuit, current would in principle be infinite; the current compliance prevents that by capping output current. Compliance protects both your DUT and the instrument.

If a reading hits compliance, the Keithley reports a magic number (9.91 × 10³⁷) in the affected channel. ResistaMet GUI detects this via the SCPI STAT word's bit 3 and flags the row in the CSV's compliance column.

2-wire vs 4-wire

In 2-wire measurement, the same two leads source current and measure voltage, so lead resistance is added to your reading. In 4-wire (Kelvin), an outer pair sources current and a separate inner pair measures voltage at the DUT — lead resistance falls out of the math because the voltmeter draws ~no current. Use 4-wire whenever sample R is comparable to or smaller than your lead resistance (anything below ~10 Ω).

The Four-Point Probe mode is always 4-wire by definition. Resistance mode lets you pick.

NPLC (Number of Power Line Cycles)

The integration time of each measurement, expressed in line-frequency cycles. NPLC = 1 means "integrate over 1 mains period" (16.67 ms at 60 Hz, 20 ms at 50 Hz) — the classic noise-rejection setting because mains hum averages to zero over an integer number of cycles.

The Keithley datasheet defines three nominal Speed buckets:

Speed	NPLC	Trade-off
Fast	0.01	~25× faster than Normal; offset spec relaxed by 0.05% of range (0.5% for special ranges)
Medium	0.1	~3× faster than Normal; offset spec relaxed by 0.005% (0.05% for special ranges)
Normal	1.0	Baseline; clean mains rejection
High accuracy	10	~10× slower than Normal; tightest reading

ResistaMet GUI accepts arbitrary NPLC values; the accuracy module interpolates as a step function (≥ 0.5 → Normal modifier, 0.05–0.5 → Medium, < 0.05 → Fast). For rigor below 1 PLC, pin NPLC to one of the three canonical values.

NPLC also drives the live-readout's significant-figure count: ≥ 0.5 → 6 figs (6½-digit hardware), 0.05–0.5 → 5 figs, < 0.05 → 4 figs.

Auto-zero

The Keithley periodically re-references its internal ADC against ground and a precision reference, then subtracts that offset from subsequent readings. Three modes:

• on — auto-zero before every reading. Most accurate; 3× slower because each reading takes 3 integrations.
• once — auto-zero once at run start, then never. ~3× speed-up. Acceptable for sensor work where the signal of interest is R(t) or V(t) and absolute-zero drift over a single run is below the noise floor.
• off — never auto-zero. Fastest, but readings drift as the instrument warms or cools.

Default is once, good for typical sensor sessions. Switch to on for absolute-accuracy runs longer than ~30 min, or for runs where you need long-term comparability. 4PP and vdP modes force on regardless of the shared setting.

Hardware averaging filter

The Keithley can average N raw readings internally and return one result, in either:

• Repeat mode — takes N readings, averages, returns one number, repeats. Best for stable signals.
• Moving mode — running average over the last N readings. Better for trending or changing signals.

Lives in the Keithley itself (no per-reading GPIB round-trip), so it's much faster than averaging client-side. Default filter_count = 5 with filter_enabled = True. Set higher for noisier signals; 1 disables.

Enhanced R mode

When measuring resistance, the Keithley 2400 family has a dedicated higher-accuracy column in its datasheet (the "Enhanced Accuracy" column on p. 7 of 1KW-2798-3, April 2021). It's active when:

• Source-readback is ON (ResistaMet always requests :FORM:ELEM VOLT,CURR,RES,STAT so V and I come back alongside R)
• Offset-compensated ohms is ON (:SENS:RES:OCOM ON)

The "Offset-compensated ohms" technique sources +I, measures V₊; sources −I, measures V₋; reports R = (V₊ − V₋) / (2·I). This cancels any thermoelectric EMF in the circuit (the steady DC offset that fights you when measuring low-R DUTs).

Cost: each reading takes ~2× longer (two source/measure cycles instead of one). The sampling-rate cap halves automatically when Enhanced R is on.

Win: at low R (V-offset-dominated regime, ~ < 200 Ω) the published σ_R is dramatically tighter than V/I propagation gives. The Enhanced R table covers 20 Ω – 200 MΩ; ResistaMet falls back to V/I propagation outside that range.

ON by default because a precision-measurement tool should ship the accurate default. Fast scans where you don't care about EMF cancellation can turn it off in the Resistance tab.

Sheet resistance vs resistivity vs conductivity

• Sheet resistance Rs (Ω/□, "ohms per square") — the resistance of a thin square of material, independent of square size. Material property + thickness combined. For a 4PP measurement: Rs = K · α · (V/I) where K is the geometric factor (4.5324 for an infinite half-plane and the classic Smits factor; modified by ASTM F84 corrections for finite geometries) and α is the finite-sample correction.
• Resistivity ρ (Ω·cm) — material property only. ρ = Rs · t where t is the film thickness, so requires knowing thickness.
• Conductivity σ (S/cm) — σ = 1/ρ.

The 4PP tab computes all three from V, I, geometric factors, and thickness. The vdP tab computes Rs first and then ρ from the F76 chain.

Four-point probe (4PP) quick anatomy

Four collinear probe tips, evenly spaced (default spacing s = 0.1016 cm = 40 mil for the Signatone SP4). Outer two source current, inner two measure voltage. Standards-aligned with ASTM F84-02; correction factors F₂ (geometry), F(w/S) (thickness), F_T (temperature, for n-/p-type silicon) compose multiplicatively to give the effective K.

Van der Pauw (vdP) quick anatomy

Sheet resistance of arbitrary-shape samples with four periphery contacts. Procedure per ASTM F76-08 Method A: cable up four geometries in sequence (the GUI walks you through each), source +I then −I at each, for 8 voltage readings total. The implicit f(Q) factor is solved numerically. F76 §11.1 flags the sample as non-homogeneous when |ρ_A − ρ_B|/ρ_avg > 10%.

The per-geometry bar chart on the vdP result panel color-codes the four R values by deviation from the mean: green for |R_i − mean|/mean < 3% (clearly uniform), orange for < 10% (within the F76 gate but worth watching), red for ≥ 10% (outlier; sample probably bad).

Compliance "magic number" (9.91 × 10³⁷)

When a reading exceeds range or hits compliance, the Keithley 2400 family returns +9.91E+37 for that channel as a sentinel — IEEE 754 "not-a-number"-ish. ResistaMet detects this via the SCPI STAT word's bit 3 and falls back to a threshold check for older firmwares. Detection bumps the row's compliance column to a flag string.

Touch-safety warning

IEC 61010-1 sets the SELV (Safety Extra-Low Voltage) upper bound at 30 V DC — below this, electric-shock hazard is negligible under any realistic skin-resistance condition. The 2400/2410/2425/2430/2450 can compliance-clamp at 60–1100 V depending on model, well above SELV.

Before any run whose gating voltage reaches the threshold, ResistaMet shows a warning modal. The gating voltage depends on mode:

• Voltage Source mode: the sourced V (vsource_voltage)
• Resistance / Current Source / Four-Point / Van der Pauw: the configured V compliance — an open-circuit current source swings up to compliance, so even a 1 mA test current can put 200 V on the leads if compliance is set there
• I-V Sweep: max(|sweep_start|, |sweep_stop|) when sourcing voltage, otherwise the compliance

The threshold defaults to 30 V and is per-user-profile-configurable; set it to 0 to disable entirely. A "Don't show again" checkbox sets a sticky silence flag for the profile; Settings → Measurement has a re-enable toggle.

The status bar appends ⚡ N V live while a hazardous run is active. Warning is informational — it never blocks the measurement.

Probe power envelope

Four-point probe tips can melt or oxidize if you push too much power through them. The Signatone SP4 series tungsten-carbide tips are spec'd at ~100 mA continuous; ResistaMet ships with conservative defaults:

• fpp_power_warn_w = 10 mW — status-bar flash above this measured V·I
• fpp_power_stop_w = 100 mW — hard stop, output disabled, run aborted above this

A pre-flight check refuses to start a 4PP run if worst-case I_source × V_compliance would exceed the hard stop. These are also protective for thin films and conductive polymers, where local Joule heating can damage the sample before the probe.

Output-off mode

When the source output is disabled (between geometries in vdP, after a Stop, etc.), the Keithley can leave the output terminal in one of several states. ResistaMet uses high-impedance (:OUTP:SMOD HIMP), which gates the output through a relay — safest for delicate DUTs because there's no low-impedance path even momentarily.

Auto source delay

Source-then-measure timing matters: switch the source level, wait, then read. Too short and the reading captures the transient; too long and you're throwing away throughput. The Keithley has built-in tables of recommended delays per range; setting :SOUR:DEL:AUTO ON lets the instrument pick. ResistaMet enables this by default.

Cable null

"Measure the lead resistance once, subtract it from every subsequent reading." One button in the Resistance tab: short the leads, click Null, the measured R becomes a software offset stored as res_cable_null until cleared. Useful for thin-film resistance work where lead/probe-tip R is a meaningful fraction of the DUT R but you can't 4-wire (e.g. through a switch matrix).

Uncertainty: instrument vs statistical vs combined

Three flavors appear in the GUI and CSV:

• Instrument uncertainty (u_inst): floor from the Keithley datasheet's per-range accuracy specs. Doesn't reduce with more samples (it's systematic — every reading inherits it).
• Statistical uncertainty (u_stat): standard error of the mean, std / √N. Reduces with more samples (random component).
• Combined uncertainty (u_total = √(u_stat² + u_inst²)): the right way to roll them up under GUM §5.1.2.

The live results panel and CSV finalize metadata report all three for 4PP per-spot stats and for vdP final results.

RSS vs linear-sum propagation

When propagating uncertainty through R = V/I, the Keithley user manual sums the relative uncertainties linearly — the conservative worst-case bound, fine for spec-sheet language. ResistaMet uses root-sum-of-squares (RSS) per GUM §5.1.6 Eq. 12:

(σ_R/R)² = (σ_V/V)² + (σ_I/I)²

This is the standard treatment for uncorrelated error sources under the law of propagation of uncertainty. V and I uncertainties come from different signal paths inside the Keithley, so treating them as uncorrelated is appropriate. In V-dominated or I-dominated regimes the two methods agree to ~0.1%; in the balanced regime RSS gives ~0.71× the linear-sum answer.

Full rationale + GUM citations in the docstring of resistance_uncertainty.

Machine-local GPIB address

gpib_address is the one setting that can't be portable across PCs: a 2400 wired up as GPIB0::24::INSTR on one machine might be GPIB0::3::INSTR on another. ResistaMet stores it keyed by hostname and strips it from user-profile and global-settings writes, so a NAS-shared config.json works across lab PCs without one machine's wiring shadowing another's. Any legacy single-address slot is auto-lifted into the per-machine entry on first save.

Rate-cap predictor

The per-tab Sampling Rate spinbox is dynamically capped based on a model of the Keithley's actual per-reading time (timing.estimate_max_sample_rate_hz). The cap depends on NPLC, auto-zero mode, hardware-filter count, and (for resistance mode) the Enhanced-accuracy / offset-compensated-ohms checkbox.

When you type a rate above what the instrument can sustain, the spinbox clamps to the cap and the status bar reads (e.g.):

"15.0 Hz is above what the instrument sustains right now (8.5 Hz). To get there: set auto_zero to 'once' (re-zeros are cached during the run)."

The suggestion tries the cheapest accuracy-cost change first (auto_zero ON→ONCE), then filter_count reductions, then NPLC reductions. The tooltip on the spinbox always shows the current ceiling and the next cheapest change to raise it. Bench-validated within ~5% across 27 (NPLC, auto-zero, filter-count) combinations.

Wong palette

The 8-color colorblind-safe palette from Wong, Nature Methods (2011). Used throughout ResistaMet's live readouts and plot canvases so V/I/R/P live-readout labels match their trace colors:

• V — blue #0072B2
• I — bluish green #009E73
• R — vermillion #D55E00
• P — orange #E69F00

You can override these in Settings → Display if you have a strong preference, but the Wong defaults are accessible by design.