← All writing

The first time a measurement lied to me badly enough to cost a day, it was a probe. A circuit that behaved fine until I looked at it, and misbehaved whenever I did. The design was correct. The probe was the bug.

That’s the thing nobody tells you clearly enough at the start: a probe is not a wire. It’s a resistor, a capacitor and an inductor that you deliberately hang across a node in a working circuit, and then ask to report back honestly on what it just disturbed. Every probe choice is a trade between how little you disturb the node and how much of the signal survives the trip to the scope.

Active and passive probes sit at two ends of that trade. This is what each one actually is, what it costs, and how I decide.

The one number that matters: tip capacitance

Skip everything else for a moment. If you take one thing from this article, take this.

A probe’s tip capacitance sits in parallel with your node, and its impedance is

which means the loading gets ten times worse for every decade you climb in frequency. That single fact is why probe selection stops being optional somewhere around a few tens of megahertz, and it’s the reason active probes exist at all.

Put numbers on it. A run-of-the-mill 10X passive probe presents around 10 pF at the tip. A single-ended active FET probe presents under 1 pF — the Tektronix TAP3500, for instance, warrants 0.8 pF.

At 100 MHz that ordinary passive probe is a 159 Ω shunt hanging off your node. Not 10 MΩ. One hundred and fifty-nine ohms. The 10 MΩ printed on the probe body is a DC specification, and it stopped being true somewhere around a kilohertz.

Tektronix make the same point in their probe primer from the other direction: a standard 10 MΩ passive probe with 11 pF of tip capacitance has a reactance of about 290 Ω at 50 MHz, and depending on the source impedance this loading “could have a major effect on the signal amplitude, and it could even affect the operation of the circuit being probed.”

That last clause is the one that cost me the day.

What a passive probe actually is

A passive probe contains no transistors and no amplifier — just wires, connectors, and, where attenuation or compensation is needed, resistors and capacitors. That simplicity is why they’re rugged, cheap, and shipped free with every scope.

The classic 10X probe is a frequency-compensated voltage divider. The tip holds a 9 MΩ resistor. The scope’s input is 1 MΩ. Nine plus one makes ten, so the signal arrives at the scope divided by ten, and the node sees 10 MΩ at DC.

The interesting part is what happens once you add the capacitance that is unavoidably there anyway.

Why compensation exists

The divider has to divide by ten at every frequency, not just at DC. The resistive arm divides by ten because 9 MΩ and 1 MΩ say so. The capacitive arm divides according to reactance, which is the inverse of capacitance — so to get the same ratio, the tip capacitor has to be nine times smaller than the scope-side capacitance. Both arms agree when

Turning the trimmer until that holds is exactly what probe compensation is. Rohde & Schwarz describe it as matching “the oscilloscope’s inherent input capacitance with the capacitance in the probe tip,” which is the same statement in words. Get it wrong and square waves come back with drooping or overshooting tops, and every amplitude and rise-time number you take is wrong by an amount that varies with frequency.

The good consequence: 10X divides the capacitance too

Work out what the node actually sees. The tip capacitor $C_1$ is in series with the scope-side capacitance $C_2$, and the compensation condition already told us $C_2 = 9C_1$:

The 10:1 divider divides the capacitance by ten as well as the voltage. Around 100 pF of cable and scope input becomes around 10 pF at the tip. Real probes land a little higher because of stray capacitance in the tip and body, but that’s the mechanism.

Which is why 1X mode is a trap. In 1X the divider is bypassed and the node gets the full cable-plus-scope capacitance. The Tektronix P2220 — an ordinary bundled probe — spells the consequence out in its own specification table:

Same probe. Flick one switch and you throw away 97% of the bandwidth and multiply the capacitive loading by six. The only good reason to reach for 1X is a signal so small that dividing by ten would bury it in the scope’s noise floor — and even then, ask whether the real answer is a better probe rather than a worse divider.

Modern passive probes are much better than the reputation

Worth saying plainly: “passive probe” does not have to mean 15 pF. Tektronix’s TPP series specifies under 4 pF and up to 1 GHz on a 10 MΩ, 300 V CAT II passive probe. That’s passive-probe ruggedness with loading in the same order as an entry-level active probe, and the datasheet makes a point of saying that low capacitance is “more forgiving of longer ground leads” — which, as we’ll get to, may be the bigger practical win. Rohde & Schwarz put the class as a whole at up to 500 MHz and up to 1000 V.

So the honest framing isn’t “passive is slow.” It’s “cheap passive is slow.”

What an active probe actually is

An active probe puts an amplifier — almost always a FET input stage — right at the tip, inside the probe head.

Two things follow from buffering at the tip.

The node only sees the FET’s input. A gate is a very small capacitor. Tektronix quote “a few picofarads down to less than 1 pF” for FET inputs generally; the TAP3500 datasheet warrants 0.8 pF and 40 kΩ, and Rohde & Schwarz put their single-ended active probes under 1 pF as well. Because $X_C = 1/2\pi f C$ stays high far longer, the probe’s input impedance stays high far longer, and specified bandwidths climb from a few hundred megahertz to several gigahertz.

The cable stops being part of your node. In a passive probe the coax and the scope input hang off the node through the divider. In an active probe the buffer’s low-impedance output drives a properly matched 50 Ω line into a 50 Ω scope input. Whatever the cable is doing, it’s doing it on the far side of an amplifier.

The picture that puts it together

Here’s what the three common probe classes actually present to your node across frequency. This is the plot I wish someone had shown me on day one.

Three things are worth staring at.

• The passive probe’s 10 MΩ is gone by about 1.6 kHz. That’s where $1/2\pi RC$ puts the corner for 10 MΩ and 10 pF. Above it, the trace is a straight capacitive slope all the way out.
• The active probe is the worse load at low frequency. 40 kΩ is not 10 MΩ. We’ll come back to what that costs you, because it’s the trap that catches people who’ve just been handed their first active probe.
• The dashed 500 Ω line beats the passive probe above about 32 MHz. A resistor and a length of coax outperforms your bundled 10X probe at high frequency, as long as your node can survive 500 Ω. That isn’t a trick — it’s a real technique, and it’s two sections down.

What the loading actually does to your measurement

Impedance curves are abstract. Here are the three concrete failure modes.

It stretches rise times

Any real source has some resistance. Add the probe’s capacitance to whatever is already on the node, and the RC network integrates your edge. The 10% to 90% rise time of an RC network is always

Tektronix work an example: a pulse generator with 50 Ω of source resistance and 20 pF of load capacitance has a 2.2 ns rise time. Probe it with an ordinary 11 pF probe and the total becomes 31 pF, so the measured rise time becomes 3.4 ns. The edge did not change. Your measurement of it did, by 55%.

Their quick estimate is worth memorising, because it needs no calculator:

It moves poles in your circuit

This is the one that changes behaviour rather than just the reading. A probe’s tip capacitance lands on the node and creates a pole with the node’s own impedance:

On a 1 kΩ node, a 10 pF passive probe puts a pole at 15.9 MHz. A 0.8 pF active probe puts it at 199 MHz. If your circuit has any business happening between those two numbers — a feedback loop, an oscillator, a comparator that’s now slow enough to double-trigger — the passive probe doesn’t measure your circuit. It builds you a different one.

Crystal and MEMS oscillator circuits are the classic casualty. SiTime’s application note on probing oscillator outputs points out that their parts are rated for a 15 pF output load, so a typical passive probe alone can push the node past its specification before you count what’s already on the board. Hence their blunt guidance: active probes for clock and high-speed measurements, passive probes for low-speed debug and DC.

It divides your amplitude

At high frequency the probe’s reactance and the source resistance form a plain divider. Electronic Design measured the effect at 70 MHz: a passive probe’s impedance had fallen to roughly 150 Ω while an active probe was still around 2.5 kΩ. Against a source of a few hundred ohms, that’s the difference between reading the signal and reading a fraction of it.

Where active probes lose

Active probes are not simply better. They’re a different trade, and the bill comes due in five places.

Dynamic range, which is brutally small

The TAP3500’s linear input dynamic range is ±4 V, with a ±10 V DC offset range to slide that window around, and ±30 V as the absolute non-destructive limit. Tektronix put the class generally at ±0.6 V to ±10 V linear range with withstand voltages as low as ±40 V.

Compare that with the passive probe on the same bench: 300 V_RMS CAT II, and it doesn’t care. You can’t measure from millivolts to tens of volts on an active probe the way you can on a passive one, and touching a rail that’s outside the window doesn’t just clip the trace — it can destroy the probe. They’re also static-sensitive, which passive probes essentially aren’t.

An active probe on an unknown board is a bet. Take a passive probe, find out what the node does, then decide.

DC loading, which is worse than passive

Look at the impedance plot again. That active probe’s DC input resistance is 40 kΩ, not 10 MΩ. (It varies a lot by model — SiTime quote a 30 kΩ to 10 MΩ spread across the market, and Rohde & Schwarz build theirs at 1 MΩ — so read the datasheet rather than assuming.)

What 40 kΩ does to a high-impedance node is not subtle. Take a divider of two 100 kΩ resistors from a 3.3 V rail, sitting at 1.65 V:

The active probe didn’t measure that node. It rebuilt it. This is the single most common surprise for someone moving up to active probes, and it’s why “active probe” is not a synonym for “better probe.”

Noise

An amplifier at the tip is an amplifier, and it contributes its own noise to the measurement. Tektronix specify it for active probes as tangential noise, which runs about twice the RMS noise figure. On small signals close to the scope’s own floor, that contribution is real and it’s worth checking the number before you blame your design for a noisy rail.

Cost, and the fact that it can break

A passive probe comes in the box. An active probe is a serious purchase, and Electronic Design’s treatment of the trade doesn’t dance around the price. It also plugs into a probe interface, so it has to be the right one for your scope.

Propagation delay you have to deskew

The TAP3500 specifies 5.3 ns of propagation delay. A passive probe on the next channel has a different one. If you’re measuring timing between two channels — setup and hold, a clock-to-data relationship, anything where the answer is a time interval — you must deskew, or the difference between the probes will show up in your result and look exactly like a real timing error.

The third option people forget: the resistor-divider probe

There’s a class of passive probe that gets almost no bench time despite being both the cheapest and, in the right situation, the fastest thing you own. It goes by several names: Z₀ probe, transmission-line probe, low-impedance probe, resistor-divider probe.

It’s a resistor at the end of a piece of 50 Ω coax, terminated in the scope’s 50 Ω input. A 450 Ω series resistor plus the scope’s 50 Ω gives 500 Ω total, so the divider ratio is 10:1 and the node sees 500 Ω. Electronic Design put the useful range at 450 Ω to 950 Ω for 10:1 and 20:1 versions. Tektronix note that these “50 Ω probes” reach several gigahertz with rise times of 100 ps or better.

What makes it fast is that there’s essentially no capacitance to speak of — just the resistor’s own parasitics — and no unterminated cable hanging off the node.

What makes it dangerous is that the load is now resistive and permanent. 500 Ω to the probe’s reference, at DC as well as at 5 GHz. On a 50 Ω transmission line or a stiff CMOS driver, that’s a tolerable insertion loss you can calibrate out. On a high-impedance node, a bias network, or anything current-starved, it’s a short circuit with a nice name.

The three columns aren’t a ranking. They’re three different answers to “what can this node afford to give up?” — capacitance, DC current, or bandwidth.

The ground lead ruins all of them

Here’s the part that matters more than the passive-versus-active argument, and gets a tenth of the attention.

Your probe needs a return path. That return path is a loop of wire, and a loop of wire is an inductor. That inductance sits in series with the probe’s input capacitance and forms a series resonant circuit whose only damping is the probe’s own input resistance — which, at 10 MΩ, is no damping at all.

The frequency is just the series resonance:

Tektronix state the result for the standard case: an 11 pF passive probe with a six-inch ground lead rings at about 140 MHz. SiTime measured the same physics independently on an oscillator output — a 16 pF probe with a ground loop of roughly 200 nH produced a 13 dB peak at about 90 MHz, which is exactly what the formula gives. Two vendors, two setups, one equation.

Two consequences that engineers get wrong in opposite directions.

On a slow scope, the ringing hides. Tektronix make the point precisely: at 140 MHz, a 100 MHz scope simply can’t see it, so the trace looks clean. Put the same probe with the same lead on a 200 MHz scope and the ringing appears — and people blame the faster scope, or worse, blame their board. The ringing was always there. You just bought an instrument good enough to show it to you.

Lower tip capacitance pushes the resonance up and out. That same 120 nH loop with 0.8 pF instead of 11 pF resonates near 500 MHz instead of 140 MHz. This is the underrated benefit of a low-capacitance probe, and it’s why the TPP datasheet advertises being “more forgiving of longer ground leads,” and why Tektronix note that active probes tolerate longer ground leads than passive ones.

The practical version of all of this is short and unglamorous. Bin the six-inch alligator lead for anything fast. Use the ground spring that came in the probe kit and lands on a nearby ground pad millimetres from the tip. Where you’ll probe a point repeatedly, fit a proper test point — the TPP series ships an MMCX tip cartridge that mates to a connector on the board, which brings the shield right up to the node and makes the ground loop essentially zero. SiTime’s guidance is the same: contact the probe’s ground collar rather than running a separate wire, and keep the loop small.

Differential probes, briefly

Plenty of what you actually want to measure isn’t referenced to ground at all — a serial data pair, a current-sense shunt, a gate drive floating on a switch node.

There’s a tempting shortcut: two single-ended probes and the scope’s A−B maths. Tektronix call it “usually the least desirable method,” and their reasons are worth knowing. You have two long, separate signal paths through two channels, so any delay mismatch between them shows up in the difference. Worse, the common-mode rejection of a two-channel subtraction falls apart with frequency, because it depends on the two channels matching. The result is a signal that looks far noisier than it is.

A differential probe does the subtraction in one amplifier at the tip, before anything can go wrong. Rohde & Schwarz’s range runs 200 MHz to 4.5 GHz at under 1 pF input capacitance. Most differential probes are active, which means they inherit the whole active trade — limited range, needs power, expensive — plus a common-mode rejection ratio you now have to read on the datasheet.

How I actually choose

There’s no ranking. There’s a sequence of questions, and the first one is about the node, not the probe.

1. What’s the highest frequency I care about? Not the clock rate — the harmonic content. Tektronix’s rule of thumb is that the probe bandwidth should be three to five times the fundamental, so the first few harmonics survive with their relative amplitudes intact. A 50 MHz square wave has significant content past 100 MHz.
2. What’s the node’s impedance there? If the probe’s $X_C$ isn’t at least ten times the node impedance at that frequency, I’ve already chosen: I need a lower capacitance probe.
3. What’s the DC voltage, and could it be worse than I think? Anything above about 10 V, anything mains-referenced, anything unknown — passive, and that isn’t negotiable. The CAT rating exists for a reason.
4. Can the node afford real current? If yes, and I need serious bandwidth, the Z₀ probe is cheap and superb. If no, it’s off the table entirely.
5. Is the signal ground-referenced? If not, differential probe. Not two probes and subtraction.
6. Only then: is my ground return short enough to not invent the very artefact I’m about to go and debug?

A few more things that are always true, regardless of which probe wins:

• Derate the bandwidth for amplitude accuracy. At its rated bandwidth a scope is 3 dB down, which is 30% of amplitude error. Tektronix’s guidance: for amplitude measurements within 3%, pick a scope with three to five times the bandwidth of your highest frequency of interest.
• Convert bandwidth to rise time when that’s what you’re measuring. $t_r \approx 0.35/BW$ below 1 GHz, with the constant creeping to 0.40–0.45 for faster instruments.
• Combine system rise times in quadrature — but only for active probes. $t_{r,sys}^2 \approx t_{r,probe}^2 + t_{r,scope}^2$. Tektronix explicitly say the relationship is more complex for passive probes and that this formula should not be used for them.
• “Bandwidth at the probe tip” is measured from a 25 Ω source. That’s the industry test setup — 50 Ω source terminated in 50 Ω. Real circuits are rarely 25 Ω, so expect less than the spec sheet in real use, especially on higher-impedance nodes.

The thing to walk away with

Passive probes trade capacitance for robustness, range and price. Active probes trade robustness, range and price for capacitance. Z₀ probes trade DC current for both. All three are correct answers to different nodes, and reaching for the same probe every time is how you end up debugging your own instrument.

But the honest summary is that most of the measurements I’ve seen go wrong weren’t lost on the passive-versus-active decision at all. They were lost on six inches of ground wire, or a probe nobody had compensated, or a 10 MΩ specification that stopped being true four decades below the frequency in question.

Know what your probe presents at the frequency you care about. Keep the loop short. Then trust the trace.