TVS diodes and the surge
A TVS diode does nothing at all until the voltage goes somewhere it should not, and then it conducts tens of amps for a millisecond to stop it. The number that decides whether it saved your board is not the one in its part number.
Everything in this chapter so far has been about current going the wrong way or too much of it. A transient is the other failure: voltage, arriving faster than anything can react to, from a motor switching off in the next room, a long cable picking up a nearby lightning strike, or a finger on a dry day.
A transient voltage suppressor is a diode built to be destroyed slowly. Below its standoff voltage it is invisible. Above its breakdown voltage it conducts hard, and it is designed to take the whole pulse rather than pass it on.
Three voltages, and only one of them matters
Take the SMAJ5.0A, a 400 W part in an SMA package that costs a few cents:
- Standoff, 5.0 V. Below this it does nothing. Leakage in the microamps.
- Breakdown, 6.4 V minimum. Where it starts conducting in earnest.
- Clamping, 9.2 V maximum, at 43.5 A. What it actually puts on your rail while it is absorbing the pulse.
The name says 5.0. The number your board experiences is 9.2.
Switch the part being protected and send the surge again. A wide-input buck converter rated to 40 V does not care. A 5 V LDO with a 6 V absolute maximum survives. An ESP32 pin, absolute maximum 3.6 V, does not — and nothing about the TVS working correctly changes that.
Choose a TVS by its clamping voltage against the absolute maximum of what is behind it. Not by the number in its name, which is the voltage it ignores. Those two figures are nearly a factor of two apart on every TVS made, and the gap is where the mistakes live.
The window
Two inequalities, and people get the first right and the second wrong.
The standoff voltage must clear your rail. Including its tolerance and its ripple. If the rail ever reaches the standoff voltage the TVS conducts a little, warms, conducts more, and eventually fails — not dramatically, just as a board that got hot on Tuesday for no reason.
The clamping voltage must fit under the absolute maximum. If it will not, either pick a lower TVS — which the first rule may not allow — or put something in series to share the pulse. A resistor, a polyfuse, a ferrite, an inductor: anything with impedance turns some of the transient into a voltage across itself rather than across your chip.
Where they go, in practice
On a DC input jack. A unidirectional TVS from the positive rail to ground, right at the connector. This is the classic use and it is worth doing on anything with a cable longer than a metre.
On data lines that leave the board. RS-485, CAN, a long I2C run. These need a bidirectional part (the "CA" suffix on an SMAJ), because the signal itself swings both ways.
On USB D+ and D−. A dedicated low-capacitance ESD array rather than a power TVS. A 400 W SMA part has tens of picofarads of capacitance across it, which on a 480 Mbit/s data line is a short circuit as far as the signal is concerned. Most ESP32 boards already have one of these next to the socket.
What it is not
A TVS is not a regulator and not a fuse. It absorbs energy for about a millisecond and it does that by getting extremely hot very quickly. A sustained overvoltage — an 18 V adapter in a 5 V socket — puts it into conduction and keeps it there, and it will short itself out permanently, which takes the rest of the board down with it or blows the input fuse, depending on what else you fitted.
That failure mode is deliberate and it is usually the right one: a shorted TVS is a board that needs a repair, and a board that needed a repair is better than a board that caught fire.
Edit this page — content/fundamentals/power/tvs-and-the-surge.mdx
Discuss this article
Ask about this page. The answer stays here, on the page it belongs to, for whoever hits the same wall next.