Zener Reverse-Current Ratings in H-Bridges: The Freewheeling Path Nobody Sizes
Size an H-bridge Zener freewheeling path by repetition and official IZM, not peak current. 1N4749A allows 38 mA; the family PDF has no pulse curve.

The rating that usually kills a Zener in an H-bridge is not the peak current everyone writes on the schematic. It is the average dissipation from events that keep repeating. Official Good-Ark numbers make that hard to miss. The 1N4749A 24 V, ±5% zener, in the same 1 W DO-41 family as 1N4733A, has a 38 mA maximum zener current. At 24 V that is already about 0.9 W, which is the entire official dissipation class at a 75 °C lead temperature. A labeled model that dumps 0.5 J ten times per second is 5 W of average power. That is not a 1 W zener problem. It is a part-family problem. The same official file also does not publish a peak-pulse-power curve, so a single “modest” event cannot be checked the way a TVS is checked against a 10/1000 µs row.
(Last modified date: November 18, 2026)
An H-bridge has two current paths: the switching path that is usually sized first, and the freewheeling or clamp path that absorbs or returns energy when a winding is released. A discrete freewheel diode or a MOSFET body diode conducts forward, often back into the supply rail. A Zener placed to clamp the coil conducts in reverse breakdown and turns that energy into heat in the package. The two jobs use different ratings, and a part chosen only on a forward or “24 V” headline can fail on the reverse-current duty, the repetition, and the heat in the loop.
The freewheeling-path checklist before a Zener is approved
Work these checks in order. The first failure in the list is the reason not to freeze the part, even if a bench pulse looked acceptable.
- Name the path and the polarity. Is the device a forward freewheel diode (or body diode) back to the rail, a Zener (or zener-plus-diode) clamp across the coil, or both, depending on the bridge state?
- Separate inductive energy in the winding, ½ LI2, from any mechanical energy the motor still has when the torque direction changes. Kinetic energy can be much larger than the inductance term, and it must not be ignored in a braking or reverse event.
- Write the repetition rate, not only the peak current. Convert stored energy per event times events per second into average power, and state the assumptions.
- Compare that average power, and the current that produces it, with the official, derated dissipation and the maximum zener current IZM at the actual ambient and mounting.
- Open the official file and look for a pulse-power or surge waveform rating. If the file does not publish one, do not treat the zener as a TVS. The 1N4728A family file used below does not publish a 10/1000 µs or similar pulse-power row.
- Check the opposite polarity. In one direction the clamp may be in breakdown; in the other it is an ordinary forward diode. The residual voltage at the peak current appears across the switch. The 1N4728A family file does not publish a forward-voltage figure, so do not invent one at motor current.
- Confirm the thermal path under the real drive pattern: case or board temperature after the events have been repeating long enough to settle, not after a single bench pulse.
- If average dissipation or an unpublished pulse energy is outside the zener class, move to a device that publishes a pulse curve, or change the clamp architecture so the energy returns to the supply instead of being burned in a small package.
Why repetition, not the event peak, usually sets the part
A single inductive event deposits energy E in the clamp. If the zener holds a breakdown voltage VZ for a discharge time t, a rough average current during that event is E / (VZ × t). That arithmetic is a model: real current is a decaying waveform set by inductance, resistance, and the clamp voltage, not a rectangle. The official limit that must then be checked is not a guessed peak, but whether the package can shed the heat when the event repeats.
Average power is E times the event rate. A labeled 0.5 J event at 24 V, if it were discharged in 5 ms, has a model average of about 4.2 A during the event and about 100 W of instantaneous dissipation while it lasts. That already cannot be checked against the 1N4749A file, because the file publishes 1 W at TL = 75 °C and IZM = 38 mA, and no pulse-energy rating. If the same 0.5 J were repeated ten times per second, the average power would be 5 W, more than five times the official 1 W class even before ambient derating. A cooling-fan duty that repeats much less often, or stores much less energy, may still fit a derated 1 W part. A steering-assist or PWM duty that repeats many times per second usually will not. The original “single event is modest, repetition is not” intuition is right; the missing step is to stop treating an unpublished pulse as if the part had already passed it.
| What is being checked | Official 1N4749A figure | What a labeled 0.5 J model does |
|---|---|---|
| Continuous dissipation | 1 W at TL = 75 °C; typical RθJA 170 °C/W; TJ 175 °C | One event every 2 s is 0.25 W average, inside the class after derating only if the mounting matches the note. Ten events per second are 5 W average, outside the class. |
| Maximum zener current IZM | 38 mA | A 5 ms discharge at 24 V is a model average near 4 A during the event, two orders of magnitude above IZM. IZM is a continuous maximum, not a pulse rating. |
| Pulse-power or surge waveform | Not published on this official file | There is no official 10/1000 µs or 8/20 µs row to compare the event with. Do not fill that gap with a TVS number from another family. |
| Voltage window at the test current | 22.80–25.20 V (typical 24 V) at 10.5 mA; ZZT 25 Ω maximum | The clamp voltage at a much higher current is not the 10.5 mA test point. Dynamic resistance raises the voltage, and the official file does not give a high-current clamp curve. |
The 1N4728A thru 1N4764A datasheet (Rev. B) is the source of those official rows. It is a 1 W, 3.3 V to 100 V family in DO-41 (DO-204AL), with ±5% voltage tolerance, a typical 170 °C/W junction-to-ambient path, and a 175 °C junction limit. The ratings note says they are valid provided the leads, at 9.5 mm from the case, are kept at ambient temperature. A first-order ambient derating using (175 °C − TA) / 170 °C/W leaves about 0.62 W at 70 °C, or roughly 26 mA at 24 V, below the 38 mA IZM headline. The 5.1 V member of the same file, 1N4733A, has a higher 178 mA IZM only because the voltage is lower; it is not a 24 V clamp. Suffix and test-current language is in Zener datasheet language. Voltage steps in this family are mapped in matching Zener part numbers from 3.3 V to 12 V and in 1N4728A in practice.

DO-41 (DO-204AL) is the official outline for 1N4749A. The product page is 1N4749A, and listings start at the Zener catalog. A 350 mW SOT-23 comparison class, which is even smaller on continuous dissipation, is the AEC-Q101 qualified ABZX84Bxxx family; ABZX84B24 is the 24 V member of that file, rated 350 mW. Thermal-path arithmetic is covered in the rectifier thermal design guide.
SOT-23 is the official outline on the ABZX84B12 product page, a listed member of the same ABZX84B file. Do not treat this 350 mW class as a substitute for a 1 W clamp, and do not treat either class as a TVS.
Forward freewheel, Zener clamp, and the voltage the switch actually sees
A plain freewheel diode clamps the released winding close to one diode drop, usually back into the rail. The energy is not all burned in the diode; some of it is returned to the supply, and the current decays more slowly because the voltage across the inductance is small. A Zener clamp is chosen precisely because it holds a higher reverse voltage, so the current decays faster and the overshoot is bounded. That speed is paid for as heat in the zener: power during the event is roughly VZ times the decaying current. Raising the official clamp voltage shortens the event and raises the voltage the switch and the insulation must stand. Lowering it lengthens the event and reduces the spike. The trade is real, but it does not create a pulse rating that the datasheet omitted.
The opposite polarity still matters. When the bridge drives the winding the other way, the same device may conduct forward. The forward drop at the peak current is then the residual voltage across the switch. Because the official 1N4728A file does not publish VF, that drop has to come from a measurement or from a different official file that actually states it. The BZT52Bxx datasheet and the ABZX84B file both list 0.9 V at only 10 mA, which is not a motor-current figure and must not be scaled as if it were. A clamp that protects the coil but leaves the transistor exposed has only moved the problem. The broader zener-versus-surge distinction is in TVS vs Zener and in Zener, TVS, and ESD diodes on a DC rail.
Fan, seat, and steering duties, read against the same official limits
The same checklist maps onto three different dominating terms. A small cooling-fan motor that coasts occasionally may store little energy and repeat at a modest rate; if the average power and the current stay inside a derated 1N4749A, a 24 V 1 W zener can be a reasonable clamp. A power-seat motor that moves rarely but with high current is a high-energy, low-repetition case. The official file still does not give a single-event joule rating, so a large rare pulse is not “automatically acceptable” just because the average power looks small; it is a reason to use a part with a published pulse curve, or to confirm the waveform on the bench against a rating that actually exists. A steering-assist or other high-rate duty that repeats many times per second is usually an average-power problem: the 5 W model above is already outside the 1 W class, and PWM at a still higher rate is worse. Knee, breakdown, and the three datasheet checks are summarized in Zener diodes in plain language.
The closing rule is the same as the opening one. The freewheeling path is a real current-carrying path. Size it for polarity, stored energy, repetition, derated dissipation, IZM, and whatever pulse rating the official file actually publishes. A part sized only on a forward headline or a single bench pulse can survive the lab and fail in the first week of the real drive pattern.
FAQ
Does a Zener conduct forward or in breakdown in a clamp circuit?
Both, depending on the polarity. Across a coil, the flyback event drives it into reverse breakdown, and that is the clamping action. In the opposite polarity it behaves as an ordinary forward diode. The official 1N4728A family file does not publish a forward-voltage figure, so the drop at motor current has to be measured or taken from a file that states it.
What decides whether a clamp diode survives repeated flyback?
The repetition rate, converted into average dissipation, compared with the derated package capability at the real ambient. Peak current alone is not enough. A high-rate PWM or steering duty loads the same part very differently from an occasional seat-motor event.
How do I size a Zener clamp for a motor H-bridge?
Start from the stored energy per event, including mechanical energy if the motor is reversing, and from the event rate. Convert that into average power and current. Compare those figures with official, derated dissipation and IZM. Then look for a published pulse rating; if the file does not have one, do not assume TVS-like surge capacity.
Why does the forward voltage still matter in a clamp path?
Because the device may have to conduct in the opposite polarity, and the drop at the peak current appears across the switching device. A clamp that protects the coil but exposes the transistor has moved the problem rather than solved it.
Does the official 1N4728A family file include a pulse-power rating?
No. The Rev. B file publishes dissipation, IZM, test-current voltage windows, and dynamic resistance, but not a 10/1000 µs or 8/20 µs peak-pulse-power row. A single-event joule figure cannot be checked against that file the way a TVS can be checked against its waveform rating.
.png)
