Does a TVS Fail Short or Open? Why One Shorted TVS Can Take Down the BMS MOSFETs and the Whole Pack
Intek Technology (INTEK, www.intek.vip) is an authorized distributor of AMSEMI. Datasheets, samples and FAE review are available for every standard and foldback -N 5.0SMDJ part referenced below.
One-sentence conclusion: a TVS fails short, not open - it turns into a resistor of tens of milliohms and stays connected where it was. What that costs depends entirely on which side it sits: across the cell tabs, inside the MOSFETs, the fault loop closes locally between the battery and the TVS, the BMS sees nothing and cannot switch it off, and the whole pack drains itself; at the PACK terminal, outside the MOSFETs, the fault current has to cross the MOSFETs and the shunt before the BMS gets a chance to act. Where the TVS sits decides the safety margin after failure more than how big it is - and the main fuse takes part in neither.
flowchart TD
A["Failure starts<br/>surge overload or long-term slight conduction"] --> B["Junction temperature reaches the limit<br/>local silicon melting"]
B --> C["Low-resistance eutectic path forms<br/>part becomes tens of milliohms (short)"]
C --> D{"Which side is this TVS on?"}
D -- "Cell tabs<br/>(inside the MOSFETs)" --> E["Fault loop closes locally<br/>bypasses MOSFETs and shunt"]
E --> F["No signal reaches the BMS<br/>no trip, no alarm"]
F --> G["Pack drains continuously<br/>cells over-discharged beyond recovery"]
D -- "PACK terminal<br/>(outside the MOSFETs)" --> H["Fault current crosses MOSFETs and shunt<br/>enters the overcurrent window"]
H --> I{"Is protection fast enough?"}
I -- "trip within MOSFET withstand" --> J["Back-to-back MOSFETs block both ways<br/>fault reported, system stops"]
I -- "slower than MOSFET withstand" --> K["MOSFETs destroyed first<br/>same event as the TVS"]
D -- "Sampling line<br/>(with series resistor)" --> L["Current limited to milliamps<br/>shows up as a bad cell reading"]1. Why the failure is nearly always a short
A TVS is a silicon avalanche device, and it fails thermally: overcurrent or long-term slight conduction drives the junction temperature to its limit, the silicon melts locally, and the molten region forms a low-resistance eutectic path with the metal layer. When it cools, that path stays. The part does not return to being a diode, and it does not become an open circuit - it becomes a resistor of tens of milliohms.
Open-circuit failure exists, but on a different path: the bond wire melts first and the energy is too low to melt silicon. That combination needs a very small current for a very long time, which the BMS main circuit practically never delivers. Solder fatigue and moisture ingress can also open a part, but that is a reliability mechanism rather than an overload failure.
The design consequence is direct:
The protection assumption must be written as "it will become a short", not "it will disappear". Most failure reviews only ask how much surge the TVS can absorb, never what the board does after it cannot. That gap is where the complaint "fitting a TVS made it less reliable" comes from - those boards did not get protection, they got several potential short circuits.
The attribution method for blown-MOSFET cases is covered in the four-check root-cause method and in the layer-by-layer failure analysis. This article answers the later question: it has already shorted - who cuts the current now.
2. The key rule: the side it sits on decides whether the BMS can save the pack
This is the table worth remembering. Same VRWM 54 V part, same short, same pack - a different position, an order of magnitude difference in outcome.
| TVS position | Fault current through MOSFETs and shunt? | Can the BMS detect and switch off? | Typical outcome |
|---|---|---|---|
| Cell tabs (BAT+ to BAT-, inside the MOSFETs) | No. The loop closes locally between battery and TVS | No, no signal at all | Pack drained through tens of milliohms, cells over-discharged, board carbonised, whole pack lost |
| PACK terminal (PACK+ to PACK-, outside the MOSFETs) | Yes, must cross the MOSFETs and the shunt | Yes, enters the overcurrent / short-circuit window | Fault reported and system stopped; one TVS and possibly the MOSFETs lost |
| AFE sampling line (series resistor, then TVS) | Yes, but limited to milliamps by the resistor | Not directly, but exposed as a sampling fault | One cell reading offset, balancing misbehaves, board survives |
Three points:
First, the cell-tab position is the most dangerous, and it looks the most reasonable. "A TVS should sit as close to the battery as possible" is true for the surge injection point and exactly wrong for the failure outcome. Surge enters at the PACK terminal; mounting the clamp at the cell tabs makes the surge travel extra copper and MOSFETs before it is clamped, raising the residual voltage - and when the part shorts, the fault loop conveniently bypasses every component that could have detected it. A surge-protection TVS belongs at the entry of the port it protects, which means the PACK side.
Second, switching off only cuts the current if the MOSFETs are back-to-back. A single MOSFET, once off, still conducts in one direction through its body diode. With a shorted TVS, the fault current keeps flowing through that body diode, and hundreds of amps will destroy the diode before the die. Only a charge MOSFET and a discharge MOSFET in back-to-back series, with opposing body diodes, both off, blocks both directions - that is what "cutting the current" actually means. It is also a physical reason BMS designs use back-to-back devices, beyond reverse-connection protection.
Third, where the part must sit at cell level, the series resistor does the work, not the BMS. Cell sampling lines are the classic case: the clamp has to sit at the sampling input, so the series limiting resistor is the only effective buffer - it drops the fault current from kiloamps to milliamps and turns "burnt board" into "one bad cell reading". The constraints on leakage and junction capacitance are covered in does a BMS sampling line need a TVS.
3. Fault current is kiloamps, not leakage
People often picture a shorted TVS as a small leakage that slowly drains the battery. That is half true - the slow half only happens where the current has been limited.
A shorted TVS on a 54 V class pack sees a total loop impedance made of the short's residual resistance, copper and traces, two MOSFETs in conduction and the shunt, in the tens of milliohms. At 20 milliohms: 54 V / 0.020 ohm is about 2700 A. Cell internal resistance, harness and contact resistance pull that down, but it still lands in the hundreds of amps to kiloamp range.
On the time axis:
| Element | Order of magnitude | Note |
|---|---|---|
| MOSFET short-circuit withstand (SCSOA) | Tens of microseconds | The window is narrow at low voltage; follow the SOA curve in the datasheet |
| BMS short-circuit response | Hundreds of microseconds to milliseconds | Includes sampling filter, decision delay and gate turn-off |
| Main fuse melting | Milliseconds to seconds (faster end at high current) | Selected for rated current; strongly current dependent |
The conclusion is hard: the MOSFET is destroyed before the BMS has reacted. What the field reports as "the TVS blew and the MOSFETs blew" is not two incidents but two casualties of one event - which went first is unobservable and irrelevant. Treating "replace the TVS" as the repair is how the same board comes back a second time.
One datasheet fact worth stating: no TVS datasheet specifies how long it survives a short. Ipp and Ppp are single-pulse figures; continuous short-circuit withstand is not a parameter. The problem must be solved at system level and cannot be solved by choosing a tougher TVS.
4. Why a main fuse cannot help at either end
"Just add a fuse" is the usual answer. Work the numbers and it is clear why that fails - the fuse is squeezed between two contradictory requirements.
Requirement one: it must not melt on a legitimate surge. The fuse melting I2t must exceed the TVS single-pulse I2t. For a 5.0SMDJ54CA-N with Ipp 57.5 A at 10/1000us, the single-pulse I2t works out to the order of 2 A2s on an exponential-decay estimate (an engineering estimate, not a datasheet value).
Requirement two: it must melt fast when the TVS shorts. A 100 A class main fuse selected for load current has a melting I2t in the tens of thousands of A2s (follow the manufacturer I2t curve).
| Item | I2t order of magnitude | Relation to one TVS surge |
|---|---|---|
| 5.0SMDJ54CA-N, one 10/1000us surge | About 2 A2s (estimate) | Baseline |
| 100 A class main fuse, melting I2t | Tens of thousands of A2s | Three to four orders higher: blind to the surge |
| Actual fault current with a shorted TVS | Hundreds of amps to kiloamps | Melting still takes milliseconds: slower than the MOSFET window |
It helps at neither end. The fuse is designed to protect the harness and the cells; its scope never included a diode a few millimetres across. Expecting it to cover the TVS pulls a device two orders of magnitude away into the same protection chain.
What about a resettable PTC? Its boundaries are narrower still:
- It trips thermally, taking seconds, and still leaks after tripping - it does not fully open;
- Its I2t is smaller than a one-shot fuse, so a legitimate surge trips it, and it has thermal memory - after repeated trips the resistance drifts up, raising loop impedance and hurting efficiency;
- On a BMS it belongs at module or cell level as a temperature and overcurrent buffer, not as the main-circuit surge and TVS protection device.
The real division of labour on the main circuit - contactor coil flyback, bus, precharge, gate supply - is broken down in five transient types on the BMS main circuit.
5. The three real defences
Since the fuse cannot do it, three defences remain, and the order matters.
One: position. Surge-protection TVS at the PACK terminal, never across the cell tabs; sampling-line TVS with a series resistor. This costs nothing but one decision at layout time, and it is the only measure that converts "unrecoverable after a short" into "recoverable after a short".
Two: derating, to remove the most common starting point. In field data, most shorted TVS parts were not killed by a single surge - they died of heat from long-term slight conduction. VRWM was picked too low, so at full charge, during balancing, on a single overcharged cell or after a wrong-charger event the part sits at the edge of breakdown; leakage climbs exponentially with temperature (commonly described as doubling per 10 degrees C) and the accumulated heat ends in a crystallised short. The grading rule is explicit:
VRWM at least the maximum bus voltage times 1.15, then check VBR and Vc at or below 0.85 times the downstream measured typical withstand. That three-stage window is worked grade by grade for 23S and 24S packs and for 13S to 22S two-wheeler packs. Leakage itself is treated in TVS leakage and quiescent current - that article covers leakage that is within spec but eats standby time; this one covers the same curve's later stage, where leakage ends in a short.
Three: monitoring, to make an invisible short visible. A shorted TVS at the battery side shows up as excessive self-discharge at rest. Three layers are available:
- Sleep-state quiescent current monitoring: after the BMS sleeps, the residual current should be in the microamp range; a jump to milliamps is an alarm;
- Pack self-discharge rate comparison: leave a batch at rest for seven days and start with the unit whose voltage falls fastest;
- End-of-line AOI plus a pre-power impedance spot check at the TVS locations, to keep bad joints and shorted incoming parts off the line.
6. Field diagnosis in four multimeter steps
When a TVS is suspected of being shorted, follow this order and save yourself a detour.
Step one: power down and lift one leg. In-circuit measurement is always skewed by the surrounding network - shunt, balancing resistors and AFE input impedance all pull the reading. At minimum lift one pad.
Step two: use the resistance range, not the diode test. A good bidirectional TVS reads above one megohm in both directions (use the 20 megohm range; many diode tests cannot resolve that). Near zero in both directions means it has failed short; infinite in both means open (rare).
Step three: for a unidirectional part, respect the direction. About 0.6 V forward is normal, above one megohm reverse; conduction in reverse means failure.
Step four: separate cell self-discharge from a shorted TVS. At pack level the two look alike but the physics differ: a shorted TVS is a metallic path, so the voltage falls at a nearly constant rate and barely responds to temperature; cell self-discharge is electrochemical and clearly temperature dependent. The direct route is to disconnect the BMS from the cells and measure the TVS alone - one minute settles it.
The step most often skipped: after removing the shorted TVS, check the MOSFETs. At minimum check the body diode for damage from the reverse fault current, Rds(on) against a fresh part, and the gate for breakdown. Replacing only the TVS without checking the MOSFETs is the most common source of repeat failures in this repair.
7. Selection: low clamping gains a second meaning here
Lower clamping is usually framed as headroom for the downstream MOSFET. In this context it has a second meaning: at the same surge current, a lower Vc means less dissipation inside the TVS itself, a lower junction temperature, and more distance from the crystallised short.
Take the 54 V grade (10/1000us convention, datasheet 8060014 for the standard part, 8060055 for the foldback -N part):
| Item | AMSEMI 5.0SMDJ54CA (standard) | AMSEMI 5.0SMDJ54CA-N (foldback) |
|---|---|---|
| VRWM | 54 V | 54 V |
| VBR (min to max) | Identical in this grade | 60.0 to 66.3 V |
| Maximum Vc at Ipp | About 87.1 V | 69.6 V |
| Ipp (10/1000us) | 57.5 A (5 kW tier) | 57.5 A (5 kW tier) |
Same VBR, same Ipp within the grade, Vc about 20 percent lower. The three power tiers (3 kW, 5 kW, 8 kW) share identical VBR and Vc at a given voltage grade and differ only in Ipp (54 V grade: 34.4 / 57.5 / 91.8 A) - stepping up the power tier never lowers the clamp, it only makes the part harder to destroy, which under a short-circuit risk framing is a meaningful reason to choose it. The full range and cross-reference are in the AMSEMI foldback TVS model list.
Two convention reminders:
- AMSEMI 5.0SMDJ figures (8060014 and 8060055) are all 10/1000us; the datasheets contain no 8/20us data. Littelfuse 8x20us figures belong to the Littelfuse convention and the two must not be converted or subtracted; compare percentage reduction only.
- For automotive projects use the 8000 W tier 8.0SMDJ**CA-N (VRWM 48 to 90 V, 11 grades, AEC-Q101, -H suffix on automotive part numbers). That tier stops at 90 V; the 95 V and 100 V grades exist only in the 5 kW series.
FAQ
Q1: Does a TVS diode fail short or open?
Almost always short. The failure is thermal: overcurrent or long-term slight conduction drives the junction to its limit, the silicon melts locally and forms a low-resistance eutectic path with the metal, leaving a resistor of tens of milliohms in place. Open failure needs the bond wire to melt first with too little energy to melt silicon, which is rare. With one leg lifted, a good bidirectional TVS reads above one megohm both ways; near zero both ways means shorted.
Q2: Can the BMS detect a shorted TVS and protect the pack?
Depends on the side. Across the cell tabs, inside the MOSFETs, the loop closes locally, no current crosses the MOSFETs or the shunt, the BMS gets no signal and the pack drains until the cells are irreversibly damaged. At the PACK terminal, outside the MOSFETs, the fault current must cross the MOSFETs and the shunt, which is the only way the BMS can trip. Position decides the safety margin after failure more than power rating does.
Q3: Why do the MOSFETs blow after a TVS shorts, and why does the fuse not blow?
One event, not two. A shorted TVS on a 50 V class pack sees tens of milliohms, so the fault current is hundreds of amps to kiloamps; MOSFET withstand is around ten microseconds while BMS response is hundreds of microseconds to milliseconds, so the MOSFET dies first. A main fuse sized for load current has a melting I2t in the tens of thousands of A2s against roughly 2 A2s for one 5.0SMDJ54CA-N surge at 10/1000us - blind to the surge, too slow for the short.
Q4: How do you stop a shorted TVS from destroying the whole pack?
Three defences: position (surge TVS at the PACK terminal, never across the cell tabs; series resistors on sampling lines to limit fault current to milliamps), derating (VRWM at full-charge voltage times 1.15, so the part never sits in slight conduction), and monitoring (sleep-state quiescent current and pack self-discharge trending). When replacing a shorted TVS, always check the MOSFET body diode, Rds(on) and gate.
Contact Us
TVS protection has two halves: the surge arrives and the part holds, and the part does not hold and the board still lives. Almost all industry discussion stops at the first half, while the money is lost in the second - a diode costing cents takes a full set of cells and a board with it.
Send us your cell count and chemistry, full-charge and balance voltages, current MOSFET platform and part numbers, TVS mounting position (cell side / PACK side / sampling line), and the field failure signature (discharge rate, teardown photos), and we will work three things in one pass: the TVS grade with the VBR/Vc double-gate check, the safety margin of the failure position, and the series-resistor and TVS values for sampling lines. Datasheets, samples and FAE review for AMSEMI 5.0SMDJ standard and -N foldback parts are available - see our contact page.
Shenzhen Intek Technology Co., Ltd - authorized distributor of AMSEMI TVS and HOPERF CMT digital isolators Phone / WeChat: 136-3264-8484 Address: Room 4 A11, Building 2, No. 309 Ping'an Avenue, Pinghu Subdistrict, Longgang District, Shenzhen Website: www.intek.vip
Disclaimer: This article is technical content for industry communication and does not constitute selection, purchasing or other commercial advice. The fault-current magnitude, the I2t magnitudes, the MOSFET short-circuit withstand time and the BMS response time are engineering order-of-magnitude estimates used to convey scale, not specifications of any particular device; actual values follow the cell, MOSFET, fuse and BMS documentation and system-level testing. No TVS datasheet specifies continuous short-circuit withstand. AMSEMI 5.0SMDJ figures (datasheet 8060014 standard / 8060055 foldback -N) follow the 10/1000us convention with no 8/20us data; Littelfuse 8x20us figures belong to the Littelfuse convention and the two must not be converted or subtracted. The rules "VRWM at least maximum bus voltage times 1.15" and "Vc at or below 0.85 times downstream measured typical withstand" are engineering conventions, not mandatory standards. Work on battery packs and high voltage must be performed by qualified professionals, and a failed pack should be isolated and assessed before teardown.