Protection circuit for devices powered on a car

[Foaly]

Careful with the LTC surge stoppers, they have weaknesses.
Jr. engineers designed them-in on 24V truck-powered systems and products started failing.
Traced it down to -ve voltage transients destroying the IC and mosfet. I wouldn't use them again.

Instead I use 6.6kW automotive TVS found in ECM's: 8SMA27.
It's big enough to take load-dump and clear a 15A fuse if it shorts.

The schematic I copied from the datasheet features a few additional protection diodes to protect the LTC and the MOSFET on the front line from big surges (greater than the ~100V that they can withstand). I attached a screenshot. From what I understand that should make the whole thing pretty robust even to large surges, and in the worst case scenario the fuse should blow up before any damage is done.
The problem with diodes is that they are not precise enough for my application, I only have a few volts of margin between the normal voltage and the maximum voltage admissible by the load. This figure from Linear shows exactly what I needed (with 16V clamp instead of 27V) :

I wonder how things are going to change now that more and more cars are using switching power supplies to supply the 12V rail.
http://techno-fandom.org/~hobbit/cars/ginv/i1mech.html

I guess that would be the end of load dumps? Your article clearly states that the alternator is not connected to the 12V rail, but I'm not sure if there are other sources of surges.

[Foaly]

So I finished this project and I thought I'd share the result in case someone stumbles upon this post with the same problematic.

The PCBs came out great from OSHPark (as always), I ordered the 0.8mm 2oz version to improve current capacity and heat transfer. I found two flaws in the design :

1/ First, the FDB33N25 that I used for the front-line MOSFET had a Rdson too high (87mOhm) and without a heatsink I couldn't pass more than 4A through it before it started to heat up way too much. After looking that up I understood that the high Rdson is due to the high Vds_max by construction, so I replaced both high-side FETs with two IPB017N08N5 (Vds 80V, Rdson 1.7mOhm) which were the same package and pin-compatible. Even at ~13A (the circuit is designed to cut at ~15A) the new FETs wouldn't go higher that ~45°C still without a heatsing, which was way more acceptable.
Of course that left the circuit susceptible to dumps higher than 80V because of the lower Vds of the new FETs (and I wasn't sure so I checked, obviously they fail closed-circuit). To prevent this I changed the 1.5K200A TVS diode on the input to a 1.5KE51A which should protect the transistor and not conduct (=blow the fuse) too easily.
So this one is solved.

2/ Secondly, I didn't take into account the hysteresis of the UV/OV pins. My battery is ~12.7V when charged and I set the trimmer of the UVLO to cut below ~12.4V to prevent overdischarge. However, because of this hysteresis, 12.7V wasn't enough to enable the chip from a disabled state and I needed to start the engine (to boost the input to 14V) to enable it (after that, I could stop the engine and it would still work at 12.7V down to 12.4V as it was supposed to).
My first idea was to lower the impedance of R1 (the resistor on the high side of the divider of UV) for a few undreds of milliseconds after power up by putting a capacitor across it. I did some simulations and tests on a breadboard, but there was something else to take into account : boosting the UV voltage this way would also boost the OV voltage, and the chip would lock itself up on the overvoltage condition. I fixed this problem by using another capacitor on the low side of the divider, with a higher value than the first one, to allow UV to rise faster than OV. Simulations and oscilloscope measurements showed it worked well in theory : in practice, the "spike" on the UV pin caused by the fist capacitor was too high and too narrow, so I simply put a resistor in series to fix both of these issues. Soldering these additional components across the existing ones on the pcb I had was not what you would call state-of-the-art, but it works well.

Finally, I mounted the circuit on a piece of copper using thermal tape to act as a heatsink and ground link, 3D-printed a case, and plugged everything on the wire harness I had already installed in my car. The whole system works really well, the only problem I have is (as was discussed earlier in this topic) that cranking, or even preheating the Diesel engine, resets the output due to the UVLO. But that's not really a big issue, and anyway I was already planning another circuit (a power multiplexer) that will be able to solve this in case I need it.