3dBi: I made a slight mistake in the post above. The A input triggers on edge, not level, so some specific outputs may be different. But the general message still holds: a race condition is present.
If you insist on using CD4538, you may try adding a resistor and a capacitor on the A line, so the signal is delayed a bit: Rd
V_ign o--/\/\/---o-------A[CD4538]
|
o--/\/\/-.
| Rb |
Cd --- |
--- |
| |
V V
Rd/Cd introduce a slight delay. Rb is a bleed resistor (1MΩ or so), so the capacitor is discharged after ignition is off. A bit of experimentation needed to tell, what values work. Will not work at all, at least not reliably, if CD4538 can’t deal with slow edges. The datasheet tells nothing about that. A Schmitt trigger may be required to address that issue.
You may also take advantage of low standby current of CD4538: in two dozen nanoamp range. So it may be permanently powered on, with ¬CD kept high. But that has two issues. First: it’s a typical value, not the maximum one. The maximum is in 1000× of that. Which is still not that much if you use your car regluarly, but also not something to completely ignore. For production you would need to test each device and discard those, which draw too much energy. For single implementation it’s of less concern, since most chips will be near the typical range, but you may get unlucky. The second issue is: voltage regulator will take some current too. So unless you already have that available and always on, this solution is a no-go.
But I would support Benta’s idea to choose a different solution.
Also, since this is automotive, remember that all lines — inputs, power supply and outputs — may see a lot of noise, including high voltage spikes. The design should take that into account.