1. How accurate do you really need this time?
Not very. 2 minutes to 6 minutes would be fine ... it's just a soft-power switch.
2. How accurate is the actual threshold? I'm guessing it's not 2.00V precision. (Also, does it have hysteresis?)
No, not very. The spec is 0.6V max off, 2.0V min on for the logic levels. It doesn't state hysteresis and I imagine it might not have it, but poor behavior in the middle region is okay (except if the output voltage went above spec, which it shouldn't be allowed to do.)
3. How accurate is the internal current sink? Do they specify limits?
Double-checked that:
At 0.6V, -1.0µA min, 0.01µA nom., 1.0µA max.
At 2.0V, 0.1µA nom., 1.0µA max.
At 24V, 0.5µA nom., 2.5µA max.
So yeah, not very tightly specified. If the voltage cutoff works, a much larger capacitor with a high-value bleeder resistor might be a better choice.
4. You didn't mention what voltage you're considering. Values around 3.3uF are plenty available at 16V+ ratings, X7R, probably in 0805 to 1210 packages.
Battery-powered, 9V specifically, so there are all kinds of X7Rs to use.
In any case, if C is dropping off at high voltages, why do you care, as long as it's repeatable? Use a 4.7uF 10V, or even a 6.8uF 6.3V.
That's the idea, although according to the
linked article, the voltage coefficient on X7Rs will
likely drop the capacitance to 20% of its nominal value. Even with a sloppy target of 2 minutes to 5 minutes, calculating the discharge characteristics seems a bit unknown.
Does the device have an MCU ? If so, do it in software. If not, maybe use a CD4060 oscillator/counter
The enable pin current is likely to vary a lot between parts and over temp, so dangerous to rely on it.
Alas, no MCU otherwise that would have been my first choice too. I was not sure if the enable pin/capacitor was a good idea, and it looks like it's leaning toward not good enough. I was looking for a cheap solution to avoid a power switch, and to make sure the batteries wouldn't bleed dry (wet gooey mess, actually) if someone left it on too long. I might be able to use a MOSFET with a capacitor discharge instead which would be reliable enough and offer just enough accuracy. Stay tuned!