Back on the PIC, dissecting the datasheet, this is what I have for the LF13K50

• Sleep mode: 24 nA
• Watchdog Timer: 450 nA
• Timer1 Oscillator: 790 nA @ 32 kHz

So using the watchdog timer to wake up the pic should lead to extremely low power consumption, provided that I don't need to wake the pic often, and that it won't do much when it'll be awake (turning back to sleep rapidly).

I could even experiment the use of timer1 driven by the 32k crystal instead of the watchdog to give much more decent time accuracy on my measurements intervals.

My power rail is 3.3V so I can use the LF model. I could reduce even more consumption by lower it down to 2.8V using another regulator. 2.5V or 1.8V would be even better, but then the temperature sensor TMP100 is off specs (2.7V min).

Looks good on the paper, now onto the program to get this. I still need to fix the USB side first though...