← Back to the chapter · Previous detail · Next detail →
Technical detail · Edge Acquisition: Power and Gateway Functions
Bound a missing DONE and test cold-start failures
| Design boundary | Question to prove |
|---|---|
| Rail rise | Does voltage settle before the MCU or sensor leaves reset, including battery and cold-temperature impedance? |
| Cold boot | Can all required state be reconstructed without retained RAM? |
DONE timing | Is DONE asserted only after flash writes, radio hand-off, and output-safe state complete? |
| Failed transaction | Does firmware stop after a bounded retry count and record loss honestly rather than keeping the rail on forever? |
| Hung firmware | Does the programmed maximum-on interval bound a missing DONE, and is a shorter independent watchdog also required? |
| Manual/service wake | Can a technician wake the unit without creating an unbounded bypass current? |
For the TPL5110, a missing DONE does not leave the load on indefinitely. The device holds the load on for the programmed interval minus roughly 50 ms, then forces DRV high for the final 50 ms before the next cycle. That maximum-on window is a hardware recovery boundary, but it may be far longer than the application’s safe fault time. Choose the interval so legitimate cold boot, sensing, and radio retries fit, then add an independent watchdog when a hung actuator or high-current radio must be cut off sooner. Record both the normal DONE time and the forced-off time in the fault test.