Choosing an MCU when your industrial sensor must sleep for years
Pipeline operators and grain handlers across South Africa increasingly ask for wireless nodes that report hourly yet survive three to five years on a primary cell without a solar panel. The marketing sheets for ultra-low-power MCUs all cite impressive stop-mode figures; your firmware and passives determine whether those numbers appear on your bench.
Start with a wake budget, not a part number. List every event that must pull the CPU out of sleep—RTC tick, magnet reed, accelerometer interrupt, radio TX window—and estimate duration and current for each. If radio dominates, picking the MCU with the lowest deep-sleep nanoamps saves little compared to choosing a front-end module with a sane duty cycle.
Leakage on GPIO matters when sensors share rails with the radio. We have seen STM32L families behave well while an external level shifter leaked microamps because enable pins floated during sleep. Schematic review belongs in MCU selection, not after layout.
On-chip ADC reference drift with temperature can force shorter wake intervals than your link budget requires, which looks like a sensor problem in the field. Budget one week of firmware time to characterise RH or pressure channels across −10 °C to 45 °C in a cheap environmental chamber before you freeze the BOM.
Finally, plan for South African logistics: if your chosen part has twelve-week lead time, a pin-compatible second source may cost more per unit but less in calendar time. Document the sleep entry sequence for both parts in the same repository branch so you do not fork power management at the last minute.
Need help narrowing parts for a specific sensor chain? Describe your wake budget and we will suggest a short list without pushing a single vendor.