Power Management in IoT: How to Make a Device Last Years on One Battery
In a utility cupboard somewhere in Manchester, a gas meter fitted in 2015 is still sending a reading every night. Same battery. Nobody has opened the case.
Meanwhile, the ESP32 sensor you built last weekend — the one on the windowsill reporting temperature to a dashboard — is dead. Three days on two AA cells.
The two devices aren't as different as they look. Both are a microcontroller, a sensor and a radio. Both wake up, take a measurement, send a few bytes and go back to sleep. The difference between three days and fifteen years isn't a better battery — it's what the device does with the 99.8% of its life when nothing is happening.
That's what power management in IoT actually is: not a component you buy, but a set of decisions about sleeping, waking and talking. π
The Short Version
Every battery-life question comes down to one division: capacity divided by average current. A pair of AA lithium cells holds about 3,000 mAh. Draw 3 mA on average and you get six weeks. Draw 30 Β΅A and you get eleven years. The word doing the work is average — and a device that reports every fifteen minutes is awake for 0.2% of its life. So the average is set by two unrelated numbers:
- The floor — what the device draws while asleep. Ten microamps costs about 88 mAh a year, under 3% of that AA pair.
- The bump — what each wake-up burns. One extra second awake at 100 mA, every fifteen minutes, costs nearly 1,000 mAh a year. A third of the battery, from a single second.
The mental model that actually works: a battery-powered device is a sleeping device that occasionally wakes. Design the wake-up, not the runtime.
The Radio: Where the Battery Actually Goes
Same board, same 3,000 mAh battery, same fifteen-minute schedule — swap only the radio:
- ESP32 Wi-Fi, naive connect — scan, associate, DHCP, TCP, TLS: three to eight seconds at ~100 mA per wake. About seven months.
- NB-IoT with PSM — several seconds of radio time per report. About a year; report once a day instead and it runs into the 2030s, like the meter in the cupboard.
- LoRaWAN Class A — ~120 mA, but for 50–200 ms, with no connection to establish. Over a decade, limited by the cell's own ageing.
- BLE advertising — a few milliseconds at 5–6 mA, the reading broadcast inside the advertisement itself. Over a decade, again shelf-life limited.
The lesson isn't "never use Wi-Fi". It's that the radio's connection model matters more than its transmit power.
The Silent Leaks
Most first builds die here, and it's rarely the microcontroller. It's the parts that came with the board:
- The power LED — 1–3 mA, permanently. Two hundred times the chip's deep-sleep current.
- The AMS1117 regulator — about 5 mA quiescent, just existing. A modern TPS7A02 draws 25 nanoamps.
- The USB-serial chip — a few hundred microamps, cable or no cable.
- The battery divider — two 100 kΞ© resistors across 3.7 V: 18 Β΅A, all day.
- Sensors that never sleep — a GPS module left powered is 20–30 mA continuously.
Add them up and the sleeping "10 Β΅A" dev board is actually drawing 7–8 mA. That's not a rounding error; it's a different product.
Batteries Lie (a Little) — Briefly
The capacity on the label was measured under a gentle constant load at room temperature — the opposite of an IoT device's nothing-then-120-mA-pulse pattern. Coin cells hate pulses and need a capacitor beside them. Alkaline collapses in the cold. Lithium AA holds a flat 1.5 V to −40 °C and is the outdoor default. Lithium thionyl chloride is what the meters use — twenty-year shelf life. And at −20 °C most chemistries deliver half their rated capacity: a five-year sensor in the lab is a two-year sensor in a Scandinavian car park.
None of which you can design on paper. A multimeter can't read a signal that spends fifteen minutes at 10 Β΅A and fifty milliseconds at 120 mA; a dynamic current profiler like Nordic's Power Profiler Kit II shows you the real curve for your real board. The area under it is your battery life.
π‘ Final Thought
A well-designed IoT device isn't a device that does a lot with a little. It's a device that does almost nothing, almost all the time, and does the little it does with great care. The engineering is in the restraint.
The gas meter and the dead sensor are built from the same parts. One was designed around its sleep and the other around its features. The battery didn't decide which one lasted; the designer did, mostly in the microamps nobody was looking at.
The battery is never the problem. What you do with it is.
→ Full breakdown: sleep-mode ladders and event-driven wake-ups, the radio-by-radio arithmetic, every battery chemistry compared, how to measure the real board, where things stand in 2026 (Wi-Fi 6 TWT, ambient IoT, the EU battery rules), the honest trade-offs, and the complete builder's and buyer's guide: Read the deep dive
Follow for more IoT hardware deep dives — part of my ongoing 101-story series. π¬
Comments
Post a Comment