The usual first estimate goes like this: the battery holds so many milliamp-hours, the device draws so many milliamps when it is on, divide one by the other. For a device that sleeps most of the time (a sensor, a tracker, a remote, a smart vent) that estimate is wrong twice. It ignores the current drawn while the device is “off”, which can be most of the energy, and it trusts a capacity figure that was measured under kinder conditions than yours.
The arithmetic that actually matters
Average current is the sleep current plus the charge each wake-up costs, spread over the time between wake-ups:
Iavg = Isleep + Qevent ÷ T
Battery life is then capacity divided by Iavg. To make that concrete with published numbers: a Texas Instruments white paper on coin cells tested a radio wake-up of 8 mA for 2 ms, 15 mA for 1 ms and 8 mA for 2 ms. That is 47 µC per event, and the paper notes the profile is deliberately heavier than a real Bluetooth Low Energy one. The same paper puts typical sleep current “in the 1uA range”, while its own test load slept at 100 µA. Take a CR2032 coin cell with a typical capacity of 235 mAh and choose the wake interval yourself:
| Wake interval (your input) | Average current, 1 µA sleep | Life on paper | Average current, 100 µA sleep | Life on paper |
|---|---|---|---|---|
| 1 second | 48 µA | ~200 days | 147 µA | ~67 days |
| 10 seconds | 5.7 µA | ~4.7 years | 105 µA | ~94 days |
| 1 minute | 1.8 µA | ~15 years | 101 µA | ~97 days |
| 10 minutes | 1.1 µA | ~25 years | 100 µA | ~98 days |
Read the right-hand columns first. With 100 µA of sleep current, slowing the radio from every 10 seconds to every 10 minutes buys you four days: the cell lasts about three months either way. With 1 µA, the interval is everything, and an on-paper life of a decade or more should be read as “the battery is no longer the constraint”, not as a promise. The wake-up routine you spent a week optimizing can be worth less than one pull-up resistor left conducting.
Where sleep current hides
Sleep current is rarely the microcontroller’s headline datasheet figure. It is the sum of everything still connected to the battery:
- the regulator’s quiescent current, paid every second of the product’s life
- pull-up and divider resistors with voltage across them
- sensors, LEDs and level shifters that were never switched off
- floating or misconfigured GPIO pins
- debug interfaces left enabled in production firmware
- protection and monitoring circuits: for lithium-ion packs, Battery University notes the protection circuit adds about 3% per month to self-discharge
Large systems are no exception. On the high-voltage packs we designed for an electric pusher trailer, an aftermarket battery management system monitors the modules. Anything that watches a battery is also a load on it, and it belongs in the budget.
The capacity figure is optimistic too
- It was measured gently. Coin-cell capacity is typically rated at a low, steady current and counted down to 2.0 V (TI). The CR2032 above is rated into a 15 kΩ load, about 0.19 mA (datasheet).
- Pulses cost capacity. Under TI’s pulsed test, cells from some vendors, including some branded ones, delivered only about 50% of rated capacity. Adding a capacitor in parallel with the cell improved the worst performers by more than 40%.
- Cold makes it worse. Internal resistance rises as temperature falls, which TI notes makes that capacitor even more important at low temperatures.
- Your cutoff is not the datasheet’s. If your circuit browns out above 2.0 V, you never see the tail of the discharge curve.
- Rechargeables leak. Battery University puts lithium-ion self-discharge at about 5% in the first 24 hours, then 1–2% per month, with the rate roughly doubling for every 10 °C rise.
Not every product is a sleeper
Some products are dominated by active loads instead. On the powered respirator helmet we designed, the big consumers are a blower, Peltier coolers and an ultrasonic humidifier. There the lever is mechanical. A large, cap-shaped HEPA filter reduced flow restriction, and with it the motor speed and the noise that comes with it. Know which kind of product you have before deciding what to optimize.
Either way, model the real duty cycle instead of assuming one. On a smart-vent system, hardware-in-the-loop kinematics modeling raised device battery life by 20%.
What to do this week
- Measure sleep current on your real board, fully assembled, with a meter that resolves microamps, across the whole sleep period rather than the first second of it.
- Capture one wake-up on a scope or power analyzer and integrate it to get Qevent in microcoulombs.
- Re-run the estimate pessimistically: half the rated capacity, at your coldest operating temperature. If the product still meets its promise, ship the number.
Plug your own numbers into our battery life calculator. If you want a second pair of eyes on your power budget, send us a brief.