Where the microamps really go in low-power design
Good low-power design is mostly leak hunting: regulator quiescent draw, pull-ups, floating pins and the sensor that never sleeps. Where we look first.

Most battery-life problems we are asked to look at are not about the microcontroller. The datasheet promised a few microamps in sleep, the spreadsheet promised three years on a cell, and the first prototype is flat in five weeks. Good low-power design is less about picking the “right” chip and more about finding every small, constant drain that nobody put in the budget.
This post is for founders and engineers with a battery-powered device that is not lasting as long as it should, or one that is still on the whiteboard. Here is where we look first.
Start with an honest current budget
Battery life is an average-current problem. A device that sleeps most of the time and wakes briefly has two contributions: the sleep floor, which runs all day, and the active bursts, which are short but large. Both need to be on paper before you can tell which one to attack.
A simple worked example. Suppose a sensor node sleeps at 5 µA, then wakes every 60 seconds for 200 ms at 20 mA to measure and transmit.
- Active contribution: 20 mA × (0.2 s / 60 s) ≈ 67 µA
- Sleep contribution: 5 µA
- Average: roughly 72 µA
On a 2,000 mAh cell that is about 27,800 hours, a little over three years, before derating. Now add one 10 kΩ pull-up resistor that sits pulled low during sleep at 3.3 V. That is 330 µA on its own, and the same device now lasts under seven months. One resistor, more than five times worse. This is the pattern we see again and again: the floor is rarely what the microcontroller datasheet says, because the microcontroller is only one of many things connected to the battery.
Always derate. Usable capacity depends on temperature, discharge profile, cut-off voltage and self-discharge. For a first estimate we take 70–80% of nameplate capacity and treat anything better as margin.
The regulator nobody checked
The power path is the first place we look, because its losses never sleep. A linear regulator’s quiescent current is drawn whether the load is asleep or not. A classic general-purpose part such as the AMS1117 draws in the region of 5 mA of quiescent current. That alone would empty a 2,000 mAh cell in under three weeks with the rest of the board switched off.
Low-quiescent parts change the picture completely. The MCP1700 family is around 1.6 µA typical, and newer nano-power LDOs such as the TPS7A02 are specified in tens of nanoamps. For switching regulators, check the efficiency curve at your sleep current, not at 100 mA.
Other power-path items to check:
- Battery voltage dividers. A 100 kΩ / 100 kΩ divider across a 4.2 V cell draws about 21 µA permanently. Use megaohm values with a small capacitor at the ADC input, or gate the divider with a MOSFET so it is only connected during measurement.
- Power LEDs. Obvious, yet still the most common thing we desolder from first prototypes.
Microcontroller sleep modes in low-power design
Chip choice does matter, but less than the sleep mode you actually reach and what you keep alive in it. Typical figures for the chip alone, from the datasheets, with an RTC running:
| Part | Mode | Typical current |
|---|---|---|
| STM32L4 | Stop 2 | around 1–2 µA |
| nRF52840 | System ON, RTC wake | low single-digit µA |
| ESP32 family | Deep sleep, RTC timer | roughly 7–15 µA depending on variant |
Those numbers assume everything else is configured correctly. In practice the gap between datasheet and board comes from a few repeat offenders: peripherals left clocked, a debug interface still enabled, RAM retention you do not need, or a wake source that fires far more often than intended. On ESP32 development boards, the USB-to-UART bridge and the onboard regulator commonly draw more than the module itself, so measuring a dev board tells you little about your product.
Wake frequency is the other lever. Every wake costs a startup ramp: oscillator start-up, flash wait, perhaps a radio calibration. Batching measurements and transmitting less often usually saves more than shaving a microamp off the floor. We cover the radio side of that trade-off in BLE or LoRa? Picking the radio before picking the chip.
GPIOs, pull-ups and floating pins
Pins are where the microamps hide in plain sight. Three patterns account for most of what we find:
- Pull-ups held low. An I²C bus, a button or a chip-select that sits low during sleep turns its pull-up into a permanent load. At 3.3 V, 10 kΩ costs 330 µA and 100 kΩ costs 33 µA.
- Floating inputs. A CMOS input left floating can sit near mid-rail, where both input transistors partly conduct. The draw is unpredictable and can change with humidity or a hand near the board. Unused pins should be set to analogue mode or tied to a defined level.
- Back-powering. When a peripheral’s supply is switched off but the MCU still drives its lines high, current flows through the peripheral’s protection diodes. The floor rises and the part may not reset cleanly. Drive those lines low or set them to high impedance before cutting the rail.
The sensor that never sleeps
Peripherals are where the budget most often goes wrong, because each part has its own idea of “idle”. Many digital environmental sensors sleep below a microamp, but only once they have been commanded into sleep or forced mode. Leave one in continuous mode and it runs all day.
Things we check on every board:
- SPI NOR flash. Standby is typically in the tens of microamps; deep power-down is around a microamp. The firmware has to send the power-down command, and wake it before the next access.
- Accelerometers. Wake-on-motion modes are often a few microamps, while higher output data rates can be ten to a hundred times more. Pick the lowest rate that still catches the event.
- Analogue front ends and PIR sensors. These often need continuous power to work at all. Budget them as part of the floor, not as an active load.
- GNSS and cellular modules. Use their documented low-power or power-saving modes, or gate their supply entirely, and account for the energy cost of re-acquisition when they come back.
Where a part has no good sleep mode, a load switch on its rail is usually the simplest fix. It costs a few cents and a GPIO, and turns an unknown idle current into almost nothing.
Measuring microamps properly
You cannot fix what you cannot see, and a multimeter in series is a poor tool here. Its burden voltage can brown out the device, and it averages away the spikes. A shunt and an oscilloscope struggle with the dynamic range between a 2 µA floor and a 200 mA radio burst.
Dedicated source-measure tools such as the Nordic Power Profiler Kit II, Joulescope or Otii Arc handle that range and log charge over time, which is exactly the number you need for a battery estimate. Our routine is simple: measure the bare board with firmware held in its lowest state, then add one subsystem at a time and record the floor after each. When the floor jumps, you know which block did it. We also repeat key measurements at temperature, because leakage and self-discharge both rise as it gets warmer, as part of the wider testing and verification work we do.
A checklist before you commit to a battery
Before choosing a cell size or promising a battery life to anyone, we would want these answered:
- A written current budget with sleep floor, active bursts and wake frequency, derated to 70–80% of capacity.
- A regulator with quiescent current in the low microamps or below, checked at your actual sleep load.
- No resistor dividers, pull-ups or LEDs permanently across the supply without a reason.
- Every unused GPIO in a defined state, and no peripheral back-powered through its I/O lines.
- Every sensor, flash chip and module explicitly put to sleep by firmware, or on a switched rail.
- A measured floor on your own board, not a dev kit, with a proper current profiler.
- Battery life recalculated from measured numbers, not datasheet ones.
None of this is exotic; it is careful bookkeeping on real hardware. The microamps are always somewhere, and it is much cheaper to find them on the first prototype than in the field.