Power Design for Battery Products
Power Design for Battery Products: Budget First, Components Second
The battery is a budget, not a box. A field sensor on a CR2032: the headline budget says 513 days, the measured budget says 44. The gap is sleep current, regulator losses, and a battery that is not a constant, and the fix is arithmetic, not a bigger cell. The power budget, the BMS, and the rules that make battery life honest.
First, a simple picture
A weekly wage, in cash, in an envelope.
On Monday you get the envelope. It has to last until Sunday. You write down what you spend: rent, food, the bus. If you write down the rent wrong, if you write down what you wish the rent was, the envelope is empty by Thursday, and you don't know why, because the arithmetic was honest and the numbers were not.
A battery is the envelope. The capacity, the milliamp-hours printed on the cell, is the cash in it. Every part of the product spends from it: the sensor, the processor, the radio, and the parts you forgot about, the regulator that is always awake, the pin that is floating, the sensor that never quite powers down.
Power budgeting is writing down what each part really spends, in the right units, and adding it up before you pick the parts. Pick the parts first and the budget second, and the envelope is empty by Thursday.
This article is the arithmetic: what power budgeting actually means, why the spec-sheet battery life is almost never the real battery life, and what the battery management system (BMS), the guard that keeps the battery inside its safe limits, has to do.
The budget: one formula, four traps
The whole of power budgeting is one formula:
A milliamp-hour (mAh) is a unit of charge: a 230 mAh cell can deliver 230 mA for one hour, or 1 mA for 230 hours, or, in theory, any combination that multiplies to 230. Average current is the current the product draws, averaged over its whole cycle: awake and asleep, transmitting and waiting. The formula is simple. The traps are in the two numbers that go into it.
Trap 1: the average current is not the active current. A field device transmits for seconds and sleeps for hours, so the current it draws while doing nothing dominates the total. Duty cycling, spending as much time as possible in the deepest sleep, is the most powerful technique in low-power design, and it works by arithmetic: sleep at 2 µA for 59 seconds and wake at 15 mA for 1 second, and the average is only ~252 µA, a hundred times less than the active current, because the device is asleep 98% of the time. The rule is the same one Faststream Engineering states plainly: a few microamps of avoidable sleep current outweigh a year of transmissions. The energy is spent in the quiescent state, not the one you watch.
Trap 2: the datasheet sleep current is not the board's sleep current. Datasheets quote 2 µA for the chip in deep sleep. A real board with debug circuitry may draw 200 µA, a hundred times more, because the debug chip is awake, a sensor was never powered down, a pin is floating, or the regulator's quiescent current (the current a regulator draws just to stay alive, typically 5–50 µA for an LDO) was never counted. A floating pin, an unconnected GPIO with neither pull-up nor pull-down, can oscillate, and the oscillating input stage draws excess current. None of this is in the chip's datasheet. All of it is in the board's measured sleep current, and nobody measures it until the field returns start.
Trap 3: the battery is not a constant. Datasheet capacities assume a specific discharge rate and temperature. Real capacity can be 30–50% lower at high pulse currents or cold temperatures. A coin cell delivers less capacity in the cold, critical for outdoor and cold-chain sensors, and a cell that comfortably lasts on the bench can fail in winter if the current pulses were not modelled. Size for the worst-case temperature, not the room-temperature datasheet.
Trap 4: the radio is the easy part, and the expensive part, at the same time. The radio dominates energy on most sensor nodes, but only because it is the loudest spender, not the biggest one. One BLE advertising packet costs about 30 µJ (8–15 mA for 1–3 ms). One LoRa packet costs about 2,000 µJ. One Wi-Fi packet costs about 30,000 µJ, a thousand times the BLE packet, for the same data. The protocol choice is a power decision, and it is made before the first line of code.
Average power, not capacity: the wearable proof
The cleanest proof that battery life is set by average power, not by cell capacity, is the wearable.
The Apple Watch Series 10 (46 mm) has a 1.266 Wh cell, confirmed by teardown, against Apple's official 18-hour rating. That is an average draw of about 70 mW. Low Power Mode halves the average to roughly 35 mW and doubles the runtime.
The Oura Ring 4, with no display, averages about 0.5 mW on a cell of about 26 mAh, one twelfth of the watch cell, and lasts 8 days. The 140× spread on similar cells comes from PMIC quiescent current (a nanoamp contest), power gating, and duty cycling. Ship-mode current in leading wearable PMICs (power management ICs, the chips that convert battery voltage into the rails each circuit needs, control charging, and disconnect blocks that are not in use): 35 nA for Nordic's nPM2100, under 50 nA for TI's BQ25120A, 370 nA for Nordic's nPM1300. The nPM2100's boost converter runs at 95% efficiency at 50 mA and 90.5% at a 10 µA load, because the device spends most of its life asleep, light-load efficiency matters more than peak efficiency.
The lesson, in one line: a bigger battery buys weeks. A smaller sleep current buys years.
The sleep hierarchy: how deep can you sleep?
Most ARM Cortex-M and ESP32-class chips offer layered sleep, and each layer trades current for wake latency:
| Mode | Typical current | RAM retained | Wake latency |
|---|---|---|---|
| Run | milliamps | Yes | , |
| Idle / WFI | hundreds of µA | Yes | microseconds |
| Light sleep | tens of µA | Yes | milliseconds |
| Deep sleep | single-digit µA | partial / none | tens of milliseconds |
Enter the deepest sleep your wake latency budget allows. A sensor reporting every 15 minutes has no reason to stay in light sleep between readings. And power down sensors explicitly, I2C devices often have a sleep register, and a software sleep bit is not always enough; a load switch (a MOSFET that cuts power to a whole subsystem) is the honest way to turn something off. A floating TX line or an always-on USB-serial bridge can cost hundreds of microamps, more than the sensor itself.
DEMO
Example Labs is a made-up company we use for worked examples.
This product is made up. Call it the Example Labs field sensor: a small weather sensor on a fence post, reporting temperature and humidity every 15 minutes, on a single CR2032 coin cell, 230 mAh. It carries the event camera's cousin (a sensor that wakes on change), the Kalman filter from Article #4 (fusing the temperature readings), and the PID loop from Article #3 (driving a tiny vent). The demo's arithmetic is the article's own, verified against the sourced inputs; the sensor is a worked example, not a measured product.
Step 1, The headline budget (the datasheet numbers). Sleep: 2 µA, the chip's datasheet deep-sleep figure. Wake: 15 mA for 1 second, every 15 minutes, to read the sensors and send one BLE packet. The product page says "over a year on one coin cell." The arithmetic agrees. The arithmetic is wrong.
Step 2, The measured budget (the board, on a bench). The board is measured with a shunt ammeter, and the sleep current is 200 µA, not 2 µA, the debug circuitry is awake, the humidity sensor was never powered down, one GPIO is floating, and the LDO's quiescent current was never counted. The product that was supposed to run 1.4 years dies in six weeks. This is the same shape as the real freezer sensor in the field: supposed to run two years on a CR2032, died in eleven weeks, because the firmware left the radio listening between uploads and the accelerometer stayed on, and nobody measured the actual sleep current until the field returns started.
Step 3, The cold, on top. The fence post is outside, and winter comes. At cold temperatures and pulse loads, the real capacity is 30% lower: 230 mAh becomes 161 mAh. The sensor dies in a month, in the cold, exactly when the data matters most.
Step 4, The fix (the same board, the sleep current hunted). The debug circuitry is removed from the production board, the humidity sensor is put behind a load switch, the floating pin gets a pull-down, and the LDO is swapped for a low-Iq part. Measured sleep current: 5 µA. No bigger battery. No new radio. The sleep current was the product.
The BMS: the guard at the edge of the box
A battery is not just a budget. It is a chemical system with a safe operating area, the box of voltage, current, and temperature inside which the cells can operate without damage or safety risk, and the battery management system is the electronic brain that keeps the cells inside it. Without a BMS, a lithium battery cannot be safely charged or discharged: it will overcharge, over-discharge, overheat, or drift out of balance, and any of those permanently damages the cells or causes fire.
The safe operating area the BMS guards
| Limit | NMC / NCA | LiFePO4 | Outside it |
|---|---|---|---|
| Cell voltage | 2.5 to 4.20 V | 2.5 to 3.65 V | Above: lithium plates on the anode. Below: copper dissolves |
| Charge current | Within the rated C-rate | Within the rated C-rate | Heating, plating |
| Charge temperature | Above 0 °C | Above 0 °C | Lithium plating, permanent damage |
Overvoltage protection (OVP). The BMS disconnects the charger when any cell reaches its maximum safe voltage, 4.20 V for NMC/NCA chemistry. Charging above 4.2 V causes lithium plating on the anode: lithium metal deposits instead of intercalating, irreversibly reducing capacity and creating internal short-circuit risk that can lead to thermal runaway. Most BMS boards trigger OVP at 4.20–4.25 V per cell. Charging to a lower cutoff, 4.10–4.15 V, extends cycle life at the cost of 10–15% usable capacity per charge. That trade is a design decision, not an accident.
Undervoltage protection (UVP). The BMS disconnects the load when any cell drops below its minimum safe voltage, typically 2.5–3.0 V for NMC/NCA. Deep discharge causes copper dissolution from the anode current collector, which can create internal short circuits on the next charge cycle. Most boards set UVP at 2.8–3.0 V, because the cell voltage drops steeply below 3.0 V and the remaining energy is minimal, the small range gained from discharging to 2.5 V is not worth the accelerated degradation.
Overcurrent protection (OCP). The BMS limits discharge current to the cells' continuous rating, and its short-circuit protection triggers in microseconds at much higher currents (100+ A) to catch hard shorts rather than moderate overloads.
Cell balancing. No two cells are perfectly identical, they have slightly different capacities, internal resistances, and self-discharge rates. Without balancing, the weakest cell limits the whole pack: the BMS triggers UVP when the weakest cell reaches 2.8–3.0 V, even if the other cells still have usable energy. Passive balancing bleeds excess charge from the highest-voltage cells through a shunt resistor until they match the lowest. It wastes a little energy to save a lot of capacity.
Temperature, and the 0 °C rule. The BMS halts charging below 0 °C. Charging below freezing causes lithium plating, permanent anode damage that cannot be reversed. For LiFePO4 cells, the safe range is 2.5 V to 3.65 V per cell; going outside it, even briefly, causes permanent capacity loss. A Grade-A LiFePO4 cell can deliver 4,000 to 6,000 charge-discharge cycles under ideal conditions, but only if the BMS keeps it inside the box for all of them.
State of charge and state of health. The BMS also measures: SOC (state of charge, the percentage full, target accuracy ±3% over lifetime, estimated by coulomb counting, integrating current over time, corrected by an open-circuit-voltage measurement whenever the pack rests for more than 30 minutes, with a Kalman filter update step limiting cumulative drift to under 2%) and SOH (state of health, the capacity fade and internal-resistance growth; end of life is typically defined as SOH below 80% for EV applications or below 70% for grid storage). The Kalman filter in the BMS is the same mathematics as Article #4, the estimator, guarding the battery instead of the position.
The standards behind it. BMS design is governed by ISO 26262 (automotive functional safety), IEC 62619 (industrial lithium-ion safety), and UL 1973 (stationary storage). The BMS market was valued at $7.9 billion in 2024 and is projected to reach $22.3 billion by 2032, a vendor-guide figure, labelled as such, because every lithium pack needs one, and the margin between safe operation and thermal runaway gets thinner as energy densities pass 270 Wh/kg.
The BMS's own power. One trap the guides all name and the budgets often miss: the BMS itself draws current, its monitoring, its balancing, its protection FETs. On a small pack, the BMS quiescent current is a line in the budget, next to the regulator and the floating pin. A guard that eats the envelope is a guard that needs budgeting too.
The decision test
Four questions. They sort the products whose power budget is honest from the products that need a power audit first:
- Has the sleep current been measured on the real board, with a shunt ammeter, not read from a datasheet? No, the budget is a wish, and the field returns will be the measurement. Yes, the budget has a foundation.
- Does the budget include the regulator's quiescent current, the BMS's own draw, and the self-discharge of the cell? No, the budget is missing lines, and the missing lines are the ones that dominate. Yes, the budget is complete.
- Is the capacity derated for the coldest temperature the product will see? No, the product dies in winter, exactly when the data matters most. Yes, the budget survives the season.
- Can the battery be replaced or serviced without throwing the product away? No, the product is disposable, and the site's own rule says instruments are made to be recalibrated, not replaced. Yes, the product is a product, not a consumable.
If the answers are yes, yes, yes, yes, the budget is honest, and the audit is a final check, not a rescue. If any answer is no, the job is a power audit, and the honest thing is to do it before the boards are ordered, not after the first hundred die in the field. And if the answer to question 1 is "we measured it, and it's 200 µA", the problem is the design, not the battery, and a bigger cell will buy weeks, not years.
"But the datasheet says two years", and two more fair doubts
"The datasheet says two years." The datasheet says the chip sleeps at 2 µA. It does not say the board sleeps at 2 µA, the regulator draws nothing, or the battery holds 230 mAh in the cold. The datasheet's numbers lie politely; the firmware decides whether the device lasts a season or a month. The honest sentence is "two years, measured on the production board, at the coldest temperature, with the debug circuitry removed." Anything shorter is a wish.
"We'll just use a bigger battery." A bigger battery buys weeks, and it buys them at the cost of size, weight, and money, and on a wearable, the battery is already the biggest part. The Oura Ring runs 8 days on a 26 mAh cell because its average draw is 0.5 mW, not because its cell is large. The fix that buys years is the sleep current, and the sleep current is free: it is the load switch, the pull-down, the regulator swap, the sensor powered down. Do the arithmetic first; buy the bigger cell second, if the arithmetic still says you need it.
"Three years of battery life can't be tested in three years." True, and that is why the proof is not a test, it is a projection: measure the energy of each event, combine it with the duty cycle and a temperature-aware battery model, and project. A projection is only as honest as the measured sleep current behind it. The audit's job is to make the projection honest, measured sleep current, measured event energies, derated capacity, and then to say, in writing, what the battery life is and what it depends on.
Sources
- Embedded ACE, "Low-Power Design Techniques"
What we used
Battery life formula (capacity ÷ average current); duty cycle example (sleep 2 µA for 59 s, wake 15 mA for 1 s → ~252 µA average); CR2032 230 mAh datasheet capacity, real capacity 30–50% lower at high pulse currents or cold; datasheet sleep 2 µA vs real board 200 µA with debug circuitry; LDO quiescent 5–50 µA; floating pins oscillate and draw excess current; radio energy per packet: BLE ~30 µJ, LoRa ~2,000 µJ, Wi-Fi ~30,000 µJ
- Michael Samuel Naeem, "Power Management for Battery Devices" (September 2025)
What we used
Freezer sensor story (supposed two years on a CR2032, died in eleven weeks; radio left listening, accelerometer left on; nobody measured sleep current until field returns); MCU sleep hierarchy (run mA, idle hundreds of µA, light sleep tens of µA, deep sleep single-digit µA); power down sensors explicitly, load switch MOSFET; floating TX line or always-on USB-serial bridge costs hundreds of µA; coin cells deliver less capacity in the cold
- Faststream Engineering, "Designing for Years on One Battery" (August 2026)
What we used
A few microamps of avoidable sleep current outweigh a year of transmissions; the energy is spent between the transmissions; cold cuts capacity and sags pulses; a three-year life is proven by measuring event energy, duty cycle, and a temperature-aware battery model, not by testing for three years
- VLSI Korea, "Wearable Semiconductors 3: The Secret of Multi-Day Battery" (November 2026)
What we used
Apple Watch Series 10 (46 mm): 1.266 Wh cell (teardown), 18-hour rating, ~70 mW average, Low Power Mode ~35 mW; Oura Ring 4: ~26 mAh cell, ~0.5 mW average, 8 days, 140× spread; PMIC ship modes: Nordic nPM2100 35 nA, TI BQ25120A under 50 nA, Nordic nPM1300 370 nA ship / 500 nA hibernate; nPM2100 boost 95% at 50 mA, 90.5% at 10 µA load; nPM1300 buck up to 93%
- eBike Engineering, "BMS Selection and Wiring"
What we used
OVP 4.20–4.25 V per cell (NMC/NCA); charging above 4.2 V causes lithium plating, irreversible capacity loss, internal short risk; lower cutoff 4.10–4.15 V extends cycle life at 10–15% capacity cost; UVP 2.5–3.0 V, most boards 2.8–3.0 V; deep discharge causes copper dissolution and internal shorts on next charge; OCP at rated continuous current, short-circuit protection in microseconds at 100+ A; weakest cell limits pack capacity; passive balancing bleeds excess through a shunt resistor
- InnovChip, "BMS Design Guide"
What we used
Cell voltage monitoring ±5 mV; temperature ±1 °C; SOC target accuracy ±3%; SOH end of life below 80% (EV) or below 70% (grid storage); coulomb counting with OCV correction after >30 min rest and a Kalman filter update, drift under 2%; standards ISO 26262, IEC 62619, UL 1973; BMS market $7.9B (2024) → $22.3B (2032), vendor-guide figure; NMC 811 energy density beyond 270 Wh/kg
- Sunlith Energy, "Battery Management System (BMS) Explained"
What we used
BMS halts charging below 0 °C (lithium plating, permanent damage); LiFePO4 safe range 2.5–3.65 V per cell, outside it even briefly causes permanent capacity loss
- OCS OORJA, "Battery Management System (BMS) Explained"
What we used
Grade-A LiFePO4 4,000–6,000 cycles under ideal conditions; a poor BMS degrades good cells, a great BMS extends the life of average cells