The number that decides whether your battery sensor lasts a week or a decade isn’t on the front page of any datasheet. It’s hiding in a row called quiescent current — the single spec that decides battery life — and most ‘low power’ designs get it wrong
- Quiescent current (IQ) is what your voltage regulator burns doing nothing — and in a sleeping IoT device, “nothing” is 99% of its life.
- The AMS1117 on most dev boards wastes ~5 mA of IQ. That alone flattens a coin cell in under two days.
- Even the LDOs people call “low IQ” — MCP1700 (1.6 µA), HT7333 (4 µA) — draw 8× to 20× more than a true nanopower part.
- At 200 nA, the regulator nearly drops out of your power budget — the charge it would have wasted goes to extra operating cycles. In a typical coin-cell sensor that’s a real, countable extra year of runtime.
What is quiescent current, really?
Quiescent current — written IQ — is the current a voltage regulator consumes just to stay alive, with no load on its output. It’s the tax the regulator charges your battery 24 hours a day, whether your circuit is transmitting, sleeping, or doing absolutely nothing. People mix it up with operating current (the current that flows to your actual load), but they’re different animals: operating current is work you asked for; quiescent current is overhead you didn’t.
For a device that’s always plugged into the wall, IQ is a rounding error nobody cares about. For a battery-powered sensor that wakes for two seconds every fifteen minutes and sleeps the rest, it’s the whole ballgame. That’s the question every embedded designer eventually asks: does the regulator drain the battery even when the device is asleep? Yes. Continuously. And if you picked the wrong regulator, it drains it fast.
Why IQ dominates battery life when your device sleeps 99% of the time
Take a typical wireless sensor node — an ESP32 reading a sensor and pushing the value over Wi-Fi. Done properly with deep sleep, the MCU is awake for maybe 2 seconds and asleep for the next 898. During that sleep window the ESP32 itself might draw 10–150 µA. If your regulator is also quietly burning 5 mA of quiescent current underneath it, the regulator isn’t the supporting act — it’s the entire power budget. You optimised your firmware for deep sleep and then handed all those savings back to a 1990s linear regulator.
This is why “how to make an ESP32 battery last longer” almost always comes back to two things: spend as much time in deep sleep as possible, and feed the board from a low quiescent current LDO. The first is firmware. The second is a parts-bin decision you make once and live with for the life of the product.
The real cost of the AMS1117, MCP1700 and friends in a coin-cell design
Let’s stop talking in adjectives and put numbers on it. Here’s how the popular 3.3V regulators actually compare on the spec that matters for battery life, with everything pulled from manufacturer datasheets:
| Regulator | Quiescent current (typ) | Max output | Dropout | Best for |
|---|---|---|---|---|
| NP-7A0333 (TI TPS7A03) | 200 nA | 200 mA | ~250 mV | Nanopower / always-on / energy harvesting |
| XC6206 | 1.0 µA | ~200 mA | 250 mV | Cheap low-power |
| MCP1700 | 1.6 µA | 250 mA | 178 mV | Coin-cell hobby designs |
| HT7333 | 4 µA | 250 mA | 170 mV | Low-power TO-92 builds |
| AP2112K | 55 µA | 600 mA | 250 mV | General 3.3V, not sleep-critical |
| AMS1117-3.3 | ~5 mA | 800 mA | ~1.1 V | Mains/USB-powered only |
Notice the spread: from 200 nA to 5 mA is a factor of 25,000. That’s not a tweak, it’s a different category of design. And notice that the AP2112K — a perfectly good regulator you’ll find on countless breakout boards — sits at 55 µA. Fine on USB; a battery-killer in a coin-cell node.
Worked example: how long does a CR2032 last on quiescent current alone?
A CR2032 coin cell holds roughly 225 mAh of usable charge. If we ignore the load entirely and ask the brutal question — how long would the battery last just feeding the regulator’s quiescent current? — the answer is devastating for the wrong part:
Two things are true here, and it’s worth being precise about both. First, the “128 years” is fiction — no coin cell lives that long. A CR2032 self-discharges at roughly 1% per year and has a shelf life around ten years, and that self-discharge is a parallel drain you can’t design away. It sets an absolute calendar ceiling on any battery, no matter how quiet your regulator is.
Second — and this is the part that actually matters in a real product — every microamp you save in quiescent is a microamp left for real work. A battery holds a fixed amount of energy, and whatever the regulator burns idling is charge your load never gets to use. Drop the quiescent draw and you free that charge up for more wake-ups, more transmissions, more operating cycles before the cell is empty. In a deeply duty-cycled sensor that’s a concrete gain in usable life: a 1 µA LDO might run the product for two years where a 200 nA part stretches the same cell to two and a half. The size of the gain depends on how sleep-dominated your design is — the more time it spends idle, the larger the share quiescent current occupies in your budget, and the more a nanopower part pays you back.
So the honest summary is this: lowering IQ almost always helps — it adds real operating cycles and extends usable life — right up until quiescent current is already far below both your averaged active draw and the battery’s own self-discharge. Only then do further reductions stop mattering, because you’ve hit the calendar ceiling. For most coin-cell and energy-harvesting designs you’re nowhere near that point, which is exactly why a regulator this quiet is worth reaching for.
A real-world example: the same sensor, six different LDOs
Quiescent-current-only math is useful for shock value, but a real product also has a load. So let’s model an actual low-power sensor node and change only the regulator, leaving everything else identical, and see how long it really runs.
- Battery: one CR2032, ~225 mAh usable.
- Duty cycle: the device wakes every 30 minutes, stays active for 0.5 s drawing 10 mA (sensor sample + a short radio burst), and deep-sleeps the rest of the time. That averages to ~2.78 µA of active current.
- MCU deep-sleep current: 1.5 µA.
- Capacitor & PCB leakage: ~0.3 µA. Real ceramic capacitors do leak a little, but every module here uses the same class of MLCCs, so this is treated as identical for all of them — it cancels out of the comparison and only the LDO changes.
- Cell self-discharge: ~0.26 µA (about 1% per year).
That gives a fixed, LDO-independent base load of ~4.84 µA. The only variable between the rows below is each regulator’s quiescent current.
| Regulator | IQ | Total avg current | Estimated real life | vs MCP1700 |
|---|---|---|---|---|
| NP-7A0333 | 200 nA | ~5.04 µA | ~5.1 years | +28% |
| XC6206 | 1.0 µA | ~5.84 µA | ~4.4 years | +10% |
| MCP1700 | 1.6 µA | ~6.44 µA | ~4.0 years | baseline |
| HT7333 | 4 µA | ~8.84 µA | ~2.9 years | −27% |
| AP2112K | 55 µA | ~59.8 µA | ~5 months | −89% |
| AMS1117-3.3 | ~5 mA | ~5 mA | ~2 days | −99.9% |
Two honest caveats. First, the size of the win depends entirely on how sleep-dominated your design is. Make the device wake more often, stay awake longer, or draw more current when active, and the averaged active current grows until it dwarfs any quiescent difference — at which point all the low-IQ parts converge and the choice stops mattering. Second, push the duty cycle the other way — wake every few hours, sip a milliamp for a few milliseconds — and the gap widens further in nanopower’s favour, until you finally bump the battery’s self-discharge ceiling. The 30-minute example above sits in the sweet spot where most real coin-cell sensors live, and there the 200 nA part buys you a genuine, countable extra year.
How to extend battery life in an IoT sensor
If you’re working out how to extend battery life in an IoT sensor, the practical checklist is short, and the order matters:
- Sleep aggressively. Keep the MCU in deep sleep and wake only long enough to read, transmit, and drop back down.
- Pick a low quiescent current LDO. The regulator runs 24/7, so choose IQ in the microamp or nanoamp range — not the milliamp range of an AMS1117.
- Kill always-on loads. Onboard power LEDs, stray pull-ups and unused peripherals quietly add up; remove or gate them.
- Power sensors on demand. Switch sensor rails on only while sampling instead of biasing them continuously.
- Budget across the whole duty cycle. Work out the average current over one full sleep-plus-wake cycle, then add the cell’s self-discharge before you estimate runtime.
Firmware handles the first point; the regulator decides the second. The energy you claw back from quiescent current is energy your load gets to spend — which is why steps 1 and 2 together, not either alone, are what turn a sensor that lasts weeks into one that lasts years.
Where nanopower actually matters
You don’t need 200 nA for everything. If your product runs from USB or a wall adapter, an AP2112K or even an AMS1117 is fine and cheaper. Nanopower earns its keep when the device spends most of its life asleep or harvesting tiny amounts of energy:
- ESP32 / STM32 / nRF deep-sleep sensor nodes — where you’ve already crushed the firmware power budget and the regulator is the last thing standing between you and a year of runtime.
- Coin-cell and primary-battery devices — door/window sensors, asset tags, beacons, medical wearables.
- Energy harvesting — solar, thermal, RF or piezo sources that deliver microamps, where a 55 µA regulator would eat the entire harvest.
- Always-on backup rails — real-time clocks, supervisors and SRAM retention that must never fully power down.
Frequently asked questions
Does a voltage regulator drain the battery when the device is off?
Yes. As long as the battery is connected, the regulator draws its quiescent current continuously, even with no load. This is exactly why IQ matters so much in battery designs — for a device that’s “off” or asleep most of the time, quiescent current is the dominant drain.
How much does quiescent current affect battery life?
Enormously, for low-duty-cycle devices. A regulator at 5 mA versus one at 200 nA is a 25,000× difference in standby drain. On a coin cell, that’s the difference between roughly two days and several years of usable life.
Why is the AMS1117 bad for battery projects?
The AMS1117 is an old bipolar linear regulator with ~5 mA of quiescent current and a high (~1.1 V) dropout. It’s robust and cheap for mains- or USB-powered boards, but on a battery it wastes more current sitting idle than a well-designed sensor uses on average. It’s the single most common reason a “low-power” prototype dies in days.
What is the lowest quiescent current LDO you can buy?
Production nanopower LDOs reach down to the tens of nanoamps. The Texas Instruments TPS7A02/TPS7A03 family used in the Ramanu NP-7A0333 module runs at about 200 nA typical, with around 3 nA in shutdown — low enough that the battery’s own self-discharge becomes the limiting factor.
The short version
If you’re building anything that runs on a coin cell or a small battery and sleeps most of the time, look past the headline output current and find the quiescent current row. That one number, more than almost anything else on the page, decides your runtime. Get it into the microamps and you’re doing well; get it into the nanoamps and the regulator all but disappears from your power budget.
Built for the nanoamp end of that scale
The Ramanu TrueVolt NP-7A0333 is a 3.3V / 200mA module built on the genuine TI TPS7A03 — 200 nA quiescent current, 3 nA shutdown, Samsung 4.7µF ceramics, Diotec SD103 reverse-current protection, and a 7805-compatible pinout that drops straight into your existing footprint.
Specifications cited from manufacturer datasheets (Texas Instruments TPS7A03, Microchip MCP1700, Holtek HT7333, Diodes Inc AP2112, Torex XC6206, AMS1117). Quiescent-current figures are typical at 25°C; consult each datasheet for guaranteed limits over temperature. Battery-life figures are first-order estimates based on a 225 mAh CR2032 with the stated duty cycle and base load.