KEMI Docs · Circuit Practice

Voltage Divider Mistakes and How to Avoid Them

Last updated: 2026-10-04

Vout = Vin × R2 / (R1 + R2). It is the first circuit anyone learns, which is exactly why it gets taken for granted. The formula is only exact when nothing is connected to the tap. On a real board the tap feeds an ADC, another circuit or a test probe, and the result changes the moment it does. This article collects the divider mistakes we flag again and again in design reviews, grouped by root cause, and walks through choosing resistor values for a battery-monitoring divider.

The tap is a voltage source with a resistor: Thevenin

Looking back into the divider from its tap, you see a source of Vin × R2 / (R1 + R2) with an internal resistance of R1 ∥ R2 = R1 × R2 / (R1 + R2). That is the Thevenin equivalent, and its resistance (Rth) is where every divider problem starts. Draw any current from the tap and it drops voltage across Rth, pulling the output down. Make it a habit to calculate Rth alongside the output voltage every time you design a divider.

Loading effect in numbers

Say you use R1 = 27 kΩ and R2 = 10 kΩ to bring 12 V down to about 3.3 V. Unloaded output is 12 × 10 / 37 = 3.243 V, and Rth = 27k ∥ 10k = 7.30 kΩ. Connect anything with an input resistance to the tap and that resistance appears in parallel with R2.

Load on the tap	Actual output	Error
None (unloaded)	3.243 V	0%
1 MΩ (meter, high-impedance input)	3.220 V	−0.7%
100 kΩ	3.023 V	−6.8%
47 kΩ	2.807 V	−13.4%

The rule of thumb: if the load is at least 100 times Rth, the error stays under 1%. With Rth at 7.3 kΩ here, the load needs to be 730 kΩ or more. If you can't change the load, lower the divider resistances (at the cost of more current) or add an op-amp buffer (voltage follower) after the tap. The same effect makes a 10 MΩ multimeter read a high-impedance tap low, and remembering that saves debugging time.

ADC inputs: sampling capacitors and source impedance

An MCU's built-in SAR ADC looks like a high impedance at DC, but during each conversion an internal sampling capacitor (a few pF) is switched onto the pin and charged. If it doesn't charge to the required precision within the sampling time, the reading comes out low and gets dragged toward the previously sampled channel. That is why ADC datasheets specify a maximum source impedance, usually as a table or formula versus sampling time. A common approximation is ts ≥ (Rsrc + Radc) × Cadc × ln(2^(N+2)). Assuming Cadc = 5 pF and N = 12 bits, a 110 kΩ source needs about 5.4 µs.

Fix 1: lengthen the sampling time. For slow signals like battery voltage there is no downside to setting the ADC to its longest sample time.
Fix 2: put a capacitor on the tap. A 10 to 100 nF capacitor from the ADC pin to ground acts as a charge reservoir and supplies the sampling charge. The charge-sharing error is roughly Cadc / Cext; 5 pF against 100 nF is 0.005%, about 0.2 LSB at 12 bits.
The price: slower response. The capacitor creates a time constant τ = Rth × Cext, and settling to 12-bit precision takes about 8.3τ (ln 4096). That rules this fix out for fast-changing signals, where a buffer amplifier is the answer.
Input leakage: datasheets specify ADC pin leakage anywhere from tens of nanoamps up to ±1 µA maximum, depending on the part and temperature. Leakage × Rth is a direct error, so dividers of several hundred kilohms and up can drift noticeably at high temperature.

Worked example: measuring a lithium cell

We want to measure a Li-ion cell (up to 4.2 V) with a 3.3 V MCU (12-bit ADC, 3.3 V reference). The tap must stay below 3.3 V at 4.2 V, so the ratio must be at most 3.3 / 4.2 = 0.786; with margin we pick 0.5 (R1 = R2). 4.2 V becomes 2.1 V at the pin, and 1 LSB = 3.3 / 4096 = 0.81 mV (1.61 mV referred to the battery), which is plenty of resolution. What remains is choosing the absolute resistor values.

Divider	Always-on current (4.2 V)	Drain per day	Rth	Battery-referred error at 100 nA leakage
100k + 100k	21 µA	0.50 mAh	50 kΩ	10 mV
220k + 220k	9.5 µA	0.23 mAh	110 kΩ	22 mV
1M + 1M	2.1 µA	0.05 mAh	500 kΩ	100 mV

100k + 100k uses 15 mAh a month. On a 1,000 mAh cell that is 1.5% a month, comparable to the cell's own self-discharge; on a CR2032 coin cell (about 220 mAh) the divider alone empties the battery in a little over a year. Raising the values into the megohms cuts the drain but makes leakage error and ADC source impedance worse. In this situation we usually pick one of two answers. (1) 220k + 220k with 100 nF on the ADC pin (τ = 11 ms) and the longest sampling time; that is fine for large rechargeable cells. (2) A switched divider that is only connected while measuring; that is the right answer for coin cells and multi-year products.

The switched-divider trap: low-side switching

The easy way to switch a divider off is an N-MOSFET under R2, or tying the bottom of R2 to a GPIO that is driven low only during measurement. But when that switch is off, no current flows in R2, so R1 pulls the ADC pin all the way up to battery voltage (4.2 V). On a pin above the MCU's 3.3 V supply, current flows into the internal clamp diode (a few µA with R1 = 100 kΩ). That violates the datasheet's injection-current limits, can corrupt conversions on other channels, and can back-power the MCU rail during sleep. Wherever battery voltage can exceed the MCU supply, use a high-side switch that disconnects the top of R1 (a P-MOSFET with an N-MOSFET gate driver, or a load-switch IC). After switching on, wait 8 to 9 time constants before sampling.

Why a divider is never a power supply

'I only have 12 V; can I run a 5 V module off two resistors?' comes up more often than you would think. The numbers settle it. To keep the output from sagging more than 0.25 V (5%) as the load swings from 0 to 20 mA, you need Rth ≤ 0.25 V / 20 mA = 12.5 Ω. The divider that achieves this is R1 = 30 Ω, R2 = 21.4 Ω, which passes 233 mA and burns 2.8 W on its own. You throw away 2.8 W to feed a 20 mA module, the voltage still rises above nominal when the load turns off, and if the module draws an inrush at startup the voltage sags further and it resets in a loop. Make power with an LDO or a buck converter.

Is a divider fine for level shifting?

Dividers are commonly used to bring a unidirectional 5 V signal (a UART TX, say) down to a 3.3 V input. 1 kΩ + 2 kΩ gives 3.33 V, and with 10 pF of input capacitance the time constant is about 6.7 ns, fine up to a few MHz. Scale the resistors up 10× and the time constant grows 10×, rounding off fast edges. Bidirectional open-drain buses such as I²C can't use a divider; they need a proper level shifter.

Accuracy: how resistor tolerance reaches the output

With divider ratio k = R2 / (R1 + R2), if the two resistors err in opposite directions, the worst-case output error is about 2 × (1 − k) × the resistor tolerance. For 27k / 10k (k = 0.27), that is about ±1.46% with 1% parts and about ±7.3% with 5% parts. ADC reference error adds on top; if the MCU supply is the reference, the LDO's ±1 to 2% output tolerance goes straight into the reading. When accuracy matters, combine (1) 1% or better resistors from the same family (matched temperature coefficient), (2) a dedicated voltage reference, and (3) a one-point calibration in production (apply a known voltage and store a coefficient). For most battery gauges, step 3 alone makes a big difference.

⚠ In high-voltage dividers, check the voltage rating

When dividing hundreds of volts, R1 takes nearly all of it. A 0603 chip resistor's maximum working voltage is typically around 50 to 75 V, so it can be over its rating even with the right resistance. Split R1 into several resistors in series to keep each under its rating, and for anything connected to mains, review creepage and insulation separately against the applicable safety standard (such as IEC 62368-1 or UL 62368-1).

Common mistakes at a glance

Not calculating Rth. It is the root of both loading and ADC source-impedance problems.
Feeding an ADC pin from a several-hundred-kilohm divider with no capacitor. Readings come out low and get pulled toward other channels.
Adding the capacitor and then sampling right after switching on. Without waiting out the RC time constant you read a partly charged value.
Leaving a battery-sense divider permanently connected. In coin-cell products it becomes the largest load.
Switching the divider low-side. While off, the ADC pin rises above the MCU supply.
Using a divider as a power supply. Neither efficiency nor regulation works.
Forgetting ADC reference error. Switching to 0.1% resistors does nothing if the reference is off by 2%.

Practical checklist

At maximum input, is the tap voltage within the ADC input range (usually VREF)?
Did you calculate Rth, and is the load at least 100 × Rth?
Did you check the ADC datasheet's maximum source impedance and sampling time?
For high-value dividers, is there 10 to 100 nF on the ADC pin, and is the response time acceptable?
For battery products, is the divider's always-on current in the power budget?
For a switched divider, is it high-side, and do you wait 8 to 9 τ after switching on?
Is the combined worst-case error from resistors and reference within spec? If not, is there a calibration step?
For high voltage, did you check resistor voltage ratings and creepage?

FAQ

What values should I start with?

For general signal dividers, start in the 10k to 100k range so Rth lands at a few kilohms to tens of kilohms. For always-on battery sensing, work backward from the current budget, and once values reach hundreds of kilohms add a capacitor on the ADC pin.

How big should the tap capacitor be?

To keep the worst case (a sampling capacitor still holding another channel's voltage) within 1 LSB, you need 2^N times the sampling capacitance, about 4,000× at 12 bits. For a few-pF sampler that means 22 to 47 nF or more. Beyond that, trade it against response time (τ = Rth × C).

My MCU has an internal battery-sense channel. Do I still need a divider?

Some MCUs can divide and measure their own supply internally, and if the MCU supply is what you want to measure, use that. But behind a regulator the MCU supply is not the battery voltage, so measuring the battery ahead of the regulator still needs an external divider.

When should I use a dedicated IC instead?

If you need accurate state of charge for a lithium battery in percent, voltage alone isn't enough; use a fuel-gauge IC that counts charge. Cell voltage moves too much with load and temperature.