How it works
Almost every adjustable regulator, from an LDO to a buck controller, compares a fraction of its output with an internal reference. The resistor from the output to the feedback pin, R_top, and the one from the feedback pin to ground, R_bot, set that fraction, and the loop drives the output until the feedback pin sits at V_ref. That gives V_out = V_ref·(1 + R_top/R_bot). Only the ratio of the two resistors matters for the voltage, so their absolute size is chosen separately, by the current you are willing to spend in the divider.
With standard values the ratio cannot be hit exactly. The calculator takes the divider current you choose, which fixes R_bot near V_ref/I, tries every standard R_bot within a factor of two of that, pairs each with the nearest standard values for R_top, and ranks the pairs by error. It is worth trying more than one series. A ratio of 3.125, which is what 3.3 V from a 0.8 V reference needs, falls almost exactly halfway between two steps of the E96 grid, so no E96 pair gets closer than 0.5 %. The irregular E24 values contain 75 k and 24 k, which give it exactly, and most 1 % resistor families stock those values too.
The nominal error is usually the smallest part of the spread. The reference tolerance moves the output by the same percentage. Resistor tolerance enters through the ratio and is scaled by (1 − V_ref/V_out), so it matters most when the output is several times the reference; two 1 % resistors can shift the ratio by up to about 2 %. The bias current that flows into or out of the feedback pin flows through R_top and adds I_FB·R_top. With tens of kilohms that is negligible, but with megohm resistors chosen to save quiescent current it can be the largest term. The calculator adds all three in the unfavourable direction for the worst case and shows each one alone.
The chart sweeps the divider current from 1 µA to 1 mA with everything else fixed. On the right, the band is set by the reference and resistor tolerances; on the left, it opens up as I_FB·R_top grows. Somewhere between lies a sensible compromise between standby power and accuracy, and noise pick-up on a high-impedance feedback node is another reason not to go too far left.
Design procedure
- 1Read V_ref, its tolerance over temperature, and the maximum FB bias current from the regulator datasheet.
- 2Pick a divider current, or check the datasheet for a recommended R_bot range.
- 3Choose the resistor series you actually stock, look at the best pairs, and try E96+E24 if the error is large.
- 4Check that the worst-case output range fits the load's supply tolerance; if not, tighten the resistor tolerance or choose a regulator with a better reference.
- 5If the regulator uses a feed-forward capacitor across R_top, size it for the final R_top.
Formulas
- Output voltage
- V_out = V_ref·(1 + R_top/R_bot) + I_FB·R_top
- Ideal values from the divider current
- R_bot = V_ref/I_div, R_top = R_bot·(V_out/V_ref − 1)
- Worst-case high
- V_out,max = V_ref(1 + t_ref)·(1 + R_top(1 + t_R)/(R_bot(1 − t_R))) + |I_FB|·R_top(1 + t_R)
- Worst-case low
- V_out,min = V_ref(1 − t_ref)·(1 + R_top(1 − t_R)/(R_bot(1 + t_R))) − |I_FB|·R_top(1 − t_R)
- Divider current and power
- I_div = V_ref/R_bot, P = V_out²/(R_top + R_bot)
Worked example
The defaults set 3.3 V from a 0.8 V reference with ±1 % tolerance, a 50 nA feedback bias current, 1 % resistors from the E96 series and a 50 µA divider current target. The ideal values are 50 kΩ and 16 kΩ. The best E96 pair is 35.7 kΩ over 11.5 kΩ, which gives 3.2835 V, 0.50 % low, at 70 µA; the next best, 33.2 k over 10.7 k, is 0.54 % low. Switching to E48 gives 51.1 k over 16.2 k at 0.71 % high, while E24 contains 75 k over 24 k, which is exact, and E96+E24 finds the same pair. With the E96 pair the worst-case output spans 3.200 to 3.369 V. Of that, the reference accounts for ±33 mV, the resistors for ±50 mV, and the bias current for only ±1.8 mV. Using 0.1 % resistors narrows the range to 3.244 to 3.323 V. Dropping the divider current target to 1 µA instead scales the resistors to 3.57 MΩ and 1.15 MΩ, and the 50 nA bias current alone then moves the output by 0.18 V, widening the worst case to 3.03 to 3.55 V. The pair search and the worst-case values were checked against a brute-force search in Python over hard-coded E24 and E96 tables.
Common mistakes
- Assuming the ±1 % reference is the only error, when two 1 % resistors can add almost another ±2 % for outputs well above the reference.
- Choosing megohm resistors to save quiescent current without checking the FB bias current.
- Using the typical reference voltage instead of its tolerance over temperature.
- Rounding R_top and R_bot independently to the nearest values, which can be worse than searching for a good pair.
- Routing a high-impedance feedback node next to the switch node, where it picks up noise that shifts the average output.
Frequently asked questions
Why can't the E96 series hit 3.3 V from 0.8 V exactly when E24 can?
E96 values are, after rounding, a geometric series with 96 steps per decade, so the ratio of any two values is close to a power of 10^(1/96). The ratio 3.125 sits almost exactly halfway between two such powers, which means every E96 pair misses it by about half a step. E24 values were standardized with irregular roundings such as 75 and 24, and their ratio happens to be exactly 3.125. Combining the two series, as many 1 % resistor families allow, usually gives a better match than either alone.
How large should the divider current be?
Large enough that the feedback-pin bias current is a small fraction of it, with a hundredfold margin a common starting point (TI's SLVA450 suggests it for comparator inputs), and that the feedback node is not so high-impedance that it picks up switching noise. Small enough that the divider does not dominate the standby current. For many regulators that lands between 10 µA and 200 µA; some datasheets recommend a specific R_bot range, which takes precedence.
Does the feed-forward capacitor across R_top change the DC result?
No. A capacitor across R_top adds a zero and pole to the loop response and changes transient behaviour, but at DC it carries no current, so the output voltage is still set by the resistor ratio. Its value does depend on R_top, so if you change the resistances later, recalculate the capacitor.
Which sign does I_FB have?
It depends on the input stage of the error amplifier, and datasheets often give only a magnitude. Current flowing into the pin raises the output (it adds to the current R_top must carry); current flowing out lowers it. The calculator uses the magnitude with both signs for the worst case.
How were the numbers checked?
A separate Python program tries every combination from hard-coded E24 and E96 tables, with R_bot within a factor of two of the ideal value, and computes the worst-case range with the same tolerance directions. Its best pairs and ranges match the calculator. Where several pairs are exact, as with 30k/10k, 36k/12k and 39k/13k for 5 V from 1.25 V, the calculator picks the one closest to your divider current.
References
- Texas Instruments, application report SLVA450, "Optimizing Resistor Dividers at a Comparator Input" (divider current about 100 times the input leakage, and the effect of resistor tolerance and leakage on accuracy; written for supervisor inputs, the same analysis applies to a feedback pin).
- IEC 60063, "Preferred number series for resistors and capacitors" (the E24, E48 and E96 series).
- The pair search and the worst-case ranges on this page were recomputed independently in Python by brute force.
Results are engineering estimates. Verify your design with datasheets, simulation and measurement.