100nF

TL431 shunt regulator: R1, R2 and the cathode resistor

Set a TL431 output with the I_ref × R1 term kept, then size the cathode resistor across input and load: 5 V from 9–15 V at 20 mA needs 99.6 to 187.8 Ω.

A TL431 shunt regulator takes two calculations. The divider sets the voltage: VKA = Vref(1 + R1/R2) + Iref × R1, with Vref = 2.495 V and Iref = 2 µA typical, so 10 kΩ over 10 kΩ gives 5.010 V, of which 20 mV is the reference current flowing through R1. The cathode resistor RS sets the current, and it has to satisfy two corners at once: at least 1 mA left in the TL431 at the lowest input with the heaviest load, and no more than 100 mA (or what the package can dissipate) at the highest input with no load. For a 5 V rail fed from anywhere between 9 V and 15 V and supplying up to 20 mA, those corners put RS between 99.6 Ω and 187.8 Ω. 180 Ω leaves 1.92 mA in the TL431 at the bad corner and 55.2 mA at the other.

Everything below comes from TI’s TL431/TL432 datasheet, SLVS543S, and every derived number from the same module the TL431 calculator runs, so the two agree. The datasheet prints its own divider equation three different ways, and the Iref term is where they part company.

How a TL431 shunt regulator works

The datasheet’s description (p. 1) is short: the TL431 is a three-terminal adjustable shunt regulator, and

The output voltage can be set to any value between Vref (approximately 2.5V) and 36V, with two external resistors.

Inside is a reference and an error amplifier driving a Darlington to the anode. Tie REF to the midpoint of a divider across the output and the part sinks whatever current holds that midpoint at Vref. It behaves like a Zener whose voltage you choose, and like a Zener it can only sink. Everything it sinks, and everything the load and the divider take, arrives through one resistor from the input.

Schematic of a TL431 shunt regulator. A series resistor R_S runs from the input to the output node. From that node three branches go to ground: the R1 and R2 divider whose midpoint drives the TL431 REF pin, the TL431 itself from cathode to anode, and the load. Arrows mark the current through R_S splitting into the divider current, the cathode current and the load current. For the worked example at 9 volts in with a 20 milliamp load and R_S of 180 ohms, 22.17 milliamps through R_S divides into 20 milliamps of load, 0.25 milliamps of divider and 1.92 milliamps left for the TL431.
Fig 1 — Every current in a TL431 shunt regulator passes through R_S, and the TL431 gets what the load and the divider leave. The numbers are the worked example at its worst corner, computed by the calculator's module: 9 V in, 20 mA of load, R_S = 180 Ω, leaving 1.92 mA in the cathode.

Kirchhoff’s current law at the output node does most of the design: (VIN − VKA)/RS = IKA + ILOAD + IR1. The current through RS depends only on the input voltage, not on the load. When the load draws less, the TL431 takes the difference. When the input sags with the load at full, the TL431’s share is what is left over, and if that falls below the minimum, the part stops regulating. Section 8.1 (p. 21) says so directly:

In order for this device to behave as a shunt regulator or error amplifier, >1mA (Imin(max)) must be supplied in to the cathode pin.

The electrical characteristics (pp. 6–14) give Imin, the “Minimum cathode current for regulation”, as 0.4 mA typical in every grade, with a maximum of 1 mA for the standard grade, 0.6 mA for the A and B grades in the C range and 0.7 mA in the I and Q ranges. The recommended operating conditions (§6.4, p. 5) run the cathode from 1 mA to 100 mA. Designing to 1 mA covers every grade.

Choosing R1 and R2 for the output voltage

Figure 7-2 (p. 19), the test circuit for VKA > Vref, prints the equation with R1 from cathode to REF and R2 from REF to anode:

VKA=Vref(1+R1R2)+Iref×R1V_{KA} = V_{ref}\left(1 + \frac{R_1}{R_2}\right) + I_{ref} \times R_1

Solved for R1 at a chosen R2:

R1=VKA−VrefVref/R2+IrefR_1 = \frac{V_{KA} - V_{ref}}{V_{ref}/R_2 + I_{ref}}

Vref is 2495 mV typical in every grade and Iref is 2 µA typical and 4 µA maximum, both at 25 °C with 10 mA in the cathode (pp. 6–14). With R2 = 10 kΩ, an exact 5.000 V needs R1 = 9.960 kΩ. The nearest stocked value is 10.0 kΩ, which gives 5.010 V. A 3.3 V output needs 3.201 kΩ; E96 has 3.24 kΩ, for 3.310 V (+0.30 %). A 12 V output needs 37.79 kΩ, and E96’s nearest, 37.4 kΩ, gives 11.901 V (−0.82 %). Rounding to a stocked series is often a larger error than the reference tolerance of a good grade, which is why the calculator searches R2 as well as R1. The E-series calculator lists what is available near any value, and the voltage divider calculator handles the plain divider with loading and tolerance.

R2 also sets the divider current, Vref/R2 + Iref: 0.25 mA with 10 kΩ, 2.5 mA with 1 kΩ. That current comes through RS along with the load’s, so a stiff divider costs supply current at every input voltage. A large one costs accuracy, through the second term.

The Iref × R1 term, and the datasheet’s three versions of it

The REF pin is the base of an NPN transistor. Section 8.3 (p. 22) explains why it cannot float:

However, the reference pin can not be left floating, as it needs IREF ≥ 4µA (please see Electrical Characteristics, TL431C, TL432C). This is because the reference pin is driven into an npn, which needs base current in order operate properly.

The tables it points to list 4 µA as the maximum, not a minimum, so the sentence reads best as “it needs a path for up to 4 µA”. That current flows out of the divider’s midpoint into REF. R1 therefore carries the R2 current plus Iref, and the extra drop across R1 raises the cathode by Iref × R1. Figure 7-2’s plus sign is the one Kirchhoff gives.

The datasheet does not print it consistently. Figure 9-3 (p. 26), the shunt regulator schematic, labels the output VO = (1 + R1/R2) Vref with no current term, as do the system examples on p. 28. The text beneath Figure 9-3 then offers a refinement with the opposite sign:

The cathode voltage can be more accuratel determined by taking in to account the cathode current:

Vo=(1+R1/R2)*VREF-IREF*R1

It says cathode current, but the term is the reference current, and the minus sign contradicts both the circuit and Figure 7-2. Here is what each version does to a 12 V design with R2 = 10 kΩ, each R1 evaluated with Figure 7-2’s equation at 2 µA:

  • Figure 7-2, plus sign: R1 = 37.79 kΩ, output 12.000 V.
  • Figure 9-3, no term: R1 = 38.10 kΩ, output 12.076 V.
  • Page 26, minus sign: R1 = 38.40 kΩ, output 12.154 V.

The B grade’s whole initial tolerance at 12 V is ±0.48 %, or ±58 mV. Leaving the term out with 10 kΩ resistors costs 76 mV at the typical reference current, and taking page 26’s sign doubles that.

Log-log plot of how far a 12 volt TL431 output lands above target when R1 is chosen from the simple equation that leaves out the reference current, against R2 from 1 kilohm to 100 kilohms. The error is a straight line, 0.06 percent at 1 kilohm, 0.63 percent at 10 kilohms and 6.3 percent at 100 kilohms with the typical 2 microamps, and twice that with the 4 microamp maximum. Horizontal lines mark the 25 degree reference tolerance of each grade, 0.48, 1.0 and 2.2 percent; at 10 kilohms the typical error already exceeds the B grade tolerance. A second row of labels gives the divider current, 2.5 milliamps at 1 kilohm falling to 25 microamps at 100 kilohms.
Fig 2 — The I_ref × R1 term at 12 V. R1 is solved from V_ref(1 + R1/R2) alone and the output computed with Figure 7-2's full equation. With R2 = 10 kΩ the output lands 76 mV high at the typical 2 µA, 152 mV at 4 µA: more than the B grade's whole initial tolerance. Smaller resistors shrink the error and cost divider current, shown under the axis.

The error scales with R1, so it is a choice of divider current. At R2 = 1 kΩ the omitted term is under a tenth of a per cent at 12 V; at 100 kΩ it is 6 %. The term never goes away entirely, because Iref itself varies from part to part (no minimum is printed) and over temperature by II(dev), up to 1.2 µA in the C range and 2.5 µA in the I and Q ranges. Keep it in the equation and keep R1 small enough that its uncertainty is small next to the reference tolerance.

Sizing the TL431 cathode resistor

RS has two limits, one from each corner of the operating range. It must be small enough to leave the minimum cathode current when the input is lowest and the load heaviest, and large enough that the cathode current stays under its limit when the input is highest and the load lightest:

VIN(max)−VKAIKA(max)+ILOAD(min)+IR1≤RS≤VIN(min)−VKAIKA(min)+ILOAD(max)+IR1\frac{V_{IN(max)} - V_{KA}}{I_{KA(max)} + I_{LOAD(min)} + I_{R1}} \le R_S \le \frac{V_{IN(min)} - V_{KA}}{I_{KA(min)} + I_{LOAD(max)} + I_{R1}}

The datasheet states the low-corner rule in the note to Figure 9-5 (p. 28):

R should provide cathode current ≥1mA to the TL431 at minimum V(BATT).

For the worked example (5.010 V, 9 V to 15 V in, 0 to 20 mA of load, 0.25 mA in the divider, 1 mA floor, 100 mA ceiling):

RS(max) = (9 − 5.010) / (1 mA + 20 mA + 0.2515 mA)   = 187.8 Ω
RS(min) = (15 − 5.010) / (100 mA + 0 + 0.2515 mA)    = 99.6 Ω

Pick the largest stocked value under RS(max), because it wastes the least current at every other condition. In E24 that is 180 Ω. At 9 V with 20 mA drawn it passes 22.17 mA, of which 1.92 mA reaches the TL431. At 15 V with the load disconnected it passes 55.5 mA, and the TL431 sinks 55.2 mA of it.

Plot of TL431 cathode current against the cathode resistor R_S from 60 to 400 ohms on a log axis, for a 5 volt output from 9 to 15 volts with a 0 to 20 milliamp load. Two falling curves: at 15 volts with no load the cathode current is highest and crosses the 100 milliamp limit at 99.6 ohms and the SOT-23 thermal limit of 77.5 milliamps at 128.5 ohms; at 9 volts with 20 milliamps of load it is lowest and falls through the 1 milliamp floor at 187.8 ohms, going negative beyond, where the TL431 drops out of regulation. The usable window lies between, and the chosen 180 ohm resistor gives 1.9 milliamps at the low corner and 55.2 milliamps at the high corner.
Fig 3 — The usable cathode resistors lie between two edges. The right edge is the 1 mA floor at 9 V with the full load (R_S ≤ 187.8 Ω); its left edge is the 100 mA limit at 15 V with no load (R_S ≥ 99.6 Ω), or the package, which for SOT-23 at 70 °C allows 77.5 mA and moves the edge to 128.5 Ω. 180 Ω is the largest E24 value inside it.

The two curves are far apart because both differences between the corners work in the same direction. The headroom across RS at 15 V (9.99 V) is two and a half times that at 9 V (3.99 V), and at the high corner the load’s 20 mA is no longer subtracted. The low-corner curve falls through the 1 mA floor and keeps going: past 197 Ω the input cannot even feed the load and the divider, and the output sags with the TL431 off.

When no cathode resistor works

Raise the example’s maximum load to 40 mA and the window closes: RS(max) falls to 96.7 Ω, below the 99.6 Ω that keeps the TL431 under 100 mA at 15 V with no load. No resistor satisfies both corners. For this input range the heaviest load any single resistor can serve is 38.8 mA, and 29.8 mA once the SOT-23 thermal limit from the next section replaces the 100 mA one.

The reason is structural. A shunt regulator draws its full design current from the input all the time, and whatever the load does not take, the TL431 must. Every milliamp of load capacity bought at the low corner returns as a milliamp of TL431 current when the load is removed, on top of the extra current the higher input pushes through RS. A wide input range and a wide load range together leave no window.

The datasheet’s own answers are to let the TL431 control a larger device. Figure 9-7 (p. 28) adds a transistor across the output to sink the bulk of the current, and Figure 9-5 turns the TL431 into the reference of a series regulator with a pair of pass transistors. Figure 9-9 (p. 29) sets an LM317 with it. A series regulator draws only what the load takes, and the LM317 calculator and the article on LDO dropout voltage cover the headroom such a design needs instead.

TL431 power dissipation and the package

The TL431 dissipates VKA × IKA, worst at the high corner. In the example that is 5.010 V × 55.2 mA = 277 mW. The thermal table (§6.3, p. 5) gives RθJA for each package, from 52 °C/W for SOT-89 (PK) to 259 °C/W for SC-70 (DCK), with 206 °C/W for both SOT-23 packages. The note under the recommended operating conditions turns it into a limit:

The maximum allowable power dissipation at any allowable ambient temperature is PD = (TJ(max) – TA)/θJA. Operating at the absolute maximum TJ of 150°C can affect reliability.

At 70 °C, the top of the C range, a SOT-23 may dissipate 388 mW. At 5 V that is 77.5 mA, not 100 mA, and the window’s left edge moves from 99.6 Ω to 128.5 Ω. The 180 Ω design puts the junction at 70 + 0.277 × 206 = 127 °C, inside the limit with modest margin. A TO-92 (140 °C/W) allows 571 mW under the same conditions and a SOT-89 1.54 W, which clears 100 mA at any output up to 15 V.

RS runs hotter than the TL431. It dissipates (VIN − VKA)²/RS whatever the load does, and at 15 V that is 9.99² / 180 = 554 mW, more than double a quarter-watt rating. The absolute maximum cathode current is 150 mA (§6.1), and §9.4 (p. 31) adds a warning that matters for a shunt: “In order to not exceed the maximum cathode current, be sure that the supply voltage is current limited.” RS is that limit, so it must still be there when the input is at its highest.

Reference accuracy by grade and over temperature

The grade sets the initial tolerance of Vref at 25 °C, and the temperature range sets how far it may wander:

GradeVref at 25 °CToleranceVI(dev) max, C rangeVI(dev) max, I range
B2483–2507 mV±0.48 %16 mV34 mV
A2470–2520 mV±1.0 %16 mV (SOT23-3, TL432), 25 mV (others)34 mV, 50 mV
standard2440–2550 mV±2.2 %16 mV (SOT23-3, TL432), 25 mV (others)34 mV, 50 mV

The tolerance column is the printed limit’s distance from 2495 mV; page 1 rounds it to 0.5 %, 1 % and 2 %. Every Q-range part has 34 mV. VI(dev) is a spread. The tables’ footnote defines it as “the differences between the maximum and minimum values obtained over the rated temperature range” (p. 6), and §7.1 (p. 19) converts it to an average:

∣αVref∣(ppm°C)=(VI(dev)Vref at 25 °C)×106ΔTA\left|\alpha_{V_{ref}}\right| \left(\frac{\text{ppm}}{°\text{C}}\right) = \frac{\left(\dfrac{V_{I(dev)}}{V_{ref}\ \text{at}\ 25\,°\text{C}}\right) \times 10^6}{\Delta T_A}

For the B grade in the C range that is 34 ppm/°C from the 6 mV typical spread and 92 ppm/°C from the 16 mV maximum; for a standard-grade I-range part in SOIC, 50 mV over 125 °C, it is 160 ppm/°C. The section adds a caution worth repeating: “The full-range temperature coefficient is an average and therefore any subsection of the rated operating temperature range can yield a value that is greater or less than the average.”

Two further terms do not depend on grade. ΔVref/ΔVKA is −1.4 mV/V typical and −2.7 mV/V maximum from Vref to 10 V, and −1 and −2 mV/V from 10 V to 36 V: Vref is specified with the cathode at Vref, and it sags as the cathode rises. At 5 V the worst case lowers the output by 13.5 mV. Iref adds its own uncertainty through R1. Section 9.2.2.2.2 (p. 27) lists the errors that apply above Vref: resistor accuracy, VI(dev), ΔVref/ΔVKA and dynamic impedance. The calculator adds them, each at its worst, into a band.

Diverging bar chart of the worst-case error terms of a 5.01 volt TL431A output with 1 percent resistors, 0 to 70 degrees and a cathode current between 1.9 and 55 milliamps. Initial reference tolerance is plus or minus 50 millivolts, resistor tolerance about plus or minus 50, temperature deviation plus or minus 32, reference current minus 20 to plus 32, the cathode-voltage coefficient minus 14 and zero, and dynamic impedance minus 8 and plus 45. Below, the total band for each grade: B, A and standard, each wider than the grade tolerance alone because the other terms do not depend on grade.
Fig 4 — Where the error in a 5 V TL431 output comes from, each term at its worst, computed by the calculator's vkaBand for 1 % resistors, 0 to 70 °C and the 1.9 mA to 55.2 mA cathode current of the worked example. The grade sets only the first row; past the B grade the resistors and the dynamic impedance dominate.

For the 5 V example with 1 % resistors over 0 to 70 °C, the band is 4.864 to 5.195 V for the B grade, 4.839 to 5.221 V for A and 4.779 to 5.282 V for the standard grade. With 0.1 % resistors the A grade tightens to 4.882 to 5.174 V. The grade moves only the first bar of Fig 4; the resistor term is as large as the A grade’s reference tolerance, so a B-grade part with 1 % resistors buys less than its price suggests. These are bounds, every term at its limit in the same direction, not a prediction of a typical board.

One inconsistency belongs here. The comparator section (p. 24) states that “The internal virtual reference voltage will be within the range of 2.5V±(0.5%, 1.0% or 1.5%) depending on which version is being used.” The tables and page 1 give the standard grade 2 %, and its printed limits are 2.2 % either side of 2495 mV. Take the tables.

Dynamic impedance: how much the output moves with current

|ZKA| is 0.2 Ω typical and 0.5 Ω maximum, for f ≤ 1 kHz and IKA from 1 to 100 mA. With a divider the gain multiplies it, §7.2 (p. 20):

∣z′∣=ΔVΔI≈∣ZKA∣(1+R1R2)|z'| = \frac{\Delta V}{\Delta I} \approx |Z_{KA}|\left(1 + \frac{R_1}{R_2}\right)

The 5 V divider doubles it, to 0.4 Ω typical and 1.0 Ω maximum; the 12 V divider (37.4 kΩ over 10 kΩ) takes the maximum to 2.37 Ω. The same section notes what that means for accuracy: Vref is specified at a test current, and

Any deviation from Itest can cause deviation on the output VKA.

The test current is 10 mA. The example runs from 1.92 mA to 55.2 mA, so the maximum impedance moves the output by −8 mV and +45 mV from the 10 mA value, the last bar of Fig 4.

For regulation the impedance decides everything, since in a shunt every change in load or input is a change in IKA. Removing the 20 mA load adds 20 mA to the cathode and moves the output by at most 20 mV. Raising the input from 9 V to 15 V adds 6 V / 180 Ω to the cathode, which through the divider of RS and |z’| moves the output by at most 33 mV. Figure 6-12 (p. 17) shows |ZKA| flat at the low end of its 1 kHz to 10 MHz axis and rising by roughly two decades over the upper part of it, so fast load steps meet a higher impedance than the table’s 0.5 Ω.

Output capacitor and TL431 stability

A capacitor across the output is optional, and adding one is the risk. Section 8.3 (p. 22):

Unlike many linear regulators, TL43xx is internally compensated to be stable without an output capacitor between the cathode and anode. However, if it is desired to use an output capacitor Figure 6-18 can be used as a guide to assist in choosing the correct capacitor to maintain stability.

There are two charts, and the part and package decide which applies. Figure 6-16 (p. 17) is for “All TL431 and TL431A Devices (Except for SOT23-3, SC-70, and Q-Temp Devices)” and Figure 6-18 (p. 18) for “All TL431B, TL432, SOT-23, SC-70, and Q-Temp Devices”. The example’s TL431A in SOT-23-3 falls under Figure 6-18. Each plots cathode current against load capacitance from 0.001 µF to 10 µF at 25 °C, with boundary curves for VKA = Vref, 5 V, 10 V and 15 V, and the note:

The areas under the curves represent conditions that can cause the device to oscillate.

Both charts print “Stable” on each side of the curves. Small capacitance is stable and so is large; the trouble is in the middle of the axis, and how far it reaches depends on VKA and IKA. A shunt regulator moves along the IKA axis as its load and input change, so check the capacitor at both corners of the cathode current, here 1.92 mA and 55.2 mA, not at one operating point. Section 9.2.2.2.3 (p. 27) also warns that the capacitance may not be yours to choose: “the device that receives the shunt regulator’s output voltage could present a capacitive load that is within the TL43xx region of stability”, where “region of stability” means the region the chart marks as unstable. Decoupling on the powered circuit counts. The datasheet’s own example in Table 9-2 (p. 26) specifies 10 µF, the right-hand end of the chart axis, and points to application note SLVA482 for how the charts are measured.

Sources

  • Texas Instruments, TL431, TL432 Precision Programmable Reference, SLVS543S, August 2004, revised May 2024. Absolute maximum ratings, thermal information and recommended operating conditions (p. 5); electrical characteristics by grade and temperature range (pp. 6–14); the stability boundary charts, Figures 6-16 and 6-18 (pp. 17–18); Figure 7-2’s equation and the temperature coefficient (p. 19); dynamic impedance (p. 20); the overview and feature description (pp. 21–22); the comparator overdrive note (p. 24); the shunt regulator design, Figure 9-3, Table 9-2 and the page 26 equation (pp. 26–27); the system examples (pp. 28–29); power supply recommendations (p. 31).