A user already asked a question about it at TL431 Constant current source. My SPECIFIC question is about the figure 29. Sure it is not a complete circuit, so I completed it, in order to understand it, with two resistors, R3 and R4, as place holder for possible "useful circuit". enter image description here And sure, AM1 shows that the current through it is roughly constant (I am using TINA(TM) de Texas Instruments). Also note that AM2 isn't: enter image description here So, the net effect can be accomplished without using the transistor, with a single diode, as shown here:enter image description here

And yes, not shown, but AM1 will still be constant. In fact, if we use nothing, and no connection between R3 and R4, AM1 is constant, as it should.enter image description here

So my question: what is the use of that transistor since it does not help in sinking a constant current between R3 and R4 (which was, I thought, was the purpose of the building block) and since a single diode can do the job ( maybe at a higher expense of energy though ) ?

Note that I varied R3 and R4 with 3 steps each, producing 9 cases, but similar results occur if I use a larger number of steps.

  • \$\begingroup\$ please redraw your schematic so that it follows convention .... V+ at top with R1 and R3 oriented vertically .... you will also see that the last schematic does not make sense \$\endgroup\$
    – jsotola
    Sep 1, 2018 at 20:54
  • \$\begingroup\$ Note that it is not 1.5V but V1 (label) 5V (value). \$\endgroup\$ Sep 1, 2018 at 23:22
  • \$\begingroup\$ Note that the last schema is the way to get a reference of 2.5V as in V=2.5( 1 + 0/R2) = 2.5. Also, R3 and R4 are just place holders, for some circuits using the building block (made of the other parts of the complete circuit). \$\endgroup\$ Sep 1, 2018 at 23:25
  • \$\begingroup\$ As for the pin reference numbers, I agree that they do NOT follow the pin numbers for a TO-92, but they are those of Texas Instruments, not mine, and I confirm that their supplied SPICE model works fine none the less, with this numbering, as used here. \$\endgroup\$ Sep 1, 2018 at 23:29
  • \$\begingroup\$ Also note that for AM1, we have 2.5 V / 220 Ohm = 11.364 mA working fine, I mean, the model fits the theory on that point. \$\endgroup\$ Sep 1, 2018 at 23:34

3 Answers 3


The regulator adjusts its cathode voltage to maintain a constant voltage at the reference input. It's like a reference voltage plus an op-amp with a transistor.

enter image description here

The external transistor will control current depending on the base voltage. So by controlling voltage at the emitter (where the REF input is connected), the collector current (which is about 99.5% of the emitter current) is effectively controlled to within a fraction of 1%, provided you don't run out of compliance voltage. The collector current will be approximately Vref/R2, or about 2.495V/R2

There are slight errors because the transistor gain is not infinite and the REF input draws a couple microamperes, but it's a pretty good current sink. The collector can't quite get to Vref, so the minimum collector voltage is about 2.6V, meaning the voltage across a load to +5 can't be more than about 2.4V/I.

Note that the collector current is what is controlled to be constant. The circuit block is a constant current sink. If you measure the collector current and open up R4 then the collector current should not change (unless you exceed the permissible range of voltages at the collector).

In this case, the current is 2.495V/220\$\Omega\$ = 11.34mA. The maximum load resistance (collector to +5) is about 2.4V/11.34mA = 211\$\Omega\$. When you exceed that resistance you no longer have a constant current sink. So your circuit with 1K is not a useful working example.

The emitter current as a consequence is also constant but that's pretty useless as it changes with R2 so it's more of a set voltage across a resistor.

Here is a quick LTspice simulation with a 100 ohm resistor as a load:

enter image description here

Ic(Q1):  0.0113147   device_current

Reduce the 100 ohm resistor (R3) to 0 ohms and we have:

Ic(Q1):  0.0113152   device_current

So it's a pretty good current sink. Output resistance (ideally it would be infinite) is about 2M\$\Omega\$.

Similarly, if I increase the supply voltage to 10V with R3 constant at 100 ohms, I get:

Ic(Q1):  0.0113529

So about +0.06% per volt line regulation. That could be improved by increasing R2. With R2 = 1K the line regulation is improved to +0.01%/volt and load regulation is slightly better too.

  • \$\begingroup\$ You are right. I initially though that the sinking current was not constant (when components not supposed to modify it were modified, it was greatly modified too) because the transistor was in deep saturation. \$\endgroup\$ Sep 2, 2018 at 12:02
  • \$\begingroup\$ So I tried with a second source, hopefully putting the transistor back in active, or at least, in light saturation, but I was still getting strong variations even putting 12V to the "loop" R3-R4 (instead of letting it connected at the same 5V than for the other components -- no picture included, but available on request). \$\endgroup\$ Sep 2, 2018 at 12:03
  • \$\begingroup\$ I re-tried, this morning, putting it at a higher level, and I need around 20V minimum to get to finally get a constant sinking current (with the transistor). \$\endgroup\$ Sep 2, 2018 at 12:03
  • 1
    \$\begingroup\$ So indeed the circuit works, it is just a matter that it is outside the intended range of initial values, and for a typical Raspberry/Arduino project, as you mention, at least where the voltages are "low", the initial resistors are much too large for the circuit. \$\endgroup\$ Sep 2, 2018 at 12:04
  • \$\begingroup\$ Yup- or you could say (using my circuit) the ratio R3/R2 is too high. \$\endgroup\$ Sep 2, 2018 at 12:45

Perhaps you miss-understand the use of the TL431 and how it works. At heart it is a shunt type voltage regulator, an adjustable zener diode. You supply a safe current through a resistor from a voltage source that is within the TL431's safe limits. By attaching resistors or a trim pot wiper to the control pin, and the CW and CCW pins to cathode and anode, the cathode becomes a multiple of the ratio \$(1 + R1/R2{\cdot} 2.50)\$.

By itself it can only be an adjustable voltage reference, but a very stable one. To get a current clamp or constant current sink you need a transistor to form a feedback loop. Your pdf of this part shown them in detail. Anytime you have a resistor (Re) from the control pin to the anode the cathode remains as positive as possible until the transistor (NPN) conducts enough current so that the Vdrop across the emitter resistor (Re) equals the 2.5 volt internal ref voltage of the TL431. The current is fixed or clamped at 2.5/Re.

This same circuit can act as a current clamp because it will not allow more current to flow at the NPN collector than what creates the 2.5 volt drop on the Re resistor. The current is 'locked' even if the voltage rises up to the safe limits of the transistor, which could be a 300 volt MJE340.

The NPN could be replaced with a N-channel MOSFET with a 700 volts rating. Note that the Collector/Drain voltage can be separate and much higher than what the TL431 needs to function ok. To control a MOSFET the TL431 needs a 12 to 15 volt source so the MOSFET gate can reach the +10 V saturation voltage.

There are many examples for using the TL431 in the datasheet, often using external transistors to boost or limit the current supplied to a load, as well as controlling the supply voltage.

Precision resistors are often required not just for accuracy but because they are very temperature stable, so the voltage/current you set is the same hours later, to .1% or better. Beware of large value trim pots, as they have a temperature drift of ~200 ppm C.


You haven’t included anything varying in your circuit. Of course the current will be constant.

A constant current sink draws the same current as the voltage across it varies. One way to analyze this is to determine the small-signal resistance, dV/dI. For a constant current source this will be very high.

  • \$\begingroup\$ TINA(TM) allows to vary the values as a "What if". Here, the two resistors R3 and R4 were made variable, not to produce time variation, but components variation to see how the "constant sinking current" was indeed "constant" when components not supposed to change its value were modified. \$\endgroup\$ Sep 2, 2018 at 11:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.