TL431 not giving expected voltage

[HwAoRrDk]

Am I being dumb here? Am I missing something? I designed a TL431 voltage reference into a circuit, but now that I have assembled the PCB and am measuring the voltage at the cathode, I am not seeing the 2.5V expected. I'm only getting 2.41V.

My circuit is as attached. A 1k current-limiting resistor from the 5V rail, and a 100nF capacitor in parallel on the output.

If I measure across the 1k resistor, I can see it's dropping 2.56V, showing that it's passing approx 2.6mA of current to the TL431 (assuming the resistor is bang-on spec; although it is a 5% part, but that shouldn't affect the current by anything meaningful).

Am I perhaps not giving the TL431 enough current? The datasheet says that the minimum regulating current is only 1mA, though.

The part I ordered is supposed to be a 0.5% part, but even if for some reason I ended up with a 1% part instead, 2.41V is still below the minimum 2.475V that might provide.

Or do I just have a duff part?

[HwAoRrDk]

I just tried putting a 220 ohm resistor in parallel with the 1k, for approx. 180 ohm total, which should be passing 14mA through the TL431. That raised the output to 2.48V. With 100 ohm in parallel (total 90 ohm), passing 27mA, it gives 2.49V.

So do I just need to put a lot more current through it? Are the figures in the datasheet for minimum regulating current a load of rubbish?

[ledtester]

This TI app note might be applicable...

https://www.ti.com/lit/an/slva482a/slva482a.pdf

In any case, for stability try using a larger cap, like at least 4.7uF.

[HwAoRrDk]

Ahh, I hadn't considered it might be oscillating. Didn't know that was a thing with TL431s. TIL.

I just poked it with the 'scope, and indeed it is! It's oscillating (with a perfect sawtooth waveform) at around 46kHz with a peak-to-peak voltage of 496mV.

So, if I'm interpreting that graph on the first page of that TI app note, taking the 'A' case (V_KA = V_ref), with a 100nF cap, I'm right in the middle of the unstable region.

This is an 0805 SMD capacitor, so I don't have anything as big as 4.7uF in that size, so would it be okay to go smaller? I have some 4.7nF I could use, which would appear to put me out in the stable region for 2.6mA current.

Or do I just get rid of the capacitor altogether? I only really added it as an attempt at some filtering.

[HwAoRrDk]

I changed the cap to 4.7nF, and it seems to have solved the problem. I get a nice solid 2.49V now, which is about in spec. for a 0.5% part.

Was initially concerned that I was still getting about 20mV p-p oscillation at a few hundred kHz, but when I changed to probing with a spring clip ground connection (direct to other side of the cap), that disappeared and gave me about 3mV p-p.

Should I be concerned that I may still need to reduce the cap value further to accommodate for phase margin (whatever that is) like the TI app note says? Technically, even with 4.7nF and 2.6mA current, I am still outside the stable area of the graph in figure 2.

[floobydust]

TL431's have WIDELY varying stability regions with capacitance, it depends on the manufacturer.
Why are you buying Jiangsu Changjiang and expecting a TI datasheet to be valid? It's hilarious the cut'n'paste datasheet and fonts mix 

[T3sl4co1l]

You don't need a cap on it, or almost never anyway.

In case you do, do it like this.



Don't think of it as a zener, that's simply wrong.  TL431 is better described as a three-terminal self-powered op-amp, with a massive yet suspiciously stable 2.50V input offset voltage, and an open-collector type output.  "REF" is actually -IN.  +IN is tied to GND, which is "A".  "K" is output.  Thus, all the usual stability concerns: C-loading, phase shift, compensation -- all come into play here.

Tim

[magic]

You don't need a cap on it, or almost never anyway.

In case you do, do it like this.

Some words of explanation would be in order regarding those resistors.

Not sure what value you suggest for R2 and C1, but R2 can't be too large because it conducts the reference input current which is ~2µA and varies with temperature.
I suppose R3 isolates the cathode from the capacitive load of C3, is a few ohms sufficient?

Voltage drop across R3 mustn't exceed some 0.3V maybe, because the chip stops working right if the cathode goes significantly negative with respect to the input.
https://www.eevblog.com/forum/projects/trap-for-young-players-tl431-reference-input-current-and-how-to-cook-your-tl431/

In this case there is no danger of cooking unless R2 is very low, but the other problem remains: the emitter follower must have at least some quarter volt positive collector-emitter voltage to work normally.

[T3sl4co1l]

Yeah, typical values for a voltage reference (good for a couple mA) would be R3 = 47 ohm, R2 = 1 to 10k something like that.  Less than 22 ohms I think would be in danger of not having enough effect

Roughly speaking, R2*C1 = R3*C3, give or take other C-loading and such on the ref rail.  Adjust values to get good step response, usual stuff.

Tim

[HwAoRrDk]

You don't need a cap on it, or almost never anyway.

In that case, I might remove it, 'cause then I don't need to change the PCB.

In case you do, do it like this.

I don't understand how this works. Where is the output being tapped off? From the Ref node, or where C3 is?

TL431's have WIDELY varying stability regions with capacitance, it depends on the manufacturer.
Why are you buying Jiangsu Changjiang and expecting a TI datasheet to be valid?

Because the CJ datasheet doesn't have that characteristic graph. But to satisfy my own curiosity, I did a quick survey of TL431 datasheets from Diodes, Nexperia and TI, plus the one Chinese-brand that actually had it in the datasheet, UMW. They all have pretty much the same stability curve (at least for low currents). So I'd say making an assumption about a random no-name part based on the characteristics of other parts is reasonable.

[floobydust]

The CJ datasheet is a Photoshopped Zetex datasheet, which has no stability graph at all. I note they use a 47nF for all their tests.
It gets complicated if you use a MLCC cap. Over temperature, DC bias and aging the actual capacitance you're getting is not what you'd expect. It has to work, or you'll get an oscillator. So there is a lot of risk with the cap, as well as using another manufacturer as an alternate.

[HwAoRrDk]

It gets complicated if you use a MLCC cap. Over temperature, DC bias and aging the actual capacitance you're getting is not what you'd expect. It has to work, or you'll get an oscillator. So there is a lot of risk with the cap, as well as using another manufacturer as an alternate.

For the record, I'm using a 50V X7R capacitor, a Yageo CC0805KRX7R9BB472. I just checked the characteristics, and changes in capacitance due to DC Bias and temperature (at least, over the expected usage range) are insignificant.

But, if you really expect that things may go significantly wayward due to time and environment, then I might as well remove the cap altogether.

[magic]

A capacitor could potentially be useful to reduce high frequency output impedance of the circuit, maybe in some situations.
It doesn't filter TL431 noise, because the reference is shorted straight to it and charges or discharges the cap as it pleases.

[MrAl]

I am surprised the app note does not mention anything about output cap ESR.  If you use an output cap, it should have some reasonable ESR that is NOT too low.  Sounds counterproductive but this isnt the only application like that.  A boost converter with some reasonable ESR that is NOT close to zero becomes more stable.  That's because it gives the feedback a little wiggle room to level off without being too strongly affected by the clamping action of the output cap.
In a way the output cap can cause a delay so the feedback does not reach the input as soon as the output of the driver reacts to the previous input.  With a small value resistor in series with the cap, the the feedback reaches the input in a more timely manner and allows a little tiny wiggle to persist if needed, something that the cap would shunt to ground.
There is quite a bit to this though probably cant address everything in a casual conversation.  Feedback circuits have to be analyzed very carefully.

[T3sl4co1l]

I don't understand how this works. Where is the output being tapped off? From the Ref node, or where C3 is?

The common (top) rail, with C3.  Implication being, start with any old shunt circuit (note that R2 may have a companion divider resistor below it, REF to GND), add R3 in series, and C3 can then be whatever value as needed.  Pull-up not shown either.

I am surprised the app note does not mention anything about output cap ESR.  If you use an output cap, it should have some reasonable ESR that is NOT too low.  Sounds counterproductive but this isnt the only application like that.  A boost converter with some reasonable ESR that is NOT close to zero becomes more stable.  That's because it gives the feedback a little wiggle room to level off without being too strongly affected by the clamping action of the output cap.

Note that relevant ESRs are much higher than most capacitors will have.  As just any other op-amp circuit, an "isolation" resistor of some 10s of ohms is typically required.

Less ESR may be required for values near the instability region, than for values in the middle of it; the region also varies with bias current.

The other "best answer" is this: if you need bypass for filtering purposes, TL431 simply isn't what you want, get a less noisy ref instead!  LM4041 say, or various other choices.  (Note that LM4041 is variable, like TL431, but its REF is referenced to "cathode" instead, so the divider ratio is upside-down! ).

There are also refs available, with pass type outputs (lower current consumption), and with internal filtering nodes (so you can tack on a bypass cap there, and the output pin is a voltage follower from it).

And, the related "non-answer": don't inject AC currents into your REF node to begin with!  If you are, you're probably doing it wrong.  A non-answer because it could require significant circuit revision, but revision that is nonetheless worthwhile from a design practices standpoint (if not worthwhile/applicable for a particular mostly-complete design).

For example, last time I needed 2.5V ref, I used a TL431, no bypass, and passed that straight into the MCU (AVR-DA) VREF pin (12-bit ADC).  Readings are stable and accurate.

Tim

[Terry Bites]

I think the TI and other datasheets are correct wrt to operating current. See www.ti.com/lit/zip/slvc093.

[MrAl]

I don't understand how this works. Where is the output being tapped off? From the Ref node, or where C3 is?

The common (top) rail, with C3.  Implication being, start with any old shunt circuit (note that R2 may have a companion divider resistor below it, REF to GND), add R3 in series, and C3 can then be whatever value as needed.  Pull-up not shown either.

I am surprised the app note does not mention anything about output cap ESR.  If you use an output cap, it should have some reasonable ESR that is NOT too low.  Sounds counterproductive but this isnt the only application like that.  A boost converter with some reasonable ESR that is NOT close to zero becomes more stable.  That's because it gives the feedback a little wiggle room to level off without being too strongly affected by the clamping action of the output cap.

Note that relevant ESRs are much higher than most capacitors will have.  As just any other op-amp circuit, an "isolation" resistor of some 10s of ohms is typically required.

Less ESR may be required for values near the instability region, than for values in the middle of it; the region also varies with bias current.

The other "best answer" is this: if you need bypass for filtering purposes, TL431 simply isn't what you want, get a less noisy ref instead!  LM4041 say, or various other choices.  (Note that LM4041 is variable, like TL431, but its REF is referenced to "cathode" instead, so the divider ratio is upside-down! ).

There are also refs available, with pass type outputs (lower current consumption), and with internal filtering nodes (so you can tack on a bypass cap there, and the output pin is a voltage follower from it).

And, the related "non-answer": don't inject AC currents into your REF node to begin with!  If you are, you're probably doing it wrong.  A non-answer because it could require significant circuit revision, but revision that is nonetheless worthwhile from a design practices standpoint (if not worthwhile/applicable for a particular mostly-complete design).

For example, last time I needed 2.5V ref, I used a TL431, no bypass, and passed that straight into the MCU (AVR-DA) VREF pin (12-bit ADC).  Readings are stable and accurate.

Tim

We could talk about the ESR issues some more, but more to the point it is interesting that you suggested a different voltage reference IC.  I was going to do that myself but didnt get back here before you.  The way i see it is, in this day and age if you have to fuss with the oscillation of something as simple as a voltage reference, as you say also, get a different reference.  There are far, far too many more that will work much better with no problem.  Many of them even have a MORE stable output voltage with variations in temperature, even though the 431 is not really that bad.  A quick search will most likely turn up a lot of them.

This all makes me think that maybe the 431 chip is becoming outdated.  They still use them in a lot of regulated wall warts though so they probably wont stop making them.  Just like the LM358, we hear a lot of complaints but it's still a very useful op amp.

[T3sl4co1l]

Oh absolutely; it's a sort of "ideal transistor", and a 3-terminal op-amp.  It's too useful to go away, even if it's not the best at its original purpose (cost excepted).

Tim

[floobydust]

The TL431 has been improved as TI TL431LI and Diodes Inc AS431 (which replaces their TL431 offering which had the worst stability region, your basic nightmare).
Here's some stability curves, to show the huge variations.

[MrAl]

Oh absolutely; it's a sort of "ideal transistor", and a 3-terminal op-amp.  It's too useful to go away, even if it's not the best at its original purpose (cost excepted).

Tim

It can also be used as a comparator although you have to play around a little to get different voltage trip points.
I think the TL432 part is better tolerance, maybe (verify part number).

Along these lines one of my favorite chips from the past was the LM432 Dual Op Amp with reference.  Very versatile for use in various applications because it has a voltage reference AND two op amps similar to LM358.

[HwAoRrDk]

The common (top) rail, with C3.  Implication being, start with any old shunt circuit (note that R2 may have a companion divider resistor below it, REF to GND), add R3 in series, and C3 can then be whatever value as needed.  Pull-up not shown either.

Ohh, I see. That was what was throwing me: the current-limiting (or "pull-up" as you call it) resistor is not shown. I even simulated it with only what was shown and as expected I just got the input voltage straight out, but also the TL431 oscillated like crazy.

However if I add a prior series resistor on the input it works in the simulator - using R2=1k, R3=47, C1=4.7n, C3=100n, maintaining the ratio you mentioned.

I don't think I'll bother using this, though - too many components - I'll just remove the cap from my existing circuit.

[T3sl4co1l]

It can also be used as a comparator although you have to play around a little to get different voltage trip points.
I think the TL432 part is better tolerance, maybe (verify part number).

Just reverse pinout.  The suffixes (or variants from other mfgs) have different tolerances, and also higher or lower Vmax, Vref, same or lower Imin, etc.  Some of which also have lower noise.

With 0.5% or somewhat less available, they're quite capable; that's about 8 bits of guaranteed scale, and a little trimming will get you a tad better than that, give or take allowable tempco and noise.  If you need more like 12+ bits of course, a better reference is probably called for.

Along these lines one of my favorite chips from the past was the LM432 Dual Op Amp with reference.  Very versatile for use in various applications because it has a voltage reference AND two op amps similar to LM358.

Fun yeah, have used LT6703 before.  Not that LT is usually the first thing you'll reach for, for obvious reasons, but...

Or the analog equivalent: AP4312, NCP4328, SEA05, etc. Real handy for PSUs.

Tim

[magic]

The TL431 has been improved as TI TL431LI

Not a problem for switcher applications, but something to keep in mind: they removed noise specifications from the datasheet, which means it must be noisy as hell

They also removed typical minimum current for regulation spec and reduced reference input current 10x.
The schematic is the same (although I'm not 100% sure if it's correct).

Starting to see a pattern here ...

[Dannyx]

Hey guys. Old thread, I know, but I was curious which of the info I should use when it comes to the stability graphs for the capacitive load on TL431A: this REALLY old document from TI, OR the graphs in post here #19 ? The values seem different.

[magic]

For current production parts from TI use the current datasheet from TI. Note that there appear to exist two die versions for different packages and there are two plots in the datasheet.

For other manufacturers, replace TI above with Other Manufacturer.

For variable/unknown manufacturer, stay far away from typical instability regions and pray that your margin is enough.