Trap for young players: TL431 reference input current and how to cook your TL431

[magic]

So TL431 has absolute maximum rating on reference input current: 10mA.

It may look like not a big deal: after all, the input is supposed to be high impedance going into some opamp thingy (the detailed schematic shows an emitter follower) and it would take some pathological conditions to force significant current into it.



But is it really so hard to create such pathological conditions?

The problematic part is marked in red and it's the base-collector junction of the input emitter follower. As long as cathode voltage is above reference voltage, the follower works normally. But when the reference input is driven above 2.5V, the chip starts to sink current like crazy and it is very easy to create a circuit where the cathode will be pulled below 2.5V or even 2V in such event - most applications pull up the cathode to some voltage rail with a few kΩ resistor.

And this means trouble, because the emitter follower's base-collector junction becomes forward biased and then the cathode may sink current through it from the input pin. The effect may be somewhat moderated by the fact that the output pulldown transistor is driven by the PNP current mirror which in turn is driven by that NPN cascode transistor which will saturate when cathode voltage is too low, but this limiting mechanism is borderline effective and there is so many variants of TL431 that it would be foolish to count on it.

Needless to say, I learned it the hard way. By my mistake, the chip's input was tied to a ~9V rail and it got pretty hot pretty fast
Afterwards, I found permanent shift in reference voltage by several mV.

It's also possible to create this condition in seemingly innocent circuits, such as this application example taken from the datasheet. If R1 is set to zero, both preconditions are satisfied: the reference pin is exposed to an external source of high current and the cathode is about one diode drop below 2.5V. It is quite a random chance whether the external transistor or the reference pin will end up sinking the current



In general, all circuits potentially operating TL431 at unity gain may be affected. A simple fix is to put resistance in series with the input, but mind that this generate a small voltage shift due to the 2µA input bias current.

[T3sl4co1l]

Yep, good perspective on this part is to remember it's all diodes between terminals.  So, A-C is a diode (often shown explicitly, as here), A-R (not shown in the above schematic, but present I believe either for ESD or parasitic for simply how the pin is laid out? or, is it the diode-strapped NPN plus diode so it can actually be -2Vbe?), and R-C (the B-C junction highlighted above).

Sometimes we forget that BJTs are symmetrical, if poorly so, but not so poorly that their diodes don't work!

Also relevant to applications of this: a complementary emitter follower for example, can deliver rail-to-rail output if the input is driven slightly beyond the rails.  The input will be clamped to a Vbe beyond the rails due to each B-C junction.

Another way to express the TL431, is as an ideal BJT, with 2.50V Vbe and very little tempco, minimal Ib (which is to say, very high hFE, high enough we don't care*), and a rather poor Vce(sat) in comparison, being about 1.8V.  Which we can further model as having a Baker clamp from REF to C, which is, after all, a physically correct statement -- and also gives us the exception to the hFE, i.e., like a regular BJT, hFE approaches zero in saturation; which is to say in this case, Vca won't drop any more, no matter how much current we dump into REF.

*Or maybe it's even negative, which is also fine: in terms of input resistance r_pi, it's effectively near infinity, and the real number line wraps around at infinity so whether it's near + or -infty, it's not violating continuity or anything.

Tim

[TaylorD93]

Other thing to be very wary of is stability.

If you look in the datasheet there are plots showing the allowable capacitive loads depending on the Cathode Current and Cathode-Anode Voltage. May think added extra decoupling can help, but depending on how much can cause it to be massively unstable.

Once had one ringing away at 130MHz quite happily

[mawyatt]

Most reference type inputs, even generic Op-Amp inputs, have this problem. Whether forward biasing a bipolar BC junction, the IC substrate diode, or the input protection clamp diodes. Generally it's not a good idea to have an input signal that exceeds the most positive chip feature (cathode in this case) or most negative chip feature, like substrate diode (anode substrate in this case). Some CMOS is particularly sensitive to a forward biased substrate diode, and can go into destructive SCR type latchup.

About 50 years ago we lost a very expensive custom hybrid DAC to power supply sequencing due to CMOS substrate forward biasing during startup. On another later occasion where custom hybrid high resolution integrating navigation digitizers were failing after a few hundred hours of operation. This was eventually traced to an LM108A where the input super-beta NPN BC was being slightly and momentarily forward biased during power down, slowly degrading the beta and causing a rising input bias current which in turn caused an increasing error drift. The damaging charge came from the integrating capacitor.

Best,

[magic]

That's true, but it's easy to forget when the chip has no explicit supply pin

Also, something like LM317 will take a few volts on its sense input before protection kicks in. TL431 could do the same by getting rid of the follower, but then its input bias current would be more like LM317's.

Some vendors are more honest in their datasheets
This is Wing Shing ("WS"), popular on auction sites. ON Semi also shows the trap diode.



I tested a few chips today:
- 1431T - 1990s, Asian, maybe Toshiba(?)
- AZ431 - 2000s, BCD Semiconductor (?)
- TL431CLP - modern, ON Semi
- HA17431 - 1990s, Hitachi (?)
- TL431ACLP - fairly modern, Fairchild, likely ex-Samsung

The test: cathode is open circuit, reference is pulled with 100Ω to 5V.
The 1431T clamped to 1.7V, sinking 33mA.
The next pair clamped to 3V, sinking 20mA.
The final pair clamped to 4.2V, sinking 8mA.

Then the two modern chips were subjected to the same with 12V.
ON Semi clamped to 4V, sinking 80mA.
Fairchild clamped to 8.5V, sinking 35mA.

In all cases, the cathode was 0.8~1.2V below the reference. Shorting the cathode to the reference restored normal operation, with 25mA/95mA sink current.

Conclusions: everything as predicted
They don't sink as much as they would want to, but enough to cause trouble. High variability between vendors.

The chip which I abused previously was National, BTW, but it's still soldered into its circuit (and working normally) so I left it alone. And unfortunately I ran out of Wing Shing too.

[magic]

*Or maybe it's even negative, which is also fine: in terms of input resistance r_pi, it's effectively near infinity, and the real number line wraps around at infinity so whether it's near + or -infty, it's not violating continuity or anything.

Input resistance is approximately the combined value of all those resistors in the bandgap cell times β. Possibly ~100kΩ.
The BE junctions that hold those resistors to ground contribute some more, but maybe not a lot more.
Transconductance is the inverse of output resistance at unity gain, I suppose.

Yep, good perspective on this part is to remember it's all diodes between terminals.  So, A-C is a diode (often shown explicitly, as here), A-R (not shown in the above schematic, but present I believe either for ESD or parasitic for simply how the pin is laid out? or, is it the diode-strapped NPN plus diode so it can actually be -2Vbe?), and R-C (the B-C junction highlighted above).

A→C: guaranteed to exist as a parasitic of the power transistor, if nothing else.
A→R: go and measure your TL431 Most of mine show two diodes, but ON Semi has one and Hitachi has three
R→C: turns out it not always is so simple

This whole thing reminded me about something I have seen last year when I sacrificed a few Wing Shings for decapsulation experiments. Namely, their connection to the input emitter follower was a little weird and now I think that it's an attempt at input protection.

See below. Red is the input transistor. Purple is what I now believe to be an epi-FET (N-ch JFET with grounded gate, drawn as resistor here) to limit current into the red transistor. And orange is the Diode of Death which they have deliberately reintroduced for some reason, but not as the red transistor's BC junction anymore. (Actually, I think this diode is the BE junction of a vertical PNP whose collector is the substrate/ground below - see how wimpy the collector connection coming from the cathode is).



I have no idea if anyone else uses this scheme, but it's noteworthy that the JFET introduces a single diode connection to ground. So maybe ON Semi, but surely not the others I have tested.

I also produced full schematic of the Wing Shing because this chip uses a different topology. Don't ask me what their reason was, component count is similar. The same topology, minus input protection, is seen in a no-name TL431 on Zeptobars. Meanwhile this one is a classic. Lack of consistency is the only consistency to be found on auction sites

[T3sl4co1l]

Neat!  Yeah, that explains the REF-only voltage drops... sorta.

Tim

[amyk]

On the topic of TL431, here's another one reverse-engineered: http://www.righto.com/2014/05/reverse-engineering-tl431-most-common.html

[magic]

Right, this is TI who originated the TL431.

Note no input protection - REF goes straight to Q2, but there is an undocumented diode clamping it to ground..

[David Hess]

Linear Technology has many improved industry parts where the adjustment pin is much more tolerant of high voltage, but apparently not their LT1431.