TL431 as 5V reference accuracy

[lk.dgironi]

Hi all,

For an op amp based voltage converter (-10V..10V to 0V..10V) I need a 5V reference.
The PCB which contains this circuit still has on one TL431 and 5k 0.1% resistors, so those two components will "come for free" (or at least will cost just a few bucks).

I've used one REF195 as reference but then I realize the reference voltage for the op amp converter does not need to be that accurate, cause later on the signal path I've a trim pot that center the signal.  because I yet have a TL431 on board and cause I don't need that much accuracy I'm wondering if I can use the TL431. Also I'm wondering how this reference compare with a AMS1117-5.0, which by datasheet has 2% accuracy 5V +-0.1V.

Find attached sample circuit.

[magic]

TL431 datasheet specifies minimum/maximum output voltage at 25°C and over full rated temperature range.
Multiply by 2 for 5V output and add ±0.2% due to resistor tolerance (probably negligible).
You may also need to account for reference pin input current flowing through external resistors, maybe a few mV.

Trimming with a pot is an option, but watch out for thermal drift if the circuit will work over a wide range of temperatures.

[lk.dgironi]

Thanks

Seems better the AMS1117-5.0.
Vref for TL431 (version TL431I) goes from 2495 to 2550, that is almost 2.5% *2 it's 5% almost... am I missing something?

[David Hess]

It will depend on the grade (A, B, none) of the TL431, which could yield an accuracy of 0.5%, 1%, or 2%.  The calculated temperature coefficient could be greater than 100 ppm/C for the A grade, or 150 ppm/C for the others.  So potentially 4 times better than your AMS1117-5.0.

[Whales]

Just FYI, your C20 1uF across the cathode & anode of the TL431 may make it unstable.    TL431's are really picky about capacitance, some manufacturers have different recommendations (and some don't even mention it in the datasheet at all, because stfu).  You have to fight your natural instincts of "capacitors make constant voltages better".

See this TI note:  https://www.ti.com/lit/an/slva482a/slva482a.pdf

[lk.dgironi]

Thank you.

Just FYI, your C20 1uF across the cathode & anode of the TL431 may make it unstable...

I've read that document. Maybe it's better to remove C20, to prevent oscillation, my Vka is 5V, by simulation (I haven't do the math) my Ika is 20mA. So looking at Figure 15 of TI datasheet (https://datasheet.lcsc.com/lcsc/1809111518_Texas-Instruments-TL431AIDBZR_C23892.pdf), i should put less than 20nF or more than 1.8uF to prevent oscillation.

It will depend on the grade (A, B, none) of the TL431...

I've a test circuit here, I've notice is that my R25 get's hot (70C almost). By simulation R25 takes 33mA, so it's 300mW (at 10V). Also the TL431 get pretty hot (50C  almost). Maybe I've to go back to REF5050 or REF195.

Find attached LTspiece simulation files.

[Zero999]

The TL431 has a reference voltage of 2.495V, which is is already -0.2% lower than 2.5V: twice as bad as the 0.1% resistors. For exactly 5V, put 20R in series with the top 5k resistor: R26 in the schematic attached to the OP, R5 in the LTSpice file.

[pqass]

I've a test circuit here, I've notice is that my R25 get's hot (70C almost). By simulation R25 takes 33mA, so it's 300mW (at 10V). Also the TL431 get pretty hot (50C  almost). Maybe I've to go back to REF5050 or REF195.

You could lower the current through the TL431 but then buffer it with an NPN.  Simulation here.

[magic]

The TL431 has a reference voltage of 2.495V, which is is already -0.2% lower than 2.5V: twice as bad as the 0.1% resistors. For exactly 5V, put 20R in series with the top 5k resistor: R26 in the schematic attached to the OP, R5 in the LTSpice file.

I wouldn't bother. You gain +5mV from the 2μA input current flowing through 2.5kΩ output resistance of the divider.
The output should end up close to 5V exactly, with only part-to-part variation and thermal drift left.

[Zero999]

The TL431 has a reference voltage of 2.495V, which is is already -0.2% lower than 2.5V: twice as bad as the 0.1% resistors. For exactly 5V, put 20R in series with the top 5k resistor: R26 in the schematic attached to the OP, R5 in the LTSpice file.

I wouldn't bother. You gain +5mV from the 2μA input current flowing through 2.5kΩ output resistance of the divider.

I did do the calculation, but read the wrong figure on the data sheet, so it came out too low.

The output should end up close to 5V exactly, with only part-to-part variation and thermal drift left.

You're right, but it's actually better than that, since it's an NPN transistor on the reference input, the current only flows through the upper resistor (R1 on the schematic in the data sheet) making 10mV.

[magic]

5mV input-referred × 2x closed loop gain = 10mV output-referred

This is exactly what I meant. The input current error makes the apparent reference voltage 2.5V with these resistor values.
At least on average, because the 2μA figure is only "typical". It appears to depend on β of internal NPN, so there will be some spread...

[lk.dgironi]

Thanks for clarifications!
The NPN is the way for using the TL431 keeping low temperature. Unluckily I've spacing and BOM constraints, so I prefer to add a SO8 package reference.
I'm back to REF5050 due to spacing constraints in the board. the REF5050 just need one low ESD (I'm using a tantalum that at 100kHz is 2ohm, which is slightly more what REF5050 suggest, but at lower freq it goes to lower ESR).

[Doctorandus_P]

You have to distinguish between accuracy and stability. Apparently you already have an adjustment potentiometer, so that solves the accuracy or initial calibration, but it does nothing for stability and drift with temperature or aging. Those are separate topics, and you have to handle them independently from each other. Even very expensive voltage references can have an abysmal (or mediocre) initial accuracy, but have excellent temperature and long term drift characteristics.

[lk.dgironi]

Thanks @Doctorandus_P
It's now clear. Term drift is certainly something I've to consider, indeed this PCB will be enclosed in a box inside an electrical cabinet, and there are also some power IC that dissipate some heat on this PCB. The REF5050 I'm using has 3ppm/C which is for the scope of this PCB enough I think.