Why does a resistor and diode make this circuit temperature insensitive

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
I'm in need of a reasonably precise 5V voltage reference, so I built one of the reference circuits for the Linear Technology LT1236 reference IC's, and it works great! But there is one part I don't fully grasp yet. The design notes indicate that adding a diode and resister to the circuit shown below makes it temperature insensitive. My question is, why does this work?

Also can anyone recommend an equivalent IC that's not $5/ea? :)

Data sheet is attached, DigiKey link here: http://www.digikey.com/product-detail/en/LT1236BIS8-5#PBF/LT1236BIS8-5#PBF-ND/889719

Reference circuit:

circuit.png
 

Attachments

cmartinez

Joined Jan 17, 2007
6,646
I'm in need of a reasonably precise 5V voltage reference, so I built one of the reference circuits for the Linear Technology LT1236 reference IC's, and it works great! But there is one part I don't fully grasp yet. The design notes indicate that adding a diode and resister to the circuit shown below makes it temperature insensitive. My question is, why does this work?

Also can anyone recommend an equivalent IC that's not $5/ea? :)

Data sheet is attached, DigiKey link here: http://www.digikey.com/product-detail/en/LT1236BIS8-5#PBF/LT1236BIS8-5#PBF-ND/889719

Reference circuit:

View attachment 96300
If I remember correctly, the application note itself explains why. But here's a hint, it's important that the resistor is 27k and not something else or approximate, since it has to match the impedance of the reference.
 

OBW0549

Joined Mar 2, 2015
3,102
I'm in need of a reasonably precise 5V voltage reference, so I built one of the reference circuits for the Linear Technology LT1236 reference IC's, and it works great! But there is one part I don't fully grasp yet. The design notes indicate that adding a diode and resister to the circuit shown below makes it temperature insensitive. My question is, why does this work?
This is one of those "that's just the way it is" things; it's apparently due to the way the chip internally implements trimming. The datasheet explains:
The Datasheet said:
The LT1236-5 does have an output voltage trim pin, but the TC of the nominal 4V open circuit voltage at pin 5 is about –1.7mV/°C. For the voltage trimming not to affect reference output TC, the external trim voltage must track the voltage on the trim pin.
 

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
Why dont you use a Lm7805, ic?
Thanks for the suggestion. I actually need a voltage reference for a sensor, I need to be as spot on to 5.00v as possible and it looks like the lm7805 isn't for this purpose.

If I remember correctly, the application note itself explains why. But here's a hint, it's important that the resistor is 27k and not something else or approximate, since it has to match the impedance of the reference.
Here's the except from the data sheet about it. It explains that it works, but not the theory behind it:

LT1236-5
The LT1236-5 does have an output voltage trim pin, but
the TC of the nominal 4V open circuit voltage at pin 5 is
about –1.7mV/°C. For the voltage trimming not to affect
reference output TC, the external trim voltage must track
the voltage on the trim pin. Input impedance of the trim pin
is about 100kΩ and attenuation to the output is 13:1. The
technique shown below is suggested for trimming the
output of the LT1236-5 while maintaining minimum shift
in output temperature coefficient. The R1/R2 ratio is
chosen to minimize interaction of trimming and TC shifts,
so the exact values shown should be used.
 

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
@OBW0549 - We must have been typing at the same time. :) I saw that note in the data sheet, but I'm wondering the theory behind it all and how they figured out which parts to use.
 

BillB3857

Joined Feb 28, 2009
2,506
The forward voltage drop of the diode is temperature sensitive, also. By using that voltage drop change, the trim voltage varies with temperature, thus stabilizing the output. At least that's the way I see it.
 

MikeML

Joined Oct 2, 2009
5,444
The way I read it is that without the diode-resistor, the trim-pot alone would create a temperature dependency because the external trim-pot (which remains ratiometric as a function of temperature) doesn't match the tempco of the internal circuitry.

Adding the external diode's intrinsic temperature dependency to the mix compensates for the mismatch.
 

OBW0549

Joined Mar 2, 2015
3,102
Ok LM317 then.
The LM7805 and LM317 are voltage regulators, not voltage references. There are HUGE differences between the two types of devices, both in terms of initial accuracy and in temperature stability, with regulators like the LM7805 and LM317 being between 50X and 100X worse on both counts than voltage references like the LT1236-5 and the MAX6165.

They're as different as night and day.
 

OBW0549

Joined Mar 2, 2015
3,102
Wouldn't a TL431 also do this?
The problem with the TL431 is its temperature coefficient: 92 ppm/°C, versus 5 ppm/°C for an LT1236A-5. The TL431 would be OK for a lot of applications, but only if accuracy and temperature stability are non-critical. Also, keep in mind that for the TL431, the temperature coefficients of the voltage-setting resistors (100 ppm/°C for regular 1% metal-film resistors) contributes to temperature drift, too.
 

MikeML

Joined Oct 2, 2009
5,444
.... Also, keep in mind that for the TL431, the temperature coefficients of the voltage-setting resistors (100 ppm/°C for regular 1% metal-film resistors) contributes to temperature drift, too.
However, if you use of the same type of resistors for voltage setting, the voltage at the Ref input to the TL431 is hardly effected by temperature because of tempco's of the resistors cancel in a voltage divider...

Almost the exact same situation as started this thread. It should be possible to compensate for the TL431's tempco by adding a diode to the voltage setting network...
 

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
Thanks everyone for the suggestions for other IC's, I'll check them out! I'll post a link below to a vid of the LT1236 circuit working.

As for the temperature compensation; I think this link answers it:

Adding the external diode's intrinsic temperature dependency to the mix compensates for the mismatch.
From that article:

A diode can be used as a temperature measuring device, since the forward voltage drop across the diode depends on temperature, as in a silicon bandgap temperature sensor
That's the key that I didn't know! Looking at the circuit, you have a voltage divider with a diode in series with the pot. As the diode temperature changes, the forward voltage will change, which will change the value of the voltage divider, and therefore the value on the trim pin. I'll take their word that this counteracts the chips internal changes, but knowing the diode forward voltage changes with temperature by a predictable amount answers the question of why this works. Thanks for the link!

OK don't laugh at my video, it's kind of a pitch video for guys who are waiting for this for a specific product, to show the 5.00v reference will remain stable as the 9v source battery goes dead over time. I realize there's no load and no temperature test, but it shows the LM1236 providing very stable 5.000v output with wildly varying input voltage.

 

tindel

Joined Sep 16, 2012
682
I need to be as spot on to 5.00v as possible
This is not a requirement. This is hand wavy marketing guy/gal talk.

So, 5.000V +/- 5mV initial accuracy with 650ppm (about half of the entire part spec'd temp range) + 20ppm (line regulation) + 35ppm (load regulation - assuming you're sourcing) accuracy isn't good enough? That's 5.000V +/- 8.525mV initial accuracy. That's pretty stinking good - and will provide three bits error on a 12bit ADC (2.44mV/bit).

I doubt that the temp compensation is less than 10ppm after trimming and adding the diode... if it was they would have put the resistor and diode in the die! Assuming you trim the part and still obtain 10ppm, you still have somewhere around +/-3.525mV (2 bits) error over temp, line, and load.

So what's your requirement?

[edit: miscalculation for the temperature range... oops]
 
Last edited:

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
So what's your requirement?

[edit: miscalculation for the temperature range... oops]
@tindel - I need a 5.00v supply that gives 5.00v always, summer and winter, and over the useful life of the batteries.

I make and sell a tool that people use to adjust a sensor on motors. The sensor must be powered up to be adjusted. Normally the engine ECU provides 5v to power the sensor, but in some cases it's more convenient to use an external power supply than it is to power up the ECU. I'm designing the external power supply, and that is what this 5v reference is for.

Adjustments of .01v make a real and significant difference in engine performance, so I need a supply that is above all consistent to at least .01v. Not only consistent tool to tool, but consistent whether it's 110F in Arizona in summer, or 10F in Minnesota in winter. If I can provide a tool that consistently provides 5.00v, from every tool, summer or winter, then the buyers will perceive my tool as technically superior to my competitors and I will sell more tools. And this is the end goal.

Sometimes it is easier, and better marketing sense, to make a product that meets the buyers expectations, than it is to educate a community of non-technical buyers why their perceptions aren't necessarily correct. Especially when you are in fact providing a better tool by doing so.
 

tindel

Joined Sep 16, 2012
682
@tindel - I need a 5.00v supply that gives 5.00v always, summer and winter, and over the useful life of the batteries.

Adjustments of .01v make a real and significant difference in engine performance, so I need a supply that is above all consistent to at least .01v.
So it sound like your requirement is 5.000V +/- 10mV? THAT is a requirement! I have shown in my last post that the reference will meet this requirement, under worst case conditions, without having to add any parts - no matter what part of the world you're in. Again, I'll stress that adding these parts will most likely make your tempco worse than if you had left them out of the circuit.
 

Thread Starter

MrSoftware

Joined Oct 29, 2013
1,615
Again, I'll stress that adding these parts will most likely make your tempco worse than if you had left them out of the circuit.
Ahhh somehow I missed that point in your prior post, interesting thought! I'll have to make a couple up, one with and one without the extra bits and see what they do after being warmed up and put in the freezer. Thanks for posting that, I was just taking their word that the extra bits were required for temperature compensation.
 
Top