Playing with LM723 voltage regulator - Vref etc.

[iMo]

While reading the recent posts on LM/uA723 voltage regulator I started to search through my various junk-boxes and was lucky to find a bag of ~30years old MAA723CN DIL14 chips.

MAA comes from TESLA (former Czechoslovakia, TESLA was an electronics giant producing almost everything).

First of all I wanted to see how these old chips change their Vref while baking them in an oven at 120-150degC temperature.
I took 16 chips, wired them all to a 16V source.

I did a simple measurement before the baking, 20minutes after the baking, and finally 12hours after, while 16V was applied continuously to all the chips during the whole procedure.

Below are the results.

Even this was NOT intended as a metrology-like measurement you may see the half of the chips moved 20mV or more after that torture.

[iMo]

While I spent last weekend with baking my 723s I could not resist to try with the self
heating 723 reference schematics from Elektor journal July/August 1978 - exactly 40 years after 

First of all I wanted to see how the stuff works in detail..

The Elektor's schematics below (I have edited) includes the pin numbers for the DIL14 version, as well as the VRef, VS, VL, RL(R5), IL markings in red for easy referencing.

How does the self-heating voltage reference work?

Voltage regulator and its setting
======================================

Let us consider a standard "setup" with Vcc=12V, RL=33ohm, Vref=7.10V, ambient T=25C.

Imagine there is no T3 (the T3 is the temperature "sensor"). The circuit then works as a standard voltage regulator, where the VL (Load voltage at RL) is set by R3/R4.

The VL will be the MAX voltage you can get at RL then, therefore it limits the max current IL and max power loss at T2 as well.

The T2 transistor is the "chip heater". The power loss (the "heat") produced by the T2 will be

P = (Vcc - VL) * IL [Watt]

We want most heat will be produced by the T2, therefore our intention is to have the VL as small as possible (not to heat the RL resistor instead). On the other hand, the temperature regulation must have enough headroom for proper operation.

The VL set to 1.93V works fine here (as the "regulation at reasonable package temperatures" works well below this value).

You may set the VL higher (ie. to 2.5V) but the initial power loss at T2 (when cold) would be in a dangerous region already.

From my experiments the "temperature regulation" with this circuit and setup works within VL=1.0-1.5V.

That creates T2's power loss of

Code: [Select]

Pmin         = (12.0 - 0.0) * 0.0 / 33 =    0mW
Preg_min     = (12.0 - 1.0) * 1.0 / 33 =  333mW
Preg_max     = (12.0 - 1.5) * 1.5 / 33 =  477mW
Pmax         = (12.0 - 1.93) * 1.93 / 33 = 589mW
The IL is

Code: [Select]

IL_p_min     = 0.0 / 33 =   0mA
IL_p_reg_min = 1.0 / 33 =  30mA
IL_p_reg_max = 1.5 / 33 =  45mA
IL_p_max     = 1.93 / 33 = 58mA

Because of that it has a sense to set the VL close to 2V and no higher, as it can smoke the T2 without the regulation.


Temperature sensor and regulation
========================================

Now, with the T3 sensor involved, the regulation works such the rising heat decreases the Vbe of the T3 (a diode with -2mV/C), and with the VS set to a certain threshold the T3 opens (it starts to conduct) thus it pulls the T2's output current down.

1. With VS set close to 0V the T3 does not regulate and you get full VL with maximum power loss at T2.

2. With VS around 0.56V it regulates, and the VL voltage in this setup changes from 1.0-1.5V based on the ambient temperature and 723 package cooling.

3. With VS set higher the regulation stops and the current IL is almost zero (no power loss at T2).

At this regulation range and with this setup the 723 package temperature is somewhere around 30-60C.

The resistor RL remains "cold", the power loss at it is always well below PLmax = 0.058^2 * 33 = 111mW.


Setting up the regulation
========================================

It seems to me the best approach is as follows (Vcc=12V, RL = 33ohm, Vref = 7.10V, Tamb=25C).

1. Use an RL=470-1k resistor (to be at the safe side), and set the VL_max to aprox 2V with R3/R4.

I've been using here R3=22k and R4=10k (my VL_max = 1.93V).

2. replace the RL=1k with RL=33ohm.

3. set the VS such the total current of this circuit is aprox 35mA with package temperature around 40-45C. Temperature below aprox 50C means you can keep your finger at the package for an unlimited time..

The VL will idle somewhere around 1.1V at the equilibrium.

I've been using here R1=22k, P1=3k3, R2=0 while setting it up. Replace the P1 with fixed resistor(s) afterwards.

Mine divider ended up with R1=22k, P1=0, R2=1k8 for Icc=31mA, VL=1.05V, Tpackage=45C at the equilibrium in this setup..

The temperature regulation is pretty agile and fast. A gentle touch on the top of the 723 package with your finger (your finger is usually around 31C) will almost immediately (within a second) rise the VL and IL.

You may monitor the regulation by watching the VL.

EDIT: Added a chart with "T2 heater power loss" against "Vcc change with RL=33ohm".

[Ian.M]

IIRC some '723 clone processes tended to leave trapped charge in SiO in the reference section, which resulted in high drift as they aged.   Baking at or near Tj_max under bias is accelerated aging, which should reduce the trapped charge and result in lower on-going drift once they've cooled and had long enough for package stresses on the die to equilibrate.   If you want to pursue this further,  I would suggest a 24H bake at 140 degC, under bias, followed by *SLOW* cooling, with no bias, over the next 24H.

[iMo]

TI's 723s may suffer as well..

[iMo]

.. If you want to pursue this further,  I would suggest a 24H bake at 140 degC, under bias, followed by *SLOW* cooling, with no bias, over the next 24H.

I do not know, frankly. I do not have such facilities handy. 24H slow cooling seems to me you would need a programmable oven.
Moreover, spending so much energy may cost more than a handful of new LM399ties 

Also I am still not convinced the Vref in those 723 (any variety) is of such quality, even people claim it. It would make sense if we were able to get, say, sub-mV stability. Not sure an ..723 is of such parameters, however..

[Ian.M]

Even a couple of hours would be better than no bake.

The slower you cool them the lower the residual thermal stresses.   If you cant do 24H on a -5 deg C per hour temperature profile, bury the test board in a tin of dry clean sand so you've got a lot of thermal mass, with a thermocouple in contact with the board so you can check the temperature, and immediately you shut off the oven, pull it out, and wrap it in rock-wool then bubblewrap and put it in an expanded polystyrene box with a tight fitting lid and let it cool naturally.

They certainly aren't going to be metrology grade references, but the extra stability comes fairly cheap if you can process 20 at a time.  The test jig is pretty simple - just power it via V+ and V- with Vref tied to N.I. and Vout tied to Inv. with all other pins (including Vc) open (a Muntzed version of the datasheet's Basic Low Voltage Regulator circuit).   A 0.1uF ceramic for decoupling wouldn't be a bad idea per every four chips.   That should be easy enough to build on stripboard with cheap 14 pin DIL sockets.

Once you've burned in a reasonable sample, you could set up a data acquisition system with some analog switches and bench DMM that can be PC controlled to do a long-term test in comparison with a control sample that you haven't baked.  The results should be interesting.

[Wolfgang]

Hi,

I am one of the LM723 grillers, too. See

https://electronicprojectsforfun.wordpress.com/silly-circuits/silly-circuits-a-heated-lm723-reference/

The larger initial drift is real. I use devices from ST, and they show the same effect. Watching a long-term experiment with about 95°C die temperature
showed about 20-25mV drift in about 2 months. The logging has not yet finished, but it seems this is slowing down now.

I also have some vintage military chips on hand (with -55 to +125°C range) and I will try them out later. Of course, for that money you could buy a LM399, too.

[iMo]

@Wolfgang: I tried with the simplest Elektor's version of the self heating - as there were some doubts among people here whether it actually works. And it works fine, imho.
Not sure the 723's Vref qualities justify your more complex version. I haven't seen any practical measurements or results so far, except at your page there (siily-circuits).

I saw your first results there with the +2mV drift and maybe 1.5-2mVpp noise. The noise here is similar. Would be great to see your results after longer period of time.

PS: not sure the through-hole mounting you are using is the best one as these 723s in epoxy DIL14 packages seem to be sensitive to a mechanical pressure. I saw here maybe 1-2mV off when pushed at the chip firmly while it was sitting in a precise socket. Try with yours..

@Ian.M: I've been using a simple method of the wiring: a string of loose packages (aprox 15mm apart) with their Vcc and Vss soldered to 16V rails (kynar wire). No decoupling, no sockets. I may try baking again with the worst packages to see if they still drift. Cooling is a problem, but the clean sand is something I was thinking already. But still it will not cool down in 24H . Also more digits would be handy..

[Ian.M]

No, you wont get 24H cooldown , unless you use a stainless vacuum flask that can withstand 150 deg C instead of a tin for the sand, and probably not even then, but the slower the cooldown, the less mechanical stress as the epoxy has more time while its softened by the elevated temperature to creep to relax stress from differential thermal contraction, so *anything* is better than rapid cooling. 

It may be worth looking for a suitable stainless vacuum flask and turning it into a low power oven you can control with PWM from an Arduino.

[Wolfgang]

@imo
Thanks for the comments. My long-term test is still ongoing.
The argument that a solder mount of a thru-hole plastic component could cause mechanical stress is valid. The same applies to SMD packages, however, only TO-99 or CERDIP packages would be more immune to this. Just for curiosity, I will try some of those, too. This means a new PCB, however.
Another suspicion I have is a dependency on humidity. To clarify this, I have bought some Honeywell humidity sensors that I will include in the measurements.
Sorry that you all have to wait for so long to get results 

[iMo]

@Wolfgang: happy to hear your measurement is running. We all expect the final results will be much better than the ones usually seen in the "Metrology" section

PS: Added above a chart with 723 Self-heating Thermostat - "Vcc (the power supply of the 723)" vs. "T2 (the heater) power loss" with RL=33ohm. It includes an estimation of the package temperature. Otherwise the setup as above [Vcc is set, then VL measured after about 60secs, free air, T_amb=25C].