Self biased Zener reference using Wilson current mirrors

[Dabbot]

I saw a video by @w2aew where he demonstrates the Wilson current mirror operating with little to no change across different burden voltages:

I decided to try a self biased Zener voltage reference using two Wilson current mirrors to see how it performs, and thought I'd share.

The circuit here is biasing a 5.1V Zener diode close to where its temperature coefficients cancel out (set by R3). Q1/Q2, and Q5/Q6 are matched by hand. I built the circuit up on a breadboard and supplied either 9V or 18V while measuring the output, and found it differed by 1-2mv at most.

I noticed that when I used my DMM to measure the voltage across R2, the upper mirror went into oscillation, hence the presence of C1. This is quite possibly because it's on a breadboard. C1 could be reduced, or even taken out entirely if the circuit is soldered.

[moffy]

That's not a bad idea, using the reference/zener to create it's own fixed bias current with power supply rejection built in.

[tggzzz]

Good to see someone showing imagination and inventiveness, and realising solderless breadboards have problems

Your schematic doesn't show any PSU decoupling capacitor, which would be necessary on a virtually any construction technique.

Another way of self-biassing a zener is inside an opamp feedback loop (TAoE3, p676)

[Dabbot]

Your schematic doesn't show any PSU decoupling capacitor, which would be necessary on a virtually any construction technique.

Not pictured.
It was the first thing I checked when I noticed something funny about the top current source.

Another way of self-biassing a zener is inside an opamp feedback loop (TAoE3, p676)

I have the remnants of a zener and op-amp circuit beside this one. What is the case where you would need a startup circuit? I assume if you're using an op-amp with PNP input stage and negative rail referenced zener, it's not necessary? I was using a LM358 and didn't have any issues with startup.

[mawyatt]

Clever use of the Wilson Current Sources, well done

One could also investigate using a ~6.2V Zener with TC ~+2mv/C and take the output from the base pair connection of Q5 and Q6, thus the output voltage would be Vz + Vbe which should have ~ 0 TC due to Vbe ~ -2mv/C. Use R3 to adjust Zener current for best overall TC.

Best

[PCB.Wiz]

One could also investigate using a ~6.2V Zener with TC ~+2mv/C and take the output from the base pair connection of Q5 and Q6, thus the output voltage would be Vz + Vbe which should have ~ 0 TC due to Vbe ~ -2mv/C. Use R3 to adjust Zener current for best overall TC.

Yes, tho usually if chasing temperature compensation, you get two diodes in the same package.
The Nexperia PLVA26xx is an example of dual diodes designed for this use.

Attached data sheet plots show how TC varies slightly with Zener current, and another from TI showing the average TC effects of Zener + diode(s) in series. 

[mawyatt]

Clever use of the Wilson Current Sources, well done

One could also investigate using a ~6.2V Zener with TC ~+2mv/C and take the output from the base pair connection of Q5 and Q6, thus the output voltage would be Vz + Vbe which should have ~ 0 TC due to Vbe ~ -2mv/C. Use R3 to adjust Zener current for best overall TC.

Best

Taking the Output Voltage from bases of Q5 and Q6 leaves a temperature gradient which can't be corrected alone by R3 in changes in simulation.

Here's an adaptation utilizing a 6.2V Zener and adding a Vbe junction in series for TC compensation. With this configuration the Vout Positive TC can be adjusted by R3 and the Negative TC by R4 and as shown can be balanced to producing the classic bowed Vout curve.

There's all sorts of reasons why this isn't practical for use as a TC Compensated Reference, however an interesting circuit concept to play around with.

Anyway, thought some folks might be interested.

Best

[iMo]

..Here's an adaptation utilizing a 6.2V Zener and adding a Vbe junction in series for TC compensation..

The zeners in the LTspice (ie in your above sim) usually do not have the specific params defined needed for their TC..

PS: Here is a model I experimented with in past (in some posts I did re Vrefs sims). Example only.

.MODEL BZX84C6V2imo D(TRS1=7m TRS2=10u IS=180.4E-15 N=1.228 RS=.09 IKF=1 XTI=3 EG=1.11 CJO=149.5E-12 M=.3207 VJ=.6676 FC=.5 BV=6.326 IBV=.654 TBV1=.38E-3 TT=418.4E-9 Vpk=6.2 mfg=Zetex type=zener)

[RoGeorge]

Nice trick with Q7 to help adjusting both the +/- temp coeficients, thanks. 

I'm tempted to think about the initial circuit (upper and lower half) as being two current sources in series.  Thumb rule is to avoid putting current sources in series (or voltage sources in parallel) or else they'll fight with each other.  I wonder if there will be a bigger benefit in using only half of the schematic.  I'm tempted to believe that will be more stable, but that's only a guess, I didn't try.

I wonder which one will be more beneficial, two mirrors like it is now, or a single mirror, and with the extra transistors used to make Darlingtons or Sziclay transistor combination instead of Q3 (so to have bigger Beta and less Ib in Q3).  Oh, and a red LED instead of Zener, for low noise. 

[mawyatt]

..Here's an adaptation utilizing a 6.2V Zener and adding a Vbe junction in series for TC compensation..

The zeners in the LTspice (ie in your above sim) usually do not have the specific params defined needed for their TC..

PS: Here is a model I experimented with in past (in some posts I did re Vrefs sims). Example only.

.MODEL BZX84C6V2imo D(TRS1=7m TRS2=10u IS=180.4E-15 N=1.228 RS=.09 IKF=1 XTI=3 EG=1.11 CJO=149.5E-12 M=.3207 VJ=.6676 FC=.5 BV=6.326 IBV=.654 TBV1=.38E-3 TT=418.4E-9 Vpk=6.2 mfg=Zetex type=zener)

This is just a limitation of the model not necessarily LTspice, the same would be for any simulator including Cadence.

Creating a specific model for a specific diode as you've done would be in order if one was considering actually building this circuit, then you would still be up against not having all the circuit elements on a common thermal platform such as a silicon die, so basically likely a waste of time, why we indicating "playing around"

If we were still designing chips, this might be an interesting circuit to throw onto a test chip, but those chip designing days have long since passed

Best

[iMo]

The zeners in the LTspice (ie in your above sim) usually do not have the specific params defined needed for their TC..

Meaning - in "LTspice's libraries".. So you have to find the params and add them to the LTspice's zener models. Otherwise your zener will not show typical zener's TC in your sim..

[mawyatt]

Nice trick with Q7 to help adjusting both the +/- temp coeficients, thanks. 

I'm tempted to think about the initial circuit (upper and lower half) as being two current sources in series.  Thumb rule is to avoid putting current sources in series (or voltage sources in parallel) or else they'll fight with each other.  I wonder if there will be a bigger benefit in using only half of the schematic.  I'm tempted to believe that will be more stable, but that's only a guess, I didn't try.

I wonder which one will be more beneficial, two mirrors like it is now, or a single mirror, and with the extra transistors used to make Darlingtons or Sziclay transistor combination instead of Q3 (so to have bigger Beta and less Ib in Q3).  Oh, and a red LED instead of Zener, for low noise. 

The current sources are really mirrors and actually not having the high Z outputs driving each other. The top Current Mirror has a low Z input impedance on the right side as one "sees" a low Z due to the 2 Vbe junctions. The top Mirrors drives the lower mirror with it's high Z output driving the low Z input of the bottom mirror as "seen" on the left side thru 2 Vbes and the Zener.

So this isn't like having 2 high Z current sources in series but more like having the high Z Current Mirror drive the lower Z of the opposite mirror which in turn has its output driving the 1st Mirror low Z.

One potential point is that this circuit may not start up because of the coupled mirrors driving each other and may need some sort of start-up circuit (bleed resistor).

Note as iMo mentioned the Zener model used was just grabbed from the library, and any "serious" work should 1st start off with a fully developed Zener model, same is true for the transistors.

Anyway, still a fun circuit to play around with

Best

[David Hess]

Last time I checked, a couple years ago, temperature compensated reference zener diodes were about the same price as equivalent LT1236 or LT1021 buried zener series references.
         
$3.47   1N823   5.9-6.5V   50 ppm/C
$3.98   1N825   5.9-6.5V   20 ppm/C
$5.68   1N827   5.9-6.5V   10 ppm/C
$8.00   1N829   5.9-6.5V   5 ppm/C
                     
$6.66   LT1236BC   5.0V      5 ppm/C
$7.29   LT1236BC   10V      5 ppm/C
$8.93   LT1021DC   5.0V      3 ppm/C
$8.93   LT1021DC   10V      5 ppm/C

[RoGeorge]

Tried out of curiosity a minimalistic version, where the second mirror was replaced with an emitter voltage follower.


Click to enlarge.

- at low frequency (100Hz Vcc ripple) it has about the same performance as the schematic with 2 current mirrors (~112dB rejection of Vcc).
- at higher frequency (e.g. 1MHz Vcc ripple) it behaves much better because of the lack of C1.  If the C1 is added here, too, it is only marginally better than the circuit with 2 mirrors.
- this mirror + follower version uses 4 transistors instead of 6, so if it were to make use of 6 BJTs, we could try a 3rd schematic with composed transistors for a higher gain (and thus an even bigger rejection), and indeed the 3rd version has a bigger Vcc rejection.


Click to enlarge.

The Vcc ripple was added as another AC source in series with the Vcc.  AC response is to estimate the performance of all 3 versions (in terms of Vcc ripple rejection).

At least in simulation, the transistors from the second mirror doesn't seem to contribute with any improvement, but I didn't try in practice, and didn't investigate the stability.  Didn't try any noise simulation either (for the idea of using a LED for its low noise, instead of the Zener).

Another interesting test was to look at the thermal drift.


Click to enlarge.

While the Vcc ripple rejection is stellar in all 3 versions, the thermal drift is expected to produce much large deviations.  Will worth the effort of tuning the resistor for minimal thermal coefficient, or to put the circuit in an thermostat oven.

[mawyatt]

Like your 4 transistor version

Best

[magic]

I don't like its thermal drift
But it could be corrected with a diode on top of the zener, or using a lower voltage zener maybe.

Note that PSRR may be unrealistic in simulation due to perfect matching.

[PCB.Wiz]

Last time I checked, a couple years ago, temperature compensated reference zener diodes were about the same price as equivalent LT1236 or LT1021 buried zener series references.
         
$3.47   1N823   5.9-6.5V   50 ppm/C
$3.98   1N825   5.9-6.5V   20 ppm/C
$5.68   1N827   5.9-6.5V   10 ppm/C
$8.00   1N829   5.9-6.5V   5 ppm/C
                     
$6.66   LT1236BC   5.0V      5 ppm/C
$7.29   LT1236BC   10V      5 ppm/C
$8.93   LT1021DC   5.0V      3 ppm/C
$8.93   LT1021DC   10V      5 ppm/C

Yes, and note the other issue with Zener based circuits is the relatively poor tolerance.
Two diodes in series is a common thermal fix for Zeners, eg the PLVA26xx series do this, but they only spec 3% as I guess they expect post-trim in usage.

You can get Dual Zeners down to 2% in single packages - eg BZB84-B6V2,215, but 1% Zeners only come as singles
The Nexperia BZT52H-A6V2-QX is 1%, and comes in SOD123F, which maybe can thermally couple in pairs, but the price climbs for that 1% bin.
The data does suggest the cathode has best thermal coupling
Device mounted on an FR4 PCB, single-sided copper, tin-plated, mounting pad for cathode 1 cm2

[Dabbot]

Tried out of curiosity a minimalistic version, where the second mirror was replaced with an emitter voltage follower.

I built this up from components of my original circuit and it at least matched the original. ~1mv change between 9V and 18V. I like it.

While the Vcc ripple rejection is stellar in all 3 versions, the thermal drift is expected to produce much large deviations.  Will worth the effort of tuning the resistor for minimal thermal coefficient, or to put the circuit in an thermostat oven.

For a laugh I put the thing in the freezer (sans DMM) and it changed ~5mV (5.041V to 5.036V). It's 27°C at the moment, and the freezer is -18°C, so ~110uV per °C. I don't think I'll go through the effort of putting it in the oven right now. 

[tggzzz]

Tried out of curiosity a minimalistic version, where the second mirror was replaced with an emitter voltage follower.

I built this up from components of my original circuit and it at least matched the original. ~1mv change between 9V and 18V. I like it.

While the Vcc ripple rejection is stellar in all 3 versions, the thermal drift is expected to produce much large deviations.  Will worth the effort of tuning the resistor for minimal thermal coefficient, or to put the circuit in an thermostat oven.

For a laugh I put the thing in the freezer (sans DMM) and it changed ~5mV (5.041V to 5.036V). It's 27°C at the moment, and the freezer is -18°C, so ~110uV per °C. I don't think I'll go through the effort of putting it in the oven right now. 

Is it constructed on a solderless breadboard? If so, how do you take account of the poorly-defined contacts between components and internal spring contacts?

Is there any condensation on the circuit? That could unbalance it.

Anyway, a fun exercise. it would be even more of a laugh to put a solderless breadboard in an oven

[Dabbot]

Is it constructed on a solderless breadboard? If so, how do you take account of the poorly-defined contacts between components and internal spring contacts?

Is there any condensation on the circuit? That could unbalance it.

I was hoping any combination of dicky contacts and condensation would balance out across the circuit somehow. 

Anyway, a fun exercise. it would be even more of a laugh to put a solderless breadboard in an oven

I'm sure I'll stop laughing once I notice the funny taste in my next casserole.

[magic]

I wonder if this would work?

(If unstable, try a capacitor in the indicated position.)

[RoGeorge]

I wonder if this would work?

Wow! 

Indeed, the 3rd BJT in the Wilson current is to avoid the imbalance caused by Early voltage when the Vcb is different in the two arms of the mirror, except in this particular usecase (with an emitter follower between the mirror arms), the Vc will vary the same (at 0.6V apart), so the Early voltage is not a problem any more, and the 3rd transistor in the Wilson mirror becomes redundant.  I failed to see that.

I put the thing in the freezer (sans DMM) and it changed ~5mV (5.041V to 5.036V). It's 27°C at the moment, and the freezer is -18°C, so ~110uV per °C. I don't think I'll go through the effort of putting it in the oven right now. 

Impressive result, you might be very lucky with that Zener.

Never tried to measure the TC of a Zener, but now I'm curious to test various Zener diodes I might have through the scrap boxes.  I would like to log V vs T, but the DMM can only log one of the two.

I didn't need a Vref, but now I'm tempted to build one. 

[Dabbot]

I wonder if this would work?

(If unstable, try a capacitor in the indicated position.)

It works pretty well. Doesn't noticeably budge when I flip it from 9V to 18V. No capacitor required. Very nice!

[Dabbot]

Impressive result, you might be very lucky with that Zener.

I doubt it. I got some of these because their datasheet specified a min/max TC which crossed zero, which means easily tuned by current.

Never tried to measure the TC of a Zener, but now I'm curious to test various Zener diodes I might have through the scrap boxes.  I would like to log V vs T, but the DMM can only log one of the two.

You could bias a thermistor on the same board and multiplex the thermistor voltage with your test Vref. 

[PCB.Wiz]

Impressive result, you might be very lucky with that Zener.

I doubt it. I got some of these because their datasheet specified a min/max TC which crossed zero, which means easily tuned by current.

The min-max is the production spread, not the current spread.

A problem with the simple current adjust to find nominal min TC, is the knee of those lower voltage zeners is quite soft.
See the attached plot, of both Zener TC and Knee

That's why it is more common to see a Diode + Zener or VBE multiplier + Zener pairing used.
The higher voltage Zener knee is sharper, which means lower Z which also means lower noise. 

The other issue here is that in 2025, you can buy a 0.05% sub 6ppm precision reference for about $1. Lower current, and no trimming and checking needed.

You can get corrected Zener simulations to under 6ppm, but as 2mV/K  is about 330 ppm/K, you are asking for a temperature tracking of about 20 milli Kelvin.
That's why you see these on IC's wafers, with heaters.

See this thread for a modern example of a (multiple) Zener with heating control reference
https://www.eevblog.com/forum/metrology/ref80-a-competition-to-adr1001/