1N5439: Varactor Diode or Zener

Never mind about this part. I transposed the digits.

How do you balance them? A 1N5439B would have a 5% tolerance, so 12V means 11.4V to 12.6V. You are trying to run then at 250mA each, but the dynamic impedance at 100mA is 2.5 ohms, so the worst case voltage difference would put most of the current through the lower voltage part.

Putting zeners in parallel is a really terrible idea. Use a single bigger zener, or a power transistor plus a zener.

There's a better way to design that bit of the circuit

All diodes act varactors for small reverse bias. Not many are specified to be used that way, but people do use all sorts of didoes as cheap varactors.

Yes, i am waiting for some 3.3 ohms resistors to put in series with the zeners. I can also pretest the current for each zener. I can probably order/pretest 1% tolerance.

They are physically big and also expensive. The 5W zeners are good size and price.

But expensive. 5W 1N5439B are around $0.2 each. 10W 1N2976B are around $3 each. I'll get some to try out anyway.

Nobody sane does,

A power transistor and a cheap zener is probably going to be cheaper, but you would have to think about what you were doing.

This single device (12V 5A) might work:

Actually, i don't care if they are 5% or 10%. All i care about is the trigger voltage (turn on voltage). Out of the 32 sets, they are all fairly close to 12V. They are soldered together with lots of solder as heat sink. So, it's OK if one of them get beyond 5W, as long as four of them stay under 20W. I don't care about some failure, as long as I pre-sort and pre-test them.

Power rating is dependent on lead/body temperature.

Keep them cool by heatsinking their metal details.

Short leads on 5W axial part - lots of PCB copper.

RL

BZY93 (12V 5A) should handle up to 60W. My worst case is 1A to 2A.

I keep the long leads and lots of solder on it. Perhaps also solder some copper foils on the leads as heat sink.

I also put some butter on it to smell and see the smoke. Butter is a good heat sink grease and easy to get.

It's not "trigger voltage" but "breakdown voltage", For a 12V device it is actually the avalanche voltage.

You need to work how close together you need to get the breakdown voltages to get a current distribution that doesn't push any one of them above 5W.

At least with 12V zeners breakdown voltage rises as they get hotter, but not enough that you can rely on it to get a safe current distribution with unselected diodes.

There should be a mathematical model for voltage versus reverse current for the kind of avalanche diodes that you are using - LT Spice has obviously got one for modelling reverse biased Zener/avalanche diodes, but google won't find it for me.It probably would if I could find the right question to ask.

I measure the voltage to trigger the avalanche breakdown by pushing them to around 12.6V and let it settle down back to 12V. Most of them settle back to around 12.1V to 12.2V.

I will measure the actual current distributions for group of four zeners.

I trust actual measurements rather than mathematical model.

But, diodes typically fail in short circuit; what good is a Zener breakdown tolerance against that kind of behavior? Resistors is good, fuses have some knowable resistance...

As you should. But electronics is all about finding mathematical models that work well enough to let you keep track of what is going on, and put together circuits that won't blow up.

Yes, but mathematical models indicate that they should all be the same, with identical current distributions. What I need is to measure the manufacturing variations between them.

I use 3P3O3 (1/6Wx3) resistors to limit the current and slow blow the shorted zeners. If 3 Ohm 3 is too high, I can go with 2O2.

Don't be silly. The mathematical model requires you to plug in the actual breakdown voltage for each device, which you have to measure or specify. If you go in for really accurate modelling you may have to allow for varations in the cross-sectional area for each device - they are made to be identical but never are - and variations in the doping profile.

The main utility of mathematical model is that it allows you to interpolate accurately between a limited number of precise measurements.

In your case it would let you work out how far the current distribution between two parallel diodes would vary as applied voltage changed.

But how? The parallel zeners all have the same voltage on them; so, variations between them is zero in the math model.

I have four voltmeters on segments of 8S4P. Right now, they are well balanced at 95.2V, 95.5V, 96.5V and 96.2V, or less than 0.1V from 12V on average. When there is big variation within the segment, then i check the individual module. Mathenmatically, they should be 96.0V, 96.0V, 96.0V and 96.0V.

The mathematical models have adjustable parameters which let a generalised avalanche diode breakdown model replicate the performance of different examples of diodes with the same part numbers, but slightly different breakdown voltages. The breakdown voltage is the first adjustable parameter you need to adjust.

You should have four different models, each with a slightly different breakdown voltage parameter, each one modelling a different diode.

You don't seem to be understanding what mathematical modelling is intended to be doing for you.

Trigger/Threshold/Target avalanche breakdown voltages are within 11.9V and 12.1V, well within 1% to 2% of stated voltage. Even if currents/powers are off 10% to 20%, it is well within thermal rating. This is just using same batch without pre-sorting. Pre-sorting and pre-selecting should be able to improve it further.