low noise voltage ref. diodes

I am looking for a roughly 5V buried zener (low noise PSU application) with the lowest possible low frequency noise. a little drift is ok, Id rather use a raw diode than one of the ICs that combine buried zener+buffer with active temperature compensation like LM329 or REF02.

1N6083 looks promising from these people: http://www.knoxsemiconductor.com/catalog/LVA343A-3100ASERIES.PDF

but then I read that they just merged with another similar small company.:
http://www.knoxsemiconductor.com/acquisition.html

another diode specialist, some interesting stuff on their webpage, namely this quote "COMPANY NEWS: Aeroflex Announces Acquisition of MicroMetrics".

any other tips for a nice low noise zener?

mike p

That noise density shown in the tables for Knox looks pretty high especially when you consider that it is measured in a band 1-3 kHz. I presume that the noise 1/f-ish corner is somewhere around there, so you are going to be a lot worse off at lower frequencies.

What I have used for very low 1/f noise references is a complex and possibly impractical arrangement: I select a JFET with good low frequency noise performance and a reasonably long channel, and find the zero tempco point for a given drain source voltage and drain current (bias with a resistor in the source to common, with gate to ground). Then I use that arrangement as a current source, sometimes cascoding with another higher pinchoff voltage FET, and develop the ref voltage across another resistor. Or, you can use the gate-source voltage, or use another resistor to ground fed from the whole shebang.

A part Toshiba used to make, the 2SK373, would be a good fit. But there's enough spread that it's a PITA to adjust for a production-quantity design.

This beat by a large margin the performance of the fancy temp-controlled buried zener parts like the LM199. But obviously it was quite cumbersome.

For a poor man's version with pretty decent noise and reasonably low tempco, a current source based on a decent bipolar and biased with a red GaAsP "standard red" LED is not too bad. You still have something that has to be used with something else to generate the equivalent to a voltage reference, but often that is not too difficult if it's for a power supply.

well, that is something else to consider, but I havent given up with the buried zener quite yet.

The plan was (positive version) to use a NPN current source biased by the reference diode to pull current thru an HLMP-6000 LED, anode to the + rail. the LED voltage biases a PNP current source that sets up the current in the reference diode, aiming for the zero-tempco point. a few hundred thousand ohm resistor from the LED cathode to ground helps ensure startup. the input rejection of this circuit is impressive. a little L-C filtering is required at the input to extend the rejection to several GHz. just a ferrite chip and ceramic cap. this is followed by an RC filter (using a large value, low voltage AO cap) to a JFET (LSK389) input discrete error amp. for filtering and a gradual turn-on. the only stuff that isn't filtered into oblivion is the very low freq noise. It might not even matter, I should get an assortment of regular zeners and see how they perform?

mike p

I'd probably look at something from the 1N821-1N829 range, in case one can still get them. Low 1/f noise asks for a well-controlled process, and these probably are from such.

Other zeners with good processes are (presumably) those used as noise diodes. Ironically those give you lower noise than ordinary zeners, as they are optimised for flat power density spectrum which means excess noise (mostly a 1/f phenomena for zeners) must be reduced as far as possible (and the white noise is AFAIK just inherent for given breakdown voltage and reverse current). Look at those from www.noisecom.com--some are specified down to 0.1 Hz. Don't be surprised though if they turn up as the most expensive component in your design.

My last and cheapest suggestion would be to use a low-noise transistor with reverse-biased base-emitter junction. That will give about 6.2 V (no idea about the production spread though)--for low tempco you could add a forward bias diode (again probably best from a low-noise transistor) in series.

Samuel

One thing I noticed, and was corroborated in an old Motorola databook section on zeners: there is a voltage around the transition between the two breakdown processes of avalanche and zener, where the tempco is zero. It's about 4.8 to 4.9V IIRC.

Unfortunately, it seems the highest noise tends to occur there as well, although I didn't so much explore the spectrum. The blurbs on buried zeners talk about minimizing surface effects, and the experiments I did were with conventional parts. But I think the noise effect was associated with the two mechanisms having similar influence.

I also thought about a roll-your-own bandgap reference using really big bipolars like the low rbb' parts and lots of current. A lot of work and not very efficient either. And we're reduced to NOS on the great parts AFAIK now.

It seems the highest noise tends to occur there as well.

Click to expand...

According to M & F noise actually still rises above 5 V; there's a big step of about two orders of magnitude from 4 V to 5 V though. At 3 V and below data indicates noise well below 1 uV/sqrt(Hz); at 6 V and above it is around 30 uV/sqrt(Hz). This is for 250 uA reverse current--I'd say at higher currents the figures will drop.

Samuel

I had forgotten they discussed that. One thing that may have skewed my results on the bench was trying to coax a zero tempco V out of nominal 5.1V parts, hence reducing their operating currents to where the noise gets more Poissonian.

[quote author="Samuel Groner"]I'd probably look at something from the 1N821-1N829 range, in case one can still get them.[/quote]

http://www.newark.com/09F4807/semic...merican-power-devices-1n821a&_requestid=90496

almost 20k! i will try, thanks for the idea

I use a noisecom RF noise source all the time actually. too pricey though as you suggested

Ill try to measure the noise of a reverse B-E zener breakdown but Im all Labbed out right now. you'd think that if that worked it would be used more often in integrated references.

mike

Mercury battery?

Im starting to think the new AD XFET parts are the way to go, very low .1Hz-10Hz noise and not TOO expensive. ADR445:
http://www.analog.com/static/imported-files/data_sheets/ADR440_441_443_444_445.pdf

Ah yes I was trying to remember those. An interesting technique. The midband noise is still a bit high but amenable to filtering. The low frequency noise is good compared to other refs, although I think my I-source-developed approach still beats it by a large margin, but at high cost and complexity.

It would be fun to have a reference voltage source noise-off. I remember it was mighty hard to measure the low frequency stuff, and I wound up estimating from the 1/f noise curves of the FET what the likely noise was when I had to write a paper.

FWIW, other groups at the time did in fact use a battery as their reference.

[quote author="Samuel Groner"]
According to M & F noise actually still rises above 5 V; there's a big step of about two orders of magnitude from 4 V to 5 V though.
Samuel[/quote]

more precisely, there is a big jump when the device works in avalanche mode, which is somewhere around 5V for the devices M+F tested. the noise character in avalanche is strange, with discrete "steps" in noise level, random in time. I wonder if all zeners changeover at about the same voltage, or if it varies alot with device type.

I got some 1N823, mil spec 6.3V to test. I read something in a paper that suggested screening hi-rel parts by testing for /1f noise, because they found a correlation between long term reliability and low initial 1/f noise. I have no idea if these parts, or any hi-rel zeners, are actually screened this way, but at least it is something.

I tried to compare the 1N823 to the el-cheapo 4.7V BZX84 that I was using. I dont have the equipment to filter .1Hz-10Hz. I suppose I could do it digitally, but the lowest noise preamp I have right now is built into a sennheiser UPM-550 voltmeter. it is limited on the low side to 10Hz. but anyway, wideband noise is just about as low as I can measure for both devices, about -93dBu at 1MHz BW. good enough for a PSU, for now anyway.


mike p

That's right---the avalanche process is inherently noisy. And I think the breakdown mode is closely correlated with breakdown voltage.*

I was going to mention that, although Motchenbacher and Fitchen do advise against avalanche breakdown diodes for biasing altogether, you can do fairly well with brute-force bypassing of them with big caps. Also, although as Samuel points out the noise continues to rise with higher voltage after the big jump into avalanche territory, the noise as a percentage of the developed voltage falls off a bit: eyeballing the graph on page 182 of M&F, I see what looks like 30uV/sq rt Hz at 10V, 70uV/sq rt Hz at 100V. Of course 100V may not be a particularly convenient value!


*M&F say 7V or less for Zener action, but that sounds way high. The other consideration of low-voltage diodes is their impedance is often rather higher, a less-sharp "knee" in the I-V curve, which makes them poorer for use in a regulator.

While we're here...has anyone measured (or specified) noise voltage density from 5W diodes, particularly higher-voltage ones 82-100V? These are pretty commonly used in emitter-follower-type regulators for tubed circuits.

If they're as noisy as I suspect, I'll do the CCS into a resistor instead.

Peace,
Paul

Haven't looked at their noise Paul. Expectations would be, if they were operated at the same current density they would have lower noise, very roughly, according to the linear dimension of the junction (say the diameter) since they are like paralleling smaller diodes, the noise falling off as the square root of the number of such diodes.

Motchenbacher and Fitchen remind that the noise properties are highly process-dependent though.

Could I make a heater (oven) to improve keep the diode or Vref chip ?

I need V stability within 5ppm.

The LM399/199 has a heater built-in. My old HP6119a psu has a heated ref. diode inside a foam filled case.

Is there a standard method used to regulate the "heat" ?

I have several AD58x V reference chips, but they're all the lowest grade (highest tempco).
Or would I just seal them up away from moving air, and trim as necessary depending on the ambient temperature of my lab?

=FB=

[quote author="Freq Band"]I need V stability within 5ppm.[/quote]
That's...non-trivial.

What exactly do you mean by '5ppm stability'? Over one measurement session? Over the equipment's lifetime? Compared to absolute voltage standards?

[quote author="Freq Band"]I have several AD58x V reference chips, but they're all the lowest grade (highest tempco).[/quote]
Those will drift quite a bit (spec says 15ppm/1000hours typical).

[quote author="Freq Band"]Or would I just seal them up away from moving air, and trim as necessary depending on the ambient temperature of my lab?[/quote]
How will you trim them? Are you sure (enough) that a linear approximation is sufficient to stay within 5ppm?

[quote author="Freq Band"]Is there a standard method used to regulate the "heat" ?[/quote]
There are several ways, from a PTC heater (crude) to double ovens (complex). One of the most common applications is stabilizing a crystal oscillator; you can find more that you'd ever want to know about ovens and temperature stabilization techniques by Googling ocxo or crystal oven.

JDB.
[heating the diode isn't going to do you any favours noise-wise, either]

What exactly do you mean by '5ppm stability'?

Click to expand...

Sorry, I meant 5ppm V drift over time (1/2 hour).

Maybe you should see my other thread in the Brewery, it is more precise, and I don't want to hijack this thread.
Thanks JD.

http://www.groupdiy.com/index.php?topic=28189

=FB=

... There's no way to iron out the distortion out of a zener voltage referenced supply?

[edit] :?: