Noisy NPN SOT-23 reverse breakdown BE junction (Audio)

Hi, George:-

It's just a one-off or two-off, but I went and made a PCB and 3D printed case for it to stay in practice.

Yes, I'm sure I can generate the PRNG digital pre-recording and apply filters pretty easily in some program like Scilab etc.

Scilab, Matlab etc. import and export .wav files.

Just out of curiousity and despite this thread title does anyone know if a PNP reverse BE breakdown behaves differently to an NPN - is the noise spikiness different?

piglet

Almost certainly yes, though I haven't measured it. Silicon is one of the best avalanche materials because the ionization cross-section of holes is very small compared with that of electrons, about 1/100 iirc.

That means that once an avalanche begins, all the electrons arrive at the anode at nearly the same time, whereas the holes are smeared out by the transit time, depending on where in the depletion region they were generated, but that causes no issues because they don't contribute to the avalanche. Thus the avalanche is well-behaved and dies out in an orderly way.

Materials such as GaAs are horrible--the electron and hole cross-sections are nearly equal, so the holes cause ionization as they head for the cathode, and the avalanche bounces back and forth for a long time until it dies out stochastically.

For a noise source, it might be the case that a reverse-biased LED makes a better noise diode than a zener. (Some types sure do make nice high-Z switches.)

I tend to lean toward fundamental noise sources myself.

Cheers

Phil Hobbs

Huh, and I would have guessed they were the same. In both cases it's a p-n junction. Does the doping density at the ends make a difference? I figured the avalanche happens in the depletion region.

If you didn't care too much about speed, There are these GaP red leds that can be biased as spads. (you then just need a clean light source.) But maximum count rate is the pits... ~10 kHz... And the led capacitance ~20 pF along with typ. 100k ohm quench resistor is a ~2 us RC time. You do get nice ~1 volt (RC)spikes for each 'event'.

Oh here's an interesting theory type question. Say I take my slow spad and trigger a one shot. And to make it easy three time scales that differ by a factor of 100. The one shot has a width of 10 ns. t_w (with no sharp edges, we could talk about sharp edge case too.) The spad has a dead time of 1 us. t_d and the average count rate is 10 kHz... one every 100 us. t_r (Oh I've got to let my spectrum analyzer average for a long time... several seconds... in 100 seconds I'd only have 10^6 events.)

In broad strokes what would a plot of the noise density look like? (I've made a sketch... will add later. my biggest unknown is the slope of the first roll-off, I put it in as first order, 20dB/ dec.)

George H. broad strokes... I've got a corner at 100 kHz. ~ 1/(2*pi*t_d)

It's still a p-n junction, but in real life pnp's tend to have higher b-e breakdown voltage than npn's. That would tend to make them more avalanche than zener.

Thanks, I've never tried to zener (rev. bias) either. For a TH 1n5xxx zener I found more voltage was more noise... at least around the knee. In theory you should be able to look at noise and DC current and figure out the avalanche gain.

Sure, but the geometry is different. The BE junction is harder to deplete than the CB junction because the emitter doping is much higher, and the base region is very thin.

I don't know what the overall effect of that will be on the avalanche behaviour, but I expect they'll be quite different.

Cheers

Phil Hobbs

sorta related,

How aboyt Incandescent light bulb. A lamp running white hot should produce plenty of thermal noise. (a while)

Avalanche is a free gain-producing preamplification on any noisy process, and to get a true thermal noise source, you want broadband.

So, an excellent source is a waveguide containing a gas discharge tube; there's many gigahertz range, and good statistics, for such a device. It's good enough, but not convenient, so mainly seen over 50 GHz (according to a Keysight app note).

You can't really run it white hot, because tungsten melts at ~3400K.(*) The Johnson noise power is kT/2 per hertz, so a tungsten bulb is only 10 dB noisier than a room-temperature resistor.

Cheers

Phil Hobbs

(*)There are '3400K' bulbs, but that's because the IR emissivity of hot tungsten is fairly low, shifting the colour temperature up by a couple of hundred degrees. They really run around 3200.

kT per hertz, so a tungsten bulb is only 10

Phil Hobbs

I tried to look at the noise from a tungsten lamp (small ~28V bulb) Three 'problems'.

That I believe. It's a lot easier to use a quieter amplifier to get that 10 dB.

Cheers

Phil Hobbs