How long does it take for a diode to conduct?

Heh heh heh.

Answer: "A long time, if it's hanging off the end of a long wire looped through the air".

This is what I get for getting all caught up in project OVER THE PHONE, without demanding pictures.

I swear, you guys are all talk.

Here's a 1N914:

It kind of looks like the area under the overshoot curve is sort of constant.

John

I find this threat rather interesting, as I learnt that the conduction of a diode was instant, for either minor, or majority carrier conduction. And if anything the potential difference across the junction actually increased over time due to stored carriers until equilibrium. So I've been waiting here for an explanation of why a diode should take time to turn on, assuming of course it's not down to lead inductance. After all with 700 Amps you don't need many nH!!

I don't know _why_, but for the one diode data sheet that I found that listed forward recovery, t_fr was about five times t_rr.

As stated in the follow-up, the real problem was a combination of inductance and trying to diagnose problems over the phone.

El 12-04-12 1:16, John Larkin escribió:

Hello John,

Very nice pictures! I do recognize them (for a 1N4448). I first "discovered" this in a symmetrical clipper driven from 10 MHz, peak current about 100..200 mA.

First I blamed parasitic inductance, probe issues, common mode, etc. But in the end I (after trying various setups) am sure it was the forward recovery time of the diode.

Also the reverse recovery of the other diode could be seen very well, even with a 350 MHz oscilloscope.

With kind regards,

Hey, if I set the pulse generator to make +5 and -5, you can see the reverse recovery, too...

I should try some higher voltage diodes. They are more of a PIN structure, so should be different.

A real step-recovery diode would be fun, too. Pity I have a day job.

In the forward direction, you start at zero current, so it takes a while to pump in carriers. In the reverse direction, the diode is nearly a short, so the generator pumps in a lot of current immediately, which clears the carriers out. Like this:

Something hand-wavey like that.

Sorry I missed the explanation, but was just as interested why anyone would measure it, it's not something I would see as useful except taking into account series inductance, not like turn-off times which are diode type and structure dependent.

If the diode is there to prevent flyback from destroying circuit elements, then you'd care deeply about t_fr. So there's reason to not only want it measured, but guaranteed.

I assume that you were using high speed differential probes coupled around a problem diode ? It's very hard to see these problems doing common measuring. Problems with in node area's tend to hide themselves form the common world.

Jamie

I think it's called recombination in P N, metal types like schottkys do not seem to exhibit this problem.

I've never had that must issues working out these problems. Most likely cause I have not pushed it to the limits as you are.

Jamie

Didn't really matter in this case -- the quality of measurements is overcome by the repeatability of the measurement. The same transient signal appears on all nodes, regardless of probe ground clip orientation and location, and the use of ferrite beads to cut out common-mode sneak paths (often, probe grounds will pick up stray noise even when the probe is short-circuited). Like I said, the signal was pretty much the same no matter where it was measured.

As far as in-node problems, I've found that a compact inductive loop probe does wonders. Another, very useful way of looking at it is an uncalibrated amprobe: set your scope to integrate and it becomes a crude, freehand Rogowski coil. I've measured the waveform of current going through vias in this way -- it would be some feat to accomplish that with a voltage probe!

Tim

I've recently taken an IGBT module onto the bench for laboratory testing. I already know these things have criminal inductances, but man, the diodes are slow, too. The co-pack diodes easily double the apparent inductance in the package, which is already a grotesque 20nH or so.

What astonishes me is, manufacturers don't even give a crap about it. They must not know their wires have length, or something. I'm amazed they can even be tested to measure the datasheet parameters, what few data they have. Hard switching just once nearly ought to explode them.

Tim

Sorry havn't really been following thread, so don't know if this has already been posted:

I think we are talking cross purposes. I haven't seen anywhere in the literature that says a native diode inherently has a mechanism where it takes time for a junction to conduct. It leads me to conclude that a turn-on time is package dependent, through package inductance, skin effect or whatever other mechanism may come into play. I hear what you're saying, and from a designer's point of view t-fr must be taken into account, I was trying to establish its cause.

oss

of

many

ng

Yeah, Phil H. posted that earlier. Very similar pics to what John L. just measured. I was looking through my two 'practical' solid state books (Streetman and Sze) but could find no discussion of the forward turn on time. (They both discuss the turn off time.) It's interesting that different diodes (of the same type) show such different turn on times. If I'm remembering my basic diode physics correctly, then you have to wait for the minority carriers to diffuse to the junction. Since the diffusion coefficient is pretty much the same, I would speculate that the difference is due to different 'lengths' in the devices... somewhat different doping profiles?? Though I'm now speaking way above my knowledge base.

George H.

No, it's a minority carrier effect. A Schottky in the same package on the same board won't show the same effect.

Cheers

Phil Hobbs

It really, really does. See the pics I posted. The higher the voltage rating, the wider the intrinsic region, the slower the turn-on, even in the same package.

Yes...

However, measurement error/inductance must be involved for a multi-hundred volt transient.

Bert