some pics

It has DC outputs on pin jacks as well as the SMA one. I expect that the gold brick is an InGaAs photodiode with a 50-ohm resistor on the anode and a bias tee on the cathode, to apply bias and allow them to measure the DC photocurrent. In the process they need to not screw up the settling transient at late times, while not turning the PD to lava if you send in too much light. (There's probably 10V of reverse bias, so once you get into the milliamps, the PD dissipation can be significant.)

So the cathode circuit will need a medium-speed TIA with carefully controlled input impedance, a current-limited bias generator, and a diff amp to ground-reference the TIA's output.

Cheers

Phil Hobbs

Seems like a lot of parts. Competitive boxes have one battery and a cap. The gain is low, so it's probably not an avalanche diode.

I wish some patriot would bootleg the SD-series schematics. And the

No, it's a regular photodiode. But it's a really small one, which makes it electrically and thermally delicate.

The 1180x team were really bird-dogging their step responses, at both short and long times.

Ditto.

Cheers

Phil Hobbs

Can you get close to a mA from a PD that does 8 GHz into 50 ohms? I was blasting a big PD the other day.. measuring laser power And my PD saturated at ~10 mA

That was with a laser spot smaller than the PD area, so it might be a "local" saturation effect.

George H.

Reverse bias helps a lot. Other things that help:

Thick epi. Reduces lateral voltage drops and local carrier depletion, at the expense of short wavelength response. (You basically only see the light that makes it through the epi before being absorbed.)

Large spot size. Reduces the local current density and injection level.

Good transimpedance amps. Keeping the voltage swing across the diode under control gets rid of a lot of nonlinear capacitance funnies.

Cheers

Phil Hobbs

Yeah this is a PD->TIA that I made before reading your book. No bias on the PD. A DC measurement in this case. I'll have to try bashing in a bias.

George H.

Reverse bias helps, because it makes depletion region wider? I guess I don't know what causes the saturation.

George H.

Reverse bias drops the junction capacitance radically. And after that effect levels out, more bias still sweeps charges out faster, which continues to speed things up.

Recombination due to high level injection. The recombination rate goes as n_H*n_E.

With enough carriers, you shield out the built-in E field in the junction, so the carriers just diffuse around till they recombine.

A big reverse bias pulls carriers out faster, so it takes more light to make that happen.

Speed in PDs is limited by RC delays (R is mostly in the epi) and by diffusion in the undepleted regions. Some PDs speed up amazingly when you put on enough bias to deplete them right up to the contact.

There are two types of ohmic contact: (a) a heavily-doped region and (b) a Schottky barrier of negative height. I don't know if anyone uses (b) in PDs, but it ought to make them faster.

Cheers

Phil Hobbs

I should imagine the contact itself should have an interesting shape; a fractal of the right type ought to have minimum ESR and maximum transparency (small optical cross section, and if extending to very small scales, thin enough wires or short enough segments to be transparent to relevant wavelengths too).

Same applies to transistors, which always bothered me; the only conclusion I can think of is, they /want/ power transistors to perform poorly. Perhaps so designers don't accidentally get them singing at 200MHz+. In principle, there should be very little difference between, say, a 2N7002, or an IRF540, and comparably rated RF parts.

Tim

Doesn't work that way with metal, which is why PDs use the top epi to conduct photocurrent to contacts at the edge. You can use interdigitated Schottky contacts, but you have to really bias it up in order to get the opposed diodes to conduct, and you lose sensitivity due to the metal area.

(I'm doing optical antenna design right at the moment.)

The doping profiles are really different though. Getting high breakdown voltage slows down the diffusion-limited speed.

Cheers

Phil Hobbs

Great, that makes sense. By the epi layer you mean the top, p-doped layer of the Photodiode. (just checking) So here's another question. You've often said that you speed things up, by cranking up the reverse bias such the epi layer starts to deplete. Now if that happens shouldn't that increase the resistance of the epi layer? (or do you have to deplete right to the layer... but no further.)

George H.

....

...

I'm always impressed by the quantum efficiency of InGaAs photodiodes.

At 1550nm the resulting responsively usually works out to close to 1A/W. Which is convenient for back of the envelope calculations.

In this case the sensitivity shown on page 11 is 25mV/mW this is consistent with a diode terminated directly with a 50ohm resistor at the diode and a 50ohm termination at the other end of the cable with no amplification, as Phil said.

They state the sensitivity as 30mV/mW at 1330nm - I'm a bit surprised why it is more than at 1550nm, usually I have seen a rising response up the cutoff at ~1700nm. Page 12 shows a peak at ~1350nm.

I agree it seems a lot of bits to do the biasing and low frequency optical power measurement.

kevin

The specs say that the output impedance is 1K nom, so they don't terminate internally. It ohms just about 1K with power off.

The only user-accessible output is the SMA connector. Mysterious.

It's amusing to look at microphotos of transistor dice, the frequency DOES affect the emitter and base connection geometry. Most important, though, is the effect in SCRs (and LASCRs). Speed matters a LOT when you're putting kiloamps into the wafer, and turn-on time is dominated by a dI/dt limit (comply or things melt). The gates look like christmas-tree fractals sometimes. For LASCRs, the 'gate' is a set of holes through the metallization, and the dI/dt limit basically goes away if you can flash-lamp the turnon signal through all of 'em.

I'm not really a semiconductor physics guy, but I don't think so. You have metal contacts at both ends, so there's an effectively infinite reservoir of both electrons and holes, and a fully depleted epi has a whole bunch of available travelling-wave states for them to move in and out of.

Fully depleted is quite different from intrinsic.

Cheers

Phil Hobbs

If it had a QE of 100%, it would reach 1 A/W at 1.23 um. It has a lot to do with the epi absorption, junction thickness, and AR coating (if any).

Keeping the junction thin makes the device faster, but hurts the long-wavelength efficiency because a lot of the light doesn't get absorbed within the depletion zone (where the carriers can be collected).

Cheers

Phil Hobbs

Nah, there are one or two pin jacks for the DC stuff too, aren't there? There are on my SD-46es.

Cheers

Phil Hobbs

This is an SD43. There are lugs on the board but they don't come out. Maybe they used the same PCB but didn't bring out the slow signals for some reason.

Still mysterious.

The '43 is 8 GHz and works at 850 nm, where many other o/e converters are blind.