SPICE model for photodiodes including transit time effects?

The Eurocon paper from Lasovic et al? That almost had me clobbered once, but then again I am not exactly the big whizbang math genius.

Not all of that but maybe this helps in the search:

L. Ravezzi, G. F. Dalla Betta, D. Stoppa, A. Simoni, "A versatile photodiode SPICE model for optical microsystem simulation" in MICROELECTRONICS JOURNAL, v. 31, n. 4 (2000), p. 277-282

The go-to guy for photoelectric stuff on this group is a guy name Phil Hobbs; he seems to -- oh, wait.

(Sorry -- couldn't resist)

Laziness has been a bit of a hobby of mine for awhile now. ;)

Cheers

Phil Hobbs

"Laziness is the mother of invention."

Thanks. That's sort of what I'm doing now, except that the model I'm using includes shot noise. Their data table doesn't make a lot of sense--they've got the device capacitances listed backwards, for one thing, but even so, the n+/p sub device is still too slow.

Those all seem to be RC models decorated with various things like noise sources and lead inductance, which is the sort of thing I'm doing now.

The reason I care is that larger photodiodes have different behaviour for photocurrent generated in different places. For obvious reasons they can't have a nice thick metal layer over top of the epi like ordinary diodes. The epi is heavily doped to improve its conductivity, so there's not much E field inside it, and the electrons have to travel mostly by diffusion. Heavy doping also reduces the carrier mobility.

That tends to make regions of the diode far from the contact slow down quadratically with increasing diameter, which is something I care about a lot. The quadratic slowdown is masked in datasheet curves, because the RC delay is going quadratically with diameter as well. However, the RC delay is position-independent, so there's a big difference in performance for things like noise cancellers, which rely on the photocurrent being a faithful replica of the incident beam power no matter where it happens to land.

There are Schottky-barrier photodiodes and double Schottky barrier (interdigitated) photodiodes, which have metal top layers, but not in any decent sizes like 1-3 mm diameter. Some clever sub-wavelength interdigitated structure could probably be nearly independent of position and polarization, but now is not the time to launch a research project on photodiode design, though it might be fun eventually.

Cheers

Phil Hobbs

Or the father, as my mum used to say. (Of course she may have been biased.) ;)

Cheers

Phil Hobbs

If I were any better at device-level modeling I'd offer myself as a resource to do the work. As it is I think I could do a dynamite job if it was in LTSpice and if you had a white paper that spelled out the effects, which you just needed to be translated into SPICE-ese.

But only given those two conditions: even if you had the clear and concise paper but not LTSpice, I'd have to decline.

Thanks!

I'm using LTspice, but haven't found the clear and concise paper yet. :(

Cheers

Phil

It also lowers the average speed so it's probably not just a concern to engineers trying to cancel noise.

A group of scientists from Japan had that patented and I wouldn't be surprised if some company started making them. Usually they are called interdigital electrodes.

... oh, this might be it:

Hi Phil, Do you have a link to that paper? Maybe I can decipher it.

Or, better yet, describe the effect and maybe I can directly model it.

(I need something fun to do :-) ...Jim Thompson

That's true. I'm just a couple of orders of magnitude more sensitive to it, which makes me the photodetection equivalent of the miner's canary. ;)

Interdigitated PDs have been around for 30 years that I know about--classically, they have two opposing Schottky barriers, so you have to bias them above reach-through (when the potential well between the barriers is tipped over enough by the applied E field that the minimum reaches the position of one of the barriers, which then effectively goes away).

The Hamamatsu ones are doing something different, because they work at low bias voltages. I was talking mostly about using a sub-wavelength grating for the fingers, which seems like it could be insensitive to polarization and position.

Cheers

Phil Hobbs

Hi, Jim,

Thanks. What I'm trying to get a handle on is the effect of transit time in the epi, which is position-dependent, vs. the substrate, which basically isn't.

Small photodiodes are RC-limited, but bigger ones aren't, and I need to figure out what's going on. I can do C-V measurements to get the doping density vs depth, but the lateral effects are harder to get a handle on. (If I were a silicon device guy, I'd be further ahead than I am--I just know enough to be dangerous, and not enough to be really clear about it.)

So something like

contact p+ epi-------------------------------p+ epi--------- nu (very low doped N) nu nu nu N++ contact metal_metal_metal_metal_metal_metal

where the top layer is both an RC transmission line (which looks like diffusion) and also really diffusive, due to the low fields inside.

If there's a standard way to model transit time and diffusive transport in SPICE, that would be a win. Modelling high level injection in the nu layer would be interesting too, but maybe that's too much work and too many adjustable parameters.

Thanks again

Phil Hobbs

The paper is at

(or

Cheers

Phil Hobbs

Thanks! The only Lazovic paper I found was a horrible scan with a watermark that obscured all the important stuff. I'll take a look and see if I can decipher it. ...Jim Thompson

Thanks. I think my stackup was probably backwards--on reflection, it would be pretty perverse to use p-type epi, since its carrier mobility is so much lower.

Cheers

Phil Hobbs

Is the top contact just around the edge? In that case, the p+ epi resistance is max for light that hits the center, less closer to the edge. For slow pulses! Yikes.

Does anybody apply a contact grid to the top, like a solar cell?

Worse than that. Lots of them have a contact only at one edge, which (assuming diffusion dominates) makes them 4 times slower at the opposite edge than in the centre.

Various people have made PDs with indium-tin oxide electrodes, or very thin metal on top of the epi. It costs you some efficiency, and ITO isn't that great a conductor anyway. (It's a pure hole conductor, for one thing.)

A subwavelength grid, properly designed, could do a pretty good job, but might be a bit narrowband and would certainly be more money, because you'd need real masks. Photodiodes can practically be made with a stencil and spray paint.

Cheers

Phil Hobbs

I don't know of any such models, not having looked, but it occurs to me that the issue has been extensively researched by CERN, both for direct radiation detectors, and for CCDs.

The CCD master is Janesick:

Scientific Charge-Coupled Devices, James R. Janesick, SPIE Press, 2001,

This may be a source of ideas.

Joe Gwinn

Thanks, I have JJ's book. (We corresponded on the CCD mailing list for some years, on and off.) CCDs are a different regime though--small and slow, vs. big and fast.

Cheers

Phil Hobbs