PD Photocurrent rise/fall time an unexpectedly strong function of wavelength?!

[evb149]

This is a mere curiosity since it has been a while since I read about / did design with photodiodes and I was surprised to see a data sheet which indicated that a reverse biased silicon photodiode would have a LOT (50:1?) longer rise & fall time across an external 50R load resistor depending on wavelength. 

The device in question "typically" has the slowest rise/fall times at around 940nm, its peak sensitivity wavelength, whereas it
has much faster (1/50th) rise/fall times around 540nm where the photocurrent sensitivity is about 50% of the peak value.
So only one octave difference in wavelength, one octave difference in photocurrent sensitivity, but a 50x worse transition time at the peak sensitivity wavelength.

I don't recall ever seeing anything like that empirically listed in a photodiode data sheet nor do I recall any theoretical reason why
it'd be so in a case such as the above.

I don't think there are any phosphors or odd wavelength dependent factors about the device besides being a generic Si PIN photodiode.

I didn't ask the OEM or look at wherever they detail their test conditions specification for enlightenment, and it's of no real importance in this ad hoc case, I'm just wondering if I'm not thinking of some general truth that'd extend this to other devices and situations which I should know.

EDIT: The only things that comes to mind (and I doubt it's relevant or even the case to a significant degree here) is if the actual photoelectrons are 'hotter' / faster when a short wavelength generates them than a long wavelength and may have more mobility / speed; also it'd be possible the short wavelength phototons are generally absorbed at a different depth than the long wavelength ones and that could lead to a different diffusion time until they're externalized and contribute to the photo-current.   Interesting ideas, but somehow they don't feel satisfactory here and if they were significant to this degree I would think I would remember hearing about this level of time dispersion before since the effect if true is HUGE.

[741]

That is interesting and likely not "well known" - would you be able to post a link to the datasheet please? I don't think A.D. do photodiodes, but if they did I'd post an RAQ (Rarely Asked Question).
Cheers
Stephen

[Kleinstein]

The wavelength effects on how deep the light goes into to bulk matrial. At 940 nm (and even more with longer wavelengths liek 980 nm) much of the light is absorbt deep in the bulk material, while the active PIN structure is close to the surface of maybe a few µm. With PIN diodes this depth depends on the type, but often the low dopded zone is still relatively thin and not making up the bulk. The electron hole pairs have to first move (in a relatively weak field, gradient) to the junction before they actually contribute to the photo current.
With 540 nm most of the light is absorbed in front or inside the junction. This makes the photocurrent to get effective nearly instantaneous.  The front layer (often N doped) is usually quite thin.

For the different wavelengths the relevant effect is changing the absorbtion in the silicon. The absorbtion goes down quite a lot when approaching the band edge.

[jonpaul]

please post diode ID or spec sheet

All photo sensitive devices have a curve of current per unit illumination vs wavelength

Risetime into a capacitive load dv/dt= I/C

As excitation wavelength changes, so will Conversion to current, thus low current gives slow Risetime.


Suggest you read a few papers or a textbook, EGG, UDT, Tektronix have great light measurement notes

Jon

[evb149]

Thank you all very much for the responses!

Here's the spec sheet which motivated my question attached (hopefully) to this.

[evb149]

Coincidentally just earlier today I noticed and was reminded they do make these things with
integrated PDs and a TIA, though since they're so integrated and tailored for a specific purpose I assume there'd be
no similar thing noted in the DS.  I'll have to download some of their DSs and have a look, though.

https://www.analog.com/en/product-category/photo-detector-integrated-transimpedance-amplifiers.html

That is interesting and likely not "well known" - would you be able to post a link to the datasheet please? I don't think A.D. do photodiodes, but if they did I'd post an RAQ (Rarely Asked Question).
Cheers
Stephen

[evb149]

Thank you very much for the interesting explanation!

I vaguely remembered hearing something about spectrally dependent absorbtion depth in semiconductors but I couldn't recall for sure if that was noted about camera image sensors or photodiodes or solar cells or what and also what the real world effects of that were for this case.

Very interesting...

The wavelength effects on how deep the light goes into to bulk matrial. At 940 nm (and even more with longer wavelengths liek 980 nm) much of the light is absorbt deep in the bulk material, while the active PIN structure is close to the surface of maybe a few µm. With PIN diodes this depth depends on the type, but often the low dopded zone is still relatively thin and not making up the bulk. The electron hole pairs have to first move (in a relatively weak field, gradient) to the junction before they actually contribute to the photo current.
With 540 nm most of the light is absorbed in front or inside the junction. This makes the photocurrent to get effective nearly instantaneous.  The front layer (often N doped) is usually quite thin.

For the different wavelengths the relevant effect is changing the absorbtion in the silicon. The absorbtion goes down quite a lot when approaching the band edge.

[evb149]

Thank you very much!  Oh, yes, that's a very good point; I thought a little bit if somehow it could have
to do with the capacitance but I discounted the idea before thinking it through because I just assumed they'd be
using comparable irradiance levels for the 530nm and 940nm tests, so there'd only be the
~ 1:2 sensitivity vs lambda difference between the photocurrents and in the favor of the slower response having higher presumed photocurrent. 

But it was only a bad assumption that the radiance would be anywhere near equal, so if they used a considerably weaker
intensity source for the 940nm test then for the reason of maybe much less photocurrent the response time there could be
even further slowed because of the capacitance.

I'll have to see if I can find the test conditions document to tell what their test setup is and if the irradiance of the two test
wavelengths differ much or not.

please post diode ID or spec sheet

All photo sensitive devices have a curve of current per unit illumination vs wavelength

Risetime into a capacitive load dv/dt= I/C

As excitation wavelength changes, so will Conversion to current, thus low current gives slow Risetime.


Suggest you read a few papers or a textbook, EGG, UDT, Tektronix have great light measurement notes

Jon

[jonpaul]

Rebonjour EVB149, You may have some misconception  re light intensity, radiance, and the efficiency of conversion, ie, the unit of light energy or intensity vs diode outp current.

Please check the total capacive load diode, connectors, cables, probe and scope , you may have 10..50 pF.

The diode current can be measured or calculated.

The quantum efficiency of a photo emissive device is a strong function of wavelength and varied between a maximum and zero.

Plank, Wein, Einstein etc.  worked out this physics in 1900..1920s.

Work at BTL in 1950s...1960s characterized solid state photo diodes and transistors.

Fast photo sensitive devices are common in nuclear research, 1 nS is quite feasible.

Enjoy,

Bon courage

Jon

[Marco]

Risetime into a capacitive load dv/dt= I/C

That's unlikely to be relevant in the context of a 50 Ohm loaded reverse biased photodiode. It's generally not rising to a fixed voltage, it's rising to the equilibrium voltage which is only a small fraction of the bias voltage. With the simple current source + parallel capacitor model, an instantaneous increase of the photocurrent should lead to a predictable risetime regardless of the magnitude of the change.

A simple google search shows Thorlabs giving a nice image of the cause to go together with Kleinstein's explanation. The effect seems only relevant to long wavelengths, since the depletion region is almost right on top of the photodiode.

Conceivably a manufacturer could also thin their die before final assembly to mostly get rid of the effect, if photons bounce off the reverse electrode and are straight back in the depletion region the effect will almost disappear.

[tom66]

Risetime into a capacitive load dv/dt= I/C

That's unlikely to be relevant in the context of a 50 Ohm loaded reverse biased photodiode. It's generally not rising to a fixed voltage, it's rising to the equilibrium voltage which is only a small fraction of the bias voltage. With the simple current source + parallel capacitor model, an instantaneous increase of the photocurrent should lead to a predictable risetime regardless of the magnitude of the change.

Not only that, but the change in photocurrent isn't that significant between 530nm and 940nm (roughly 2:1) but the rise time varies by 100:1, so it's clearly not as simple as dv/dt.

[jonpaul]

repeat experiment into a transconductance amp or 1 Ohm R so only the current is sensed and voltage change is negligable

Jon

[jonpaul]

Check the curves of Quantum Efficiency vs wavelength.

QE 1.0 1 photon >>1 electron.

It is rarely even close to 1.0 at the peak of curve. It is a property of the material.

https://en.wikipedia.org/wiki/Quantum_efficiency
Jon

[mawyatt]

I vaguely remembered hearing something about spectrally dependent absorbtion depth in semiconductors but I couldn't recall for sure if that was noted about camera image sensors or photodiodes or solar cells or what and also what the real world effects of that were for this case.

Very interesting...

Foveon (later acquired by Sigma) made use of this wavelength dependent depth with their unique image sensor which created a "3D" pixel where Blue was captured at the top surface, Green in a mid-layer and Red photons on the bottom surface of the pixel. The original sensors were made at National Semiconductor on a Silicon bipolar process line.

Interesting Carver Mead was one of the original investors in Foveon.

Best,

[mag_therm]

Comments here are  about the actual quantum efficiency versus wavelength rather than the rise and fall times although they may be related.
Single photodiodes are held reverse biased and the current variation with light level is used.
Camera sensors have the pd reset/pulled negative just prior to the integrating period with shutter open.
Color information is by a tri-color mask over the sensor which otherwise has all cells equal.
The sensor photodiode levels are written to a raw HEX file prior to any processing.

I extracted data for a few pixels to see the levels of the 3 colors.
https://app.box.com/s/473axtfr6pm43m4pi0et5i97blovdlt5

Note the readings for the LAB 50,0,0 grey image. The raw levels will be multiplied by the manufacturers' correction matrix and should result in
that neutral image. (before color temperature correction and user preference etc)
Firstly, the cameras use 2 times more Green wells than Red and Blue.
This is because the s/n ratio of green is best.
Secondly Red has the lowest levels, along with worst s/n.

The sensor quantum efficiencies versus wavelength  are adjusted by manufacturers for best fidelity. Generally the peak is at about 0.65 um
the blue side rolls off by the choice of coating of the front illuminated sensor.
The red side rolls off (I think unavoidably) because of recombination deeper in the cell before the electrons reach the storage.

My ref here for the above comments is from 2011 and might be getting a bit dated:
Holst and Lomheim "cmos/ccd Sensors and Camera Systems"

[Kleinstein]

In the red range photodiodes can reach a quantum efficiency quite close to 1. The main point there is reflection and shading from the contact areas. For the blue range the refraction index of silicon changes quite a bit when the direct band-gap is reached. This makes the anti reflective coating less effective and makes most solar cells and photodiodes look blue. The surface layer may also have nore recombination that the bulk material, which can also slightly reduce the QE in the deep blue.

For the NIR side the QE goes down as the lighte enters ever deeper into the buld and thus a slightly increased properbility for reconbination (and thus loss in current). The design and purity of the material effects how fast to QE drops to the long wavelength end. An additional drop can also come from decreasing effectiveness of the anti reflective coating as the refractive index also changes at the long wavelength end. By itself the reduced QE does not effect the speed. It is the extra distance in the bulk of the material that takes additional time. To generate a current from the electron hole pairs generated deep in the bulk, it need to build up a concentration of minority carriers to drive them to the junction. This acts a bit like additional capacitance.

Even though quite fast as a photodetector most photodiodes are slow diodes showing really slow reverse recovery. They may be an option as an RF PIN diode (act as a variable resistor / swtich for high frequency AC) for the lower frequency end, down to some 20 kHz. Chances are photodiode are among the slowest Si diodes one can get, as they use quite pure material. With a high qualtiy solar cell (kind of close to a photodiode) I have measured a reverse recovery time of some 50 µs.

[jonpaul]

Good cat: UDT: (now OSI) 

App notes, slection chart, specs on a wide varierty,

https://peshkin.mech.northwestern.edu/datasheets/Optoelectronics/Photodiode_UDT_catalog.pdf

https://www.osioptoelectronics.com/default.aspx

Jon

[StillTrying]

This is a mere curiosity since it has been a while since I read about / did design with photodiodes and I was surprised to see a data sheet which indicated that a reverse biased silicon photodiode would have a LOT (50:1?) longer rise & fall time across an external 50R load resistor depending on wavelength.

I've read that it can be up to 1000 times slower.

When I was flashing LEDs at a photodiode redish colours including IR had slower edges and curved tops, I don't know whether that was the LEDs or the PD.
The fastest I got was with 200mA though super bright blue or green.
https://www.eevblog.com/forum/chat/20w-halogen-bulb-viewed-by-a-photodiode/msg2411274/#msg2411274

To reduce the intensity of the blue or green I often shone the LED at the side of the SFH 213 PD, I was expecting the fast edges to be ruined, but that didn't seem to make any difference to the speed - I could still get <20ns rise and fall and a flat top.

[Kleinstein]

Some LEDs in the red / IR range can also be slow with rise / fall times in the 1 µs range. One can test the speed of the LED by looking at the reverse recovery of the LED. The delay / time constant of the LED is similar to the revese recovery time.

[TomKatt]

This discussion is well beyond my comprehension, but it reminds me of CuriousMarc's repair video of the photo chopper circuit in his vintage HP plotter...

Love that channel!

[Kleinstein]

Those LDRs used in the photo choppers are a bit different from the ones to just measure brightness: they are faster and this comes with reduced sensitivity. The LDRs get good sensitity from long lived traps, that capture electrons or holes and this way conserve the extra photoconductivity for a longer time. Without that the photo conductive effect is rather low sensitivty.

The tricky part is to get some medium speed trapping centers to get enough sensitivity, but also only few long lived trapping centers that give a slow tail.  Quite often there are a few very slow parts and it takes really long (could be hours) to get back all the way to the dark resistivity.

[RJSV]

Yes, Kleinstein, I'm trying to study / comprehend that reverse recovery concept, doing work on solar cells, in some other EEVBLOG posts.
   I think I'm seeing some hidiously long responses, like 1.2 mSec but also confusing is a little 4-pin IC doing comparator switched output. Going to take a little time to comprehend.

[Kleinstein]

Relatively pure silicon usually has rather few deep traps, though one can not totally exclude them. In photo conduction of germanium I have seen a slow component  - at low temperature it got really slow, but at room temerature it was not so bad. Looking at the reverse recovery at different temperature sometimes runs under the fancy name of deep level transisent spectroscopy.

Besides the electronic effect there can also a thermal effects. 1.2 ms is about in the time scale for thermal equilibration for a normal thickness silicon wafer. When driven in forward direction the heat generation is not that uniform across the thickness of the solar cell - there are even regions (e.g. the contacts if there is not much recombination at the contracts) that see cooling from peltier effect. At university I did my diploma thesis about the heat production in this case. In addition there can be lateral concentration of the current - this would tend to be slower to get thermal equilibrium.

[RJSV]

Glad to see that 1.2 mSec delayed photo- diode isn't out of bounds.  I've got Tektronics scope right here in the study room, and all the parts, (just busy), so I should be emphasizing doing all those needed checks.
Thanks, lots.

[RJSV]

   (Some more info, on Solar Cell interface):
  The 1.2 mSec delay I quoted was an average, as both switching directions happen, via 'inverter' chain, in my implementation of a classical Edge Detect; that's with several 'gates' feeding in serial order, into a two input logical 'AND'.
   In some related trials, I did quite easily manage to simply connect one of those 'garden light' gates as a straight replacement to a regular push button, in one TOY keyboard.  That works to play a piano note, or to cause a canned song to play.
Interesting part of that is the solar light, as a 'gate' does work in EITHER connection config...That is, if we assume that toy does key scanning by pulling down a 'row', while monitoring a set of 8 columns in typical style.  One direction of connecting not super good, but reversed worked quite well.  I'm a bit concerned, about possibly supplying excess voltage, in that substitution, as solar cell unit, (don't forget, output is 222 khz), can put out actual voltage, vs a passive dome push-button, but seems to work, by way of small flashlight.
Kind of weird, that it works in both directions of hook-up.
  I need some more time, for more results, as I need to construct (the optical edge detect, using several gates)
that being on the test bench.  Plus, I need to get a periodic action, for best oscilloscope capture, of those optical rise and fall times.  That, to characterize for actual switching time, as evb149 has been describing, the large delay seen in photo-voltaic output, in one direction of switching.  Thanks.