Circuit review: Transimpedance amplifier to measure shot noise in a photodiode

[Joakim]

Hi electronics enthusiasts!

In my research lab, we are currently investigating the quantum nature of shot noise in photodiodes, and I am designing the transimpedance amplifier circuit needed to be able to measure the small shot noise current produced by a photodiode when a LED is illuminating it. The LED is "mated" to the photodiode with a heat shrink tube in a Chinese finger trap-type contraption. I would like to minimize Johnson-Nyquist noise from the circuit and have the Poissonian noise dominate. It has proven to be rather difficult to get this circuit to work properly on both a breadboard and a protoboard, and therefore I believe I have to spin a proper board for this.


Previous attempts:
So far, I have assembled the very same circuit on both a breadboard and on a protoboard, with different results.

Breadboard:
The diode's anode is connected to ground and the cathode is connected to the amplifiers inverting input. Furthermore, the amplifier is connected to a single supply with a supply voltage of 5V. The output of the amplifier is fed back through a 1Mohm feedback resistor, and there is no feedback capacitor. The output of the amplifier is also connected to a Keysight frequency counter (DC coupled, 1MOhm input impedance). When light of the same wavelength as the photodiode's central wavelength is incident on the photodiode, the circuit outputs pulses of around 2 volts, and the amount of pulses per unit time is definitely increasing with increased illumination, but nonmonotonically. Obviously, this circuit had all the shortcomings of being assembled on a breadboard together with using ~1 meter long cables with banana plugs from the power supplies, so I am not sure how much noise is coupled to the circuit here.

After measuring the number of pulses per 10ms integration window for 100000 windows, when driving the LED with different currents using a Keithley 2400, I plot the distribution of samples. To this, I fit a Poissonian distribution with the same mean value as the mean value from the recorded data, and I observe that the distribution of the counts do not match. This leads me to think that what I am sampling is, in fact, not the fluctuations in the photocurrent.

What makes me suspicious here is that the pulses are well-formed and "discrete", while I would expect the output voltage to be continuous with a varying amplitude. Many implementations that I have seen use some kind of comparator circuit to digitize the current into pulses. Perhaps the amplifier is oscillating?

Protoboard:
Assuming that parasitics are the cause of the pulses I soldered a protoboard with the very same circuit without any success. The circuit instead oscillates at 90 MHz and does not respond to light. I also tried using a L7805CV to supply the amplifier (following the application example in the datasheet), hoping that any noise from the switched benchtop power supplies would be less without success. A naivë thought is that perhaps parasitics in the breadboard actually stabilized the circuit.


Schematic:
So, taking what I have figured out from prototyping this circuit, I have drawn up a circuit to be manufactured. Below follow my design choices and my thoughts about the circuit.

Mainly, the circuit is based on the LTC6268 FET-input Op Amp that I already have in my parts drawer. My understanding is that low input bias current amplifiers are good choices for transimpedance amplifiers.

• I have opted to use a gain of 1 MOhm, but as the photodiode datasheet does not specify responsivity, I cannot really estimate what gain is needed. Perhaps this resistor is easily swapped.
• As I don't know whether to run the photodiode in photovoltaic or photoconductive mode, I have added J2 to be able to switch between negative bias and ground.
• The amplifier can run off either a single supply or dual supply, so I have included the jumper J3 to be able to switch between connecting VEE to GND or to VEE_IN.
• The "Power ports"-block denotes placeholders for whatever physical connector/pad I will use for the power supplies.
• Decoupling capacitors are chosen somewhat arbitrarily based on this StackExchange post
• The output is a 50 Ohm SMA, but this can be anything "reasonable" like a scope probe or another IC (DAC)
• The amplifier does not need to supply any significant current, as anything I would connect its output to (DAC/DSO) would be high impedance.

The problem:
The last board I designed (with a great amount of help from the users at EEVBlog!) worked quite well, and the challenge there was high speed and high current capabilities. Here, the challenge is to isolate one noise phenomenon with a very low current, and measure it adequately.

Questions:

• Is this a reasonable design to accomplish what I am trying to do?
• I would like to have as high gain as possible to be able to detect the photocurrent, but not too much to have thermal noise in the resistor be the dominant source of noise. Perhaps 1 MOhm is too low gain here?
• It seems like the feedback capacitor is important, but I calculate (using equation 17 here) it to be around 3.5e-14 F, which seems very low
• I am afraid that using jumpers for the bias voltage and for the supply rails is non-ideal and could introduce unwanted noise. Do you believe this could be the case here?

Datasheets

• LTC6268 Ultra-Low Bias Current FET Input Op Amp
• TEFD4300 Silicon PIN Photodiode

Sincerely,
A graduate student who's once again lost in the analog forest 

[mawyatt]

In a transimpedance type configuration the photo diode is acting as a shunt capacitance across the - opamp input. With the large feedback resistor R the phase shift produced is enough for instability (Bode Analysis) with additional parasitic capacitances. Adding an optimized feedback cap across the feedback R is the usual simple solution.

Regarding isolating the two noise sources, the Shot Noise is proportional to squareroot I, while Thermal Voltage Noise of a resistor is proportional to square root R. So forcing I thru a (ideal) transimpedance amp will produce a result that favors higher feedback R since the output will be root-sum-squared of both {[R*(Shot Noise)]^2 + 4kTR} for normalized bandwidth. So separation between two noise sources increases as square root of R. This assumes no contribution from the (ideal) transimpedance amp.

Here a link that may be beneficial.

https://web.mit.edu/dvp/Public/noise-paper.pdf

Best,

[tggzzz]

There are several theoretical and practical that might jointly and severally cause the effects the OP has mentioned. I won't try to diagnose them, but I will point to information that will help him.

He is exactly the target audience of that classic book "The Art of Electronics" by Horowitz and Hill. It discusses this kind of topic as one of many in electronics, and is a solid starting point for deeper understanding of electronics. If his workplace doesn't have a copy, then they should get one. The "x-Chapters" sequel probably isn't necessary for this problem, but it is another useful resource.

Phil Hobbs' "Building electro-optical systems : making it all work" is directly relevant since "[it is] intended in the first instance for use by oppressed graduate students".

For generic signal and noise considerations, Henry Ott's "Noise Reduction Techniques in Electronic Systems" is well regarded.

If the OP uses solderless breadboard for a prototype, then they can expect to spend more time debugging that than debugging their design. For breadboarding, I would suggest Manhattan construction as per https://entertaininghacks.wordpress.com/2020/07/22/prototyping-circuits-easy-cheap-fast-reliable-techniques/

[Conrad Hoffman]

Burr Brown had a couple good app notes.

https://www.chem.uci.edu/~unicorn/243/papers/BurrBrown.pdf
http://nic.vajn.icu/PDF/Burr-Brown/apnotes/AB-057.pdf
also-
https://www.ti.com/lit/an/sboa061/sboa061.pdf?ts=1669642321632&ref_url=https%253A%252F%252Fwww.google.com%252F
https://www.ti.com/lit/an/sboa060/sboa060.pdf?ts=1669647060428
https://www.analog.com/media/en/reference-design-documentation/design-notes/dn399f.pdf
https://www.analog.com/media/en/technical-documentation/application-notes/an47fa.pdf (the big AN47 that everybody should read regardless)
https://www.electronicdesign.com/technologies/analog/article/21806128/matched-jfets-improve-photodiode-amplifier

That should keep you busy for at least a few minutes!

[Marco]

Do you know from principle the LED shot noise is less than the photodiode?

Isn't shot noise mostly relevant at high bandwidth? This isn't a high bandwidth circuit.

Why DC couple? You're presumably interested in relative noise levels, not signal amplitude. AC coupling and ignoring the signal near the transitions makes life easier.

[Someone]

Isn't shot noise mostly relevant at high bandwidth? This isn't a high bandwidth circuit.

Why DC couple? You're presumably interested in relative noise levels, not signal amplitude. AC coupling and ignoring the signal near the transitions makes life easier.

Shot noise can be dominant at any bandwidth, and is mostly seen at "low" bandwidth when other noise sources are easier to control.

But the other point is the same as my confusion. Transimpedance is usually referring to proportional amplification of the photodiode current, but the OP talks about "counting" of which photon counting is a valid operational mode but usually relies on extremely non-linear amplification (and quenching). Generally linear mode will get larger dynamic ranges which seems to be the interest of the experiment.

In my research lab, we are currently investigating the quantum nature of shot noise in photodiodes, and I am designing the transimpedance amplifier circuit needed to be able to measure the small shot noise current produced by a photodiode when a LED is illuminating it.

As above, go and read some books on the topic. There is one so narrowly specific it should be obvious:
Graeme, J. (1995). Photodiode amplifiers: op amp solutions. McGraw-Hill, Inc
Rather than over-fitting shapes, there is a well established industry method to assess noise contributions:
https://harvestimaging.com/blog/?p=1034

[moffy]

Perhaps I am missing something, but the photodiode in the PDF schematic feeds into the +Ve input of the OpAmp, which makes the OpAmp a unity gain buffer. There is also no DC path for the +Ve input which probably means that the input cap will go quite -Ve (due to photodiode current, assuming -ve bias). Also the 1M resistor does nothing, perhaps you meant the photodiode to go into the -Ve input, in which case you definitely need a compensating cap in parallel with the 1M resistor.

[mawyatt]

Didn't see the schematic pdf!!

That is not a correct use of the transimpedance setup, diode should be to - input and generally + input grounded, altho sometimes to a + bias voltage.

BTW the advantage of this transimpedance setup is that the - input to the op amp is at virtual ground, and the diode anode is also at AC ground. This keeps the dynamic AC voltage across the diode at ~ zero and no capacitance displacement current, thus removing the diode capacitance effects to a first order for dynamics at low frequencies.

Thanks moffy for spotting this for the OP.

Best,

[Terry Bites]

That's not a transimpedance amplifier, its a buffer with unity gain and gagging to oscillate.
Use this widget to design a real one https://tools.analog.com/en/photodiode/
and/ or www.ti.com/tool/CIRCUIT060018#overview
They will help with noise analysis too.
I found this:

[jmelson]

Right, transimpedance requires the signal to be applied to the inverting input.  Note that the intrinsic noise is worst at zero bias, it is quite interesting to see the noise drop as reverse bias is applied to the junction.  Remember to provide LOTS of filtering to the bias supply input.  That is easy to do with several poles of RC filter as the bias current will be VERY small.
Jon

[Kleinstein]

Besides the noise if the photdiode there is also noise from the light source. In many cases the shot noise is actually more from the light source and not the photo-diode.
There are special low noise laser sources that in combination with a reasonable good detector (good quatum efficiency) can get well below the classical shot noise limit. So the photodiodes produce less shot noise than one may think and the classical equation gives.  AFAIR there is an additional factor of  (1- quantum efficiency) for coherent light that is often ignored.

[Marco]

Shot noise can be dominant at any bandwidth

Dominant, but not very interesting. With laser light the shot noise tells you something, with current through a PN junction what's really there to measure? Yep it's shot noise?