Photodiode Preamplifier: Ultra-low Light, High-Speed Detection

David,

For noise reasons you generally want as much gain as possible in the preamp, however this is complicated by the somewhat large capacitance of your detector which will result in peaking of the opamp noise gain at higher frequencies.  As a result, you may need to use a lower gain (smaller feedback resistor in the TIA) so that you can place a larger feedback capacitor to reduce peaking and still have sufficient bandwidth.  TI published a number of very good app notes that discuss the issue in detail:

http://www.ti.com/lit/pdf/SBOA060

http://www.ti.com/litv/pdf/sboa122

Also, Jerald Graeme has a really good book that includes a lot of the information from the app notes.

http://www.amazon.com/Photodiode-Amplifiers-OP-AMP-Solutions/dp/007024247X

Another good source is Hobbs book...

http://www.amazon.com/Building-Electro-Optical-Systems-Making-Applied/dp/0470402296

I have not done any analysis (i.e. I'm winging a quick answer here) on your problem but my guess would be a *starting point* for a quick noise analysis would be a 1 Meg feedback resistor and 22pF fb cap (~7kHz bw) followed by a 10X voltage gain.  This reduces the noise gain peaking to ~6 (compared to 50 or so if you tried to get all of the gain from the first stage while preserving the bandwidth).  This at least will let you know how difficult the problem is.  I'm playing a little loose with BW here as I don't know what your requirement really is (how much error can you tolerate at 1kHz?).  Depending on what you find you may be able to increase the gain in the TIA a little and not hurt the noise...a full analysis is needed.  I'm a little concerned as a gain of 10 million is only 100uV for your low end signal.  There must be some more gain somewhere...

The real analysis will have to look at the actual pole and zero locations in the noise gain and also include the opamp you use.  How much phase margin do you need (or overshoot can you tolerate)?  Are the signals pulses or sinusoidal?  OPA627/OPA827 are good but somewhat pricy and have input bias currents little too close to your input current for comfort.  And you may have to use something else depending on your supply voltage requirements.  TI has a number of good parts. OPA140 is a good part but also has too much bias current.  OPA129 looks tempting and might work.  It has excellent bias current and current noise specs.  Voltage noise is a little high but might be OK.  Look for low input bias current and low current noise.  You may be able to relax voltage noise a little as I don't think your detector capacitance is high enough to where voltage noise will be the issue.  A full analysis will reveal the dominant sources.

Sorry for the half-baked comment but the app notes should give you a good starting point for doing some real engineering.  Plus I'm sure the TI guys will jump in and help.  TIA design is a little tricky as there are several tradeoffs that have to be balanced.  A properly designed simulation can speed the process but don't skip the theory as that will help the design process converge.  Also, don't skimp on circuit construction and quality decoupling.  I often remove the copper from under the inverting node of the opamp to reduce stray capacitance. 

Best of Luck!

Tim

Thank you so much for posting. Half-baked is better than nothing!

I have been through the app notes you posted (been through just about every app note I could find on TIA's and photodiodes). I've also read and re-read quite a bit of the Graeme book (though its not very easily digested :) ).

I started reading the Hobbs book. It is a much easier read than Graeme's. I actually came across one of the articles he did on front ends and it seems to be included in the book as well (a PDF of the book is available online).

I did a full simulation of the TIA front end in MATLAB (which was quite a learning experience). Current noise was always the dominating factor. (btw, Is there a way to search TI amps by current noise density? I could never figure that out...)

But, in the end, some new information on the project might render all my previous efforts useless. It is likely that to get my 0.1nA current I will need an array of 20 large area photodiodes for a total capacitance of 60nF. Is it still possible to get above 1kHz with this large capacitance?

Graeme suggests a boostrap and buffer circuit in the Wideband chapter which supposedly makes the bandwidth independent of the photodiode capacitance. Hobbs suggests a cascode configuration. Will I need to do a full analysis of both circuits to see which is better?

Oh, and one more thing...

What do authors mean when they say "full signal swing" as in:

"

Since the full signal swing appears across the detector capacitance Cd , the output rolls off starting at...

"

I've seen this phrase so many times and am still not sure what it means.

This may be prohibitive, but if you use an array of photodiodes, is it possible to put a preamp on each?  I currently have a design where a 2cm x 2cm detector is mounted on one side of a pcb and the preamp (plus a bunch of other circuitry) is on the opposite side.  Depending on bias, the diode capacitance is 500-1000pF and the bandwidth is close to 40 kHz.  Then use a summing amplifiers to gang the signals together.  Actually works for you as 1) with less capacitance on the preamp you likely have better SNR at that point, 2) assuming all diodes are seeing the same signal the summing output has N x the signal and 3) assuming uncorrelated noise, the summing output has SQRT(N) noise increase. 

60nF is a lot.  The largest I have worked with is closer to 5 nF.  I do recall a design (possibly by Jim Williams) that worked for 5 nF devices.  I don't know if it was in one of his books or on Linear Technologies web site.

Another question...are you reverse biasing these detectors?  You pay a little price in noise but also will reduce the capacitance of the diodes.  That may help.

Tim

I don't think having an amplifier for each photodiode would be too difficult. The current design uses 19 large area photodiodes (14.88mm²) with each a capacitance of 3nF. The engineer who designed the system says about 0.1nA is generated when light is incident and that is the figure I am referencing.

The system is operating at a chopping frequency (hopefully above 1kHz) and DC will be filtered out so the diodes will certainly be reverse biased to reduce capacitance and increase bandwidth. That is, unless there are better designs that don't require reverse biasing and can somehow isolate the photodiode capacitance from the BW equation (like Graeme suggests).

A design with 19 or so small area photodiodes was considered but the small area (0.77mm²) most likely means that the amount of light incident will be around 350x less which might make the signal undetectable.

Also, what are your thoughts on photodiodes with built-in preamps? Are they any good?

I only have experience integrated preamps from Hamamatsu and that experience is somewhat dated (20 years ago).  But at that time they worked well.

Tim

Alright, thanks again for all the help. Much appreciated! I am going to dig into both books and see if I can come up with something. Seems like I have quite a few options.

Hey, just an update...

It turns out the 60nF array was composed of quite older components. Newer photodiodes with 15mm² have much less capacitance than 3nF. I chose two 100mm² to get the job done and so far they have been working quite well. Less capacitance, more responsivity, and less components in the design. The tradeoff being large response times as well as higher dark current. Since our design is based on chopped light and our chopped frequency probably won't get very far beyond 100kHz or so, both of these tradeoffs didn't end up being much of an issue.