High frequency photodiode transimpedance amplifier design

[ConnerP]

Greetings, I'm a physicist way in over my head with some circuit design and could use some help. We want to retrofit a high speed photodiode amplifier into an existing atomic force microscope. The present amplifier has a bandwidth of 900 kHz and but we are interesting in signals up to 10 MHz. I have already taken a stab at it but could really use some help.

Constraints:
Available power: +-15V
Signal fed to 10V ADC
Diode sensitivity: 0.36A/W
Diode capacitance: 45pF (0V reverse bias), 9pF (5V reverse bias)
Laser: <=1mW @ 690nm

My attempted solution uses the LT1226, the fastest amplifier with +-15V rails I could find. It is a simple low gain transimpedance amplifier for the first stage. The gain has been set low so that the compensation doesn't wreck the bandwidth. The second stage is pure gain to boost the expected max signal (roughly) up to 10V, no compensation needed since the 80R shields it from seeing any capacitance (I think).

My main concern is, will this thing be stable? So far as I can tell, yes, but maybe I missed something. But as I mentioned, I am a total noob in this area, so I am looking forward to hearing any and all feedback y'all have to offer. If you have tips on how to potentially go faster, I'm all ears.

[tggzzz]

There is useful information in The Art of Electronics 3, plus in Phil Hobbs' book Building Electro-optical Systems: Making it All Work.

[moffy]

Download LTSpice from the Analog Devices website. They have models built in for the LT1226 and you can test stability. Got a feeling the first stage will oscillate because the LT1226 is only stable for gains of 25 or higher.
The THS4031 or something similar might be a better choice for the first stage. It is low noise and unity gain stable.
Also look at the LTC6268 data sheet. It has a 100MHz BW transimpedance amplifier with a 20k feed back resistor. It is only +/- 2.5v supplies but might be worth a couple of voltage regulators to get the performance.

[David Hess]

Got a feeling the first stage will oscillate because the LT1226 is only stable for gains of 25 or higher.

If Cin/Cf >= 25 then it will still be stable.  Cf is required to compensate for the effects of the input shunt capacitance which would otherwise lower stability even with a gain of 1 amplifier.

[Marco]

Can't you just use the LTC6560? A completely integrated solution sidesteps some problems with high bandwidth opamps.

[moffy]

Hess, thanks for  the explanation. With 1pf vs 9pf it definitely would oscillate, unless the capacitance across the photodiode is increased. Could replace the 1pf with Bob Pease's two pieces of twisted insulated wire, each twist added about 0.1pf, but it's a bit esoteric. Opamps that are only stable for high gains are often hard to tame, and can go unstable if you look at them wrong. Just my opinion though.

[LaserSteve]

Please see attached... The Magic of Dr. Phil Hobbs.
His book is even more detailed on the subject.
There is more on his web page.
He does answer questions in Sci.electronics.design newsgroup

Steve

[TimFox]

When I was working with photodiode amplifiers for x-ray detection, I found this book very useful:  "Photodiode Amplifiers Op Amp Solutions"  (McGraw-Hill, 1995)
https://www.abebooks.com/Photodiode-Amplifiers-OP-AMP-Solutions-Jerald/30358566730/bd?cm_mmc=ggl-_-US_Shopp_Trade-_-used-_-naa&gclid=EAIaIQobChMIu6y2k_j86QIVFk2GCh1vvwHBEAQYBSABEgKPAfD_BwE
One in a series of useful books by Jerald Graeme of Burr-Brown.

[David Hess]

Hess, thanks for  the explanation. With 1pf vs 9pf it definitely would oscillate, unless the capacitance across the photodiode is increased. Could replace the 1pf with Bob Pease's two pieces of twisted insulated wire, each twist added about 0.1pf, but it's a bit esoteric. Opamps that are only stable for high gains are often hard to tame, and can go unstable if you look at them wrong. Just my opinion though.

I assumed the 1 picofarad feedback capacitor was a placeholder for a trimmer capacitor which is what I would use because the photodiode capacitance can vary from unit to unit and the fixture capacitance is poorly defined.

I remember Bob's Pease's recommendation to use twisted wire which in the past was known as a "gimmick" capacitor.  I have also used a short length of small diameter coaxial cable and trimmed the open end; this has the advantage of being shielded.

[ejeffrey]

What everyone said but also definitely recommend Phil Hobbs's book for anyone that picks up a photodiode.

[moffy]

I have also used a short length of small diameter coaxial cable and trimmed the open end; this has the advantage of being shielded.

Cool.

[arivalagan13]

I'm currently designing a TIA for Faraday cup detectors.
This Texas Instruments presentation was helpful to clear the air with TIA stability problems.
https://training.ti.com/high-speed-transimpedence-amplifier-design-flow

And, Phil Hobbs's book was exceptionally good.

And, Art of Electronics - The X Chapters book has a dedicated section on TIA which kind of consolidates what could be achieved best.

This Texas Instruments presentation might also help, for sure.

[David Hess]

Also look for Burr-Brown application notes on the subject.

[bobaruni]

I'm not sure if using that particular op amp in that way will give you stability, in the LT1226 datasheet they show a photodiode amplifier on the first page with a gain of 25 where the photodiode is connected to the non inverting input and I assume the main reason is to maintain stability.
The other thing I would mention is that the signal to noise ratio will be low and this may or may not matter depending on your application.

It's been a long time since I have worked on a high speed TIA, I last used a OPA659 and OPA657 with good results and while these op amps will not run on +-15V rails, they don't need to if they are followed by a second gain stage that can run on +-15v rails or better yet an ADC with a matching voltage range to the first op amp.
Note these are both JFET input, I tend to only use FET or CMOS input amplifiers for TIAs as I have had current noise and offset problems with bipolar input op amps in low light situations.

Having said this, the ideal scenario in terms of signal to noise ratio is to have as much gain as possible in the first stage so that the feedback resistor is as high as possible while still permitting the desired bandwidth.

There are some good online calculators for calculating noise such as: http://www.jensign.com/noise/noisecalculator.html
An also bandwidth and stability: http://www.jensign.com/stability/calculator.html
These calculations are key to designing with stability and determining the required feedback capacitor.

Bob.

[Marco]

Still don't really see how high speed opamps make much sense ... it will be hard to improve on LTC6560 on any metric.

[awallin]

Here's a few python scripts that compute transimpedance, bandwidth, and noise for a simple one op-amp TIA:
https://github.com/aewallin/TIASim
comparisons against real-world measurements show good(ish) agreement with a few OPA657 designs.

if anyone has insight on how to compute stability (vs oscillations), phase-margin, etc then that would be a good addition to these scripts.

[Marco]

With 33K TIA gain, it's johnson-noise limited

Seems unlikely, datasheet current noise 2.6 pA/rtHz, Johnson-Noise 0.7 pA/rtHz.

[Marco]

To really get close to Johnson Noise for these larger feedback resistors and at high bandwidth I don't see any other way than using a discrete JFETs or GaAs FETs common source amplifier. By themselves JFETs at least won't really have the bandwidth to be able to handle a 33k feedback resistor from drain to gate though, so you'll need an extra non-inverting amplifier.

Maybe a GaAs FET could do it with just a resistor from drain to gate and no extra amplification stages? (Well, a resistor plus something to add some voltage drop.) Dunno.

PS. all assuming inputs without huge capacitances, where Phil Hobbs wizardry comes into play.

[nfmax]

Although I wasn't personally responsible for the design, I have seen a GaAs HEMT followed by a CFA used successfully with ~100k feedback resistors giving bandwidths around 50MHz to 100MHz, from a low-capacitance APD detector. DC performance is terrible, but that didn't matter for the application.

[awallin]

FWIW, I added a model for the OPA818 op-amp to TIASim, and then compared the TIASim bandwidth-predictions to those in the TI datasheet.
When I add an extra sqrt(2) factor to the computation of the optimal C_F (in parallel with R_F) I get reasonable agreement.

opamp https://www.ti.com/lit/ds/symlink/opa818.pdf
script: https://github.com/aewallin/TIASim/blob/master/opa818_bw_figure.py

so this (maybe) adds some confidence to the predictions for BW and noise give by TIASim.

AW

[ConnerP]

Hello everyone, thanks for all of the feedback on this topic! I was quite busy the past month and didn't have time to revisit this topic until now. I've taken your inputs into consideration and made another attempt using the OPA818 followed by a secondary voltage gain stage. To achieve the precise small Cf value, I used a capacitance T network as described in the OPA818 sheet. So far as I can tell, this looks to check all the boxes, including stability, even if the parts are a bit pricey (after multiplying by 4 channels). Would y'all mind providing a second round of feedback?

I've included the stability simulation for just the OPA818 and the overall amplitude/phase response of the complete circuit with secondary gain stage. In the stability simulation green is open loop gain, blue is loop gain, and red is the combined response.

 OPA818_stability_test.asc (2.41 kB - downloaded 95 times.) OPA818_with_secondary.asc (2.99 kB - downloaded 62 times.)

[jonroger]

I highly recommend using the Analog Devices design wizard:

https://tools.analog.com/en/photodiode/