Front end transimpedance amplifier for low capacitance APD

Hi Julien,

Since you are interested in lowest noise, you will find the equation to get started on page 11 of the OPA657 datasheet ( http://focus.ti.com/lit/ds/symlink/opa657.pdf).  This equation calculates the equivalent current input noise for a transimpedance amplifier.  Plugging in different amplifiers and comsidering the input capacitance of the amplifier, you will have:

                         Cd                               Ieq

OPA657          6.2pF                          17.35pA/rtHz

OPA846          4.8pF                           5.9pA/rtHz

OPA847          4.7pF                           5.26pA/rtHz

To calculate these, I considered a parasitic Cf of 0.2pF and a 1kohm feedback resistor.  This tells me that the post filter limiting the bandwidth of frequency F needs to be less than 160MHz.  I selected 150MHz for easy comparison.  This is also 3 times the minimum bandwidth requirement for the 20ns pulse.

As you can see from the equation, the lower the feedback value, the larger the nosie contribution from the voltage nosie will be.  As such a FET input amplifier such as the OPA656 or OPA657 will not provide the lowest possible noise.  The lowets nosie is normally achieved with a decompensated bipolar amplifier such as the OPA846 or OPA847.

To achieve a fully differential output, it may be possible to use the THS4520 amplifier, but the voltage noise is higher than that of the OPA847, 2nV/rtHz (THS4520) vs. 0.85nV/rtHz (OPA847), and as such the resulting noise would be higher than for a single-ended solution.

Best regards,

Xavier

Thanks a lot, Xavier, for your help !

I considered also tje OPA847 at the very beginning of my studies.However, I was somewhat afraid that the bandwidth of that product may be too low for 1 kohm transimpedence. THe rise-time of my signal is roughly 2 ns.... I would be glad not to cut it too severely.

You are welcome.

For a charge change of 10fC in 2ns with a trasimpedance gain of 1000, you will have a slew rate of 2.5V/us.  For a charge change of 2pC in 2ns and a transimpedance gain of 1000, the slew rate requirement for the amplifier is going to be 500V/us.  The OPA846 is going to have marginal slew rate with 625V/us, but the OPA847 should have enough with 950V/us slew rate.

Slew rate is something important, ok. But if the bandwidth of the device is below, let's say 200-300 MHz, I'll spoil my signal, won't I ?

You are correct, small signal bandwidth is important to maintain the fidelity of the pulse.

I attached 2 TINA-TI simulation file.  (The software can be downloaded at http://focus.ti.com/docs/toolsw/folders/print/tina-ti.html.)

The first file contains the OPA657 model.  It seems that I cannot exceed 200MHz bandwidth when doing an AC analysis.  For the AC analysis, the model shows no oscillation.  When doing a transient analysis with a 1mA 20ns pulse, the circuit oscillates.

The second file is the same circuit with the OPA847.  The AC response analysis indicates a 500MHz bandwidth and shows not sign of instability during a transient response.

Note that a post-filter set lower than 160MHz may be required to limit the noise.  This filter may also degrade the pulse quality.

The OPA847 seems the best choice for noise as well as achieveable bandwidth for your transimepdance gain.

you are of great help Xavier !!

THanks again !

So I go for the OPA847.

I intended to put a simple LC filter at the output of the opamp, right before a 50 ohms resistor.

Would you suggest to use a more sophisticated setup ?

I meant, a LC filter connected to the ground

I was thinking at a minimum RC filter.  LC filter should be fine and provide a lower brickwall equivalent bandwidth filter.

If stability becomes an issue due to the load seen by the amplifier, the SBAA108 application note may be of some help  See link: http://focus.ti.com/general/docs/litabsmultiplefilelist.tsp?literatureNumber=sbaa108a

You were very helpful  at the beginning of the  week...
 Must go for the next very stupid question ....

In the design of transimpedence amplifier, the sensor is always connected to the inverting input.

I don't get the reason for that ? As far as I understand the transimpedence stuff,  it should work as well on the other input , correct ?

I order some evaluation boards for the OPA847 and the design for the board is made such as the signal goes to the non-inverting input of the amplifier.  Actually, I wondered whether I would really have to modify the circuit.

Hello Xavier,

for my application, i´ve chosen the OPA657 as first stage, followed by the THS3202. I also like to detect a pulse from a photo diode (in my case a SI-PIN diode). The problem is, that i get a very low current from the diode - around 1nA. That´s the reason why I´ve chosen OPA657, because of it´s small input current noise (I think this should be smaller than my input signal). The main thing I like to detect is the fall time of my pulse. This should be around 1ns. The pulse width is not the problem in my case (a few hundred ns it necessary).

The input capacitance of my diode is 1.6pF. Do I need to add 5.2pF for the amplifier? I´m not sure if I´m right with my suggestions, but I think the largest bandwidth and the smallest detectable current is possible in my configuration.

Are my considerations correct or do you also recommend another amplifier?

Thank you in advance. Best regards.

Andreas

Hello Andreas,

Your selection of a FET amplifier such as the OPA657 is correct.  Depending on the transimepdance gain and desired bandwidth, you may also want to consider the OPA656.

For the source capacitance, you will have to add the amplifier's capacitance to that of the diode.

Once you have completed your design, you can send me your schematic for review at x-ramus2@ti.com if you want.

Hello Xavier,

do you know, if the spice model in Tina  contains the amplifier input capacitance, or do i need to put an extra capacitor at the input?

Is it correct that i need 350 Mhz of BW (0.35/1ns) to assign the correct fall time of my pulse?

Andreas,

Yes, this is correct.  To achieve 1ns resolution and ensure that it comes from the pulse and not the amplifier, you will need  actually more than 350MHz.

Hi Andrea,

The OPA657 spice model Rev E contains the parasitic capacitance, but the input resistance of 1Gohm is not modeled.  Other revision of the spice model model do not nodel the input capacitance.