The use of opamps as a transimpedance amplifier is well known and a good analysis of the noise behavior of them is in the old Burr Brown Application Bulletin AB-076 from 1994. This is still available from Texas Instruments’ web site as sboa060 – simply search for that on their website. Improving the signal to noise and speed of such transimpedance amplifiers has also been discussed by Dr. Philip C. D. Hobbs in his books, papers and articles such as http://electrooptical.net/www/frontends/frontends.pdf and http://electrooptical.net/www/canceller/klugesandhacks6c.pdf This blog looks at some of these potential improvements.

One of the noise problems you will face with transimpedance amplifiers is caused by the effect of the photodiode capacitance, particularly if the photodiode is large and therefore has a high capacitance even when reverse biased. This causes a voltage “noise gain” from a corner frequency determined by the photodiode capacitance (plus opamp input capacitance) and transimpedance feedback resistor. The circuit used to illustrate this is shown below:

where the current source and 100pf capacitor represent a large area photodiode and the bandwidth and noise are simulated below:

The rise in output noise is largely caused by the combination of the photodiode capacitance combined with the feedback resistance providing amplification of the opamp noise voltage as the frequency increases. As mentioned in the various analyses of the transimpedance amplifier, the voltage noise of the opamp is multiplied by the non-inverting gain of the configuration. At low frequencies that gain is unity so isn’t an issue but once the combined effect of the photodiode capacitance and feedback resistor start to take effect the gain increases and the noise rises at 20dB per decade. This can be better seen in isolation by simulating the voltage gain of exactly the same circuit but using a voltage source on the non-inverting input without the current source.

Here you can see the gain is 0dB at low frequencies but then rises, peaking at just over 40dB before falling again as the closed loop bandwidth is reached. The 3dB point of the rise is 15.6kHz. The product of the photodiode capacitance of 100pF and the 100kW feedback resistance is 15.9kHz which is not a coincidence. How disastrous this noise gain is depends on the voltage noise of the opamp and the closed loop bandwidth of your configuration. A lower capacitance photodiode will help but that may not be an option. Also, the opamp input capacitance must also be added to the photodiode capacitance in performing the calculation although in this case it is small compared to the photodiode capacitance.

One design improvement proposed by Dr. Hobbs is adding a common base transistor (cascode) to isolate the opamp from the photodiode capacitance. This works to an extent as shown below, but some bandwidth is lost in the process:

The total noise is improved by more than a factor of three but the bandwidth reduces from 2.2MHz to 842kHz because the bandwidth now has the transistor as an additional limiting component. Taking the original amplifier and limiting the bandwidth to 842kHz would have reduced the noise to 590µV anyway so the “real” improvement is less than a factor of two rather than the factor of three.

An additional problem is producing the current for the cascode. There could be little or no static current in the photodiode so an extra 100k has been added in this example to bias the transistor (R2) and provide consistent transistor biasing. This provides the bias current for the cascode but also shifts the opamp output to 1.75V where it was 0V before. This is not necessarily desirable so an alternative biasing method is required to avoid saturating the opamp output. The circuit below provides this:

The additional resistor R3 provides a route for the cascode current without it having to come from the opamp feedback resistor and Q2 provides a voltage drop equivalent to Q1 so the voltage drops across R2 and R3 are roughly the same. Note that each time you add something there is a penalty of some sort, even if there is also a benefit. The extra resistors and transistors that we are adding increase the noise (except Q2, which could be decoupled where it connects to R3 anyway). This can be seen below, comparing the cascode with and without the extra biasing components.

With the extra resistor the noise increases although it doesn’t have a major impact on total noise. The increased noise is not resistor noise but because the gain viewed from the non-inverting input is now not unity due to the extra resistor. This amplifies the opamp voltage noise even at low frequencies. The reduced bandwidth is still there so you have to accept that in order to reduce the noise. Reducing the resistance to increase the current in the transistor and increase the bandwidth actually makes the noise worse and you will never be able to get the bandwidth back to where it was without the cascode transistor present, at least with the opamp shown. The benefit of using the cascode transistor and the component values chosen will depend on the design objective and how much flexibility you have (if any) in selecting the photodiode capacitance and system bandwidth.`