Low Noise, Adjustable Gain, Photodiode Amplifier

Photodiodes are optoelectronical devices widely used in optics. They convert incident light into an electrical current in a linear way over several decades of input optical power and can be used to detect signals as large as a few mW/cm² down to less than a pW/cm². One of the major disadvantages of photodiodes however is that the current they produce is extremely small with typical conversion gains of about 0.5 A per optical Watts. A notable exception to this are the Avalanche photodiodes which have conversion gains typically one hundred to one thousand times the conversion gains of normal photodiodes but they require relatively large bias voltages which make them hazardous to work with. I will therefore stick here to conventional photodiodes which represent already a great part of the detectors available at most optical suppliers (Thorlabs, Edmund Optics, Newport…). Please note that there exist different optical sensor devices such as photoconductive materials (including lead sulphide and lead selenide which are very useful in the Infrared region), pyroelectric detector, thermopiles, bolometers…

Photodiodes are semiconductor devices that are sensitive to light. When a photon of enough energy hits the photodiode junction, it promotes an electron to the conduction bands which then generates a current. The energy required to promote an electron to the conduction band depends on the material. For instance, silicon is sensitive to photons of energy higher than 1100 nm. However, if the photon is too energetic, it will not penetrate enough the photodiode material to reach the junction. This then produces absorption bands that are specific to each diode depending on its construction. Most photodiodes are made of silicon which is usually sensitive from 200 nm to 1100 nm but other diodes have sensitivities in the UV (e.g. GaP photodiodes) and other in the infrared (e.g. MCT photodiodes).

Because photodiodes generate currents, they must be used with voltage-to-current converters. A typical transimpedance circuit for driving a photodiode is given in Figure 1.

There are two degrees of freedom in the circuit of Figure 1: the gain resistor R and the bias voltage V_bias. The gain sets, as its name suggest, the scaling factor between voltage at the output of the circuit and the current generated by the photodiode. Resistors from 1 kΩ to 10 MΩ are frequent choices for gain with silicon photodiodes. The bias voltage is something that is dependent on the application and system parameters. Values from 0 (no bias) to 20 Volts are frequent and will influence both the speed of the photodiode response, its maximum (saturation) output and its noise. Larger bias will have larger saturation currents and faster response but at the expense of higher noise generated by the photodiode. On the contrary, unbiased diodes are typical for low-noise, low-frequency applications. Maximum operation speed at no-bias will be fixed by the capacitance of the photodiode but an unbiased photodiode can usually be used up to a few hundredths of kHz.

The circuit proposed here, and shown in Figure 2, accepts a large variety of photodiodes inputs and convert their currents to voltages with conversion gains from 10³ V/A up to 5×10⁶ V/A through an eight positions selector switch. It was designed to support both unbiased and biased photodiodes with a switch to allow for a 0V or 5V bias. Moreover, the circuit can drive a 50Ω load output up to 5 Volts. It was designed for low noise applications and I successfully measured signals of 15 pWatts with it using an affordable Thorlabs FDS100 photodiode. The Gerber files of the PCB are given at the end of the post so that you can reproduce the circuit yourself.

The circuit has three major modes of operations. It can accept (1) a TO-5 or TO-39 photodiode directly mounted on the board (such as Thorlabs FDS010, FDS100, FGA21, FGA10, FDG03, FGAP71) (2) a cathode-grounded mounted photodiode with or without a 5V bias (warning: only the Thorlabs SM05PD7A, SM05PD2A, SM05PD1A, SM05PD4A, SM1PD2A, SM1PD1A photodiodes seems to accept the 5V bias), (3) any Thorlabs biased photodetector such as the DET36A (warning: check that the bias is set to zero before plugging the photodetector or you may damage it permanently!). Obviously, any other photodiodes can also be adapted to the circuit by using a BNC cable with the central connector connected to the anode of the photodiode and the outer connector connected to the cathode of the photodiode. Be sure to check the bias switch before plugging the photodiode and also check the manufacturer recommended maximum bias voltage for the photodiode you are using.

The active part of the circuit is shown in Figure 3. It consists of the transimpedance amplifier of Figure 1 (U1:A) with gain selectable by the 8:1 analog switch U2 (DG408) driven by the BCD switch SW1 (with its pull-down resistors R9-11). The source can be fed either through the BNC input J1 or via the TO-5 socket shown as a photodiode with grounded shield in the figure. In case of a TO-39 photodiode, the leg spacing is the same but there is no shield pin for the casing (which is without consequence for the compatibility). A RC low pass filter made by R20 and C15 filters the 5V bias voltage fed to the photodiode. The bias voltage can also be applied to the guard of the BNC connector J1 through the jumper although on Figure 2 I decided to replace it by a small switch because I did not order the proper jumpers. Finally, the output voltage of the transimpedance amplifier is sent to a 50Ω line drive made by U3 (LT1010) and U1:B. The role of U1:B is to cancel the high offset voltage of the LT1010. Also, the circuit is mounted as a unity gain inverser because the output of the transimpedance amplifier will always be a negative voltage and we would like to have an overall positive output. The bandwidth is ultimately controlled by R13 and C3 (160 kHz) although it can be lowered by the gain of the transimpedance amplifier and the photodiode used as well as its mode of operation (biased or unbiased).

The power source part of the circuit is shown in Figure 4. A great part of the effort of the design of the photodiode amplifier presented here was to reduce the noise as much as possible because it is this noise that will ultimately fix the smallest signal measurable with the photodiode. It was therefore important to filter as much as possible the noise coming from the power supply. As usual, I used two TC962 (U4 and U5) to produce the split power supply with enough current capacity for the 50Ω load except that this time I used a +15 Volts input and two LDO regulators (U6 and U7) to clip any ripple in the supply rail and produce the clean ±12 Volts required by the circuit. Similarly, a 5V LDO regulator (U8) produces the 5V bias voltage. Please note that these LDO have some error in the output voltage and that they will not produce exactly ±12 Volts and +5 Volts. Be sure therefore to always check the actual bias voltage value before applying it to your photodiode, especially if the maximum bias voltage is of 5 Volts precisely. If you do not have a +15 Volts power supply, you can eventually work with a +12 Volts one and replace U6 and U7 by their ±9 Volts counter parts (MC78M09 and MC79M09). Please note that I did not test it yet so I cannot guarantee that it will work exactly the same, especially in terms of output voltage swing.

The circuit was assembled using standard 1206 components and tested using a Thorlabs FDS100 photodiode and a Thorlabs DET36A detector that is an already biased version of the FDS100 photodiode. I recommend to first check all the voltages of the circuit before testing a photodiode. Also, I recommend to use a cheap photodiode such as the Thorlabs FDS100 in a first place to prevent damaging a more expensive photodiode in case something was not assembled properly.

The performances of the circuit were extremely satisfactory and met the expectations. As an example, the following video shows the response of the photodiode amplifier output of a LED driven with a different kinds of waveforms:

The noise of the circuit was measured with a 100 MHz oscilloscope with no photodiode plugged and a 50Ω terminator in at various gains. The results are shown in Figure 5. The overall oscilloscope noise floor is on the order of 300 µVolts RMS. At gains higher than 40 dB, some noise start to appear and reach 3.4 mVolts RMS at 70 dB. However, the noise structure clearly consists of an EM pickup and shielding the circuit with an aluminium foil reduced the noise to the oscilloscope noise floor back again. I would therefore recommend putting the circuit in a shielded box if possible. If you choose to do so however, do not connect the second BNC guard to the mass unless you do not plan to use the biasing at all.

The same experiment was again run with a Thorlabs FDS100 photodiode in the TO-5 socket (still with the aluminium shield wrapped around). The results are presented on Figure 6 with some ringing of about 1 kHz appearing at 70 dB. Interrestingly, the Thorlabs PDA36A did not show the same pattern although it is made of the same silicon photodiode and the noise remained at the oscilloscope noise floor. I was not able to explain this phenomenon yet but a possible answer would be ripple current at the output of the LDO. I would therefore recommend using Thorlabs own biased or unbiased photodiode for small signals.

As an order of magnitude, for a Thorlabs FDS100 photodiode such as the one used in the PDA36A, the responsivity is about 0.4 A/W in the visible range. With a 70 dB gain, a 300 µVolts signal such as the one measured (not considering that the actual noise is probably lower than that), this already represent a Noise Equivalent Power of 16 pWatts which corresponds to about 1 nWatts/cm² for this diode. Although the gain could be increased slightly, it is not recommended to go above a ×10 gain because the noise will ultimately be limited to the Johnson-Nyquist thermal noise of the feedback resistor. As an order of magnitude, with a 4.7 MΩ resistor such as the one used for the 70 dB measurements, and with a 150 kHz bandwidth at 20°C, the thermal noise of the resistors should already be about 110 µVolts, which is very close to the measurement performed here. Also, the diode itself will have its own noise (typically on the order of 10^-14 Watts/√Hz for the FDS100 at 900 nm and 20 Volts bias according to Thorlabs datasheets) and the amplifiers will have their own noise too. Slightly stronger gains may however be required for low signal application with an analog to digital conversion of ~1 mV step readings.

Finally, a thermography of the circuit was taken with a FLIR one camera. The results are shown in Figure 7. It shows a slight hot spot near the LT1010 element but of only 40°C. Some care must then probably be taken if the circuit is put in a small box but it should not be a major issue otherwise.

In conclusion, the circuit performed extremely satisfactorily with good linearity and low noise outputs. It can be a good low-cost alternative to optical manufacturer photodiode measurement units which are usually quite expensive for the amateur.