Photodiode Feedback Resistors: Find When Photon Noise Limits Further Improvement

System boundary

A nonmultiplying photodiode and linear transimpedance front end, with stationary illumination and a defined observation bandwidth. Avalanche detectors, photon-counting electronics and complete optical instrument performance are not specified.

Integration interfaces

Integration risks

Separate the wanted optical change from total detected current

An instrument may report a small modulated optical signal on top of a much larger steady background. The amplifier nevertheless receives both currents. Subtracting a recorded background level removes its estimated mean; it does not reverse the random arrival events that occurred during the illuminated measurement. Name the signal current used in the numerator of a signal-to-noise calculation separately from the total current that produces noise.

Use a nonmultiplying photodiode model only when it matches the selected detector and operating mode. Detector dark current, shunt resistance and amplifier contributions remain separate inputs. Do not transfer an avalanche gain or excess-noise factor to an ordinary photodiode, or assume the simple model describes every light source. Correlated source fluctuations and modulation interference require additional terms rather than being relabeled photon shot noise.

Compare the two contributions as input-current densities

For independent Poisson arrival statistics without internal multiplication, the one-sided shot-noise current density is the square root of 2qItot. Here Itot is the sum of the applicable positive current magnitudes producing independent arrival noise. A resistor Rf at absolute temperature T contributes input-referred thermal current density equal to the square root of 4kT/Rf. Both quantities have units of amperes per square root hertz.

These are noise amplitudes, so independent contributions combine through their squared values. Setting the two squared densities equal gives crossover current Ic = 2kT/(qRf). This is a comparison of two selected mechanisms, not the complete front-end noise floor. The resistor's excess noise under bias, amplifier voltage and current noise, detector shunt noise and environmental interference must be assessed separately.

Calculate the available improvement before requesting a different resistor

For an illustrative 1 megohm feedback resistor at 300 kelvin, thermal current density is approximately 128.7 femtoamperes per square root hertz. The equal-density photocurrent is approximately 51.70 nanoamperes. At a total detected current of 1 microampere, shot-noise density is approximately 566.1 femtoamperes per square root hertz, already larger than this resistor contribution.

Combining only those two independent terms gives approximately 580.5 femtoamperes per square root hertz. Even an imaginary removal of all feedback-resistor thermal noise would reduce that total only to 566.1, an amplitude improvement of about 2.5 percent. A tighter DC resistance tolerance does not perform that removal. These assumed numbers identify a possible design priority; they are not measured ChipSimple noise data or a detector specification.

Do not mistake higher output noise for worse input sensitivity

Within the low-frequency region where the transimpedance magnitude is approximately Rf, multiplying an input-current density by Rf gives its output-voltage density. The resistor term then becomes the square root of 4kTRf. Thus a larger resistor reduces its input-current noise density while increasing its output-voltage noise density. Those statements are consistent because the signal gain also increases.

A comparison between networks must therefore retain the same referred measurand. With the assumed 1 megohm resistor, the two individual output densities are approximately 0.1287 and 0.5661 microvolts per square root hertz. Above the flat transimpedance region, multiply each contribution by its actual transfer function instead of using a constant resistance. Amplifier voltage noise can follow a different noise gain from detector current.

Check whether optical background consumes both noise and headroom

Reducing unwanted light before the detector can reduce its arrival-noise contribution as well as the mean amplifier output. Subtracting that light's mean digitally only addresses the latter measurement offset, and does not recover analog headroom already lost to saturation. A nominally small final displayed signal is therefore not proof that the front end operates at a small total current.

Record the detector current with the intended optical background present and with the specified optical signal added. Consider whether baffles, wavelength selection or a different optical operating condition are allowed by the instrument design. Such changes belong to the optical owner and must preserve the wanted measurand. Do not hide a light-source fluctuation by attributing every illuminated variance increase to the Poisson model.

Preserve bandwidth when testing a noise improvement

A quieter time trace can result simply from a narrower filter. State the transfer function and equivalent noise bandwidth when integrating densities into RMS values. For a flat density passed through an ideal rectangular bandwidth B, RMS current is density times square root B. A real one-pole filter does not have equivalent noise bandwidth equal to its corner frequency, and sampled noise requires the actual acquisition chain.

Increasing Rf may change the feedback pole, stability compensation and response time. Compare configurations at the required signal bandwidth rather than accepting a slower response as an unexplained improvement in resistor quality. If the instrument needs faster optical events, an apparently attractive low-noise high-resistance option may not satisfy the timing requirement. Preserve raw bandwidth settings with every result.

Use illumination dependence as evidence, not as a single-cause verdict

Under controlled stationary conditions, compare blocked-input and illuminated records while keeping gain, bandwidth and acquisition unchanged. Plot input-referred variance against the applicable total current rather than comparing two screenshots at different display scales. A linear variance term can be consistent with shot noise, but source intensity fluctuations, drift or nonlinear amplifier behavior can invalidate that interpretation.

Check saturation and clipping before fitting a noise model. Repeatability of the mean, proper detector bias and a stable reference channel help distinguish optical instability from electronics. Keep any background-reference measurement's uncertainty in the final estimate: subtracting two noisy readings generally adds their independent variances. A noise subtraction must never create a negative physical variance and then be reported as an exceptionally quiet device.

Specify the network contribution the customer actually needs

The useful handoff contains the total-current range, wanted signal, optical background, feedback impedance, resistor temperature and bandwidth. Include the assigned noise contribution and how it will be measured, not only a nominal resistance and tolerance. State whether the analysis is a screening calculation or a validated front-end budget so purchasing does not mistake a model boundary for a guaranteed component result.

For a custom ceramic thick-film feedback network, provide the circuit and loading conditions for drawing review. ChipSimple can assess the defined passive network; the optical and analog owners establish detector behavior and instrument sensitivity. Select a resistor improvement only when its predicted contribution matters in the connected system, and retain the headroom and speed checks when changing resistance or feedback capacitance.