Photodiode Dark Current

At times, empirical data becomes both useful and necessary.

I wanted to establish a practical lower measurement limit for a group of photodiodes.

The following represents measurements made with an Agilent 3458A of a Hamamatsu G6854-01 InGaAs photodiode. The diode was optically sealed with zero-bias. Measurements were collected for just over 8 hours – 8PM to 4AM – overnight when facility operations were relatively quiet. Such noise is very close to an “ideal” Gaussian distribution. There are approximately 1000 samples/hour (actual: 951 samp/hr; 15.8 samp/min; 8463 samples in 8.9 hrs ~ 3.8 sec/sample)

The instrument had been left on for more than 48 hours and the photodiode was loosely enclosed in a box large enough to assume negligible variations due to self-heating but small enough to assume constant environment temperature. The actual temperature was not recorded but was in the neighborhood of 72°F. The instrument was factory-fresh and recently calibrated. Accuracy on the highest resolution range (100nA FSR; 1 pA resolution at 10ppm of reading plus 400 ppm of range)

With data points “joined”:

I did a bit of analysis using Mathematica. Here are the results:

Number of samples: 8463

Basic statistics of photodiode dark current:

Note that an ideal Gaussian Distribution has skewness of 0 and kurtosis of 3.00.

With this information in hand, I can make a few estimates of measurement limitations*.

Although there are methods to extract signal from noise in noisy environments, these often require synchronization techniques. For straight-forward signal extraction, it is very good to have at least 20 dB SNR. In the ideal case, this implies a minimum signal of … call it 200 pA. It is usually possible to squeeze 6-decades of current-mode information out of a photodetector – within the limits of the manufacturer’s recommendations of course.

So this particular photodiode might be usable for signals from 200 pA up to 200 μA … though I’d not like to stress the current that high in such a sensitive device. I’d prefer to limit the maximum signal to perhaps 20 μA – still a 5-decade range.

A 5-decade range just seems to cry out for a logarithmic amplifier, doesn’t it?

*I note this data is obtained by integrating over a 3.8 s sample period. If the detector is moving at a velocity of 200 m/s, this is the equivalent of integrating over a spatial period of 760 m. A time-sample interval of 5 s would be the equivalent of a spatial-sample period of 1 km.