[Laser] Photodiode testing.

[Steve J. Noll]

If it's of any interest, after serving as senior test engineer for a 
major optoelectronics manufacturer I can offer some insight into how 
silicon photodiodes are tested.

Dark Current is the single most important and commonly tested parameter 
and indicator of photodiode health.
In fact, for 'jellybean' parts such as the Osram SFH203P, this may be 
the only 100% test.
This can be done at wafer level on an autoprober. Bad die are either 
inked or wafer mapped to indicate not to be used.
Dark current is the current that flows when the photodiode is reversed 
biased with some predetermined DC voltage.

Although the data sheet may show dozens of specs these are pretty much 
never 100% tested, some may be sampled, sometimes only at wafer level, 
and some may never be tested as they are 'by design."

Shunt Resistance is calculated from dark current measured at 10 mV bias, 
ideally +10mV and -10mV, but more often at just one polarity.
Yeah, it's usually a pretty tiny current! Keithley Picoammeters and 
Source-Measure Units rule. I came to love SMUs, i.e.: Keithley 236. Wish 
I had one at home.
Shunt Resistance is important to some applications. For jellybean parts 
might just be sampled.

Responsivity, properly 'Absolute Spectral Responsivity.' Best measured 
with a spot of light that is completely contained within the active 
area. Flooded measurements are error-prone and avoided whenever possible.
Light is usually from a monochromator or a narrow band dielectric 
filter, and is calibrated for intensity using a calibrated photodetector.
The actual amount of light is usually unimportant and rarely specified 
by a customer. Too little and measurement is noisy, to much an detector 
is saturated.
Calibrated photodiodes from NIST are about $5K. We made our own and had 
them calibrated by NRC (Canada's NIST) for $1100.
We used monochromators (a VIS/IR, and a UV with a deuterium lamp) with a 
chopper and lock-in amp, and a proprietary spot generation system for 
custom tests.
Monochromator/lock-in tests are time consuming, never used for jellybean 
parts but reserved for expensive custom, or military, or space devices.
I built an automated  filter wheel-based machine for more common 
responsivity tests that can generate responsivity curves from 254 nm to 
1550 nm. Still, you'd never test inexpensive parts 100% this way, just 
samples.
I also adapted a wafer prober to make responsivity measurements (single 
wavelength) along with the more common dark current, shunt resistance, 
and Vf measurements.
Responsivity is measured in Amps per Watt (A/W.)Â  The Watts used for 
the measurement would be in the order of microwatts, and often less.

Capacitance - again, usually just sampled. Usually measured at 1 MHz. 
Often at zero bias and also with a reverse bias applied. Boonton, old 
HP, and CV plotters are of use.
Pretty rare to see a photodiode fail for this, would likely take a big 
screw-up in wafer fab for that to happen.

Forward Voltage (Vf) is measured often just to see if the wirebonds are 
intact, or during wafer probing to check that the probe needle is making 
contact.

Noise Equivalent Power (NEP) never measured for jellybean parts. 
Actually, almost never really 'measured' but calculated from dark 
current, shunt resistance, and responsivity.
The problem is measuring the requisite Shot Noise and Johnson Noise is 
too difficult, especially when the spec is for zero bias. There just 
isn't enough noise over the test equipment noise to measure.
I can give anyone interested the formula, and that for the related 
Detectivity (D*).

BTW, where I just retired from, Advanced Photonix, Camarillo, we over 
the years bought several of our competitors.
It was shocking to see their lack of test capability. Two did not have 
the equipment to make reliable responsivity measurements.
The third, when I saw green light emitted by their monochromator set to 
1000 nm, was oblivious to the second order output, and this setup was 
run by their PhD physicist!
It would not be too difficult for an enterprising Amateur to assemble a 
better equipped test lab as compared to some of the opto manufacturers 
I've seen.

Rise Time. Again, unless spec'd by the customer, like for a custom, 
military, or space part, is rarely measured. When measured just need a 
fast LED or laser and an o'scope.
Very desirable to make this measurement with a small spot, not flooded, 
for shortest rise time. Often measured into a 50-ohm load, so the signal 
is really quite tiny.

Breakdown (Vbr) - Usually checked by applying 10 uA from a Source 
Measure Unit with voltage limit set to the breakdown spec. Meter will 
indicate that 10 uA flow couldn't be reached at that voltage, so it passes.
Can actually measure where breakdown occurs, but that's not commonly done.

Active Area - may be always specified, but not measured on jellybean 
parts. Pretty much defined by the mask set. Many military and space 
parts require measuring even 100%, that's why they're so expensive.
Measured with a small spot of light with device mounted on an XY stage. 
Example would be a 1 mil spot, a Newport stage with digital mike heads. 
Set signal for a 100% reading at center of detector then move it 
left/right,etc., to find 50% points.
At same time Uniformity can be measured. Again, not something usually 
measured even on a sample basis for commercial parts.

Short Circuit Current is measured flooded. Usually done in a 'light 
tower' which is just a tall cabinet with a flood light at the top. The 
color temperature is specified, usually 2850 Kelvin (Minolta makes 
meters for this) and the intensity in Footcandles.
Color temperature is set by the lamp voltage, intensity by the distance 
from the lamp to the DUT. Current measured with a picoammeter or a low 
voltage burden DMM.
Short Circuit Current is sometimes measured using a solar simulator. 
Actual example is a solar simulator with an Air Mass Zero (AM0) filter 
to simulate outer space used to test photodiode sensors for satellites.
Calibrating this setup requires a photodiode calibrated against one that 
was flown in a high altitude balloon! I know of only one cal lab that 
does this, I don't think NIST or NRC does this cal.

Avalanche Photodiodes require individual special testing. Among other 
things, to find their operating bias for the specified gain, i.e.: 100.
The large area models I was involved in got 100% automated testing, 100% 
burn-in, and shipped with individual data sheets. Even the less 
expensive small area APDs are 100% tested and burned-in.

Some non-opto and non-electrical tests: PIND (Particle Impact Noise 
Detection test) for metal can packaged TO (Transistor Outline) parts. 
Part is shaken while a sensitive microphone listens for loose specs of 
stuff inside. Common for space and other critical parts.
BTW, for TO parts the die is mounted on a 'header', the metal disk with 
glass-sealed wires and capped with a 'window cap' that is welded on with 
a cap welder. The cap welder is enclosed in a glove box flushed with 
nitrogen so the finished part has an inert atmosphere inside.
This is usually a manual, one-at-a-time process. TO parts are often 
subjected to leak tests, either bubble or helium. Parts are submerged in 
a heated liquid (not water) and observed for bubbles.
Parts are 'bombed' in a container pressurized with helium and then 
checked for helium leaking out using a mass spectrometer tuned for helium.

Then there's the difficult tests, some of the above done at temperature, 
like at -55C for military & space. It's a challenge to get a 1 mil spot 
of light on a detector at -55C with nothing fogging up!

There's a lot more than silicon that goes into a photodiode, the testing 
cost can be much more than the material cost.

FWIW, LED testing is much simpler.

73,
*Steve J. Noll, WA6EJO
http://www.qrz.com/db/WA6EJO
http://www.linkedin.com/in/stevejnoll
*