Lenslet Array
The APD arrays require lenslets to be placed in front of them so that they may experience an effective fill-factor near 100% (as opposed to the straight 5-8% one gets from their sparse design).
Given that the lenslet is forced to have a 100-micron spacing (to match the APDs), and assuming we want to pack all the light into a 25-micron diameter, there is a very tight relationship between the field-of-view produced by the 4×4 array and the focal length of the lenslets. The full relation for the angular size of one pixel in the sky (σ) is: σ = as/Df, where a is the pitch of the lenslet (100 microns), s is the pupil image size on the APD, D is tha diameter of the telescope aperture, and f is the focal length of the lenslet. The commercially available lenslet from Adaptive Optics Associates (in particular, the 0100-0.5-S) is almost ideal for the task, with a 0.5-mm focal length. For a 25-micron pupil size, this translates to a pixel size of 0.29 arcseconds, for a total array size of 1.18 arcsec (small!). We could do a bit better by getting a custom 0.4-mm focal length (for $8,000 instead of $1250 for the catalog item). But the initial design should probably use the 0.5-mm lenslet. We can have this mounted on a 1-mm thick square, 1-inch across.
Mounting the lenslet is tricky. The APDs are packaged in 40-pin DIPs, sitting in a well with no built-in protection. Thus the lenslet array can act as a protective window. The bumps have to be mounted on the underside. And in fact, they need to sit slightly lower than they would if simply butted onto the DIP surface. So we need to machine off some tiny bit of material on the outside of the substrate so it'll nest properly. We should also get a very clear idea of the condition of the array as we'll receive it. The lenslet is said to be a 120×120 array, which would span 12mm square. What is the epoxy like outside of this 12mm? Is this customizable?
As for positioning, it sounds hard but shouldn't be that bad. We only have to get the angular alignment right over 4 elements: let's say 3 microns over 300 microns, or 10 milliradians or 0.6 degrees. As for translation, any offset here can be removed (to first order) by tilting the whole arrangement: a 6° tilt would move the spot by 50 microns. If we get within 10 microns, we should be able to perfectly center with very small angular adjustments. I'm thinking that an appropriate microscope setup should enable us to judge the registration rather well.