Interesting.
I recently ran across, on LinkedIn, an outfit making
"photo-pile" cells for delivering power across fiber optic
cable from a laser diode source. They demonstrated
pretty high efficiency (at least, from optical power
in to voltaic power out - laser diode efficiency is
Somebody Else's Problem) and a range of voltages
by series-connecting cells. They used a GaAs photo-
diode which gives them a higher output voltage per
cell than silicon.
Now two things occur to me here. One is, stacking
of photodiodes. Not, perhaps, practical in a junction
isolated (cheap) technology but not, depending on
details of well depth vs absorption length, out of the
question. Entirely practical in a SOI technology (like
TowerJazz CA18HB, a 1.8V/5V PDSOI with a ~1.5um
film thickness, - and this is only one example). You
could stack (electrically) any number of diodes you
like, provided you can optically couple to them all.
OnSemi has a neat backside-illuminated imager flow
that they run at Gresham, a 110-nm (if I recall)
CMOS technology that they thin post-fab, used for
visible-range star tracker applications on satellites
(and perhaps other platforms). Again it's an SOI
but now without a "handle" , very thin, and the
ability to throw the light in from the back gives you
a free hand with frontside interconnect, zero concern
for shadowing and so on.
The other branch of my thought is, using (as I mentioned
in the beginning) a GaAs photodiode would give you from
a single cell, a higher voltage and likely a higher efficiency
(hence higher available power) from the cell(s). If the goal
of this project includes that the power receiver be
homogeneously (silicon) integrated, perhaps that precludes
such options. But if heterogeneous (e.g. 2.5D chip stacking)
is acceptable maybe you can improve the power picture
(and even do it with zero chip area impact - receiver sits
atop the silicon chip, maybe within the I/O pad ring, as
opposed to consuming mucho silicon area for the optical
input region). There exist commercial GaAs PIN diodes that
are optimized for fiber optic receiver use - and some with
"multi-lane" (like 4x1) fiber cables, which if the connections
were not common-anode might then be "stackable" too -
point being this is commercial, cost-squeezed stuff that may
be already flip-chip compatible, cheap, multi-vendor-sourced
and available to you.