Aliexpress solar cells as photodiodes--not

[Phil Hobbs]

So I have this project looking to measure babies' blood oxygenation
noninvasively, i.e. using an optical sensor looking through the mom's
abdomen.

The idea is to make the business end cheap--ideally disposable. Come
with me, if you will, on a trip down memory alley.

Circa 1992, my friend and colleague Ted van Kessel and I did an
interesting semiconductor process control instrument for DRAM fab at
IBM, Burlington VT. (This was back in the 0.5-micron days, when optical
inspection was competitive.)

At the time, photoresist was generally acid catalyzed, i.e. it developed
something like photographic film. The litho tool (wafer stepper)
exposed the resist, liberating a bit of acid. Then the wafer went onto
a hot plate so that the acid could act like developer, breaking a bunch
more bonds and rendering the image developable.

The resulting line width depended on both the exposure dose and the
temperature/duration of the bake step. So Ted and I built this gizmo to
look at the diffraction pattern of the latent image as it developed on
the hot plate, and lift the wafer off it when the diffracted beam
strength was just right. That way we had a closed-loop method for
controlling line width in litho, shazam. (Turned out the fab folks
didn't want it, but I digress.)

IBM's DRAM cells were arranged in a hexagonal pattern, so when you
shined a LED vertically down on the wafer, you got a
hexagonally-symmetric optical diffraction pattern from the latent image
in the resist, with some contribution from the lower layers (previously
fabricated). Ordinarily you'd only need one diffracted order for a
measurement like that, but to correct for diffraction from the
underlying structure we needed clean +-1 orders in at least one of the
three symmetry axes of the hexagonal pattern. Unfortunately, there was
no way to control the orientation of the wafer on the hot plate, because
previously there was no reason to care about it, so the diffraction
orders could be anywhere in azimuth.

We wound up with seven 1x3-inch solar cells arranged like a 360-degree
poker hand around the vertical axis (i.e. with a bit of a taper in the
direction away from the wafer). With sevenfold symmetry, regardless of
how the wafer was oriented, we got clean measurements of at least one
+-1 order pair.

Those cells worked fine up to about 20 kHz, running into a
common-emitter stage followed by a regular op amp TIA. All the cathodes
were connected to the summing junction, and the anodes were multiplexed
to ground using open-drain outputs of a zero-power PAL (PALCE16V8Z).
(Zero-power PALs didn't push power supply noise out their outputs when
in open-drain mode.) So probably 20 nF or so.

Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go. I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.

Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?

Cheers

Phil Hobb

[Rhydian]

Post by Phil Hobbs
So I have this project looking to measure babies' blood oxygenation
noninvasively, i.e. using an optical sensor looking through the mom's
abdomen.
The idea is to make the business end cheap--ideally disposable. Come
with me, if you will, on a trip down memory alley.
Circa 1992, my friend and colleague Ted van Kessel and I did an
interesting semiconductor process control instrument for DRAM fab at
IBM, Burlington VT. (This was back in the 0.5-micron days, when optical
inspection was competitive.)
At the time, photoresist was generally acid catalyzed, i.e. it developed
something like photographic film. The litho tool (wafer stepper)
exposed the resist, liberating a bit of acid. Then the wafer went onto
a hot plate so that the acid could act like developer, breaking a bunch
more bonds and rendering the image developable.
The resulting line width depended on both the exposure dose and the
temperature/duration of the bake step. So Ted and I built this gizmo to
look at the diffraction pattern of the latent image as it developed on
the hot plate, and lift the wafer off it when the diffracted beam
strength was just right. That way we had a closed-loop method for
controlling line width in litho, shazam. (Turned out the fab folks
didn't want it, but I digress.)
IBM's DRAM cells were arranged in a hexagonal pattern, so when you
shined a LED vertically down on the wafer, you got a
hexagonally-symmetric optical diffraction pattern from the latent image
in the resist, with some contribution from the lower layers (previously
fabricated). Ordinarily you'd only need one diffracted order for a
measurement like that, but to correct for diffraction from the
underlying structure we needed clean +-1 orders in at least one of the
three symmetry axes of the hexagonal pattern. Unfortunately, there was
no way to control the orientation of the wafer on the hot plate, because
previously there was no reason to care about it, so the diffraction
orders could be anywhere in azimuth.
We wound up with seven 1x3-inch solar cells arranged like a 360-degree
poker hand around the vertical axis (i.e. with a bit of a taper in the
direction away from the wafer). With sevenfold symmetry, regardless of
how the wafer was oriented, we got clean measurements of at least one
+-1 order pair.
Those cells worked fine up to about 20 kHz, running into a
common-emitter stage followed by a regular op amp TIA. All the cathodes
were connected to the summing junction, and the anodes were multiplexed
to ground using open-drain outputs of a zero-power PAL (PALCE16V8Z).
(Zero-power PALs didn't push power supply noise out their outputs when
in open-drain mode.) So probably 20 nF or so.
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go. I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?
Cheers
Phil Hobbs

So 10 uF per square metre. That sounds like a lot, until I worked it out
for the Osram SFH2700FA I'm using for a new design. 0.59 x 0.59 mm and
4.6 pF, which works out to 13 uF per square metre. It's a PIN device,
which is supposed to reduce the capacitance. Or am I missing something?

I like the Osram parts so far, but it's not a very demanding application
bandwidth wise. It _is_ very space constrained, hence the small area.
It's the best of a similar bunch I tested for sensitivity.

[Phil Hobbs]

Post by Rhydian

Post by Phil Hobbs
So I have this project looking to measure babies' blood oxygenation
noninvasively, i.e. using an optical sensor looking through the mom's
abdomen.
The idea is to make the business end cheap--ideally disposable. Come
with me, if you will, on a trip down memory alley.
Circa 1992, my friend and colleague Ted van Kessel and I did an
interesting semiconductor process control instrument for DRAM fab at
IBM, Burlington VT. (This was back in the 0.5-micron days, when optical
inspection was competitive.)
At the time, photoresist was generally acid catalyzed, i.e. it developed
something like photographic film. The litho tool (wafer stepper)
exposed the resist, liberating a bit of acid. Then the wafer went onto
a hot plate so that the acid could act like developer, breaking a bunch
more bonds and rendering the image developable.
The resulting line width depended on both the exposure dose and the
temperature/duration of the bake step. So Ted and I built this gizmo to
look at the diffraction pattern of the latent image as it developed on
the hot plate, and lift the wafer off it when the diffracted beam
strength was just right. That way we had a closed-loop method for
controlling line width in litho, shazam. (Turned out the fab folks
didn't want it, but I digress.)
IBM's DRAM cells were arranged in a hexagonal pattern, so when you
shined a LED vertically down on the wafer, you got a
hexagonally-symmetric optical diffraction pattern from the latent image
in the resist, with some contribution from the lower layers (previously
fabricated). Ordinarily you'd only need one diffracted order for a
measurement like that, but to correct for diffraction from the
underlying structure we needed clean +-1 orders in at least one of the
three symmetry axes of the hexagonal pattern. Unfortunately, there was
no way to control the orientation of the wafer on the hot plate, because
previously there was no reason to care about it, so the diffraction
orders could be anywhere in azimuth.
We wound up with seven 1x3-inch solar cells arranged like a 360-degree
poker hand around the vertical axis (i.e. with a bit of a taper in the
direction away from the wafer). With sevenfold symmetry, regardless of
how the wafer was oriented, we got clean measurements of at least one
+-1 order pair.
Those cells worked fine up to about 20 kHz, running into a
common-emitter stage followed by a regular op amp TIA. All the cathodes
were connected to the summing junction, and the anodes were multiplexed
to ground using open-drain outputs of a zero-power PAL (PALCE16V8Z).
(Zero-power PALs didn't push power supply noise out their outputs when
in open-drain mode.) So probably 20 nF or so.
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go. I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?

So 10 uF per square metre.

I misspoke--mine are 30x50 _millimetres_. So it's more like 1
millifarad per square metre, or 100 nf/cm**2. Really good PIN
photodiodes run about 40-100 pf/cm**2 when fully depleted, about 5-7x
that at zero bias.

I'm out of the lab today, but I'll try resonating the capacitance and
see what kind of Q I get. I need to work around 220 Hz, which is far
enough from harmonics of both 50 and 60 Hz for my purposes--at 1.5 uF,
that needs a 300-mH inductor.

There's such a thing as a parametric gyrator, so it might even be
possible to use a Y5V cap and some magic to make a sufficiently-quiet
simulated inductor.

(I'll get a nanoamp of photocurrent if I'm lucky, so the Q has to be
high or I'm better off with a smaller detector.)

I don't really think that the nasty $3 AliExpress gizmo is the right
answer for the actual measurement, but it's worth checking out the
parameter space.

That sounds like a lot, until I worked it out

Post by Rhydian
for the Osram SFH2700FA I'm using for a new design. 0.59 x 0.59 mm and
4.6 pF, which works out to 13 uF per square metre. It's a PIN device,
which is supposed to reduce the capacitance. Or am I missing something?
I like the Osram parts so far, but it's not a very demanding application
bandwidth wise. It _is_ very space constrained, hence the small area.
It's the best of a similar bunch I tested for sensitivity.

Cheers

Phil Hobbs

[Brian Howie]

Post by Phil Hobbs
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?

I don't suppose reverse biasing will help much ? How about bootstrapping ?

Brian

[Phil Hobbs]

Post by Brian Howie

Post by Phil Hobbs
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?

I don't suppose reverse biasing will help much ? How about
bootstrapping ?

Thanks.

At very low photocurrents, reverse bias is not always a win even with
PIN photodiodes. You reduce the e_N*C noise by a factor of 5 or
thereabouts, but if the leakage current is 100x the photocurrent, that
winds up being worse.

For amorphous thin-film solar cells, you'd expect high doping levels to
reduce series resistance, which means that reverse bias won't reduce the
capacitance much. (The 'I' in 'PIN' refers to a thick layer of very
low-doped (intrinsic) silicon that can be fully depleted at a reasonable
voltage.)

Bootstraps and TIAs both exhibit the e_N * C mechanism, but a good JFET
(or pHEMT, at least above 1 MHz) is a lot quieter and has lower input
current and capacitance than an op amp.

Cheers

Phil Hobbs

[John Larkin]

On Tue, 18 May 2021 21:13:25 -0400, Phil Hobbs

Post by Phil Hobbs
So I have this project looking to measure babies' blood oxygenation
noninvasively, i.e. using an optical sensor looking through the mom's
abdomen.
The idea is to make the business end cheap--ideally disposable. Come
with me, if you will, on a trip down memory alley.
Circa 1992, my friend and colleague Ted van Kessel and I did an
interesting semiconductor process control instrument for DRAM fab at
IBM, Burlington VT. (This was back in the 0.5-micron days, when optical
inspection was competitive.)
At the time, photoresist was generally acid catalyzed, i.e. it developed
something like photographic film. The litho tool (wafer stepper)
exposed the resist, liberating a bit of acid. Then the wafer went onto
a hot plate so that the acid could act like developer, breaking a bunch
more bonds and rendering the image developable.
The resulting line width depended on both the exposure dose and the
temperature/duration of the bake step. So Ted and I built this gizmo to
look at the diffraction pattern of the latent image as it developed on
the hot plate, and lift the wafer off it when the diffracted beam
strength was just right. That way we had a closed-loop method for
controlling line width in litho, shazam. (Turned out the fab folks
didn't want it, but I digress.)
IBM's DRAM cells were arranged in a hexagonal pattern, so when you
shined a LED vertically down on the wafer, you got a
hexagonally-symmetric optical diffraction pattern from the latent image
in the resist, with some contribution from the lower layers (previously
fabricated). Ordinarily you'd only need one diffracted order for a
measurement like that, but to correct for diffraction from the
underlying structure we needed clean +-1 orders in at least one of the
three symmetry axes of the hexagonal pattern. Unfortunately, there was
no way to control the orientation of the wafer on the hot plate, because
previously there was no reason to care about it, so the diffraction
orders could be anywhere in azimuth.
We wound up with seven 1x3-inch solar cells arranged like a 360-degree
poker hand around the vertical axis (i.e. with a bit of a taper in the
direction away from the wafer). With sevenfold symmetry, regardless of
how the wafer was oriented, we got clean measurements of at least one
+-1 order pair.
Those cells worked fine up to about 20 kHz, running into a
common-emitter stage followed by a regular op amp TIA. All the cathodes
were connected to the summing junction, and the anodes were multiplexed
to ground using open-drain outputs of a zero-power PAL (PALCE16V8Z).
(Zero-power PALs didn't push power supply noise out their outputs when
in open-drain mode.) So probably 20 nF or so.
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go. I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?
Cheers
Phil Hobb

Lots of big ICs have on-chip bypass caps. I measured one Xilinx FPGA
that had hundreds of nF caps on various supply rails. So I don't worry
about bypassing much.

I guess you can make a lot of c when the dielectric is nanometers
thick.

[Phil Hobbs]

Post by John Larkin
On Tue, 18 May 2021 21:13:25 -0400, Phil Hobbs

Post by Phil Hobbs
So I have this project looking to measure babies' blood oxygenation
noninvasively, i.e. using an optical sensor looking through the mom's
abdomen.
The idea is to make the business end cheap--ideally disposable. Come
with me, if you will, on a trip down memory alley.
Circa 1992, my friend and colleague Ted van Kessel and I did an
interesting semiconductor process control instrument for DRAM fab at
IBM, Burlington VT. (This was back in the 0.5-micron days, when optical
inspection was competitive.)
At the time, photoresist was generally acid catalyzed, i.e. it developed
something like photographic film. The litho tool (wafer stepper)
exposed the resist, liberating a bit of acid. Then the wafer went onto
a hot plate so that the acid could act like developer, breaking a bunch
more bonds and rendering the image developable.
The resulting line width depended on both the exposure dose and the
temperature/duration of the bake step. So Ted and I built this gizmo to
look at the diffraction pattern of the latent image as it developed on
the hot plate, and lift the wafer off it when the diffracted beam
strength was just right. That way we had a closed-loop method for
controlling line width in litho, shazam. (Turned out the fab folks
didn't want it, but I digress.)
IBM's DRAM cells were arranged in a hexagonal pattern, so when you
shined a LED vertically down on the wafer, you got a
hexagonally-symmetric optical diffraction pattern from the latent image
in the resist, with some contribution from the lower layers (previously
fabricated). Ordinarily you'd only need one diffracted order for a
measurement like that, but to correct for diffraction from the
underlying structure we needed clean +-1 orders in at least one of the
three symmetry axes of the hexagonal pattern. Unfortunately, there was
no way to control the orientation of the wafer on the hot plate, because
previously there was no reason to care about it, so the diffraction
orders could be anywhere in azimuth.
We wound up with seven 1x3-inch solar cells arranged like a 360-degree
poker hand around the vertical axis (i.e. with a bit of a taper in the
direction away from the wafer). With sevenfold symmetry, regardless of
how the wafer was oriented, we got clean measurements of at least one
+-1 order pair.
Those cells worked fine up to about 20 kHz, running into a
common-emitter stage followed by a regular op amp TIA. All the cathodes
were connected to the summing junction, and the anodes were multiplexed
to ground using open-drain outputs of a zero-power PAL (PALCE16V8Z).
(Zero-power PALs didn't push power supply noise out their outputs when
in open-drain mode.) So probably 20 nF or so.
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go. I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.
Turns out that they have about 1.5_MICROFARAD_ shunt capacitance.
Where's Radio Shack when you need them?

Lots of big ICs have on-chip bypass caps. I measured one Xilinx FPGA
that had hundreds of nF caps on various supply rails. So I don't worry
about bypassing much.
I guess you can make a lot of c when the dielectric is nanometers
thick.

And has a huge epsilon as well.

Cheers

Phil "bobbing for photons" Hobbs

[Chris Jones]

Post by Phil Hobbs
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go.  I got some 30x50 cm ones
from AliExpress, which looked OK, and in fact they work fine for their
advertised use.

Why did you pick amorphous cells rather than monocrystalline silicon?

[Phil Hobbs]

Post by Chris Jones

Post by Phil Hobbs
Coming back to the fetal pulse ox gizmo, I thought it would be fun to
see how fast a modern amorphous cell could go.  I got some 30x50 cm
ones from AliExpress, which looked OK, and in fact they work fine for
their advertised use.

Why did you pick amorphous cells rather than monocrystalline silicon?

Because they're super cheap. I just tested a 25x8 mm amorphous cell
with a glass substrate (harvested from one of those solar-powered hula
dancer figurines) and it measured 5 meg, so it ain't necessarily impossible.

Cheers

Phil Hobbs