RF transistors for PD cascode

[George Herold]

I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

 

[George Herold]

On Thursday, September 12, 2019 at 12:11:27 PM UTC-4, jla...@highlandsniptechnology.com wrote:

On Thu, 12 Sep 2019 08:30:03 -0700 (PDT), George Herold
gherold@teachspin.com> wrote:

I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

Any fast cascode is going to need some keep-alive current in the
transistor. Well, unless you will always have some minimum light
level. Given some added fake-photodiode current into the cascode
transistor, and a DC coupled system, that current needs to be
subtracted out somewhere, which has its own problems.

Well some DC offset at the highest gains might be OK... maybe 100nA.
(If you care you can measure the no-light offset.)
According to Hobbs I need something that still has decent frequency response
at low current.

There's a nice graph in AoE3 fig 8.85, but no data below 100uA.

https://www.dropbox.com/s/983wckcv1o2e58l/Cascode_TIA.JPG?raw=1

You can also build a second, dummy cascode running at the same idle
current, and subtract that out later. I have a schematic for that
somewhere.

An emitter-to-emitter PNP-NPN dual cascode would be interesting too.
Tricky to bias.

I'm looking for simple.

George H.

 

On Thu, 12 Sep 2019 08:30:03 -0700 (PDT), George Herold
<gherold@teachspin.com> wrote:

I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

Any fast cascode is going to need some keep-alive current in the
transistor. Well, unless you will always have some minimum light
level. Given some added fake-photodiode current into the cascode
transistor, and a DC coupled system, that current needs to be
subtracted out somewhere, which has its own problems.

https://www.dropbox.com/s/983wckcv1o2e58l/Cascode_TIA.JPG?raw=1

You can also build a second, dummy cascode running at the same idle
current, and subtract that out later. I have a schematic for that
somewhere.

An emitter-to-emitter PNP-NPN dual cascode would be interesting too.
Tricky to bias.

 

[George Herold]

On Thursday, September 12, 2019 at 11:30:08 AM UTC-4, George Herold wrote:

I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

Hmm all the Jfets are going away too. Is it silly to think of making a
bootstrap with an opamp.

GH

 

On Thu, 12 Sep 2019 08:54:40 -0700 (PDT), George Herold
<gherold@teachspin.com> wrote:

On Thursday, September 12, 2019 at 11:30:08 AM UTC-4, George Herold wrote:
I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

Hmm all the Jfets are going away too. Is it silly to think of making a
bootstrap with an opamp.

GH

Phil says to not do that: too noisy. But OK if a little extra noise
doesn't matter.

There are still some fast NPNs and some JFETS around.

Could you bootstrap from the output of the TIA?

 

[Winfield Hill]

George Herold wrote...

On September 12, 2019, John Larkin wrote:
On 12 Sep 2019, George Herold wrote:

... playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.

(This is discussed in detail in Phil's book ..

Any fast cascode is going to need some keep-alive current in the
transistor. Well, unless you will always have some minimum light
level. Given some added fake-photodiode current into the cascode
transistor, and a DC coupled system, that current needs to be
subtracted out somewhere, which has its own problems.

We also discuss it in section 8.11.10, along with
the neccesity of maintaining sufficient transistor
fT, which is why you have to add current.

> There's a nice graph in AoE3 fig 8.85, but no data below 100uA.

At low currents you can extrapolate down to below 10uA,
by eyeballing the constant C-pi curves. The graphs,
Table 8.1, and Figure 8.39, can help you pick a good
low-capacitance transistor for use at low currents.
x-Chapters section 2x.11 has more detail and graphs.
https://www.dropbox.com/s/br9caixbtderl44/2x.9-11_BJT_beta_fT_DRAFT.pdf?dl=1


--
Thanks,
- Win

 

[Winfield Hill]

George Herold wrote...

Yeah well this is replacing a 'standard' TIA, which
is noisy as hell.

Did you try boostrapping the PD?


--
Thanks,
- Win

 

[George Herold]

On Thursday, September 12, 2019 at 12:44:39 PM UTC-4, jla...@highlandsniptechnology.com wrote:

On Thu, 12 Sep 2019 08:54:40 -0700 (PDT), George Herold
gherold@teachspin.com> wrote:

On Thursday, September 12, 2019 at 11:30:08 AM UTC-4, George Herold wrote:
I think you guys have talked about this... but I wasn't paying attention.
I started playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.
There is the BFT25A... listed 'as not for new designs' on the DK
website. What else should I order?

George H.
(This is discussed in detail in Phil's book, I've read that section a
few times, but it's only now that I'm actually building something that
I have the 'right eyes' for reading.)

Hmm all the Jfets are going away too. Is it silly to think of making a
bootstrap with an opamp.

GH

Phil says to not do that: too noisy. But OK if a little extra noise
doesn't matter.
Yeah well this is replacing a 'standard' TIA, which is noisy as hell.

There are still some fast NPNs and some JFETS around.
Name one or two please.

Could you bootstrap from the output of the TIA?

Hmm just invert the output. The bootstrap would only be as fast as the TIA...
but that might not matter. I'll try it and see.

George H.

 

[George Herold]

On Thursday, September 12, 2019 at 1:33:29 PM UTC-4, Winfield Hill wrote:

George Herold wrote...

Yeah well this is replacing a 'standard' TIA, which
is noisy as hell.

Did you try boostrapping the PD?

Well just now, (an opamp bootstrap) It does help a lot at low light level.
(The speed is only a bit better? ~300ns rise time to ~200ns)

GH

--
Thanks,
- Win

 

[Winfield Hill]

George Herold wrote...

On September 12, 2019, Winfield Hill wrote:
George Herold wrote...

Yeah well this is replacing a 'standard' TIA,
which is noisy as hell.

Did you try boostrapping the PD?

Well just now, (an opamp bootstrap) ...

Op-amp arggh, you need sub nV, you need a
BF862 or a CPH3910. Would you like me to
send you some BF862 JFETs?


--
Thanks,
- Win

 

[George Herold]

On Thursday, September 12, 2019 at 1:34:08 PM UTC-4, Winfield Hill wrote:

George Herold wrote...

On September 12, 2019, John Larkin wrote:
On 12 Sep 2019, George Herold wrote:

... playing with Phil's cascode input for TIA photodiode.
I'm using a 2N3904 which is poor w/o the bias resistor*.

(This is discussed in detail in Phil's book ..

Any fast cascode is going to need some keep-alive current in the
transistor. Well, unless you will always have some minimum light
level. Given some added fake-photodiode current into the cascode
transistor, and a DC coupled system, that current needs to be
subtracted out somewhere, which has its own problems.

We also discuss it in section 8.11.10, along with
the neccesity of maintaining sufficient transistor
fT, which is why you have to add current.
Right, that's very nice too.

There's a nice graph in AoE3 fig 8.85, but no data below 100uA.

At low currents you can extrapolate down to below 10uA,
by eyeballing the constant C-pi curves. The graphs,

OK I guess I didn't read the text. :^)

Table 8.1, and Figure 8.39, can help you pick a good
low-capacitance transistor for use at low currents.
x-Chapters section 2x.11 has more detail and graphs.
https://www.dropbox.com/s/br9caixbtderl44/2x.9-11_BJT_beta_fT_DRAFT.pdf?dl=1

Thanks,
George H.

--
Thanks,
- Win

 

[George Herold]

On Thursday, September 12, 2019 at 2:19:52 PM UTC-4, Winfield Hill wrote:

George Herold wrote...

On September 12, 2019, Winfield Hill wrote:
George Herold wrote...

Yeah well this is replacing a 'standard' TIA,
which is noisy as hell.

Did you try boostrapping the PD?

Well just now, (an opamp bootstrap) ...

Op-amp arggh, you need sub nV, you need a
BF862 or a CPH3910. Would you like me to
send you some BF862 JFETs?

No, but thanks. (I can't use an obsolete part.)
I added the cph3910 to my DK order.

The opamp is a opa2192 ~5nV/rtHz of noise. With the cascode thing in place
there is much less noise gain from the TIA amp, right? And then the noise
will mostly dominated by the feedback R. Well I haven't sat dwon and tried to
calculate the noise. (I also have to clean up the bias supply before trying to
measure it.)

George H.

--
Thanks,
- Win

 

[Winfield Hill]

George Herold wrote...

The opamp is a opa2192 ~5nV/rtHz of noise. ...

George, tidbits of info without full information
are useless for giving advice. In addition to
the opamp type, we need to know Rf, Cd and Cin,
and your desired bandwidth. If you study section
8.11, you'll see all these parameters put to use
in analyzing the TIA's performance.


--
Thanks,
- Win

 

[Winfield Hill]

George Herold wrote...

I think you guys have talked about this...
but I wasn't paying attention.
I started playing with Phil's cascode input
for TIA photodiode.

One point should be made, when looking for
ways to reduce high-frequency e_n Cin noise
in TIA, the cascode isolation approach works
well only at higher input current levels,
where you can maintain the cascode gain of
the transistor (its fT), perhaps by adding
additional bias current. But at very low
currents, the bootstrap approach can work
better. See AoE III, 8.11, pages 537-555.

Compensation of TIAs wasn't discussed much
in 8.11 (dealt with in 4.3), so we wanted to
say more, hence new x-Chapter section 4x.3

https://www.dropbox.com/s/11wp4nlumil1koq/4x.3_TIA_final-DRAFT.pdf?dl=1

Beside discussing compensation, we show a
few useful tricks. And 4x.3.7 introduces
an amazing TIA configuration, which covers
a dynamic signal range of 10^7 or more, with
a single TIA op-amp. During discussions
here on s.e.d., two other approaches came up.
See 4x.3.9 using a bootstrap, including a
1-meg TIA with 1MHz bandwidth. And 4x.3.10
with a cascode BJT. There we see again the
struggle to maintain wide amplifier bandwidth
when using a cascode at currents below 1 to
10uA, also see fT discussion in 2x.9 to 2x.11

https://www.dropbox.com/s/br9caixbtderl44/2x.9-11_BJT_beta_fT_DRAFT.pdf?dl=1


--
Thanks,
- Win

 

[George Herold]

On Thursday, September 12, 2019 at 8:57:57 PM UTC-4, Winfield Hill wrote:

George Herold wrote...

The opamp is a opa2192 ~5nV/rtHz of noise. ...

George, tidbits of info without full information
are useless for giving advice. In addition to
the opamp type, we need to know Rf, Cd and Cin,
and your desired bandwidth. If you study section
8.11, you'll see all these parameters put to use
in analyzing the TIA's performance.


--
Thanks,
- Win

Hi Win, yeah I'm redesigning this PD TIA I built years ago...
before I read Hobbs.
It's a big PIN-44*
The PD preamp has range scales from 332 ohms to 10M ohms*.
I've used all those scales.. 332 ohms the least.
Most of the applications I'm thinking of have some small
(1-10%) variation in the light intensity. (good interferometers,
are the exception I think of first.)
And there's certainly a sweet spot for 'as fast as I can'
at ~1uA. We've got an optical pumping, and with a fast enough
PD you can watch the electron spins go around in the earth's field.
And then there's laser locking. Faster and less noise should be
a win everywhere. At zero light there's still leakage in the biased PD.
(I've got to measure that still.)

George H.

*I've got nothing against reducing the PD area to meet some
frequency..
*the grayhill rotary switch puts ~5pf (I measured
something around there.) across any feedback R.
Which kills all the speed I care about. If I make some
bushing to float the switch from ground. Can I make a driven
switch? I'm driving the switch body, and no part of the active
circuit.
(I'm not at all sure that it's C to ground in the
rotary switch that is the problem.. but it seems the obvious
first thing to try.)

 

[George Herold]

On Friday, September 13, 2019 at 9:39:39 AM UTC-4, Winfield Hill wrote:

George Herold wrote...

I think you guys have talked about this...
but I wasn't paying attention.
I started playing with Phil's cascode input
for TIA photodiode.

One point should be made, when looking for
ways to reduce high-frequency e_n Cin noise
in TIA, the cascode isolation approach works
well only at higher input current levels,
where you can maintain the cascode gain of
the transistor (its fT), perhaps by adding
additional bias current. But at very low
currents, the bootstrap approach can work
better. See AoE III, 8.11, pages 537-555.

Compensation of TIAs wasn't discussed much
in 8.11 (dealt with in 4.3), so we wanted to
say more, hence new x-Chapter section 4x.3

https://www.dropbox.com/s/11wp4nlumil1koq/4x.3_TIA_final-DRAFT.pdf?dl=1

Beside discussing compensation, we show a
few useful tricks. And 4x.3.7 introduces
an amazing TIA configuration, which covers
a dynamic signal range of 10^7 or more, with
a single TIA op-amp. During discussions
here on s.e.d., two other approaches came up.
See 4x.3.9 using a bootstrap, including a
1-meg TIA with 1MHz bandwidth. And 4x.3.10
with a cascode BJT. There we see again the
struggle to maintain wide amplifier bandwidth
when using a cascode at currents below 1 to
10uA, also see fT discussion in 2x.9 to 2x.11

https://www.dropbox.com/s/br9caixbtderl44/2x.9-11_BJT_beta_fT_DRAFT.pdf?dl=1


--
Thanks,
- Win

Awesome! (I haven't read or digested much it yet.)
On the low noise end, I've been thinking about how to do a
non-Poisson-ian light source. Bunched states. It's not easy,
I was looking at a diode laser near threshold... but I think I'm
seeing mostly mode hopping 'noise'.

George H.

 

[Winfield Hill]

George Herold wrote...

On Thursday, September 12, 2019, Winfield Hill wrote:
George Herold wrote...

The opamp is a opa2192 ~5nV/rtHz of noise. ...

George, tidbits of info without full information
are useless for giving advice. In addition to
the opamp type, we need to know Rf, Cd and Cin,
and your desired bandwidth. If you study section
8.11, you'll see all these parameters put to use
in analyzing the TIA's performance.

Hi Win, yeah I'm redesigning this PD TIA I built years
ago... before I read Hobbs. It's a big PIN-44*
The PD preamp has range scales from 332 ohms to 10M ohms*.
I've used all those scales.. 332 ohms the least.
Most of the applications I'm thinking of have some small
(1-10%) variation in the light intensity (good
interferometers, are the exception I think of first).

And there's certainly a sweet spot for 'as fast as I can'
at ~1uA. We've got an optical pumping, and with a fast
enough PD you can watch the electron spins go around in
the earth's field. And then there's laser locking.
Faster and less noise should be a win everywhere.
At zero light there's still leakage in the biased PD.
(I've got to measure that still.)

Datasheet says 1nA typical dark current at -10V.

OK, in another post I spelled out the reasons you need
to use JFET bootstrap below 1uA. But somewhere above
the 1uA to 10uA point, a cascode + bootstrap, Figure
4x.40, might give lower high-frequency en-Cin noise.

If you add CMOS gain switching, detailed below, it'd
be reasonable to simply bypass a slow cascode on the
1, 10M ranges, keep high bandwidth for 1nA signals.

* I've got nothing against reducing the PD area
to meet some frequency..

You should be able to keep the large sensor, which
has more collecting area for low light levels.

*the grayhill rotary switch puts ~5pf (I measured
something around there.) across any feedback R.

Use the CMOS switch scheme in Figure 4x.29.B, to
reduce the Cf contribution from the switch to zero.
It does add a small 5pF contribution to Cin. For
more than two gain settings, add any additional
switches on the far side of the first CMOS switch.
After several ranges, with low enough Rf, convert
back to a rotary panel switch if you want.

Yes, all this goings on can add lots of parts, but
they're small, cheap and no issue for automated
SMT assembly.


--
Thanks,
- Win

 

[George Herold]

On Sunday, September 15, 2019 at 11:54:01 AM UTC-4, Winfield Hill wrote:

George Herold wrote...

On Thursday, September 12, 2019, Winfield Hill wrote:
George Herold wrote...

The opamp is a opa2192 ~5nV/rtHz of noise. ...

George, tidbits of info without full information
are useless for giving advice. In addition to
the opamp type, we need to know Rf, Cd and Cin,
and your desired bandwidth. If you study section
8.11, you'll see all these parameters put to use
in analyzing the TIA's performance.

Hi Win, yeah I'm redesigning this PD TIA I built years
ago... before I read Hobbs. It's a big PIN-44*
The PD preamp has range scales from 332 ohms to 10M ohms*.
I've used all those scales.. 332 ohms the least.
Most of the applications I'm thinking of have some small
(1-10%) variation in the light intensity (good
interferometers, are the exception I think of first).

And there's certainly a sweet spot for 'as fast as I can'
at ~1uA. We've got an optical pumping, and with a fast
enough PD you can watch the electron spins go around in
the earth's field. And then there's laser locking.
Faster and less noise should be a win everywhere.
At zero light there's still leakage in the biased PD.
(I've got to measure that still.)

Datasheet says 1nA typical dark current at -10V.

OK, in another post I spelled out the reasons you need
to use JFET bootstrap below 1uA. But somewhere above
the 1uA to 10uA point, a cascode + bootstrap, Figure
4x.40, might give lower high-frequency en-Cin noise.

If you add CMOS gain switching, detailed below, it'd
be reasonable to simply bypass a slow cascode on the
1, 10M ranges, keep high bandwidth for 1nA signals.

* I've got nothing against reducing the PD area
to meet some frequency..

You should be able to keep the large sensor, which
has more collecting area for low light levels.

*the grayhill rotary switch puts ~5pf (I measured
something around there.) across any feedback R.

Use the CMOS switch scheme in Figure 4x.29.B, to
reduce the Cf contribution from the switch to zero.
It does add a small 5pF contribution to Cin. For
more than two gain settings, add any additional
switches on the far side of the first CMOS switch.
After several ranges, with low enough Rf, convert
back to a rotary panel switch if you want.

Thanks, I was looking at those. So at the moment, I was planning on
keeping the from factor. Here's a pic
https://www.dropbox.com/s/7cw4moh9t4kkqpi/PD-pic.JPG?dl=0

I'm stuck with the big housing. I guess I could think about changing
the panel... but I was hoping to work with the existing rotary switch and
panel.

George H.

Yes, all this goings on can add lots of parts, but
they're small, cheap and no issue for automated
SMT assembly.


--
Thanks,
- Win

 

[Winfield Hill]

George Herold wrote...

I'm stuck with the big housing. I guess I could think
about changing the panel... but I was hoping to work
with the existing rotary switch and panel.

That's a huge box, much bigger than the one I use in
the RIS-617 design, which has only 3 gain positions.

Yes, all this goings on can add lots of parts, but
they're small, cheap and no issue for automated
SMT assembly.

The CMOS-switch trick to extend bandwidth takes one
sot-23-6 per gain range. To use your rotary switch,
add one wafer. Its pins will drive an 0603 resistor
and half SC70 dual diode, per gain range. Once you
get down to 100k or 332k, go back to your old switch
scheme. Tiny SMT parts should fit in that big box.


--
Thanks,
- Win