Photodiode bootstrap phase rolloff

Hi all,

I am currently working on a range-switched photodiode design for use in 
our laboratory [1]. It took me a while, but I've arrived at a design I'm 
fairly happy with (in no small part thanks to Phil Hobbs' helpful 
writings on the topic):

My current draft uses a simple BF862 source follower at the summing node 
of a ADA4817-based 2 MΩ/2 kΩ transimpedance amplifier to bootstrap the 
40 pF of photodiode capacitance (its output being AC-coupled into the PD 
bias node). On paper and in SPICE, I am getting just over 2 MHz 
bandwidth on the 2 MΩ range at reasonable noise levels (limited by the 
single BF862's e_n), assuming fairly realistic parasitics. On the lower 
gain range, however, I'm only predicting ~75 MHz of bandwidth, as the 
BF862 stage quickly loses steam driving the 40 pF load due to its output 
impedance.

This gets better when putting an active sink on the JFET source, but I 
thought I would spend some time optimizing the bootstrap design while 
I'm waiting for parts to arrive (~100 MHz bandwidth would be nice to 
have). It's not hard to come up with a circuit that provides >0.99 gain 
into the given load out to a few hundred MHz in simulation, for example 
by adding an RF(-ish) PNP transistor to make a complementary feedback 
pair, or simply tacking on an NPN emitter follower (bootstrapping the 
JFET drain for better effective gain as necessary).

However, while those bootstrap designs look good in isolation, it seems 
like they invariably render the complete circuit unstable on the 2 kΩ 
range in simulation (although they work brilliantly on the 2 MΩ range). 
I'll have to go through the maths properly, but this appears to be due 
to the two-pole rolloff caused by having two transistors in the path. On 
the 2 MΩ range this happens far above the op-amp loop bandwidth anyway, 
but this is no longer true on the 2 kΩ range. The bare BF862, on the 
other hand, has a lower gain to start with and rolls off earlier, but 
the single pole doesn't seem to degrade the phase margin enough to cause 
issues.

Building a sub-nV emitter follower that is good out to multiple GHz 
obviously can't be the right solution. I feel like I'm just missing an 
easy fix in the form of a lead-lag-type tweak in the right place, but I 
can't quite figure out where that would be, and much less how to make 
that work without degrading the noise performance too much.

Since my intuition for analog design is not great, I thought I'd ask 
here whether I missed something obvious. I can also post a sketch or 
LTspice file of the relevant part of the circuit if that helps, but I 
figured the question would be trivial to answer for the photodetector 
veterans among you, as presumably you would have run into this before.

Thanks!

  — David



[1] Quantum optics. The idea is to build small (well, small as far as 
lab gear goes – ~ 35mm x 30mm x 25mm) modules with built-in beam 
sampling optics for laser intensity monitoring and stabilisation all 
across the visible and near-IR. And yes, the range switching led to some 
rather expendable gymnastics, but the alternatives would have been even 
more painful.

On Friday, July 28, 2017 at 9:41:09 AM UTC-4, David Nadlinger wrote:
> Hi all,
> 
> I am currently working on a range-switched photodiode design for use in 
> our laboratory [1]. It took me a while, but I've arrived at a design I'm 
> fairly happy with (in no small part thanks to Phil Hobbs' helpful 
> writings on the topic):
> 
> My current draft uses a simple BF862 source follower at the summing node 
> of a ADA4817-based 2 MΩ/2 kΩ transimpedance amplifier to bootstrap the 
> 40 pF of photodiode capacitance (its output being AC-coupled into the PD 
> bias node). On paper and in SPICE, I am getting just over 2 MHz 
> bandwidth on the 2 MΩ range at reasonable noise levels (limited by the 
> single BF862's e_n), assuming fairly realistic parasitics. On the lower 
> gain range, however, I'm only predicting ~75 MHz of bandwidth, as the 
> BF862 stage quickly loses steam driving the 40 pF load due to its output 
> impedance.

Sounds nice.  (2MHz at 2 Meg ohm.)  I'm only a tadpole when 
it comes to PD circuits, so I can't help much.  I'm guessing 
a schematic would help those with more experience.  
How are you switching the gain?  
Maybe three orders of magnitude is too much. 
Could you use a second ADA4817 to bootstrap the PD? 
(I'm mostly a brute force type.)  

George H.  
> 
> This gets better when putting an active sink on the JFET source, but I 
> thought I would spend some time optimizing the bootstrap design while 
> I'm waiting for parts to arrive (~100 MHz bandwidth would be nice to 
> have). It's not hard to come up with a circuit that provides >0.99 gain 
> into the given load out to a few hundred MHz in simulation, for example 
> by adding an RF(-ish) PNP transistor to make a complementary feedback 
> pair, or simply tacking on an NPN emitter follower (bootstrapping the 
> JFET drain for better effective gain as necessary).
> 
> However, while those bootstrap designs look good in isolation, it seems 
> like they invariably render the complete circuit unstable on the 2 kΩ 
> range in simulation (although they work brilliantly on the 2 MΩ range). 
> I'll have to go through the maths properly, but this appears to be due 
> to the two-pole rolloff caused by having two transistors in the path. On 
> the 2 MΩ range this happens far above the op-amp loop bandwidth anyway, 
> but this is no longer true on the 2 kΩ range. The bare BF862, on the 
> other hand, has a lower gain to start with and rolls off earlier, but 
> the single pole doesn't seem to degrade the phase margin enough to cause 
> issues.
> 
> Building a sub-nV emitter follower that is good out to multiple GHz 
> obviously can't be the right solution. I feel like I'm just missing an 
> easy fix in the form of a lead-lag-type tweak in the right place, but I 
> can't quite figure out where that would be, and much less how to make 
> that work without degrading the noise performance too much.
> 
> Since my intuition for analog design is not great, I thought I'd ask 
> here whether I missed something obvious. I can also post a sketch or 
> LTspice file of the relevant part of the circuit if that helps, but I 
> figured the question would be trivial to answer for the photodetector 
> veterans among you, as presumably you would have run into this before.
> 
> Thanks!
> 
>   — David
> 
> 
> 
> [1] Quantum optics. The idea is to build small (well, small as far as 
> lab gear goes – ~ 35mm x 30mm x 25mm) modules with built-in beam 
> sampling optics for laser intensity monitoring and stabilisation all 
> across the visible and near-IR. And yes, the range switching led to some 
> rather expendable gymnastics, but the alternatives would have been even 
> more painful.

On Friday, July 28, 2017 at 11:42:28 AM UTC-4, George Herold wrote:
> On Friday, July 28, 2017 at 9:41:09 AM UTC-4, David Nadlinger wrote:
> > Hi all,
> > 
> > I am currently working on a range-switched photodiode design for use in 
> > our laboratory [1]. It took me a while, but I've arrived at a design I'm 
> > fairly happy with (in no small part thanks to Phil Hobbs' helpful 
> > writings on the topic):
> > 
> > My current draft uses a simple BF862 source follower at the summing node 
> > of a ADA4817-based 2 MΩ/2 kΩ transimpedance amplifier to bootstrap the 
> > 40 pF of photodiode capacitance (its output being AC-coupled into the PD 
> > bias node). On paper and in SPICE, I am getting just over 2 MHz 
> > bandwidth on the 2 MΩ range at reasonable noise levels (limited by the 
> > single BF862's e_n), assuming fairly realistic parasitics. On the lower 
> > gain range, however, I'm only predicting ~75 MHz of bandwidth, as the 
> > BF862 stage quickly loses steam driving the 40 pF load due to its output 
> > impedance.
> 
> Sounds nice.  (2MHz at 2 Meg ohm.)  I'm only a tadpole when 
> it comes to PD circuits, so I can't help much.  I'm guessing 
> a schematic would help those with more experience.  
> How are you switching the gain?  
> Maybe three orders of magnitude is too much. 
> Could you use a second ADA4817 to bootstrap the PD? 
> (I'm mostly a brute force type.)  
> 
> George H.  
Oh and what photodiode? How much is the reverse bias?  
Though I not looked at how much the capacitance changes,
you can reverse bias PD's much more than the the typical spec 
sheet claims.... that might gain you a little speed.  
GH   
> > 
> > This gets better when putting an active sink on the JFET source, but I 
> > thought I would spend some time optimizing the bootstrap design while 
> > I'm waiting for parts to arrive (~100 MHz bandwidth would be nice to 
> > have). It's not hard to come up with a circuit that provides >0.99 gain 
> > into the given load out to a few hundred MHz in simulation, for example 
> > by adding an RF(-ish) PNP transistor to make a complementary feedback 
> > pair, or simply tacking on an NPN emitter follower (bootstrapping the 
> > JFET drain for better effective gain as necessary).
> > 
> > However, while those bootstrap designs look good in isolation, it seems 
> > like they invariably render the complete circuit unstable on the 2 kΩ 
> > range in simulation (although they work brilliantly on the 2 MΩ range). 
> > I'll have to go through the maths properly, but this appears to be due 
> > to the two-pole rolloff caused by having two transistors in the path. On 
> > the 2 MΩ range this happens far above the op-amp loop bandwidth anyway, 
> > but this is no longer true on the 2 kΩ range. The bare BF862, on the 
> > other hand, has a lower gain to start with and rolls off earlier, but 
> > the single pole doesn't seem to degrade the phase margin enough to cause 
> > issues.
> > 
> > Building a sub-nV emitter follower that is good out to multiple GHz 
> > obviously can't be the right solution. I feel like I'm just missing an 
> > easy fix in the form of a lead-lag-type tweak in the right place, but I 
> > can't quite figure out where that would be, and much less how to make 
> > that work without degrading the noise performance too much.
> > 
> > Since my intuition for analog design is not great, I thought I'd ask 
> > here whether I missed something obvious. I can also post a sketch or 
> > LTspice file of the relevant part of the circuit if that helps, but I 
> > figured the question would be trivial to answer for the photodetector 
> > veterans among you, as presumably you would have run into this before.
> > 
> > Thanks!
> > 
> >   — David
> > 
> > 
> > 
> > [1] Quantum optics. The idea is to build small (well, small as far as 
> > lab gear goes – ~ 35mm x 30mm x 25mm) modules with built-in beam 
> > sampling optics for laser intensity monitoring and stabilisation all 
> > across the visible and near-IR. And yes, the range switching led to some 
> > rather expendable gymnastics, but the alternatives would have been even 
> > more painful.

One approach is to switch the whole front end and not just the feedback resistor. 

The BF862 is slower than the op amp. I've had good luck with a BFT92 PNP wraparound (shunt feedback) on the FET up to about 100 MHz. 

For faster stuff I like cascoding an ATF38143 pHEMT with a BFP640 SiGe:C bipolar.  As a follower you'd need to bootstrap the drain, i.e. AC couple the BJT's base to the pHEMT's source. Use a BLM18BB-series 5- or 10-ohm bead in series with the base to keep the BJT from oscillating. 

The result is a pretty good follower out to a gigahertz or so. 

Cheers

Phil Hobbs

On 28.07.17 5:17 PM, George Herold wrote:
> Oh and what photodiode?

If reality doesn't differ from the data sheets too much, it is going to 
be a Hamamatsu S6775; ~5 mm x 5 mm active area with 40 pF capacitance at 
-10 V bias.

 > How much is the reverse bias?
 > Though I not looked at how much the capacitance changes,
 > you can reverse bias PD's much more than the the typical spec
 > sheet claims.... that might gain you a little speed.

The PD package itself seems to be rather low capacitance, so I could 
indeed go to voltages higher than the 12-15 V I was going to use, and 
hence increase the "range" of the simple BF862 bootstrap. There are two 
drawbacks to this, though:

First, the dark current more than doubles going from 10 V to 30 V 
according to the datasheet (still waiting for the diodes to arrive to 
verify this). The shot noise from that still wouldn't be an issue, but 
it looks like the dark current will be the largest source of temperature 
drifts in the system.

(One important use of the photodiodes will be for intensity 
stabilisation, although the fact that there is a non-negligible amount 
of free space optics afterwards makes it hard to get better than 0.1% 
long-term stability anyway.)

Secondly, I would need an appropriate bias voltage source that can 
supply ~1 mA of current (full range at 2 kΩ would be ~2 mA, but 
hopefully no more than 0.5 mA of photocurrent will ever be used in 
practice). Since the PCB is going to be rather small (~ 30mm x 20mm), 
appropriately shielding a boost converter seems a bit nontrivial, so I 
would probably have to supply this externally.

Also, since there are going to be quite a few of the photodiodes in the 
experiment, I'd rather not push any parts past their documented limits 
without good reason, as I don't want to end up having to constantly fix 
the hardware.

Thanks for the suggestion, though!

  — David

On 28.07.17 4:42 PM, George Herold wrote:
> Sounds nice.  (2MHz at 2 Meg ohm.)

It certainly looks like it will turn out nicer than what you can buy off 
the people who bribe physicists with red snack boxes, yes. 1 MHz would 
also be acceptable for the application, so I'm quite optimistic that 
I'll be able to make it work.

> How are you switching the gain?

I'm using three semiconductor switches in a T configuration to switch 
the low-impedance feedback path in and out (the middle leg dragging the 
trace to ground when not used to reduce feedthrough). Getting the 
parasitics low enough to not degrade the 2 MΩ range was a bit of a 
challenge, but it turns out to be just about achievable without using 
(comparatively) huge relays.

> Maybe three orders of magnitude is too much.

Perhaps I should have made this clearer in the original post, but it's 
not actually the range switching that is the problem. I merely mentioned 
it to provide some background for the design choices I made, as some of 
the tradeoffs might be different if I only had a single 2 kΩ range to 
worry about.

> Could you use a second ADA4817 to bootstrap the PD?
> (I'm mostly a brute force type.)

I could, but this would increase the system noise by a factor of about 
five – the ~0.8 nV/rtHz voltage noise of the BF862 across the 40 pF 
photodiode capacitance already dominates the design, so the 4 nV/rtHz of 
the ADA4817 look quite bad. The input capacitance of the op-amp is also 
higher than that of a BF862 source follower, let alone a bootstrapped one.

  — David

On Sunday, July 30, 2017 at 10:27:04 AM UTC-4, David Nadlinger wrote:
> On 28.07.17 5:17 PM, George Herold wrote:
> > Oh and what photodiode?
> 
> If reality doesn't differ from the data sheets too much, it is going to 
> be a Hamamatsu S6775; ~5 mm x 5 mm active area with 40 pF capacitance at 
> -10 V bias.
> 
>  > How much is the reverse bias?
>  > Though I not looked at how much the capacitance changes,
>  > you can reverse bias PD's much more than the the typical spec
>  > sheet claims.... that might gain you a little speed.
> 
> The PD package itself seems to be rather low capacitance, so I could 
> indeed go to voltages higher than the 12-15 V I was going to use, and 
> hence increase the "range" of the simple BF862 bootstrap. There are two 
> drawbacks to this, though:
> 
> First, the dark current more than doubles going from 10 V to 30 V 
> according to the datasheet (still waiting for the diodes to arrive to 
> verify this). The shot noise from that still wouldn't be an issue, but 
> it looks like the dark current will be the largest source of temperature 
> drifts in the system.
OK, I've never looked at the dark current closely. 
 
I was thinking 40-50 volts, making it quite is a bit of work.  
(maybe some batteries?  I know, I don't really like batteries.)
> 
> (One important use of the photodiodes will be for intensity 
> stabilisation, although the fact that there is a non-negligible amount 
> of free space optics afterwards makes it hard to get better than 0.1% 
> long-term stability anyway.)
0.1% ! I often find some fingerprint/ dust spec, is right in the beam
making all sorts of bouncy noise.  
But don't you want to mostly take out the high frequency noise?  
(Well that's why I need a faster PD anyway.)

George H. 
> 
> Secondly, I would need an appropriate bias voltage source that can 
> supply ~1 mA of current (full range at 2 kΩ would be ~2 mA, but 
> hopefully no more than 0.5 mA of photocurrent will ever be used in 
> practice). Since the PCB is going to be rather small (~ 30mm x 20mm), 
> appropriately shielding a boost converter seems a bit nontrivial, so I 
> would probably have to supply this externally.
> 
> Also, since there are going to be quite a few of the photodiodes in the 
> experiment, I'd rather not push any parts past their documented limits 
> without good reason, as I don't want to end up having to constantly fix 
> the hardware.
> 
> Thanks for the suggestion, though!
> 
>   — David

On Sunday, July 30, 2017 at 10:44:23 AM UTC-4, David Nadlinger wrote:
> On 28.07.17 4:42 PM, George Herold wrote:
> > Sounds nice.  (2MHz at 2 Meg ohm.)
> 
> It certainly looks like it will turn out nicer than what you can buy off 
> the people who bribe physicists with red snack boxes, yes. 1 MHz would 
> also be acceptable for the application, so I'm quite optimistic that 
> I'll be able to make it work.
> 
> > How are you switching the gain?
> 
> I'm using three semiconductor switches in a T configuration to switch 
> the low-impedance feedback path in and out (the middle leg dragging the 
> trace to ground when not used to reduce feedthrough). Getting the 
> parasitics low enough to not degrade the 2 MΩ range was a bit of a 
> challenge, but it turns out to be just about achievable without using 
> (comparatively) huge relays.
> 
> > Maybe three orders of magnitude is too much.
> 
> Perhaps I should have made this clearer in the original post, but it's 
> not actually the range switching that is the problem. I merely mentioned 
> it to provide some background for the design choices I made, as some of 
> the tradeoffs might be different if I only had a single 2 kΩ range to 
> worry about.

Oh I was asking for my own edification.  I've been recently trying to 
speed up my PD and the switch/stray/ capacitance (grayhill rotary 
switch) is enough such that I need no compensating cap. at 100k ohm
with a 3dB point of ~800kHz (? data book is at work).. big PD ~130 pF @10V... 


> 
> > Could you use a second ADA4817 to bootstrap the PD?
> > (I'm mostly a brute force type.)
> 
> I could, but this would increase the system noise by a factor of about 
> five – the ~0.8 nV/rtHz voltage noise of the BF862 across the 40 pF 
> photodiode capacitance already dominates the design, so the 4 nV/rtHz of 
> the ADA4817 look quite bad. The input capacitance of the op-amp is also 
> higher than that of a BF862 source follower, let alone a bootstrapped one.

Huh, (Thanks)
I was thinking it'd only double the noise of the opamp.   I hadn't 
thought about the bootstrap cutting down the v_noise*C_in term,
but only about more speed.  

That does mean (I think) that if I bootstrap my circuit with an opamp
it'll only increase the noise a bit.  

4nV/rtHz is not too bad... (my opamp is much worse, 
8nV.. four times as bad in my thinking... noise should be 
expressed as a density. V^2/Hz and not these weird EE units. :^)

George H.

George h.  
> 
>   — David

David Nadlinger wrote...
>
> I am currently working on a range-switched photodiode design for use in 
> our laboratory [1]. It took me a while, but I've arrived at a design I'm 
> fairly happy with (in no small part thanks to Phil Hobbs' helpful 
> writings on the topic):
>
> My current draft uses a simple BF862 source follower at the summing node 
> of a ADA4817-based 2 MΩ/2 kΩ transimpedance amplifier to bootstrap the 
> 40 pF of photodiode capacitance (its output being AC-coupled into the PD 
> bias node). On paper and in SPICE, I am getting just over 2 MHz 
> bandwidth on the 2 MΩ range at reasonable noise levels (limited by the 
> single BF862's e_n), assuming fairly realistic parasitics. On the lower 
> gain range, however, I'm only predicting ~75 MHz of bandwidth, as the 
> BF862 stage quickly loses steam driving the 40 pF load due to its output 
> impedance.

 You want to add an emitter-follower to the BF862, to lower Zout
 and isolate it from the PD capacitance its driving.  You'll want
 a low e_n for that transistor.  For example, I use an FMMT718
 running at 2 mA.  I measure a 45MHz bandwidth for the bootstrap. 

 Please check in AoE III, chapter 8, for details on all of the
 above, valuable e_n measurements, and other TIA amp tricks. 

 BTW, I don't agree with Phil concerning these TIA bootstraps;
 one does fine for any gain over 0.95 or so, because you're
 reducing the PD's effective capacitance by 10 to 20x, and
 thus its en-C effects against the more noisy op-amp, which
 is already probably more than you need.


-- 
 Thanks,
    - Win

On 08/03/2017 04:00 PM, Winfield Hill wrote:
> David Nadlinger wrote...
>>
>> I am currently working on a range-switched photodiode design for use in 
>> our laboratory [1]. It took me a while, but I've arrived at a design I'm 
>> fairly happy with (in no small part thanks to Phil Hobbs' helpful 
>> writings on the topic):
>>
>> My current draft uses a simple BF862 source follower at the summing node 
>> of a ADA4817-based 2 MΩ/2 kΩ transimpedance amplifier to bootstrap the 
>> 40 pF of photodiode capacitance (its output being AC-coupled into the PD 
>> bias node). On paper and in SPICE, I am getting just over 2 MHz 
>> bandwidth on the 2 MΩ range at reasonable noise levels (limited by the 
>> single BF862's e_n), assuming fairly realistic parasitics. On the lower 
>> gain range, however, I'm only predicting ~75 MHz of bandwidth, as the 
>> BF862 stage quickly loses steam driving the 40 pF load due to its output 
>> impedance.
> 
>  You want to add an emitter-follower to the BF862, to lower Zout
>  and isolate it from the PD capacitance its driving.  You'll want
>  a low e_n for that transistor.  For example, I use an FMMT718
>  running at 2 mA.  I measure a 45MHz bandwidth for the bootstrap. 
> 
>  Please check in AoE III, chapter 8, for details on all of the
>  above, valuable e_n measurements, and other TIA amp tricks. 
> 
>  BTW, I don't agree with Phil concerning these TIA bootstraps;
>  one does fine for any gain over 0.95 or so, because you're
>  reducing the PD's effective capacitance by 10 to 20x, and
>  thus its en-C effects against the more noisy op-amp, which
>  is already probably more than you need.

It's a lot easier to get a smooth, predictable transfer function and low
phase  whoopdedoos with a super good bootstrap, though, particularly
with high capacitance.  Otherwise you have to try cancelling the
residual RC pole in the second stage, which generally works OK except
that the cancellation is never perfect and hence you get late-time
settling artifacts.

Plus I often like to use high slew (and thus noisy) amps such as the
LM6171 as the TIA because it greatly reduces artifacts due to slew
limiting.  In my designs the first couple of stages are much faster than
the overall TIA, because you never know what sort of nasty sharp pulses
the customer is going to send into the photodiode.

Cheers

Phil Hobbs

-- 
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net