SiGe:C transistors and Early effect...

[Phil Hobbs]

So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well. If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
<https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz). New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided). Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still. (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage. The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign! In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part. It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.

Fun.

Cheers

Phil Hobbs

BTW: SFH2400 photodiodes stink.


--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[server]

On Wed, 21 Oct 2020 22:42:03 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well. If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz). New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided). Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still. (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage. The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign! In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

The collector curves are flat, or negative.

Really a nice part. It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I could use that as a signal-pickoff emitter follower, base hung on a
50 ohm microstrip. Any guess what the equivalent loading impedance
would be from DC to maybe 4 GHz?

I\'ll get some and try the pickoff thing. Maybe a Colpitts oscillator
too.

I still need to try the SAV551 as a follower too.




--

John Larkin Highland Technology, Inc

Science teaches us to doubt.

Claude Bernard

 

[Phil Hobbs]

On 2020-10-21 23:16, jlarkin@highlandsniptechnology.com wrote:

On Wed, 21 Oct 2020 22:42:03 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well. If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz). New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided). Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still. (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage. The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign! In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.


The collector curves are flat, or negative.


Really a nice part. It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I could use that as a signal-pickoff emitter follower, base hung on a
50 ohm microstrip. Any guess what the equivalent loading impedance
would be from DC to maybe 4 GHz?

Not very different from just the pad, given C_CB ~0.2 pF and very high beta.

I\'ll get some and try the pickoff thing. Maybe a Colpitts oscillator
too.

I still need to try the SAV551 as a follower too.

Works brilliantly as a bootstrap, which is sort of a follower on steroids.

I\'m having much worse problems with the photodiodes themselves. The
SFH2400 is small and has low capacitance, but even with >10V reverse
bias, it slows down grossly around 300uA of photocurrent. (Its rise time
goes from ~10 ns to ~250 nw.)

Cheers

Phil Hobbs
>

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 2020-10-21 22:42, Phil Hobbs wrote:

So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.

I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
<https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif>

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[John Larkin]

On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

Cheers

Phil Hobbs

I\'ve seen slow bipolars and fets with negative collector current
slopes. I suspect that was bad measurement, device heating messing
with the curves.

 

[Phil Hobbs]

On 2020-10-22 19:06, John Larkin wrote:

On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

I\'ve seen slow bipolars and fets with negative collector current
slopes. I suspect that was bad measurement, device heating messing
with the curves.

Almost for sure. The Early effect is base thinning due to changes in the
width of the CB depletion zone--you can reduce that, but it would be
hard to imagine it changing sign.

The noise canceller gets its log ratio output by measuring Delta V_BE of
the main differential pair, which has the usual 2 mV/K dependence on
temperature differences. (You have to worry about the effect of the
other device--some effects add and some subtract.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Joe Gwinn]

On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistirs in a square, with diagonal units paralleled?

Joe Gwinn

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

I\'ve seen slow bipolars and fets with negative collector current
slopes. I suspect that was bad measurement, device heating messing
with the curves.

Almost for sure. The Early effect is base thinning due to changes in the
width of the CB depletion zone--you can reduce that, but it would be
hard to imagine it changing sign.

The noise canceller gets its log ratio output by measuring Delta V_BE of
the main differential pair, which has the usual 2 mV/K dependence on
temperature differences. (You have to worry about the effect of the
other device--some effects add and some subtract.)

Cheers

Phil Hobbs

 

[John Larkin]

On Fri, 23 Oct 2020 13:47:10 -0400, Joe Gwinn <joegwinn@comcast.net>
wrote:

On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistirs in a square, with diagonal units paralleled?

Joe Gwinn

With discretes? That would be interesting.

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

 

[Joe Gwinn]

On Fri, 23 Oct 2020 11:09:08 -0700, John Larkin
<jlarkin@highland_atwork_technology.com> wrote:

On Fri, 23 Oct 2020 13:47:10 -0400, Joe Gwinn <joegwinn@comcast.net
wrote:

On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistirs in a square, with diagonal units paralleled?

Joe Gwinn



With discretes? That would be interesting.

Yes. The BFP640 transistors are about 2 mm square, including leads.

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

Two dies would not be able to cancel temperature gradients. Especially
if the thermal cnnection is molding plastic, not a good thermal
condusctor lile silicon. Sort of defeats the purpose, doesn\'t it?

Four chips in a square mounted on an alumina substrate would be far
better.

Joe Gwinn

 

[whit3rd]

On Friday, October 23, 2020 at 11:09:22 AM UTC-7, John Larkin wrote:

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

That picture is possible because there was no \'hat\' over the package. Does it
get better coupling if you glue a heat spreader (the pyrolitic graphite type, or
a simple copper slug) atop the unit?

SOT143 is only one millimeter top-to-bottom, so sticking a spreader on top can
definitely give some coupling beyond what the package has intrinsically.

SOT666 from Nexperia is only 0.55mm top-to-bottom.

 

[Phil Hobbs]

On 2020-10-23 13:47, Joe Gwinn wrote:

On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistors in a square, with diagonal units paralleled?

That would help with the externally-imposed gradients, sure. It\'s less
helpful with the differential dissipation, because the thermal coupling
is heartbreakingly poor.

If you have a squint at the BC61C current mirror datasheet, you\'ll find
a cryptic note on P. 4 that indicates that the maximum input current to
the mirror for which the output device is thermally stable is 5 mA for a
5V V_CE on the output device--25 mW.

That is, at 25 mW, a 1-K temperature change will cause the output
transistor\'s current to increase enough that its temperature goes up by
another degree, and we\'re off to the races. We\'ll call the temperature
difference between the two dice Tdiff, and the thermal resistance
between them Theta_JJ. We thus have:

dPout/dT = 5V * dIout/dT

dIout/dT = ((2 mV/K) * e/kT * I_out)

dTdiff/dPout = Theta_JJ

And at the stability boundary

dTdiff/dPout * dPout/dT = 1.

Thus near room temperature the device becomes unstable when

(5V * 2 mV/K / 26 mV * 5 mA) Theta_JJ = 1

or

Theta_JJ = 520 K/W.

So even for two dice in the same SOT-343 package, the thermal coupling
between them is worse than their coupling to the outside world via the
leads. (There\'s probably a safety factor in the quoted value, but it
sure ain\'t like a monolithic pair.)

In addition there are transient problems--thermal conduction gets
quadratically slow with distance, even in the best heat conductors. The
thermal response of the BFP640 die just in its own package shows a time
constant of 2.2 ms, so coupling between devices would be slower even
than that.

Dorking the dissipation to be constant can be done as fast as you like,
_provided_ the Early voltage is high enough.

Thanks

Phil Hobbs

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

I\'ve seen slow bipolars and fets with negative collector current
slopes. I suspect that was bad measurement, device heating messing
with the curves.

Almost for sure. The Early effect is base thinning due to changes in the
width of the CB depletion zone--you can reduce that, but it would be
hard to imagine it changing sign.

The noise canceller gets its log ratio output by measuring Delta V_BE of
the main differential pair, which has the usual 2 mV/K dependence on
temperature differences. (You have to worry about the effect of the
other device--some effects add and some subtract.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 2020-10-23 14:58, Joe Gwinn wrote:

On Fri, 23 Oct 2020 11:09:08 -0700, John Larkin
jlarkin@highland_atwork_technology.com> wrote:

On Fri, 23 Oct 2020 13:47:10 -0400, Joe Gwinn <joegwinn@comcast.net
wrote:

On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistirs in a square, with diagonal units paralleled?

Joe Gwinn



With discretes? That would be interesting.

Yes. The BFP640 transistors are about 2 mm square, including leads.


Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

Two dies would not be able to cancel temperature gradients. Especially
if the thermal cnnection is molding plastic, not a good thermal
condusctor lile silicon. Sort of defeats the purpose, doesn\'t it?

On a board you can put the two of them on a paddle, which pretty well
fixes the problem.

Four chips in a square mounted on an alumina substrate would be far
better.

A bit better, for sure, but just the increased distance (and the slow
thermal diffusivity of alumina) would make it fairly disappointing at
signal frequencies.

Even SOI wafers are noticeably worse than bulk silicon for this sort of job.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 2020-10-23 15:24, whit3rd wrote:

On Friday, October 23, 2020 at 11:09:22 AM UTC-7, John Larkin wrote:

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

That picture is possible because there was no \'hat\' over the package. Does it
get better coupling if you glue a heat spreader (the pyrolitic graphite type, or
a simple copper slug) atop the unit?

SOT143 is only one millimeter top-to-bottom, so sticking a spreader on top can
definitely give some coupling beyond what the package has intrinsically.

A factor of 2 at the outside. Plastic is a really, really crappy,
really, really slow thermal conductor. That\'s a big help when you want
to use nylon screws to clamp a cold plate to a TEC, but usually annoying
the rest of the time.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[whit3rd]

On Friday, October 23, 2020 at 5:52:24 PM UTC-7, Phil Hobbs wrote:

On 2020-10-23 15:24, whit3rd wrote:
On Friday, October 23, 2020 at 11:09:22 AM UTC-7, John Larkin wrote:

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

That picture is possible because there was no \'hat\' over the package. Does it
get better coupling if you glue a heat spreader (the pyrolitic graphite type, or
a simple copper slug) atop the unit?

SOT143 is only one millimeter top-to-bottom, so sticking a spreader on top can
definitely give some coupling beyond what the package has intrinsically.

A factor of 2 at the outside. Plastic is a really, really crappy,
really, really slow thermal conductor. That\'s a big help when you want
to use nylon screws to clamp a cold plate to a TEC, but usually annoying
the rest of the time.

The \"plastic\" is a mixture, epoxy filled with some kind of material so it doesn\'t leak light.
A thin layer isn\'t too bad; STmicro 78xx regulators come in \'fullpack\' and bare-metal tabs,
with near-equal thermal resistance junction-to-case.

 

[Phil Hobbs]

On 2020-10-23 21:38, whit3rd wrote:

On Friday, October 23, 2020 at 5:52:24 PM UTC-7, Phil Hobbs wrote:
On 2020-10-23 15:24, whit3rd wrote:
On Friday, October 23, 2020 at 11:09:22 AM UTC-7, John Larkin wrote:

Most of the dual transistors being sold these days are two die in an
epoxy package, with bad thermal coupling.

https://www.dropbox.com/s/4k4bf7yfvdj0h0d/NEC_dual_transistor.jpg?raw=1

That picture is possible because there was no \'hat\' over the package. Does it
get better coupling if you glue a heat spreader (the pyrolitic graphite type, or
a simple copper slug) atop the unit?

SOT143 is only one millimeter top-to-bottom, so sticking a spreader on top can
definitely give some coupling beyond what the package has intrinsically.

A factor of 2 at the outside. Plastic is a really, really crappy,
really, really slow thermal conductor. That\'s a big help when you want
to use nylon screws to clamp a cold plate to a TEC, but usually annoying
the rest of the time.

The \"plastic\" is a mixture, epoxy filled with some kind of material so it doesn\'t leak light.
A thin layer isn\'t too bad; STmicro 78xx regulators come in \'fullpack\' and bare-metal tabs,
with near-equal thermal resistance junction-to-case.

I went through the math in another post. IC packages are generally made
of Sumitomo novolac epoxy, which like other plastics has a thermal
conductivity near 0.1 W/m/K and a thermal diffusivity hundreds of times
slower than silicon\'s.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Joe Gwinn]

On Fri, 23 Oct 2020 20:43:39 -0400, Phil Hobbs
<pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-23 13:47, Joe Gwinn wrote:
On Thu, 22 Oct 2020 20:07:22 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-22 19:06, John Larkin wrote:
On Thu, 22 Oct 2020 18:32:36 -0400, Phil Hobbs
pcdhSpamMeSenseless@electrooptical.net> wrote:

On 2020-10-21 22:42, Phil Hobbs wrote:
So I\'m doing this new laser noise canceller that I was talking about in
the \"differential signal detector\" thread.

The canceller is probably my best gizmo ever, and iw actually pretty
simple as well.  If you\'re interested, check out \"ultrasensitive laser
measurements without tears\",
https://electrooptical.net/static/media/uploads/Projects/LaserNoiseCanceller/noisecan.pdf>.
 The seven circuits in that paper use MAT04 supermatch quad NPNs, which
work great at lowish frequency (<~ 1 MHz).  New Focus has been selling
the circuit of Fig. 3 of that paper for 25 years now, without change as
far as I know.

The new one uses discrete SiGe:C transistors (Infineon BFP640/650/780,
currently undecided).  Unmatched transistors work fine, but they need to
be at the same temperature to moderate accuracy (100 mK or so for my
purposes).

To get vaguely accurate results from the canceller\'s log ratio output,
besides temperature tracking, the transistors should be closely similar.
 Same-wafer accuracy ought to be fine for most uses, and besides,
relative calibration isn\'t too hard and should sit still.  (Absolute
calibration is very hard in optics, but this is a ratio measurement.)

Today\'s tidbit: Early voltage.  The collector curves in the 45-GHz
BFP640 datasheet seem to show an indefinitely large Early voltage--at
higher current the slope even changes sign!  In contrast, your average
10-GHz NPN has an Early voltage of about 12.

In my little canceller test board, I\'m measuring an Early voltage for
the BFP640H of about 500V at 0.5 mA.

Really a nice part.  It does need a 5-ohm GHz bead in the base to keep
it well behaved, but then it really is well-behaved.

I got it all instrumented today, so tomorrow I\'ll take a bunch of data
and see what we see.


BTW: SFH2400 photodiodes stink.



I did some higher speed Early measurements using a cascode transistor to
adjust V_CE while allowing current measurement.
https://electrooptical.net/static/oldsite/www/sed/BFP640EarlyVoltage500uA_VCE_0.3-2.8V.tif

With a collector current of 0.5 mA, changing V_CE from 0.35V to 2.85V
produced a fast change of about 1.0% in I_C, for a measured Early
voltage of

V_A = 2.5V/0.01 = 250V.

As the part warmed up, about half of the peak delta decayed away with a
TC of 2.2 ms, which is why slow measurements indicated a 500-V V_A.
(The exact amount depends on the absolute power dissipation, of course.)

Doing the math, this gives a Theta_JA of about 200 W/m/K, which is
reasonable. Interestingly there was no thermal-diffusion structure on
the leading edge of the pulse--at the 10-us level there\'s no thermal
speedup visible.

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one. It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup. Makes a huge difference. (The Tek \"Vertical Amplifiers\" book is
still a good read.)

The present case is a bit more complicated, because I need to be able to
split photocurrents very accurately up to ~5 mA, in ratios from 10/90 to
90/10. A bit of degradation is OK at the edges, but in the sweet spot,
I want to get the relative error down below 10**-3 from DC to ~10 MHz
with no user adjustments.

That\'s very hard with a 100-MHz f_T MAT14, but might be easier with
45-GHz BFP640s, if the temperature tracking can be got round.

If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents. Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

Can you replicate the old IC trick of symmetrical transistors
configured to cancel temperature gradients here?

Like four transistors in a square, with diagonal units paralleled?

That would help with the externally-imposed gradients, sure. It\'s less
helpful with the differential dissipation, because the thermal coupling
is heartbreakingly poor.

Ahh. I was not thinking of that, I was thinking only of externally
imposed gradients.

If you have a squint at the BC61C current mirror datasheet, you\'ll find
a cryptic note on P. 4 that indicates that the maximum input current to
the mirror for which the output device is thermally stable is 5 mA for a
5V V_CE on the output device--25 mW.

Thermal runaway. Cute.

That is, at 25 mW, a 1-K temperature change will cause the output
transistor\'s current to increase enough that its temperature goes up by
another degree, and we\'re off to the races. We\'ll call the temperature
difference between the two dice Tdiff, and the thermal resistance
between them Theta_JJ. We thus have:

dPout/dT = 5V * dIout/dT

dIout/dT = ((2 mV/K) * e/kT * I_out)

dTdiff/dPout = Theta_JJ

And at the stability boundary

dTdiff/dPout * dPout/dT = 1.

Thus near room temperature the device becomes unstable when

(5V * 2 mV/K / 26 mV * 5 mA) Theta_JJ = 1

or

Theta_JJ = 520 K/W.

So even for two dice in the same SOT-343 package, the thermal coupling
between them is worse than their coupling to the outside world via the
leads. (There\'s probably a safety factor in the quoted value, but it
sure ain\'t like a monolithic pair.)

In addition there are transient problems--thermal conduction gets
quadratically slow with distance, even in the best heat conductors. The
thermal response of the BFP640 die just in its own package shows a time
constant of 2.2 ms, so coupling between devices would be slower even
than that.

Dorking the dissipation to be constant can be done as fast as you like,
_provided_ the Early voltage is high enough.

That makes sense, I guess. I should read up on Early voltage.
Something clever this way comes.

Joe Gwinn

Thanks

Phil Hobbs

That makes a whole lot of things possible that weren\'t before.

I took a bunch of logging data today, so I can figure out what the R_ee\'
and beta linearity contributions are doing vs. voltage and current, so
we can generate a tweaking algorithm for the real unit.

Fun.

I\'ve seen slow bipolars and fets with negative collector current
slopes. I suspect that was bad measurement, device heating messing
with the curves.

Almost for sure. The Early effect is base thinning due to changes in the
width of the CB depletion zone--you can reduce that, but it would be
hard to imagine it changing sign.

The noise canceller gets its log ratio output by measuring Delta V_BE of
the main differential pair, which has the usual 2 mV/K dependence on
temperature differences. (You have to worry about the effect of the
other device--some effects add and some subtract.)

Cheers

Phil Hobbs

 

[Okkim Atnarivik]

On 10/22/20 8:07 PM, Phil Hobbs wrote:

That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one.  It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup.  Makes a huge difference.  (The Tek \"Vertical Amplifiers\" book is
still a good read.)

....
If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents.  Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

The thermal servo trick is originally due to Tek? Interesting. I learned
it from Nikolay Ukhanski some years ago. He and his colleaques published
it, too, in a physics journal.

Regards,
Mikko

 

[Okkim Atnarivik]

On 12/8/20 8:06 PM, Okkim Atnarivik wrote:

On 10/22/20 8:07 PM, Phil Hobbs wrote:
That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one.  It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup.  Makes a huge difference.  (The Tek \"Vertical Amplifiers\" book is
still a good read.)

...
If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents.  Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

The thermal servo trick is originally due to Tek? Interesting. I learned
it from Nikolay Ukhanski some years ago. He and his colleaques published
it, too, in a physics journal.

Regards,
Mikko

The paper I\'m thinking is older and more obscure, but judging
by the abstract, doi:10.1063.12335630 looks looks like same stuff.

Regards,
Mikko

 

[Phil Hobbs]

On 12/8/20 8:06 PM, Okkim Atnarivik wrote:

On 10/22/20 8:07 PM, Phil Hobbs wrote:
That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one.  It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup.  Makes a huge difference.  (The Tek \"Vertical Amplifiers\" book is
still a good read.)

...
If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents.  Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

The thermal servo trick is originally due to Tek? Interesting. I learned
it from Nikolay Ukhanski some years ago. He and his colleaques published
it, too, in a physics journal.

Regards,
Mikko

The Tek idea is to bias the transistors at the maximum-dissipation
point, so that small excursions don\'t cause thermal transients of any
appreciable size. That\'s an open-loop thing.

The thermal servoing idea may well not be original with me, but at least
I invented it independently in about 2010 or 2011. It\'s not much use
with normal RF transistors, because their Early voltages are so low, but
SiGe devices have VA ~ 250V at signal rates. (The datasheet curves even
tilt the other way, because the thermal effect on V_BE is bigger than
the Early effect.)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com

 

[Phil Hobbs]

On 12/8/20 8:19 PM, Okkim Atnarivik wrote:

On 12/8/20 8:06 PM, Okkim Atnarivik wrote:
On 10/22/20 8:07 PM, Phil Hobbs wrote:
That\'s a good 20 times better than your average 10-GHz transistor, and
this is a 45-GHz one.  It\'s good enough (and the thermal is slow enough)
that I can mimic monolithic behaviour by equalizing the dissipation of
the two halves of the diff pair by dorking the V_CE of one side.

There is an old Tek trick, apparently re-invented by many others, to
put a bypassed resistor in the collectors of a discrete diff pair, so
the power dissipation is mostly constant with signal. That eliminated
some thermal hooks in vertical amplifiers.

Yup.  Makes a huge difference.  (The Tek \"Vertical Amplifiers\" book is
still a good read.)

...
If the transistors are at different temperatures, the splitting is
degraded because the same delta V_BE produces different ratios at
different tail currents.  Thus a symmetrical layout combined with
dorking V_CE as a function of I_C has a lot of charm.

It\'s happening at millisecond timescales, too, which makes it a good
match for some simple MCU magic.

The thermal servo trick is originally due to Tek? Interesting. I learned
it from Nikolay Ukhanski some years ago. He and his colleaques published
it, too, in a physics journal.

The paper I\'m thinking is older and more obscure, but judging
by the abstract, doi:10.1063.12335630 looks looks like same stuff.

The reference doesn\'t seem to work. Is there something after the slash?

Thanks

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC / Hobbs ElectroOptics
Optics, Electro-optics, Photonics, Analog Electronics
Briarcliff Manor NY 10510

http://electrooptical.net
http://hobbs-eo.com