250mA voltage clamp

[Winfield Hill]

Fred Bloggs wrote...

Fred Bloggs wrote ...

With that much dropout it makes no sense to let it drop out. You can keep
the LM317 in regulation by regulating to filtered Vin - 3V up to a clamped
upper limit as required. We are talking about a simple level shift and
diode clamp- and not a lot of excess current. The extra 0.5V drop is more
than paid for by the high performance of the regulator, and you have all
the protection the LM317 offers. Don't forget to add all the usual diode
protection around the IC.
View in a fixed-width font such as Courier.

.
. +--------+
. Vin>--+-----+--------------|IN OUT|-----+----+-->
. | | | | | |
. | | | ADJ | | |
. | | +--------+ [220] |
. [4.7K] [10K] | | |
. | | | | |+
. | | .--|>|--+----------+ ===
. | | | | | |
. | | | | 1N4748 | |
. | | | |< ---/ |
. +----------------+-----| Q2 // \ |
. |+ | | |\ --- |
. === | |/ | | |
. 100U +--------| Q1 | | |
. | | |> | | |
. | | | | | |
. | | [620] | | |
. | | | | | |
. | '-|>|-|>|--+ | | |
. | | | | |
. +----------------+-------+----------+----+-->
. |
. ---

That's a nice circuit. Am I right that the function of the diode is to
protect Q2's base-emitter junction from reverse breakdown?

Yes- the NPN current source keeps the base of the PNP at ~Vin-5 while
the zener clamps at 22V, so the PNP BE junction starts to reverse bias
at Vin=27V.

I like the idea of having the LM317's protection. But I'm still leaning
toward the discrete approaches, because I don't like having this much
input-output differential in the non-clamped case; it means that the piece
of equipment this is powering won't handle brownouts as well. Probably a
non-issue, but I hate for my stuff to be the weak link. I'll probably go
with one of the discrete circuits instead.

I suppose you could adapt the same approach to an LDO, pre-regulating
say 0.2V-0.5V down up until clamp level,- there are a few that handle
40V IIRC, or use a more common and cheaper one and install a zener
over-voltage bypass. If you clamp at 24V and Vin,max=35 then the LDO
only needs to hold off 11V worst case. The trick is to find one in
something easy to heatsink like a TO-220 , and the 250mA current
requirement is not an issue.

I wasn't paying attention. The purpose of the circuit was to act as
an input ripple and noise filter?

 

[Fred Bloggs]

Fred Bloggs wrote...

Fred Bloggs wrote ...


With that much dropout it makes no sense to let it drop out. You can keep
the LM317 in regulation by regulating to filtered Vin - 3V up to a clamped
upper limit as required. We are talking about a simple level shift and
diode clamp- and not a lot of excess current. The extra 0.5V drop is more
than paid for by the high performance of the regulator, and you have all
the protection the LM317 offers. Don't forget to add all the usual diode
protection around the IC.
View in a fixed-width font such as Courier.

.
. +--------+
. Vin>--+-----+--------------|IN OUT|-----+----+-->
. | | | | | |
. | | | ADJ | | |
. | | +--------+ [220] |
. [4.7K] [10K] | | |
. | | | | |+
. | | .--|>|--+----------+ ===
. | | | | | |
. | | | | 1N4748 | |
. | | | |< ---/ |
. +----------------+-----| Q2 // \ |
. |+ | | |\ --- |
. === | |/ | | |
. 100U +--------| Q1 | | |
. | | |> | | |
. | | | | | |
. | | [620] | | |
. | | | | | |
. | '-|>|-|>|--+ | | |
. | | | | |
. +----------------+-------+----------+----+-->
. |
. ---

That's a nice circuit. Am I right that the function of the diode is to
protect Q2's base-emitter junction from reverse breakdown?

Yes- the NPN current source keeps the base of the PNP at ~Vin-5 while
the zener clamps at 22V, so the PNP BE junction starts to reverse bias
at Vin=27V.

I like the idea of having the LM317's protection. But I'm still leaning
toward the discrete approaches, because I don't like having this much
input-output differential in the non-clamped case; it means that the piece
of equipment this is powering won't handle brownouts as well. Probably a
non-issue, but I hate for my stuff to be the weak link. I'll probably go
with one of the discrete circuits instead.

I suppose you could adapt the same approach to an LDO, pre-regulating
say 0.2V-0.5V down up until clamp level,- there are a few that handle
40V IIRC, or use a more common and cheaper one and install a zener
over-voltage bypass. If you clamp at 24V and Vin,max=35 then the LDO
only needs to hold off 11V worst case. The trick is to find one in
something easy to heatsink like a TO-220 , and the 250mA current
requirement is not an issue.

I wasn't paying attention. The purpose of the circuit was to act as
an input ripple and noise filter?

Yes- and then clamp the maximum output regulated voltage at 24V or so
when the unregulated gets above 27V or so.

 

[Winfield Hill]

Walter Harley wrote...

Winfield Hill wrote ...

To confirm, I hooked up an LM317 to provide 21V, with a load switched
between 75R and 1k by way of a MOSFET controlled by a pulse generator.
Then, with a scope I watched the output as the load switched on and off,
while varying the input voltage from 18V to 35V.

As Win suggests, the LM317 stays quiet and stable when its input voltage
is too low to regulate. There was no hint of ringing or noise around the
transitions. When the input voltage is too low, the LM317 looks like a
2.5V drop, in series with an ohm or so.

I was a little surprised that the drop was 2.5V, even with only 1k load
(in parallel with the ~4k divider network). From the datasheet I
expected it to be lower. I was using a JRC part rather than National,
but the JRC datasheet still suggests dropout < 2V.

Fred suggested using an LDO regulator instead of an LM317, what's
wrong with that idea? For example, LTC's LT1085 adjustable 3-term
is rated to 3A max and has about 900mV dropout at 250mA. If you're
concerned with this IC's 30V max input-output rating (a factor if
you short the output with 35V input) you can consider their LT1123,
which works with an external transistor to obtain a 250mV dropout.

http://www.linear.com/pc/productDetail.do?navId=H0,C1,C1003,C1040,C1130,P1460
Bias the GND terminal of this 5V regulator at +19V, to obtain a
24V regulated output and a 49V maximum input-voltage rating. You
can make a 19V "zener" using a common TL431 as a shunt reference.

 

[Walter Harley]

Fred suggested using an LDO regulator instead of an LM317, what's
wrong with that idea? For example, LTC's LT1085 adjustable 3-term
is rated to 3A max and has about 900mV dropout at 250mA. If you're
concerned with this IC's 30V max input-output rating (a factor if
you short the output with 35V input) you can consider their LT1123,
which works with an external transistor to obtain a 250mV dropout.

http://www.linear.com/pc/productDetail.do?navId=H0,C1,C1003,C1040,C1130,P1460
Bias the GND terminal of this 5V regulator at +19V, to obtain a
24V regulated output and a 49V maximum input-voltage rating. You
can make a 19V "zener" using a common TL431 as a shunt reference.

Nothing at all wrong with the idea. I just don't happen to have one on
hand, and I don't anticipate another parts order till January. So I was
hunting around for solutions that used only parts on hand. (I realize
that's not a useful spec, since no one else knows what parts I have on
hand.)

Here's what I came up with. It uses more parts than some of the
suggestions, but I happen to have everything on hand. This circuit appears
to work well on the bench; I haven't seen any signs of instability (as long
as the .01uF cap is there). The IRFP9240 was chosen only because I've got a
tube of them that I'll never get rid of otherwise.


----o---------------o-----+^+--------o-------
| | ||| |
.-. | === |
| | | | IRFP9240 |
| |100k PN2907A| | |
'-' | .-. |
| ___ |< | | |
o--|___|-o----| | |1k z
| 1k | |\ '-' A 22V
| | | | |
| | || | | |
| '-||---o-----o |
| || | |
| .01uF | |
| .-. |
| | | |
| | |10k |
| '-' |
| PN2222A | |
\| ___ | |
|--|___|------------)----------o
<| 10k | |
| | .-.
| | | |
| | | |470
| | '-'
| | |
----o---------------------o----------o-------


Thanks,
-walter

 

[Winfield Hill]

Walter Harley wrote...

Winfield Hill wrote ...

Nothing at all wrong with the idea. I just don't happen to have one on
hand, and I don't anticipate another parts order till January. So I was
hunting around for solutions that used only parts on hand. (I realize
that's not a useful spec, since no one else knows what parts I have on
hand.)

Here's what I came up with. It uses more parts than some of the
suggestions, but I happen to have everything on hand. This circuit
appears to work well on the bench; I haven't seen any signs of
instability (as long as the .01uF cap is there). The IRFP9240 was
chosen only because I've got a tube of them that I'll never get
rid of otherwise.

----o---------------o-----+^+--------o-------
| | ||| |
.-. | === |
| | | | IRFP9240 |
| |100k PN2907A| | |
'-' | .-. |
| ___ |< | | |
o--|___|-o----| | |1k z
| 1k | |\ '-' A 22V
| | | | |
| | || | | |
| '-||---o-----o |
| || | |
| .01uF .-. |
| | | |
| | |10k |
| PN2222A '-' |
\| ___ | |
|--|___|------------)----------o
<| 10k | |
| | .-.
| | | |
| | | |470
| | '-'
| | |
----o---------------------o----------o-------

That's not a bad approach. You can eliminate both of the 1k
resistors, they're not pulling their weight. I'd also reduce
the pn2222 base-resistor value. You can add a small pot in
series with the zener to trim the output voltage, this adds
Vbe * R/470 to the output.

The circuit has three gain elements, the pn2222 with a 100k
load, the pn2907 with a 10k load, and the MOSFET with the
output load, which usually includes a large capacitor. Each
of these elements adds a pole to the feedback loop response,
and we know two poles is enough to make an oscillator. Your
compensation cap combines two poles, which helps.

To acid-test the loop-stability of the circuit, try a range
capacitor values on the output, small to large, and test with
each one over the full range of load currents. Inject a small
low-frequency square-wave stimulus into the base of the pn2222
through a resistor, and tune the compensation to avoid ringing
on the step changes in the output voltage. You will find you
can use a smaller feedback capacitor, to speed up the response,
if you add a resistor in series with the cap (this adds a zero
in the loop response, canceling the cap's pole at a frequency,
hopefully in the region where some other element in the circuit
is adding its own pole). Let us know how it works out.

 

[Walter Harley]

[...]
----o---------------o-----+^+--------o-------
| | ||| |
.-. | === |
| | | | IRFP9240 |
| |100k PN2907A| | |
'-' | .-. |
| ___ |< | | |
o--|___|-o----| | |1k z
| 1k | |\ '-' A 22V
| | | | |
| | || | | |
| '-||---o-----o |
| || | |
| .01uF .-. |
| | | |
| | |10k |
| PN2222A '-' |
\| ___ | |
|--|___|------------)----------o
<| 10k | |
| | .-.
| | | |
| | | |470
| | '-'
| | |
----o---------------------o----------o-------

That's not a bad approach. You can eliminate both of the 1k
resistors, they're not pulling their weight. I'd also reduce
the pn2222 base-resistor value. You can add a small pot in
series with the zener to trim the output voltage, this adds
Vbe * R/470 to the output.

Thanks!

The 1k in series with the MOSFET gate is there because I was worried about
the MOSFET oscillating at RF; there might be an inch or so of wire between
the gate (or gate resistor) and the rest of the circuitry, and three or four
inches from drain to load. Should I not worry?

The 1k in series with the PN2907 base is there to limit current through the
transistors in case the PN2222 turns on hard for some reason. I confess I
don't see what could cause that, other than perhaps a momentary short across
the MOSFET.

The circuit has three gain elements, the pn2222 with a 100k
load, the pn2907 with a 10k load, and the MOSFET with the
output load, which usually includes a large capacitor. Each
of these elements adds a pole to the feedback loop response,
and we know two poles is enough to make an oscillator. Your
compensation cap combines two poles, which helps.

To acid-test the loop-stability of the circuit, try a range
capacitor values on the output, small to large, and test with
each one over the full range of load currents. Inject a small
low-frequency square-wave stimulus into the base of the pn2222
through a resistor, and tune the compensation to avoid ringing
on the step changes in the output voltage. You will find you
can use a smaller feedback capacitor, to speed up the response,
if you add a resistor in series with the cap (this adds a zero
in the loop response, canceling the cap's pole at a frequency,
hopefully in the region where some other element in the circuit
is adding its own pole). Let us know how it works out.

I will do that. It'll be a couple days before I get to it - one of my
distributors had a holiday rush and I need to go into assembly-line mode for
a little while.

Thanks,
-walter

 

[Winfield Hill]

Walter Harley wrote...

Winfield Hill wrote ...

[...]
----o---------------o-----+^+--------o-------
| | ||| |
.-. | === |
| | | | IRFP9240 |
| |100k PN2907A| | |
'-' | .-. |
| ___ |< | | |
o--|___|-o----| | |1k z
| 1k | |\ '-' A 22V
| | | | |
| | || | | |
| '-||---o-----o |
| || | |
| .01uF .-. |
| | | |
| | |10k |
| PN2222A '-' |
\| ___ | |
|--|___|------------)----------o
<| 10k | |
| | .-.
| | | |
| | | |470
| | '-'
| | |
----o---------------------o----------o-------

That's not a bad approach. You can eliminate both of the 1k
resistors, they're not pulling their weight. I'd also reduce
the pn2222 base-resistor value. You can add a small pot in
series with the zener to trim the output voltage, this adds
Vbe * R/470 to the output.

The 1k in series with the MOSFET gate is there because I was
worried about the MOSFET oscillating at RF; there might be an
inch or so of wire between the gate (or gate resistor) and the
rest of the circuitry, and three or four inches from drain to
load. Should I not worry?

Not on that account, but there is one argument for it, which
occurred to me after my posting. The two transistors with the
cap feedback act as an integrator stage, which if capacitively
loaded might be unstable at high frequencies. Your resistor
acts to isolate this stage from the MOSFET's high gate Ciss.

The 1k in series with the PN2907 base is there to limit current
through the transistors in case the PN2222 turns on hard for
some reason. I confess I don't see what could cause that,
other than perhaps a momentary short across the MOSFET.

No, that's a good reason. You can keep both resistors.

I will do that. It'll be a couple days before I get to it -
one of my distributors had a holiday rush and I need to go into
assembly-line mode for a little while.

We'll wait for your next installment.

 

[Winfield Hill]

Winfield Hill wrote...

Walter Harley wrote...

... You can keep both resistors.


We'll wait for your next installment.

The circuit's three stages bother me, they provide far more
loop gain than is required for the job, and they create new
problems with loop stabilization. For example, the irfp9240
has 1200pF of gate capacitance, which with the 10k resistor
creates a pole at 13kHz. And the MOSFET stage's gain is one
giant pole, an integrator whose gain is high but goes through
unity at 500Hz to 12.5kHz, depending on the load current,
assuming a 22uF load cap. Whew, poles all over the place.

Anyway, I'd feel more comfortable with a simple circuit.
Try this one on for size.

.. +20 to 50V PMOS 24V regulated
.. -----+------+-- S D --+------+----+------+----
.. | | G | | | |
.. | 2.2k | | | _|_+ 2.2k
.. | | | | | --- 0.5W
.. | +-----' | 18k | 22uF |
.. 4.7k | | | | |
.. | c c | gnd gnd
.. +--- b b ---+
.. | e ----+---- e |
.. \_|_ | 6.2k
.. /_\ 6.2V 1.5k |
.. | | gnd
.. gnd gnd

It has a loop gain of 2500 at 100Hz at 0.25A load, which is more
than enough, and should be nicely stable if the 22uF electrolytic
has more than 55 milli-ohms of esr, which it no doubt will have.

Zeners around 6.2V have very stable characteristics, but you can
use other values, if you change the feedback resistors.

 

[Fred Bloggs]

Here's what I came up with. It uses more parts than some of the
suggestions, but I happen to have everything on hand. This circuit appears
to work well on the bench; I haven't seen any signs of instability (as long
as the .01uF cap is there). The IRFP9240 was chosen only because I've got a
tube of them that I'll never get rid of otherwise.


----o---------------o-----+^+--------o-------
| | ||| |
.-. | === |
| | | | IRFP9240 |
| |100k PN2907A| | |
'-' | .-. |
| ___ |< | | |
o--|___|-o----| | |1k z
| 1k | |\ '-' A 22V
| | | | |
| | || | | |
| '-||---o-----o |
| || | |
| .01uF | |
| .-. |
| | | |
| | |10k |
| '-' |
| PN2222A | |
\| ___ | |
|--|___|------------)----------o
<| 10k | |
| | .-.
| | | |
| | | |470
| | '-'
| | |
----o---------------------o----------o-------

I wouldn't attempt that with all that unnecessary DC gain. A more
stream-lined stand-alone job that is stable with any capacitive loading
and reasonable wiring inductance would be more like this:
View in a fixed-width font such as Courier.

..
..
.. ___
.. .---------------|___|---.
.. | 1k |
.. | |
.. | IRFP9240 |
.. >---+------o---------+^+------o----o-----------o----->
.. | | ||| | | |
.. | .-. === | .-. |
.. | | | | === | | |
.. | | |22k | 100n| | |1.2k |
.. | '-' | | '-' |
.. | | | ___ | | |
.. | '---------o--|___|-' | .---o
.. | | 220 | | |
.. | | | | .-.
.. | | | === | |
.. +| | | 1u | | |1k
.. === .---------|-------------' | '-'
.. 1u | | | |
.. | | | '---o
.. | | | |
.. | 100n | |/ |
.. o--||--o-------|2N4410 |
.. | | |> |
.. | z | |
.. | A 12V '-------------------------o
.. | | |
.. | | .-.
.. | a | |
.. | k 1N4001 | |1.2k
.. | | '-'
.. | | |
.. >---o------o-----------------------------------'
.. |
.. '------------------------------------------------->
..
..

 

[Fred Bloggs]

A second transistor can be put to better use as a current limit-
foldback left as an exercise for the student:
View in a fixed-width font such as Courier.

..
..
.. ___
.. .---------------|___|---.
.. | 1k |
.. | |
.. ___ | IRFP9240 |
.. >---o---|___|-+----o---------+^+------o----o-----------o----->
.. | 1.3 | | ||| | | |
.. | | .-. === | .-. |
.. o-----. | | | | === | | |
.. | | | | |22k | 100n| | |1.2k |
.. | | | '-' | | '-' |
.. | >| | | | ___ | | |
.. |2N2907A|-' '---------o--|___|-' | .---o
.. | /| | 220 | | |
.. | | | | | .-.
.. | '------------------|-------. | === | |
.. +| | | | 1u | | |1k
.. === .---------|-------------' | '-'
.. 1u | | | | |
.. | | | | '---o
.. | | | | |
.. | | |/ | |
.. | -----o-------|2N4410 | |
.. | | | |> | |
.. | | z | | |
.. | 100n| A 12V '-------o-----------------o
.. | === | |
.. | | | .-.
.. | | a | |
.. | | k 1N4001 | |1.2k
.. | | | '-'
.. | | | |
.. >---o--------+-----o-----------------------------------'
.. |
.. '--------------------------------------------------------->
..
..

 

[Winfield Hill]

Winfield Hill wrote...

I'm feel more comfortable with a simple circuit. Try this one
on for size.

. +20 to 50V PMOS 24V regulated
. -----+------+-- S D --+------+----+------+----
. | | G | | | |
. | 2.2k | | | _|_+ 2.2k
. | | | | | --- 0.5W
. | +-----' | 18k | 22uF |
. 4.7k | | | | |
. | c c | gnd gnd
. +--- b b ---+
. | e ----+---- e |
. \_|_ | 6.2k
. /_\ 6.2V 1.5k |
. | | gnd
. gnd gnd

It has a loop gain of 2500 at 100Hz at 0.25A load, which is more
than enough, and should be nicely stable if the 22uF electrolytic
has more than 55 milli-ohms of esr, which it no doubt will have.

Zeners around 6.2V have very stable characteristics, but you can
use other values, if you change the feedback resistors.

A current limit can be added to this circuit to handle shorts, but
such capability means you'll need a much larger MOSFET heatsink.
Whereas 35-24V * 250mA created only 2.75 watts power dissipation,
under fault-free operation, a short-circuit fault would overheat the
MOSFET, damaging the regulator. A 400mA current limit can be added
to my circuit with only two components.

.. Low-dropout 24V regulator with 400mA current limit
..
.. +20 to 50V PMOS 24V regulated
.. -----+-- 1.6 -+-- S D -----+------+----+------+----
.. | | G | | | |
.. | b | | | _|_+ 2.2k
.. +----- e c --+ | | --- 0.5W
.. | pnp | | | | 22uF |
.. +-- 2.2k ---+--' | | gnd |
.. | | | | gnd
.. 4.7k | | 18k
.. | c c |
.. +-------- b b ---+
.. | e ----+---- e |
.. \_|_ | 6.2k
.. /_\ 6.2V 1.5k |
.. | | gnd
.. gnd gnd

In the event of a short the heatsink must handle 35*0.4 = 14 watts.
If the current limit is converted to a foldback type, which requires
two more resistors, this requirement could be reduced to say 4 watts.

 

[Walter Harley]

[...]
The circuit's three stages bother me, they provide far more
loop gain than is required for the job, and they create new
problems with loop stabilization. For example, the irfp9240
has 1200pF of gate capacitance, which with the 10k resistor
creates a pole at 13kHz. And the MOSFET stage's gain is one
giant pole, an integrator whose gain is high but goes through
unity at 500Hz to 12.5kHz, depending on the load current,
assuming a 22uF load cap. Whew, poles all over the place.

Anyway, I'd feel more comfortable with a simple circuit.
Try this one on for size.

. +20 to 50V PMOS 24V regulated
. -----+------+-- S D --+------+----+------+----
. | | G | | | |
. | 2.2k | | | _|_+ 2.2k
. | | | | | --- 0.5W
. | +-----' | 18k | 22uF |
. 4.7k | | | | |
. | c c | gnd gnd
. +--- b b ---+
. | e ----+---- e |
. \_|_ | 6.2k
. /_\ 6.2V 1.5k |
. | | gnd
. gnd gnd

It has a loop gain of 2500 at 100Hz at 0.25A load, which is more
than enough, and should be nicely stable if the 22uF electrolytic
has more than 55 milli-ohms of esr, which it no doubt will have.

Zeners around 6.2V have very stable characteristics, but you can
use other values, if you change the feedback resistors.

Posting here is generally a humbling experience for me. Thanks for your
patience.

First, re the old circuit: as it turned out, .01uF was not enough to keep it
from ringing at all load capacitances and currents. It was worst with
around 1uF of capacitance on the load; load resistance didn't seem to make
much difference. .033uF was enough to stabilize it for all loads. But your
& Fred's criticisms are of course correct.

Moving on to the above circuit: thanks, Win. That seems very simple and
straightforward. I built it, and it works nicely and shows no ringing for
any load capacitance from .01uF to 100uF.

I am having difficulty repeating your gain calculation, because I don't know
how to determine the FET's transconductance, but I will take that up on
sci.electronics.basics.

One more question: why is the second NPN's collector fed from the output
voltage, rather than from the input voltage?

Thanks again,
-walter

 

[Winfield Hill]

Walter Harley wrote...

Winfield Hill wrote ...

[...]
The circuit's three stages bother me, they provide far more
loop gain than is required for the job, and they create new
problems with loop stabilization. For example, the irfp9240
has 1200pF of gate capacitance, which with the 10k resistor
creates a pole at 13kHz. And the MOSFET stage's gain is one
giant pole, an integrator whose gain is high but goes through
unity at 500Hz to 12.5kHz, depending on the load current,
assuming a 22uF load cap. Whew, poles all over the place.

Anyway, I'd feel more comfortable with a simple circuit.
Try this one on for size.

. +20 to 50V PMOS 24V regulated
. -----+------+-- S D --+------+----+------+----
. | | G | | | |
. | 2.2k | | | _|_+ 2.2k
. | | | | | --- 0.5W
. | +-----' | 18k | 22uF |
. 4.7k | | | | |
. | c c | gnd gnd
. +--- b b ---+
. | e ----+---- e |
. \_|_ | 6.2k
. /_\ 6.2V 1.5k |
. | | gnd
. gnd gnd

It has a loop gain of 2500 at 100Hz at 0.25A load, which is more
than enough, and should be nicely stable if the 22uF electrolytic
has more than 55 milli-ohms of esr, which it no doubt will have.

Zeners around 6.2V have very stable characteristics, but you can
use other values, if you change the feedback resistors.

Moving on to the above circuit: thanks, Win. That seems very simple
and straightforward. I built it, and it works nicely and shows no
ringing for any load capacitance from .01uF to 100uF.

That's gratifying. Did you test over a range of load currents?

I am having difficulty repeating your gain calculation, because I
don't know how to determine the FET's transconductance, but I will
take that up on sci.electronics.basics.

You're not likely to get a good answer there, unless perhaps someone
was looking at AoE figure 3.24, and they dramatically extrapolated
the curve for large-area FETs. As it happens, you're operating in
the MOSFET's subthreshold region, where its transconductance acts
like that of a BJT transistor, only lower by a fixed amount, i.e.,
g_m = Ic / n Vt, where Vt = 25mV. You know, 25 ohms at 1mA, and
proportional to current. I've observed the constant n range from 3
to 10 for different manufacturer's MOSFET types. It appears to have
about the same value for different FETs and part runs within a type.
There's no hint of the issue, or of its value on most datasheets.

One more question: why is the second NPN's collector fed from the
output voltage, rather than from the input voltage?

It would work either way. The transistor has a more stable voltage
to work from this way, which really doesn't matter... The drawing
does look better this way!

 

[Walter Harley]

[...]

That's gratifying. Did you test over a range of load currents?

Just at the min and max; I didn't try intermediate points. I'll give that a
try.

You're not likely to get a good answer there, unless perhaps someone
was looking at AoE figure 3.24, and they dramatically extrapolated
the curve for large-area FETs. As it happens, you're operating in
the MOSFET's subthreshold region, where its transconductance acts
like that of a BJT transistor, only lower by a fixed amount, i.e.,
g_m = Ic / n Vt, where Vt = 25mV. You know, 25 ohms at 1mA, and
proportional to current. I've observed the constant n range from 3
to 10 for different manufacturer's MOSFET types. It appears to have
about the same value for different FETs and part runs within a type.
There's no hint of the issue, or of its value on most datasheets.

I see; I was confused about the subthreshold region. AoE 2ed. section 3.04
gives three equations for Id:

Id = 2k[(Vgs-Vt)Vds - Vds^2/2] (linear region)
Id = k(Vgs-Vt)^2 (saturation region)
Id = k[exp(Vgs-Vt)] (subthreshold region)

I was thinking that subthreshold described the far left of the Id/Vds graph,
but that's wrong, it describes the bottom (low Id, low Vgs, non-zero Vds).
Which explains why in the discussion on pp 131-132 you describe a transition
from subthreshold to saturation, without any linear in between.

I spent some time last night looking at figure 3.24, but since it shows
subthreshold being Id < 1uA I figured that didn't apply here. But you're
saying that for this large-area MOSFET, Id in the tens or hundreds of mA
still counts as subthreshold, that is, "where the channel is below the
threshold for conduction, but some current flows anyway because of a small
population of thermally energetic electrons"?

If so, then using your equation at 250mA I'd get gm = 250mA / n 25mV = 10/n
= from 1 to 3S. That would give me stage gain of G = gm * Rd = 100 to 300.
The gain of the differential stage, biased at 2mA when Vgs is 4.4V, is going
to be about Rc / 2re = 90. The resistive divider at the input divides by
3.9. So, 90 * 100 / 3.9 = 2300, I guess I can see how you get a gain of
2500. Is that the right way to do it?

Are there any good references beyond AoE that might have helped me answer
this question? (I've just been poking around IRF app notes but didn't see
anything relevant.)

Thanks,
-walter

 

[Fred Bloggs]

You're not likely to get a good answer there, unless perhaps someone
was looking at AoE figure 3.24, and they dramatically extrapolated
the curve for large-area FETs. As it happens, you're operating in
the MOSFET's subthreshold region, where its transconductance acts
like that of a BJT transistor, only lower by a fixed amount, i.e.,
g_m = Ic / n Vt, where Vt = 25mV. You know, 25 ohms at 1mA, and
proportional to current. I've observed the constant n range from 3
to 10 for different manufacturer's MOSFET types. It appears to have
about the same value for different FETs and part runs within a type.
There's no hint of the issue, or of its value on most datasheets.

Apparently the subthreshold current is modeled with a subthreshold slope
, S, which increases, as a constant, with FET channel leakage
characteristics. A commonly used fit is:
View in a fixed-width font such as Courier.

 

[Winfield Hill]

Fred Bloggs wrote...
I should add, I've measured many different MOSFETs and observed
this relationship to be accurate over seven orders of magnitude.

Apparently the subthreshold current is modeled with a subthreshold
slope, S, which increases, as a constant, with FET channel leakage
characteristics. A commonly used fit is:

. Vgs-Vt qVds
. (--------) -(----)
. S/ln(10) kT
. I = k x e x [ 1 - e ]
. sub

The notation is sloppy and the left most k is not Boltzman's
constant.

Reference please, Fred.

 

[Winfield Hill]

Winfield Hill wrote...

Fred Bloggs wrote...

I should add, I've measured many different MOSFETs and observed
this relationship to be accurate over seven orders of magnitude.

Well, five orders anyway. Drain leakage disturbs the fit on the
low end. Still, seven orders of fit is obtained if an additional
current is accounted for with a parallel leakage resistor, e.g.
5 Giga-ohms for a 1kV FET.

The transition to the saturation region, where Id = k(Vgs - Vth)^2
disturbs the fit on the high end. Looking at some data I've taken,
this starts at currents above 0.1% to 0.5% of the FET's maximum
operating current.

A 0.1% transition threshold is lower than I'd remembered, and may
put Walter's 250mA operation of an IRFP9240 well into this region,
reducing its g_m from my estimates. More detail in my answer to
Walter in this thread I plan for later today.

Apparently the subthreshold current is modeled with a subthreshold
slope, S, which increases, as a constant, with FET channel leakage
characteristics. A commonly used fit is:

. Vgs-Vt qVds
. (--------) -(----)
. S/ln(10) kT
. I = k x e x [ 1 - e ]
. sub

The notation is sloppy and the left most k is not Boltzman's
constant.

Reference please, Fred.

The first term isn't the exactly form I'd use, but it's not
in disagreement with my g_m formula. The second term dealing
with drain voltage, appears practically irrelevant over most
of the subthreshold region, where the drain current is set by
Vgs, and isn't much affected by Vds, for Vds > 200mV, etc.

 

[Fred Bloggs]

Fred Bloggs wrote...

Winfield Hill wrote:

I should add, I've measured many different MOSFETs and observed
this relationship to be accurate over seven orders of magnitude.

Apparently the subthreshold current is modeled with a subthreshold
slope, S, which increases, as a constant, with FET channel leakage
characteristics. A commonly used fit is:

. Vgs-Vt qVds
. (--------) -(----)
. S/ln(10) kT
. I = k x e x [ 1 - e ]
. sub

The notation is sloppy and the left most k is not Boltzman's
constant.

Reference please, Fred.

The best I can do is two levels of indirection removed from the
original. My original is FPGA-Based System Design by Wolf who references
Low-Power CMOS VLSI Circuit Design by Roy and Prasad, both are books.

 

[Winfield Hill]

Fred Bloggs wrote...

Fred Bloggs wrote...

Apparently the subthreshold current is modeled with a subthreshold
slope, S, which increases, as a constant, with FET channel leakage
characteristics. A commonly used fit is:

. Vgs-Vt qVds
. (--------) -(----)
. S/ln(10) kT
. I = k x e x [ 1 - e ]
. sub

The notation is sloppy and the left most k is not Boltzman's
constant.

Reference please, Fred.

The best I can do is two levels of indirection removed from the
original. My original is FPGA-Based System Design by Wolf who references
Low-Power CMOS VLSI Circuit Design by Roy and Prasad, both are books.

Thanks, I find that S/ln(10) term less helpful than my "n V_T"
term, but either form has a huge user-adjusted fudge factor.

 

[Tony Williams]

~In article <[email protected]>,

The best I can do is two levels of indirection removed from the
original. My original is FPGA-Based System Design by Wolf who
references Low-Power CMOS VLSI Circuit Design by Roy and Prasad,
both are books.

Try also.

http://ece.colorado.edu/~book/book/book/chapter7/ch7_7.htm#7_7_4

Impressive-looking tome on transistors physics......