400V regulator with a FET and an LM317L chip

[Winfield Hill]

Question: Did anyone ever build the high-voltage requlator
circuit, using a FET and an LM317L, that I posted here a few
years ago? I have just reposted it in answer to a quesiton
on sci.electronics.basics, and wonder if it works properly.

Group: sci.electronics.basics
Title: Re: Transistor regulators?
Author: Winfield Hill
Date: 22 Nov 2003 11:17:29 -0800

John Popelish wrote...

Field effect transistors are probably your best bet for this service.
They are commonly available with voltage ratings up to about 1000V and
power dissipation ratings high enough to survive a short circuit while
current limiting, if they are mounted on a large enough heat sink.

From a post of mine made a few years ago: ... a simple
high-voltage regulator that uses standard LM317 or LM317L
chips plus an easy-to-get reliable high-voltage MOSFET.

600V FET _____ LM317L 5 to 500V
IN o----+----, ,-+-----+--| |--+---+----+---o OUT
| | | | s | | | | | | 2 to 100mA
| _|_V_|_ | |_adj_| 1.2k | |
2.2M ----, | | | | | 0.47uF
1/2 W | '-||--+-----+ | ===
| | , 68pF | | | 630V
'--------+---|<|-----------|---' |
' 10V zener | 2.7 ohms
500k |
pot gnd
|
gnd

The FET needs a very serious heat sink, because 450V at
100mA is 45 watts, and under a short circuit condition the
LM317L may allow even more current to flow. You can add
a single-transistor foldback current limit if you want.

Add a 1n4002 diode backwards across the LM317 if you plan
on ever shorting out the input filter-storage capacitor.
(Always use drain resistors with high-voltage capacitors.
And always approach the open circuit with one hand behind
your back.)

The LM317L needs as much as 2.5mA to operate, and the 1.2k
resistor only takes 1mA, so this sets a 1.5mA minimum load
requirement. (Note, you could use 470-ohms instead of 1.2k,
but this would require a 200k pot with 1W of dissipation,
instead of the 0.4W maximum dissipated in the 500k pot when
it's set to about 400k for +400V out.)

---

Note: I haven't tried this circuit, since I prefer to use
HIP5600 and VB408, etc., high-voltage regulator ICs.

Thanks,
- Win

whill_at_picovolt-dot-com

 

[Robert Baer]

Question: Did anyone ever build the high-voltage requlator
circuit, using a FET and an LM317L, that I posted here a few
years ago? I have just reposted it in answer to a quesiton
on sci.electronics.basics, and wonder if it works properly.

Group: sci.electronics.basics
Title: Re: Transistor regulators?
Author: Winfield Hill
Date: 22 Nov 2003 11:17:29 -0800

John Popelish wrote...

From a post of mine made a few years ago: ... a simple
high-voltage regulator that uses standard LM317 or LM317L
chips plus an easy-to-get reliable high-voltage MOSFET.

600V FET _____ LM317L 5 to 500V
IN o----+----, ,-+-----+--| |--+---+----+---o OUT
| | | | s | | | | | | 2 to 100mA
| _|_V_|_ | |_adj_| 1.2k | |
2.2M ----, | | | | | 0.47uF
1/2 W | '-||--+-----+ | ===
| | , 68pF | | | 630V
'--------+---|<|-----------|---' |
' 10V zener | 2.7 ohms
500k |
pot gnd
|
gnd

The FET needs a very serious heat sink, because 450V at
100mA is 45 watts, and under a short circuit condition the
LM317L may allow even more current to flow. You can add
a single-transistor foldback current limit if you want.

Add a 1n4002 diode backwards across the LM317 if you plan
on ever shorting out the input filter-storage capacitor.
(Always use drain resistors with high-voltage capacitors.
And always approach the open circuit with one hand behind
your back.)

The LM317L needs as much as 2.5mA to operate, and the 1.2k
resistor only takes 1mA, so this sets a 1.5mA minimum load
requirement. (Note, you could use 470-ohms instead of 1.2k,
but this would require a 200k pot with 1W of dissipation,
instead of the 0.4W maximum dissipated in the 500k pot when
it's set to about 400k for +400V out.)

---

Note: I haven't tried this circuit, since I prefer to use
HIP5600 and VB408, etc., high-voltage regulator ICs.

Thanks,
- Win

whill_at_picovolt-dot-com

Another possible objection is that zener; at low currents (where
output voltage is high and input voltage is low): zeners at 6V and
higher tend to have negative resistance and oscillate.
From 1nA to about 5mA, the B-E junction of the ST Micro 2N2369A
behaves very well; if one uses the E-C biased so the B-E is reverse and
the B-C is forward, it is rather low TC over temperature - even to 185C
(see my posting for a "high voltage" reference i made with discretes).
BTW, check out the LR-8 by Supertex.

 

[Winfield Hill]

Robert Baer wrote...

Another possible objection is that zener; at low currents (where
output voltage is high and input voltage is low): zeners at 6V
and higher tend to have negative resistance and oscillate.
From 1nA to about 5mA, the B-E junction of the ST Micro 2N2369A
behaves very well; if one uses the E-C biased so the B-E is reverse
and the B-C is forward, it is rather low TC over temperature - even
to 185C (see my posting for a "high voltage" reference i made
with discretes).

[Robert, I looked, but couldn't find the posting you mentioned.]

The effect you mention for zener diodes is very small and of no
concern in this application. I know, having spent far too much
time researching the issue (although the measurements were fun).

I do agree that reverse Veb junctions work well at low currents.
But a persistent problem is reverse-voltage predictability, this
can vary from 5.5V to 12V for different manufacturer's parts with
the same type number. Although a given manufacturer's part, like
the ST 2N2369A you mention, may show consistency over production
lots for a few years, there's always the worrisome issue of a
change in the FAB line and in the part's characteristics. While
it's true that many manufacturers announce such changes, keeping
up with such announcements is one more chore in a busy life. In
the old days we could buy the F16H1, a 5.2V Veb zener in TO-92.

In my circuit the LM317L needs up to 2V of overhead, and the FET
could require as much as 3.5V of gate voltage at low currents, so
we barely scrape under the wire with a worst-case 5.5V transistor.
We would pass under, but I'm nervous.

BTW, check out the LR-8 by Supertex.

Indeed. Although it is hard to get and a bit on the wimpy side.
High-voltage regulators should come in TO-220 cases or better.

Thanks,
- Win

whill_at_picovolt-dot-com

 

[Fred Bloggs]

Robert Baer wrote...

Another possible objection is that zener; at low currents (where
output voltage is high and input voltage is low): zeners at 6V
and higher tend to have negative resistance and oscillate.
From 1nA to about 5mA, the B-E junction of the ST Micro 2N2369A
behaves very well; if one uses the E-C biased so the B-E is reverse
and the B-C is forward, it is rather low TC over temperature - even
to 185C (see my posting for a "high voltage" reference i made
with discretes).

[Robert, I looked, but couldn't find the posting you mentioned.]

The effect you mention for zener diodes is very small and of no
concern in this application. I know, having spent far too much
time researching the issue (although the measurements were fun).

I do agree that reverse Veb junctions work well at low currents.
But a persistent problem is reverse-voltage predictability, this
can vary from 5.5V to 12V for different manufacturer's parts with
the same type number. Although a given manufacturer's part, like
the ST 2N2369A you mention, may show consistency over production
lots for a few years, there's always the worrisome issue of a
change in the FAB line and in the part's characteristics. While
it's true that many manufacturers announce such changes, keeping
up with such announcements is one more chore in a busy life. In
the old days we could buy the F16H1, a 5.2V Veb zener in TO-92.

In my circuit the LM317L needs up to 2V of overhead, and the FET
could require as much as 3.5V of gate voltage at low currents, so
we barely scrape under the wire with a worst-case 5.5V transistor.
We would pass under, but I'm nervous.

BTW, check out the LR-8 by Supertex.

Indeed. Although it is hard to get and a bit on the wimpy side.
High-voltage regulators should come in TO-220 cases or better.

Thanks,
- Win

whill_at_picovolt-dot-com

He does have a point about that zener- seems like an enormous waste of
overhead to bring that up to speed- A boosted LM185 with constant
current bias would be better- even a 100V in/out differential is 50uA
where the zener may be hovering around 0.8 x Vz.

 

[Winfield Hill]

Fred Bloggs wrote...

He does have a point about that zener- seems like an enormous waste
of overhead to bring that up to speed- A boosted LM185 with constant
current bias would be better- even a 100V in/out differential is
50uA where the zener may be hovering around 0.8 x Vz.

I'm puzzled by your comment. A standard 10V zener is perfectly
happy running at 50uA, 10uA or even less, and it doesn't drop to
80% of its rated voltage, if that's what you mean, but even if it
did that wouldn't matter... Let's put some meat on my circuit
configured as a 400V regulator. We'll assume the raw Vdc-IN can
range from 450 to 600V. That means Z1's current may be between
18 and 270uA for outputs from 0 to 400V. What's wrong with that?
Using an LM385-adj circuit in place of Z1 to make a 7.5V drop,
etc., wouldn't make any material difference so far as I can see.

If on the other hand you imagine replacing R1 with a constant-
current source, fine, but we'll need another 600V FET and we'll
have to suffer its potentially high Coss capacitance at a high-
impedance circuit node.

.. simple 400V 100mA regulator using an LM317
.. Vdc-IN 10V
.. 450 to ,----|<|-----,
.. 500V Q1, 600V FET | _____ | 2 to 400V
.. o---+----, ,-+-----++--| LM |--++--+----+--o Vdc-OUT
.. | | | | s | |317L | | | | 0 to 100mA
.. R1 _|_V_|_ | |_adj_| 294 | |
.. 2.2M ----, | | | | | 0.47uF
.. 1/2 W | '--||--+-----+ | ===
.. | | , 68pF | | | 630V
.. '--------+---|<|------------|---' |
.. ' 10V zener | 2.7 ohms
.. Z1 100k |
.. 2W pot gnd
.. |
.. gnd

Speaking of FETs and capacitance, let's consider Q1. With a
maximum 100mA output at 0V out, and with 600V in, we see that
Q1 may have to dissipate up to 60W. That means we need a FET
with large die surface area. One candidate I happen to stock
is Infineon's SPW47N60C2. Although it's monster rated for 30A
at 100C, we don't notice that; it's the low 0.3 C/W thermal
resistance that gets our attention.

High capacitance is one price to pay for using a large-die FET:
the '47N60 has Ciss = 7000pF and Coss = 400pF max, with Vdss
above 50V. Let's try to evaluate its impact on the design.

A high Ciss means that Q1 looks like a constant-current source
to the LM317 for short time scales, which could mean poor load
transient response, at least for step load decreases. The '317
regulator has no direct way to reduce Vgs when needed, all it
can do is open up and wait for the FET drain current to charge
Ciss until it drops to the level U1 wants. For example, if the
current step is down from 100mA, we'll see 100mA through Z1 for
roughly t = dV C / I = 70ns for a -1V Vgs change. Hey, that's
not a problem, as it's faster than the LM317L response. (BTW,
the 100mA current goes through Z1.) Considering a low-current
worst case, e.g. a step down from say 4mA to the 2mA quiescent
level, we get 0.35us (for a change in Vgs of say 0.2V), which
isn't any problem either. No doubt the output capacitor will
be in control of any such events.

So we see a high capacitance for Q1 isn't bad, but I'm not so
sure we could get away with a high capacitance in place of R1.
Hmm, one nice possibility might be my favorite Supertex LND150
depletion-mode 500V MOSFET, TO-92 package, with Coss under 2pF.
But we'd have to limit Vdc-IN to 510V max.

Thanks,
- Win

whill_at_picovolt-dot-com

 

[Robert Baer]

Robert Baer wrote...

Another possible objection is that zener; at low currents (where
output voltage is high and input voltage is low): zeners at 6V
and higher tend to have negative resistance and oscillate.
From 1nA to about 5mA, the B-E junction of the ST Micro 2N2369A
behaves very well; if one uses the E-C biased so the B-E is reverse
and the B-C is forward, it is rather low TC over temperature - even
to 185C (see my posting for a "high voltage" reference i made
with discretes).

[Robert, I looked, but couldn't find the posting you mentioned.]

The effect you mention for zener diodes is very small and of no
concern in this application. I know, having spent far too much
time researching the issue (although the measurements were fun).

I do agree that reverse Veb junctions work well at low currents.
But a persistent problem is reverse-voltage predictability, this
can vary from 5.5V to 12V for different manufacturer's parts with
the same type number. Although a given manufacturer's part, like
the ST 2N2369A you mention, may show consistency over production
lots for a few years, there's always the worrisome issue of a
change in the FAB line and in the part's characteristics. While
it's true that many manufacturers announce such changes, keeping
up with such announcements is one more chore in a busy life. In
the old days we could buy the F16H1, a 5.2V Veb zener in TO-92.

In my circuit the LM317L needs up to 2V of overhead, and the FET
could require as much as 3.5V of gate voltage at low currents, so
we barely scrape under the wire with a worst-case 5.5V transistor.
We would pass under, but I'm nervous.

BTW, check out the LR-8 by Supertex.

Indeed. Although it is hard to get and a bit on the wimpy side.
High-voltage regulators should come in TO-220 cases or better.

Thanks,
- Win

whill_at_picovolt-dot-com

Sorry; i had the thought that this was a "onesies" type of project.
However, the ST 2N2369A is cheap and one could buy a few thousand,
sample test one for noise, and be done. They have been remarkeably
consistent.
I have found that almost any small-signal NPN bipolar has a reverse
B-E breakdown near 6V, and the specific V/I characteristics are almost
always negative resistance, oscillations, noise *and* voltage consistent
(varies little) for many years, for a given manufacturer.
Anyway, the resistor feeding the gate of the pass FET could be
replaced by a constant current source made with a DMOS FET: LND150 and a
100K resistor for about 14uA.
Likewise, the LM317L could be replaced with the LM285 (adjustable) for
lower standby current.
BTW, the ST 2N2369A used as mentioned gives 6.6V.
***
The posting i mentioned was Wednesday, "Towards a 'high voltage'
reference for hi temp work".

 

[Robert Baer]

Fred Bloggs wrote...

I'm puzzled by your comment. A standard 10V zener is perfectly
happy running at 50uA, 10uA or even less, and it doesn't drop to
80% of its rated voltage, if that's what you mean, but even if it
did that wouldn't matter... Let's put some meat on my circuit
configured as a 400V regulator. We'll assume the raw Vdc-IN can
range from 450 to 600V. That means Z1's current may be between
18 and 270uA for outputs from 0 to 400V. What's wrong with that?
Using an LM385-adj circuit in place of Z1 to make a 7.5V drop,
etc., wouldn't make any material difference so far as I can see.

If on the other hand you imagine replacing R1 with a constant-
current source, fine, but we'll need another 600V FET and we'll
have to suffer its potentially high Coss capacitance at a high-
impedance circuit node.

. simple 400V 100mA regulator using an LM317
. Vdc-IN 10V
. 450 to ,----|<|-----,
. 500V Q1, 600V FET | _____ | 2 to 400V
. o---+----, ,-+-----++--| LM |--++--+----+--o Vdc-OUT
. | | | | s | |317L | | | | 0 to 100mA
. R1 _|_V_|_ | |_adj_| 294 | |
. 2.2M ----, | | | | | 0.47uF
. 1/2 W | '--||--+-----+ | ===
. | | , 68pF | | | 630V
. '--------+---|<|------------|---' |
. ' 10V zener | 2.7 ohms
. Z1 100k |
. 2W pot gnd
. |
. gnd

Speaking of FETs and capacitance, let's consider Q1. With a
maximum 100mA output at 0V out, and with 600V in, we see that
Q1 may have to dissipate up to 60W. That means we need a FET
with large die surface area. One candidate I happen to stock
is Infineon's SPW47N60C2. Although it's monster rated for 30A
at 100C, we don't notice that; it's the low 0.3 C/W thermal
resistance that gets our attention.

High capacitance is one price to pay for using a large-die FET:
the '47N60 has Ciss = 7000pF and Coss = 400pF max, with Vdss
above 50V. Let's try to evaluate its impact on the design.

A high Ciss means that Q1 looks like a constant-current source
to the LM317 for short time scales, which could mean poor load
transient response, at least for step load decreases. The '317
regulator has no direct way to reduce Vgs when needed, all it
can do is open up and wait for the FET drain current to charge
Ciss until it drops to the level U1 wants. For example, if the
current step is down from 100mA, we'll see 100mA through Z1 for
roughly t = dV C / I = 70ns for a -1V Vgs change. Hey, that's
not a problem, as it's faster than the LM317L response. (BTW,
the 100mA current goes through Z1.) Considering a low-current
worst case, e.g. a step down from say 4mA to the 2mA quiescent
level, we get 0.35us (for a change in Vgs of say 0.2V), which
isn't any problem either. No doubt the output capacitor will
be in control of any such events.

So we see a high capacitance for Q1 isn't bad, but I'm not so
sure we could get away with a high capacitance in place of R1.
Hmm, one nice possibility might be my favorite Supertex LND150
depletion-mode 500V MOSFET, TO-92 package, with Coss under 2pF.
But we'd have to limit Vdc-IN to 510V max.

Thanks,
- Win

whill_at_picovolt-dot-com

You beat me to it on the LND150...
Anyway, there might be a problem in destroying the FET with a spike
voltage between the gate and the source.
In that case, either get a ST micro FET that has internal back-to-back
zener protection, or add one externally.
Be advised that this protection method can give serious troubles above
170C, especially at 185C (damn zeners and all other junctions leak like
a sieve).

 

[Robert Baer]

Robert Baer wrote...

Another possible objection is that zener; at low currents (where
output voltage is high and input voltage is low): zeners at 6V
and higher tend to have negative resistance and oscillate.
From 1nA to about 5mA, the B-E junction of the ST Micro 2N2369A
behaves very well; if one uses the E-C biased so the B-E is reverse
and the B-C is forward, it is rather low TC over temperature - even
to 185C (see my posting for a "high voltage" reference i made
with discretes).

[Robert, I looked, but couldn't find the posting you mentioned.]

The effect you mention for zener diodes is very small and of no
concern in this application. I know, having spent far too much
time researching the issue (although the measurements were fun).

I do agree that reverse Veb junctions work well at low currents.
But a persistent problem is reverse-voltage predictability, this
can vary from 5.5V to 12V for different manufacturer's parts with
the same type number. Although a given manufacturer's part, like
the ST 2N2369A you mention, may show consistency over production
lots for a few years, there's always the worrisome issue of a
change in the FAB line and in the part's characteristics. While
it's true that many manufacturers announce such changes, keeping
up with such announcements is one more chore in a busy life. In
the old days we could buy the F16H1, a 5.2V Veb zener in TO-92.

In my circuit the LM317L needs up to 2V of overhead, and the FET
could require as much as 3.5V of gate voltage at low currents, so
we barely scrape under the wire with a worst-case 5.5V transistor.
We would pass under, but I'm nervous.

BTW, check out the LR-8 by Supertex.

Indeed. Although it is hard to get and a bit on the wimpy side.
High-voltage regulators should come in TO-220 cases or better.

Thanks,
- Win

whill_at_picovolt-dot-com

Sorry; i had the thought that this was a "onesies" type of project.
However, the ST 2N2369A is cheap and one could buy a few thousand,
sample test one for noise, and be done. They have been remarkeably
consistent.
I have found that almost any small-signal NPN bipolar has a reverse
B-E breakdown near 6V, and the specific V/I characteristics are almost
always negative resistance, oscillations, noise *and* voltage consistent
(varies little) for many years, for a given manufacturer.
Anyway, the resistor feeding the gate of the pass FET could be
replaced by a constant current source made with a DMOS FET: LND150 and a
100K resistor for about 14uA.
Likewise, the LM317L could be replaced with the LM285 (adjustable) for
lower standby current.
BTW, the ST 2N2369A used as mentioned gives 6.6V.
***
The posting i mentioned was Wednesday, "Towards a 'high voltage'
reference for hi temp work".

OOPS! *wrong* control; the LM285 will not do here.
Use the ST Micro variant: the TS431.