Circuit for producing 10ns pulses of several amps

[Mr.CRC]

Hi:

I have been using MOSFET drivers to pulse LEDs at currents of up to 21A
(for 100s of ns to several microsecond pulses) and down to about 22ns
for 1A pulses into 1mm^2 power LEDs.

I can't get any faster with the drivers I've tried than about 20ns for
"parity" optical output power with the 1.0A CW max current typical of
blue 1mm^2 LEDs. (see note 1 below)


I wish to achieve 10-20ns pulses of 1-10 amps.


Three circuits that come to mind are:

1. Capacitive discharge by MOSFET switch such as the Directed Energy
PCO-7110 driver.

This circuit has the drawback of a slow trailing edge.

2. Discontinuous current mode flyback circuit. The stored current in
the inductor is switched into the LED when the MOSFET turns off.

This circuit also has a slower tail than a symmetric drive, but is
better than RC. I have gotten 30ns or so 3-4A pulses in a LTspice sim,
with 100-1000pF in parallel with diode loads.

3. Continuous current mode flyback circuit. The stored current in the
inductor is switched into the LED when the MOSFET turns off, then
shunted back through the FET when it turns back on.

This circuit produces a nice sharp pulse. I have gotten 15ns or so 3-4A
pulses in a LTspice sim, with 100-1000pF in parallel with diode loads.

At this point I have no idea if the simulated performance can be
realized with a physical circuit.

Also, much of the challenge is in the MOSFET gate drive. Hence, I keep
coming back to the fact that if the gate driver is fast enough, just
hook the LED to it and be done!

I did buy some Directed Energy (IXYS) laser diode driver assemblies to
test, but haven't gotten to spend much time with them. I still want to
be able to build my own, to meet custom mechanical requirements.

Just wondering if there are any completely different approaches to think
about?

I'm aware of transmission line pulse generation approaches, and would
consider them. But that should be a last resort. Those still require a
fast switch. So it seems all of this boils down to "how to switch
on/off several amps in 5ns or less?"


NOTES:

1. So far I've tried TC4422A and IXD630. The IXD630 is better on
paper, but with real LEDs, the TC4422A outperforms in the <100ns regime.

The way to get it to work isn't very practical for anything but lab
experimentation anyway, since to get very short pulses, I have to just
"tickle" the switching threshold of the driver by varying the amplitude
of the input pulse. The actual input pulse duration hardly even matters
below about 60ns, so I set it to 40ns and then the output pulse width
becomes a function of the input amplitude. This also varies with supply
voltage, and horribly with temperature.

 

[Nico Coesel]

Hi:

I have been using MOSFET drivers to pulse LEDs at currents of up to 21A
(for 100s of ns to several microsecond pulses) and down to about 22ns
for 1A pulses into 1mm^2 power LEDs.

What kind of MOSFETs are you using? The lower the RDSon the slower due
to the enormous gate capacitance.

 

[Mr.CRC]

What kind of MOSFETs are you using? The lower the RDSon the slower due
to the enormous gate capacitance.

So far I haven't used any, except for a IRF530 in LTspice.

The physical circuits I've used are just the MOSFET gate drivers mentioned.

But yes, in principle I would expect that some trade off btw RDSon/Qg
and speed would apply.

The Directed Energy laser diode drivers are based on their RF MOSFETs
and RF MOSFET gate driver ICs.

I have one board that is tuned to make 15ns pulses up to 50A, the
PCO-7110. But it can only do a 0.01% duty cycle. This is worth playing
with to learn about the circuit, but ultimately I'll need 50kHz rep.
rate, so about 0.05% to 0.1% duty for 10-20ns pulses.

 

[Mr.CRC]

Fast-on only if you need a short delay... the transmission line
provides the "off"

Don't you mean the rise time, rather than "short delay?"

I've used mercury-wetted switches to generate nanosecond pulse
_widths_.

It would be interesting indeed to rig that up to do 50kHz pulse
repetition frequency.

 

[Mr.CRC]

remember it's been up before, did you ever try something like this?
http://www.ixysrf.com/pdf/driver_ics/deic420.pdf

-Lasse

I have all their datasheets. Some of those RF devices are difficult to
source, except in quantities of 50, for a total of about $1500.

They do make off-the-shelf boards for fast laser diode pulsing. I have
two of them, and will be testing them.

Probably the answer to my question is, to do what IXYS RF does. But I
still want to see if there are other ways...

 

[Nico Coesel]

So far I haven't used any, except for a IRF530 in LTspice.

That one is quite old.

The physical circuits I've used are just the MOSFET gate drivers mentioned.

But yes, in principle I would expect that some trade off btw RDSon/Qg
and speed would apply.

It does apply. Last year I've designed a forward converter. A MOSFET
with a 10 times lower RDSon was less efficient. IMHO the most
efficient (and therefore fastest) MOSFET for a particular application
barely meets the specs. It will have the lowest gate capacitance.

 

[Spehro Pefhany]

That one is quite old.


It does apply. Last year I've designed a forward converter. A MOSFET
with a 10 times lower RDSon was less efficient. IMHO the most
efficient (and therefore fastest) MOSFET for a particular application
barely meets the specs. It will have the lowest gate capacitance.

How about a GaN FET with matching driver?

eg. EPC1014/LM5113

Rather prototyping-unfriendly package on the FET.

 

[Mr.CRC]

That's nice, if you want to short out the LED quickly at the end of
the pulse, and you don't mind the continuous power supply needed to
keep the current flowing in the inductor.

Yes, that is Ok. I don't have to market it. So a low voltage/high
current CC lab supply would be fine for keeping the inductor full.

Yes, it can. 10-20 ns is fairly slow. Layout will need to be tight, as
the dI/dT will be big, and every nanohenry will hurt.

Yeah, it shouldn't be too hard. Not like your crazy ps stuff. Fat
traces 8-10mils away from a sheet of copper are inherently helpful
toward avoiding inductance.

I need to start stocking and playing with 2-sided bare FR4 material, and
prototyping this sort of thing. But where I work, the cost of just
running off a 4-layer manufactured proto is no big deal.

I will be able to learn a lot from studying the Directed Energy (IXYS)
assemblies, b4 I try to make my own.

That depends on the gate capacitance. You can use paralleled TinyLogic
gates as mosfet gate drivers, sub-ns edges with a few ohms equivalent
source drive.

Hmm, I wonder how they can compete with the RF MOSFET gate drivers, when
loaded with a real gate? Also, since they are lower voltage, will the
lower threshold NMOS needed be inherently slower than a higher threshold
device?

NL37WZ16US is three brutal buffers in a can, for 12 cents. I run them
at 6.5 volts and they seem happy.

There are some multi-amp mesfets and PHEMTS and GaN fets, which have
absurdly low gate capacitances compared to mosfets. The nice thing
about these parts is that the source is usually the substrate, so you
can get a low inductance source ground by soldering the tab to the
ground plane. But you should be able to get a few ns rise/fall from
mosfets driven by TinyLogic.

These new exotic devices (maybe not so new?) are something I'll have to
study and see if they are suitable.




Thanks for the input. Or was it output? Nah, it was feedback!

 

[Tim Williams]

Hmm, I wonder how they can compete with the RF MOSFET gate drivers, when
loaded with a real gate? Also, since they are lower voltage, will the
lower threshold NMOS needed be inherently slower than a higher threshold
device?

Yes. I once tried a simple buck converter with 5V logic supply (would've
saved the bother of a 12V analog supply). Performance was terrible: "logic
level" transistors aren't any different from regular devices. Silicon is
silicon, and expecting them to have the same performance at half the gate
drive is silly, even when doubled up, with double the drive. Changed to 12V
analog supply, well worth the effort.

The only things that work well at 5V (or less) are very small geometry FETs
(which might take 3V drive to switch the load at any current you want --
there are sub-miliohm devices available), but they only stand off 10V or so.

Better off rolling your own. I've made a discrete gate drive that does the
same thing as one of those IXYS/DEI drivers, except with half the switching
time, and orders of magnitude better availability. It's not hard to do,
really; just think "how can I slam this device on really hard, then slam it
off really hard too".

Tim

 

[legg]

Hi:

I have been using MOSFET drivers to pulse LEDs at currents of up to 21A
(for 100s of ns to several microsecond pulses) and down to about 22ns
for 1A pulses into 1mm^2 power LEDs.

I can't get any faster with the drivers I've tried than about 20ns for
"parity" optical output power with the 1.0A CW max current typical of
blue 1mm^2 LEDs. (see note 1 below)


I wish to achieve 10-20ns pulses of 1-10 amps.


Three circuits that come to mind are:

1. Capacitive discharge by MOSFET switch such as the Directed Energy
PCO-7110 driver.

This circuit has the drawback of a slow trailing edge.

2. Discontinuous current mode flyback circuit. The stored current in
the inductor is switched into the LED when the MOSFET turns off.

This circuit also has a slower tail than a symmetric drive, but is
better than RC. I have gotten 30ns or so 3-4A pulses in a LTspice sim,
with 100-1000pF in parallel with diode loads.

3. Continuous current mode flyback circuit. The stored current in the
inductor is switched into the LED when the MOSFET turns off, then
shunted back through the FET when it turns back on.

This circuit produces a nice sharp pulse. I have gotten 15ns or so 3-4A
pulses in a LTspice sim, with 100-1000pF in parallel with diode loads.

At this point I have no idea if the simulated performance can be
realized with a physical circuit.

Also, much of the challenge is in the MOSFET gate drive. Hence, I keep
coming back to the fact that if the gate driver is fast enough, just
hook the LED to it and be done!

I did buy some Directed Energy (IXYS) laser diode driver assemblies to
test, but haven't gotten to spend much time with them. I still want to
be able to build my own, to meet custom mechanical requirements.

Just wondering if there are any completely different approaches to think
about?

I'm aware of transmission line pulse generation approaches, and would
consider them. But that should be a last resort. Those still require a
fast switch. So it seems all of this boils down to "how to switch
on/off several amps in 5ns or less?"


NOTES:

1. So far I've tried TC4422A and IXD630. The IXD630 is better on
paper, but with real LEDs, the TC4422A outperforms in the <100ns regime.

The way to get it to work isn't very practical for anything but lab
experimentation anyway, since to get very short pulses, I have to just
"tickle" the switching threshold of the driver by varying the amplitude
of the input pulse. The actual input pulse duration hardly even matters
below about 60ns, so I set it to 40ns and then the output pulse width
becomes a function of the input amplitude. This also varies with supply
voltage, and horribly with temperature.

Some fairly short, high current pulses can be harvested as
cross-conduction phenomena in (unloaded) gate drive circuits - their
periods being equivalent to the rise or fall time of the pre-driver
output.

Problems with this include stray inductance in the current path,
minimum capacitive loading of the pre-driver, headroom/gate threshold
relationships and techniques in tailoring the current fall time. Using
discrete parts, the cross conduction can be intentionally manipulated.

RL

 

[Fred Abse]

have been using MOSFET drivers to pulse LEDs at currents of up to 21A
(for 100s of ns to several microsecond pulses) and down to about 22ns for
1A pulses into 1mm^2 power LEDs.

I can't get any faster with the drivers I've tried than about 20ns for
"parity" optical output power with the 1.0A CW max current typical of blue
1mm^2 LEDs. (see note 1 below)


I wish to achieve 10-20ns pulses of 1-10 amps.

Avalanche transistor?

 

[Fred Abse]

It would be interesting indeed to rig that up to do 50kHz pulse repetition
frequency.

Tektronix did a pulse generator that did just that.

 

[qrk]

Hi:

I have been using MOSFET drivers to pulse LEDs at currents of up to 21A
(for 100s of ns to several microsecond pulses) and down to about 22ns
for 1A pulses into 1mm^2 power LEDs.

I can't get any faster with the drivers I've tried than about 20ns for
"parity" optical output power with the 1.0A CW max current typical of
blue 1mm^2 LEDs. (see note 1 below)


I wish to achieve 10-20ns pulses of 1-10 amps.


Three circuits that come to mind are:

1. Capacitive discharge by MOSFET switch such as the Directed Energy
PCO-7110 driver.

This circuit has the drawback of a slow trailing edge.

2. Discontinuous current mode flyback circuit. The stored current in
the inductor is switched into the LED when the MOSFET turns off.

This circuit also has a slower tail than a symmetric drive, but is
better than RC. I have gotten 30ns or so 3-4A pulses in a LTspice sim,
with 100-1000pF in parallel with diode loads.

3. Continuous current mode flyback circuit. The stored current in the
inductor is switched into the LED when the MOSFET turns off, then
shunted back through the FET when it turns back on.

This circuit produces a nice sharp pulse. I have gotten 15ns or so 3-4A
pulses in a LTspice sim, with 100-1000pF in parallel with diode loads.

At this point I have no idea if the simulated performance can be
realized with a physical circuit.

Also, much of the challenge is in the MOSFET gate drive. Hence, I keep
coming back to the fact that if the gate driver is fast enough, just
hook the LED to it and be done!

I did buy some Directed Energy (IXYS) laser diode driver assemblies to
test, but haven't gotten to spend much time with them. I still want to
be able to build my own, to meet custom mechanical requirements.

Just wondering if there are any completely different approaches to think
about?

I'm aware of transmission line pulse generation approaches, and would
consider them. But that should be a last resort. Those still require a
fast switch. So it seems all of this boils down to "how to switch
on/off several amps in 5ns or less?"


NOTES:

1. So far I've tried TC4422A and IXD630. The IXD630 is better on
paper, but with real LEDs, the TC4422A outperforms in the <100ns regime.

The way to get it to work isn't very practical for anything but lab
experimentation anyway, since to get very short pulses, I have to just
"tickle" the switching threshold of the driver by varying the amplitude
of the input pulse. The actual input pulse duration hardly even matters
below about 60ns, so I set it to 40ns and then the output pulse width
becomes a function of the input amplitude. This also varies with supply
voltage, and horribly with temperature.

Charge up a capacitor, and use a triac to dump the charge through the
load. You'll get fast rise time and a wimpy trailing edge. LEDs look
like a resistor at high currents. It's amazing how much current they
can take.

 

[Mr.CRC]

Charge up a capacitor, and use a triac to dump the charge through the
load. You'll get fast rise time and a wimpy trailing edge. LEDs look
like a resistor at high currents. It's amazing how much current they
can take.

Triac?!? You think it can do <4ns risetime?

Interesting concept. I'll keep it in mind.

 

[Jasen Betts]

Triac?!? You think it can do <4ns risetime?

closer to 4us I think.

Thyratron perhaps? Spark gaps are fast, right?

perhaps build your own thyristor from some UHF transistors?

⚂⚃ 100% natural

 

[Tim Williams]

closer to 4us I think.

Thyratron perhaps? Spark gaps are fast, right?

perhaps build your own thyristor from some UHF transistors?

Run of the mill xenon or mercury thyratrons are typically slower than their
silicon counterparts, with comparable risetime and deionization time in the
ms. Hydrogen thyratrons are well renouned for their speed, however.

As circuits go with high current, fast risetime, and dreadful tails, there's
this one. You still need a pulse generator, but that's easier than the
current gain. Some TinyLogic would take care of that.
http://cds.linear.com/docs/Application Note/an122f.pdf
'Course, Jim would've been about the only person on Earth with a stash of
stud type BJTs. YMMV.

Tim

 

[mike]

Triac?!? You think it can do<4ns risetime?

Interesting concept. I'll keep it in mind.

What's the actual specification of the pulse you're trying to generate?
This is the first mention I've seen of 4ns risetime.
falltime?
How variable does the width need to be.
50KHz rep rate was mentioned, but not sure if the context was relevant.
Do you care about jitter?
How about life of the apparatus.
Both are relevant for relays.
Is it a current pulse or a voltage pulse?
And what is required of the other? voltage compliance for current pulse
or peak current available for voltage pulse?
Does delay matter? Maybe something like a distributed amplifier
made from smaller/faster devices.

I didn't catch whether this is a test fixture or a high volume
device where cost matters a lot.


What's the actual specification of the pulse you're trying to generate
into exactly what range of loads. LED is an ambiguous definition.
The devil is in the details.

You can do some interesting things with avalanche transistors and/or
snap diodes. I've never tried it at this power level, but you don't
need much speed.

 

[Uwe Hercksen]

I've used mercury-wetted switches to generate nanosecond pulse
_widths_.

Hello,

you closed and opened the same switch within some nanoseconds?
Or did you close one switch and open another one within some nanoseconds?

Bye

 

[legg]

On Fri, 20 Jul 2012 10:52:32 -0700, "Mr.CRC"

Some fairly short, high current pulses can be harvested as
cross-conduction phenomena in (unloaded) gate drive circuits - their
periods being equivalent to the rise or fall time of the pre-driver
output.

Problems with this include stray inductance in the current path,
minimum capacitive loading of the pre-driver, headroom/gate threshold
relationships and techniques in tailoring the current fall time. Using
discrete parts, the cross conduction can be intentionally manipulated.

Talking to myself, it seems.

A quick LTspice simulation follows, using the cross-conduction alone.

RL


Version 4
SHEET 1 880 680
WIRE 192 64 64 64
WIRE 512 64 192 64
WIRE 16 80 -48 80
WIRE 512 160 512 64
WIRE -176 176 -192 176
WIRE -48 176 -48 80
WIRE -48 176 -96 176
WIRE 64 176 64 160
WIRE -192 240 -192 176
WIRE 64 256 64 240
WIRE 64 352 64 336
WIRE 192 352 192 128
WIRE 192 352 64 352
WIRE -48 432 -48 176
WIRE 16 432 -48 432
WIRE -192 480 -192 320
WIRE 64 480 64 448
WIRE 64 480 -192 480
WIRE 432 480 64 480
WIRE 512 480 512 240
WIRE 512 480 432 480
WIRE 432 496 432 480
FLAG 432 496 0
SYMBOL LED 48 176 R0
WINDOW 0 40 -8 Left 2
WINDOW 3 28 65 Left 2
SYMATTR InstName D1
SYMATTR Value QTLP690C
SYMATTR Description Diode
SYMATTR Type diode
SYMBOL ind 48 240 R0
WINDOW 0 58 38 Left 2
WINDOW 3 48 70 Left 2
SYMATTR InstName L1
SYMATTR Value 5E-8
SYMBOL cap 176 64 R0
WINDOW 0 42 18 Left 2
WINDOW 3 23 53 Left 2
SYMATTR InstName C1
SYMATTR Value 300E-12
SYMBOL res -80 160 R90
WINDOW 0 0 56 VBottom 2
WINDOW 3 32 56 VTop 2
SYMATTR InstName R1
SYMATTR Value 10R
SYMBOL nmos 16 352 R0
WINDOW 0 87 27 Left 2
WINDOW 3 57 56 Left 2
SYMATTR InstName M1
SYMATTR Value FDS6961A
SYMBOL pmos 16 160 M180
WINDOW 0 82 68 Left 2
WINDOW 3 53 40 Left 2
SYMATTR InstName M2
SYMATTR Value Si9400DY
SYMBOL voltage 512 144 R0
WINDOW 123 0 0 Left 2
WINDOW 39 24 132 Left 2
SYMATTR SpiceLine Rser=.01
SYMATTR InstName V1
SYMATTR Value 10
SYMBOL voltage -192 224 R0
WINDOW 0 -36 112 Left 2
WINDOW 3 -52 285 Left 2
WINDOW 123 0 0 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V2
SYMATTR Value PULSE(10V 0V 1E-6 1E-8 1E-8 1E-5 2E-5 10)
TEXT -248 528 Left 2 !.tran 0 11.2E-6 11E-6 1E-9 uic

 

[Uwe Hercksen]

You do it by connecting a charged, open-circuited coax stub via the
relay. The coax empties itself out in a time t = 2L/sqrt(epsilon).

Hello,

ah, the length of the pulse is determined by the coax stub only.
Pulses of some 100 ps should be possible too.
But if you want periodic pulses with a period of some ms I would expect
a jitter of some 100 ns even when the relay is driven with a good XTAL
based clock.

Cheers