Fast power LED driving

[Mr.CRC]

Hi:

I'm trying to get less than 100ns pulses at >10A, preferably up to
20-30A for 10-50ns into Cree XP-E colored LEDs.

My first attempt at fast pulsing these has been to parallel 4 of TC4422A
MOSFET drivers.

With 18V power supply, I can get 10-12A pulses. Below about 100ns they
can't produce full output amplitude. I can get to about 50ns pulses
from the LED, but with much weaker output than >100ns.


My second approach has been to use a flyback topology, with the LED tied
across the inductor in the drain circuit of a NMOS. The LED gets the
inductor current driven into it when the NMOS turns off.

I have simulated both CCM and DCM. It seems CCM works better, giving
nicer shaped pulses than letting all the inductor energy dump in the LED.

But it seems with this circuit, the main challenge to getting it to work
in reality is to drive the NMOS gate fast enough. Also, level shifting
from a 5V logic level trigger input to the gate driver's input is
needed. I have been looking into a simple complimentary BJT push-pull
pair, but not sure what to do about level translation at speed.


Thanks for any comments.


I'm reading with interest the thread "Super duper hype fast FET driver?"


Maybe I'll give Mr. Larkin a call Monday...


Good day!

 

[John Devereux]

Fun! How much capacitance do the LEDs have?


That's limited in the speed it can slam the LEDs on, since the current
in the L is all the current that's available. OTOH, the LED turnoff
will be hard.


If cost is not a big deal, look at the IXYS/DEI gate driver chips.
Heck, they may be able to drive the LEDs all by themselves.
Interesting packaging.

The people who sell mega-laser driver boxes tend to use special
flex/foil transmission lines to connect to the loads, just a few ohms
impedance. A 50 ohm line, or a 100 ohm twisted pair, can make trouble
at these speeds and currents.


Please do, preferably later in the day. I have a cute and cheap way to
drive mosfet gates really fast and hard. I'd post it publically, but
Thompson would steal it, Fields would whine about what it really
means, and Sloman would use it as an excuse to bombast about climate
change.

A pity, I'd be interested to see it.

Can't you do it just with a complimentary push-pull stage? And a one of
the tinylogix gates as a driver perhaps.

 

[Spehro Pefhany]

Hi:

I'm trying to get less than 100ns pulses at >10A, preferably up to
20-30A for 10-50ns into Cree XP-E colored LEDs.

My first attempt at fast pulsing these has been to parallel 4 of TC4422A
MOSFET drivers.

With 18V power supply, I can get 10-12A pulses. Below about 100ns they
can't produce full output amplitude. I can get to about 50ns pulses
from the LED, but with much weaker output than >100ns.


My second approach has been to use a flyback topology, with the LED tied
across the inductor in the drain circuit of a NMOS. The LED gets the
inductor current driven into it when the NMOS turns off.

I have simulated both CCM and DCM. It seems CCM works better, giving
nicer shaped pulses than letting all the inductor energy dump in the LED.

But it seems with this circuit, the main challenge to getting it to work
in reality is to drive the NMOS gate fast enough. Also, level shifting
from a 5V logic level trigger input to the gate driver's input is
needed. I have been looking into a simple complimentary BJT push-pull
pair, but not sure what to do about level translation at speed.


Thanks for any comments.


I'm reading with interest the thread "Super duper hype fast FET driver?"


Maybe I'll give Mr. Larkin a call Monday...


Good day!

Have you done any reliability studies on running that kind of pulsed
current into LEDs?


Best regards,
Spehro Pefhany

 

[Jamie]

Hi:

I'm trying to get less than 100ns pulses at >10A, preferably up to
20-30A for 10-50ns into Cree XP-E colored LEDs.

My first attempt at fast pulsing these has been to parallel 4 of TC4422A
MOSFET drivers.

With 18V power supply, I can get 10-12A pulses. Below about 100ns they
can't produce full output amplitude. I can get to about 50ns pulses
from the LED, but with much weaker output than >100ns.


My second approach has been to use a flyback topology, with the LED tied
across the inductor in the drain circuit of a NMOS. The LED gets the
inductor current driven into it when the NMOS turns off.

I have simulated both CCM and DCM. It seems CCM works better, giving
nicer shaped pulses than letting all the inductor energy dump in the LED.

But it seems with this circuit, the main challenge to getting it to work
in reality is to drive the NMOS gate fast enough. Also, level shifting
from a 5V logic level trigger input to the gate driver's input is
needed. I have been looking into a simple complimentary BJT push-pull
pair, but not sure what to do about level translation at speed.


Thanks for any comments.


I'm reading with interest the thread "Super duper hype fast FET driver?"


Maybe I'll give Mr. Larkin a call Monday...


Good day!

Hmm, where do you find LED's that can handle that current? The best I've
found is 1000ma in the CREE XP-E line.


Jamie

 

[Mr.CRC]

Have you done any reliability studies on running that kind of pulsed
current into LEDs?


Best regards,
Spehro Pefhany

I have not done a systematic study of this, but have begun considering
how I might go about it.

My limited experience based on bench testing and deploying about 4-5 LED
plus driver assemblies based on the TC4422A to several laboratories
doing high speed Schlieren imaging is that:

I have never blown an LED with 1us or less pulses at up to 10-12A.

I have put nearly the rated average thermal load of about 5W into the
LEDs with continuous pulses in the 6-9A range and up to 10us, and they
live (for at least minutes-hours).

I have blown up several blue LEDs packaged by LEDengin (I believe Cree
die, but not sure) when pulsing over 100us at high currents, perhaps 5-6A.


I have recently acquired a bunch of top bin Cree XP-E green, red, and
royal blue emitters. Also some Lumileds Rebels in cyan and royal blue.

My first thought at how to test them would be to normalize for average
power. For ex., at 1W of average power, ramp up current for a given
pulse duration (while simultaneously reducing duty cycle to keep power
constant) then see when it breaks. How to present the data I haven't
quite worked out.

Also, some long term experiments are in order, to see if at some peak
current level and thermal load, does the output gradually drop rather
than catastrophic failure. If so, can time constants be derived.

One major problem with this kind of testing is that the next batch of
dice from the same maker may behave quite different. Even for a given
experiment, it would probably be necessary to break at least a small
handful at each parameter set to get some stats.

Also, this testing would best be performed with a scientific grade
pulser, like the ones used for large laser diode bars. I think some
colleagues have these, so I could try to borrow one.

But my present setup is designed to be cheap and quick to get something
running in the lab.

I'm not sure if my manager will go for having me make this into a
research project to characterize LEDs. At least they gave me the
freedom to have developed the drivers somewhat in the background, and
now that they are popular, I can spend more time on that part.


It is remarkable that the radiance of these LEDs is now competitive with
a Xe short arc lamp, over narrow wavelength bands.

We had previously used Xe short arcs for Schlieren rather than a laser,
because the laser's coherence and thus speckle degrades image quality.
But the short arc's white light causes severe chromatic aberration in
our optical engines with quartz cylinder liners. Two options are to put
a bandpass filter in front of the Xe lamp, or find another light source.

The BP filters have only about 50% throughput. Thus, even though the Xe
radiance is much higher over its full spectrum, or even just the
visible, the LED actually wins out over a narrow range of 10-30nm!

Plus the LED can be pulsed at very high rep rates. I can get 5MHz with
full modulation from the TC4422As. Though we are doing only about
50-100kHz imaging right now.

The push to lower pulse durations is because when looking at fuel
injection sprays right near the tip, the velocity may be 600m/s. So
even 200-300ns pulses typical of high rep-rate DPSS lasers blur the motion.

The LEDs have won their place in our labs now, but we still seek to coax
as many photons from them as we can. So that's why I hammer them with
very high currents. I can get about 4-6x the light output when driving
them at 10-12A vs. the rated 1A CW. What I really hope for is to be
able to get into the 10-20W peak power range. So far the best LEDs put
out about 1.1W at 700mA. I'm hoping these will produce 10W or better
with 20-30A pulses.

In the cases where they fail, it's just 15 minutes and about $10 to
replace them!

 

[John Devereux]

I have not done a systematic study of this, but have begun considering
how I might go about it.

My limited experience based on bench testing and deploying about 4-5 LED
plus driver assemblies based on the TC4422A to several laboratories
doing high speed Schlieren imaging is that:

I have never blown an LED with 1us or less pulses at up to 10-12A.

I have not done it with the newer leds, but about 15 years ago I was
pulsing smaller IR leds well outside of their specifications like
that. (Younger and more foolish...). They worked fine for a while, but
died after a year or so in service. So any damage may not be visible on
short timescales.

I have put nearly the rated average thermal load of about 5W into the
LEDs with continuous pulses in the 6-9A range and up to 10us, and they
live (for at least minutes-hours).

I have blown up several blue LEDs packaged by LEDengin (I believe Cree
die, but not sure) when pulsing over 100us at high currents, perhaps 5-6A.


I have recently acquired a bunch of top bin Cree XP-E green, red, and
royal blue emitters. Also some Lumileds Rebels in cyan and royal blue.

My first thought at how to test them would be to normalize for average
power. For ex., at 1W of average power, ramp up current for a given
pulse duration (while simultaneously reducing duty cycle to keep power
constant) then see when it breaks. How to present the data I haven't
quite worked out.

Also, some long term experiments are in order, to see if at some peak
current level and thermal load, does the output gradually drop rather
than catastrophic failure. If so, can time constants be derived.

One major problem with this kind of testing is that the next batch of
dice from the same maker may behave quite different. Even for a given
experiment, it would probably be necessary to break at least a small
handful at each parameter set to get some stats.

Also, this testing would best be performed with a scientific grade
pulser, like the ones used for large laser diode bars. I think some
colleagues have these, so I could try to borrow one.

But my present setup is designed to be cheap and quick to get something
running in the lab.

I'm not sure if my manager will go for having me make this into a
research project to characterize LEDs. At least they gave me the
freedom to have developed the drivers somewhat in the background, and
now that they are popular, I can spend more time on that part.


It is remarkable that the radiance of these LEDs is now competitive with
a Xe short arc lamp, over narrow wavelength bands.

We had previously used Xe short arcs for Schlieren rather than a laser,
because the laser's coherence and thus speckle degrades image quality.
But the short arc's white light causes severe chromatic aberration in
our optical engines with quartz cylinder liners. Two options are to put
a bandpass filter in front of the Xe lamp, or find another light source.

The BP filters have only about 50% throughput. Thus, even though the Xe
radiance is much higher over its full spectrum, or even just the
visible, the LED actually wins out over a narrow range of 10-30nm!

Plus the LED can be pulsed at very high rep rates. I can get 5MHz with
full modulation from the TC4422As. Though we are doing only about
50-100kHz imaging right now.

The push to lower pulse durations is because when looking at fuel
injection sprays right near the tip, the velocity may be 600m/s. So
even 200-300ns pulses typical of high rep-rate DPSS lasers blur the motion.

The LEDs have won their place in our labs now, but we still seek to coax
as many photons from them as we can. So that's why I hammer them with
very high currents. I can get about 4-6x the light output when driving
them at 10-12A vs. the rated 1A CW. What I really hope for is to be
able to get into the 10-20W peak power range. So far the best LEDs put
out about 1.1W at 700mA. I'm hoping these will produce 10W or better
with 20-30A pulses.

In the cases where they fail, it's just 15 minutes and about $10 to
replace them!

If you are happy to do that then go for it!

 

[Mr.CRC]

I have not done it with the newer leds, but about 15 years ago I was
pulsing smaller IR leds well outside of their specifications like
that. (Younger and more foolish...). They worked fine for a while, but
died after a year or so in service. So any damage may not be visible on
short timescales.


If you are happy to do that then go for it!

Fortunately in our case, we have the "factory" in house, so for this
admittedly fringe application, blowing and frequently replacing the LEDs
is no problem.

These LEDs are displacing a subset of applications for >$100000 laser
systems, so at their typical price, cost is no object. We could afford
to replace them every experiment, even.


The long term longevity issue though, really does throw a wrench in the
works of any plans to commercialize an application that depends on
massive overdriving.

I've looked for commercial products similar to what we are doing and
found only a company called Visual Instrumentation. Interestingly, they
are serving a similar market to us, providing high speed lighting to
auto makers doing crash testing. But they aren't overdriving.


There is some documentation on this subject from Cree as of late:

http://www.cree.com/products/pdf/XLamp-Pulsed-Current.pdf

 

[Spehro Pefhany]

But, it can happen. Pulsed current can, for instance, cause exploding
wires (the tiny leadwires going to the die). When possible, try to keep
to the manufacturer's recommended limits. The difference
in a 20A drive and 10A drive is only gonna be three dB in light
output, after all.

You can always use multiples.


Best regards,
Spehro Pefhany

 

[Mr.CRC]

Fun! How much capacitance do the LEDs have?

Yes it is fun, because I can blast pretty colored light all over my lab
and don't have to worry about laser safety rules, goggles, etc.

I haven't measured the capacitance yet. Maybe I'll see if the intern
and I can figure out how to do that this week before I run out of him on
Thursday. Then again that will distract us even further from my summer
priority task of building the F2812+Spartan3e system...

That's limited in the speed it can slam the LEDs on, since the current
in the L is all the current that's available. OTOH, the LED turnoff
will be hard.

Why is it limited in speed? My understanding is that the inductor will
slew as fast as possible to keep current flowing, so the only limiting
factor is capacitance. A rough guess at risetime, with even 10nF and to
reach 12V on the LED with a 20A inductor current is only 6ns.

The plan was not to make a particularly efficient circuit, but just to
supply a 20-30A constant current source (voltage limited to about 2-3V)
to the flyback circuit, with hopefully low enough Rl + Ron resistance to
keep on-state power dissipation at a few watts.

My LTspice sim, even when I add an extra nF to the NMOS drain (a wild
guess at the LED cap.), can make 20ns pulses easily.

If cost is not a big deal, look at the IXYS/DEI gate driver chips.
Heck, they may be able to drive the LEDs all by themselves.
Interesting packaging.

The IXD_630 has me drooling. The only drawback is UVLO. Oh well. I
can still do CW drive with the current limiting of a lab supply. I
think this chip will be the answer to my dreams. Thanks for the tip!
Oh, and no, cost is no big deal

The people who sell mega-laser driver boxes tend to use special
flex/foil transmission lines to connect to the loads, just a few ohms
impedance. A 50 ohm line, or a 100 ohm twisted pair, can make trouble
at these speeds and currents.

I've also been aware of the DEI fast laser diode drivers previously.
I'll have to give them another read. If anything I should see if I can
buy one and learn how it works. The really fast ones I'm very curious
about.

Please do, preferably later in the day. I have a cute and cheap way to
drive mosfet gates really fast and hard. I'd post it publically, but
Thompson would steal it, Fields would whine about what it really
means, and Sloman would use it as an excuse to bombast about climate
change.

:-D

 

[Mr.CRC]

If cost is not a big deal, look at the IXYS/DEI gate driver chips.
Heck, they may be able to drive the LEDs all by themselves.
Interesting packaging.

Oh geez, now I see what the DEIC421 RF MOSFET driver can do. Sick.

But the cost on these is $49 for a min of 30 pcs. That becomes a
problem. I will have to see if we want to elevate this to a
high-priority project to buy a can of those.

Most likely, the IXD_630 drivers will suffice to get me into the <50ns
range, and be a little easier to design with.

 

[Mr.CRC]

The entire 2E product line has been EOLd!

That's an incredible bummer: we have several products that use Spartan
2E's.

All our new designs are Altera. Xilinx is just a mess.

Drat. Every time I hear something about Xilinx on these boards, it's
not good and makes me want to consider bailing.

We've got some really good people with extensive Altera experience
in-house, including a new hire that worked there and knows the guts
inside and out.

I would be happy to bail if I didn't have some existing Xilinx projects
deployed. But the thought of having to continue to work with two tool
sets is not pleasing.

Plus there are many cheap dev. boards for Xilinx. Argh! Now I have to
re-think this all over again...

Right, the capacitance is the limit. With a flyback, there's no tricks
available to boost the leading edge current.

Which at 10-30A is going to give enough speed anyway in these apps.

I ordered some of the Directed Energy drivers today, the "PCO-7110 40A
4ns Fixed Pulse Width" and the PCO-7120 12ns and up, 50A driver.

These should be fun to play with, and I can learn a lot about how they work.

I think the PCO-7120 will be suitable for our situations where we need
well under 100 ns.

For stuff that only needs 50ns and up, I think the 30A gate driver will
be worth making a custom board.

I'm also considering something really crazy: for flood illumination
where I can afford a large source area, a 1x2" matrix of about 32 LEDs
driven in groups of 2 to 4 by something like the IXD_609 in DFN on the
back side of a little tiny 1x2" board. Won't be able to dissipate heat
well, but it could pump about 60-120W of optical radiation at low duty
cycles into an engine head window.

One other possibility would be to use a Zetex avalanche transistor. 60
amps or so at 200 volts, 2 ns risetime, from a SOT23 transistor! Do
the old radar charged-line trick. That would work great as long as the
pulse rate is low.

I'll have to look into that. Can an avalanche transistor be triggered
reliably?

I discussed with the intern about this, but I don't think he got this
question answered yet.



Thanks for the input.


Good day!

 

[Mr.CRC]

Or, you can ditch the LED and use a proper gas discharge
lamp. At least, it'll turn ON quick enough (the OFF time depends
on gas mixture, and fill pressure). Very high power (multiple kilowatts)
single lamps are available, but NOT in LED technology.

I've done considerable research into Xe flash lamps and concluded that
for very high rep rates such as 10kHz and above, and very short pulse
durations such as less than 1us and all the way down to sub 100ns, that
they just aren't the answer.

Furthermore, there is jitter which would force us to use long camera
gate times which would negate some of the advantages of operating at
very high speeds, such as gating out background signal sources, and stop
motion of very high velocity phenomina.

Additionally, lamps spew in all directions, which is also a problem with
LEDs. But LEDs are closely related to laser diodes, so many of the
drive electronics and experiments we set up with LEDs are quickly
adaptable to laser diodes.

Finally, lamps are wide spectrum, so once you filter out a 20nm part of
the spectrum comparable to a colored LED, you get not so much radiance
anymore, and the LED looks pretty good. If light in the deep UV or near
IR is needed however, lamps sometimes are a good choice.

I have a great deal of fondness for lamps, as I am at heart an "arcs and
sparks" guy. But the LEDs are a great light source for some of my
purposes, and superior to flash lamps in all of the cases which are
driving my interest in LEDs.

Ultimately if we decide we want to spend more money on high rep-rate,
short pulse sources, I will head back in the direction of lasers, though
diode lasers, for cases where collimation/small apparent source size is
needed.

The key thing I have to figure out with diode lasers is to accomplish
de-coherence without loosing a large % of the photons.

 

[Numer0 Un0]

Please do, preferably later in the day. I have a cute and cheap way to
drive mosfet gates really fast and hard. I'd post it publically, but
Thompson would steal it, Fields would whine about what it really
means, and Sloman would use it as an excuse to bombast about climate
change.

And for a tit for tat (yes, you pulled this shit first), I'll say that
it is not your design.

That is the kind of treatment a fucktard like you deserves.

 

[MrTallyman]

One other possibility would be to use a Zetex

Finally. The one intelligent thing you've ever said.

 

[John Devereux]

I've done considerable research into Xe flash lamps and concluded that
for very high rep rates such as 10kHz and above, and very short pulse
durations such as less than 1us and all the way down to sub 100ns, that
they just aren't the answer.

Furthermore, there is jitter which would force us to use long camera
gate times which would negate some of the advantages of operating at
very high speeds, such as gating out background signal sources, and stop
motion of very high velocity phenomina.

Additionally, lamps spew in all directions, which is also a problem with
LEDs. But LEDs are closely related to laser diodes, so many of the
drive electronics and experiments we set up with LEDs are quickly
adaptable to laser diodes.

Finally, lamps are wide spectrum, so once you filter out a 20nm part of
the spectrum comparable to a colored LED, you get not so much radiance
anymore, and the LED looks pretty good. If light in the deep UV or near
IR is needed however, lamps sometimes are a good choice.

I have a great deal of fondness for lamps, as I am at heart an "arcs and
sparks" guy. But the LEDs are a great light source for some of my
purposes, and superior to flash lamps in all of the cases which are
driving my interest in LEDs.

Ultimately if we decide we want to spend more money on high rep-rate,
short pulse sources, I will head back in the direction of lasers, though
diode lasers, for cases where collimation/small apparent source size is
needed.

The key thing I have to figure out with diode lasers is to accomplish
de-coherence without loosing a large % of the photons.

So you don't get fringes? You can inject a strong modulating current at
UHF frequencies (>~400MHz). I.e., essentially turn the laser on and off
very rapidly.

(Phil Hobbs book was worth buying for one paragraph...)

If your pulse can be made short enough, it ought to be incoherent anyway
since it would amount to a single cycle of such modulation?

 

[Mr.CRC]

So you don't get fringes? You can inject a strong modulating current at
UHF frequencies (>~400MHz). I.e., essentially turn the laser on and off
very rapidly.

(Phil Hobbs book was worth buying for one paragraph...)

If your pulse can be made short enough, it ought to be incoherent anyway
since it would amount to a single cycle of such modulation?

That's too fast. I'm not trying to reach the lower limits of laser
diode or LED pulse widths. Just get a controllable pulse duration less
than 100ns, preferably in the 10-100ns range. This problem is basically
solved at this point, as my research and input received here has led to
several different options for driving light emitters, including some
inexpensive commercial modules already on order.

However, a modulation technique might still be useful for a laser diode
for making it less coherent.

AFIAK, the way this works is that the current modulation basically
modulates the wavelength, so that on long time scales (times longer than
the modulation period) there is effectively random phase, thus
incoherence. Right?

I'm trying to find a time to discuss this subject, and whether it can be
accomplished in other ways such as MM fibers, with one of the very
bright light bulbs at my work...

Unfortunately, every time I look up the subject of "laser despeckle" I
get nothing but a list of patents.

 

[John Devereux]

That's too fast. I'm not trying to reach the lower limits of laser
diode or LED pulse widths. Just get a controllable pulse duration less
than 100ns, preferably in the 10-100ns range. This problem is basically
solved at this point, as my research and input received here has led to
several different options for driving light emitters, including some
inexpensive commercial modules already on order.

However, a modulation technique might still be useful for a laser diode
for making it less coherent.

AFIAK, the way this works is that the current modulation basically
modulates the wavelength, so that on long time scales (times longer than
the modulation period) there is effectively random phase, thus
incoherence. Right?

Not sure, but I don't think so. I think it's more like the oscillations
in different parts of the chip dont have time to build up and
synchronise before they are quenched.

Perhaps Phil will chip in at some point to tell us.

 

[Mr.CRC]

The laser oscillation builds up from noise on each cycle of the
modulation, so there is no phase correlation between pulses. That makes
the laser essentially completely immune to mode hops and self-locking
due to back reflections. It also spreads it out to ~ 1 THz wide.

Cheers

Phil Hobbs

Is this in your book?

Perhaps I'm becoming convinced that I should buy it.

 

[josephkk]

The laser oscillation builds up from noise on each cycle of the
modulation, so there is no phase correlation between pulses. That makes
the laser essentially completely immune to mode hops and self-locking
due to back reflections. It also spreads it out to ~ 1 THz wide.

Cheers

Phil Hobbs

And i would hazard the guess that pumping it with a narrower source (or
two) isn't nearly as easy as it sounds.

?-)