Low Pulse Width LED Driver

[D_Money22]

I am working on developing a circuit for a testing apparatus for our lab. The circuit needs to be able to drive an LED with relatively constant current (This isn't extremely important, but we definitely don't want any kind of voltage instability to damage the LED) general peak current ~0.5A. The main function of the circuit is to produce short square pulses, with pulse width as low as 25ns and optimal rise/fall times of only a few ns. The circuit receives an external trigger from a precision delay generator. Right now the circuit I have (This is my first time ever working anywhere near these timescales, so I'm not totally sure what to do) involves an RC delay leading into one input of an opamp. The external trigger leads into the base of a transistor, discharging the capacitor and pulling the input low. At the end of the trigger pulse the cap charges, with the reference voltage applied to the second input determining your pulse width. With a fast opamp this could work, however the fact that pulse width is equal to the time for the capacitor to charge plus the trigger width rather than a single value independent of the trigger is not ideal. An alternative I am considering would be some sort of integrated laser driver (Just clarifying, this is not a laser diode, just a general LED). If anyone has any ideas or resources that would be incredibly helpful.

Thanks
-Daniel

[mikeselectricstuff]

If peak current is only 0.5A and pulse is short, just use a resistor to limit the current, and keep inductance and stray capacitance to a minimum. It might also help to actively short the LED at the end of the pulse to discharge its capacitance.
Power dissipation will be negligible. Use a high enough supply voltage to get the current as consistent as you need.
 

[james_s]

You could also look at laser diode drivers with modulation input. Those have to provide tight current control with minimal overshoot and fast response. There are a number of simple designs out there.

[Marco]

AFAICS an astable monovibrator from a fast logic family will suffice for the timing.

[capt bullshot]

As Marco pointed out, use an suitable IC, not an Op-Amp.
As a starting point, look at the 74HC123 which provides an edge sensitive trigger, it won't elongate its output pulse width by the trigger pulse width. You may want to search for some faster device families, 25ns pulse width might be too short for the standard HC series '123.

[vealmike]

25ns @ 0.5A could be a challenge as with all high speed stuff, it's the edge rates that you need to pay attention to.

I need to make an assumption about what you really mean by a 25ns pulse. You will need to correct these and redo the maths.

A 25ns pulse from your controller is fed into the LED driver, which must turn the LED on and off again.

Allow 5ns for switching (20% to 80%), 5ns to switch on, 5ns to switch off and your light pulse is ~15ns.

60% (80%-20%) of 0.5A is 0.3A.
0.3A in 5ns gives a dI/dt of 60MA/s!


I'll let you work out how much inductance will scupper this for you.

Personally I'd consider some other form of modulation. Can't really comment on what as we don't know what the system is.

I'd also look at what's available commercially rather than re-inventing the wheel. For example, pluggable SFP+ modules transmit data at 16Gb/s over optical fibre (OK, much lower power).
NRZ encoding keeps the maximum data frequency to 8GHz. I'll wager these LEDs are not 100% AM modulated at this frequency.

They may also contain chips that you could re-use, or at least crib ideas from.

[capt bullshot]

Based on this paper:
http://cmosedu.com/jbaker/papers/1992/MST31992.pdf
I was easily able to create pulses of 20A with faster than 100A/us slopes (I did this just because i could do so and I wanted to test the slew rates of current probes)
The message is: You can always get over inductance limits by using higher voltages and higher resitance
OK - 500V and cascode connected MOSFETs is a bit oversized here, but it shows the way.

[mikeselectricstuff]

High-speed MOSFET driver chips may be a good solution for generating the drive.
The LED Vf isn't going to vary much so a couple of volts dropped over a low-inductance resistor ought to be enough for current limiting.

e.g.

http://www.ixysic.com/home/pdfs.nsf/www/IXD_604.pdf/$file/IXD_604.pdf

~10ns rise/fall time into 1nF, 4A peak current

[bobaruni]

Also consider the rise/fall time of the LED itself which can be tens of nanoseconds.

[mikeselectricstuff]

Also consider the rise/fall time of the LED itself which can be tens of nanoseconds.

I think that will mostly just be the effect of capacitance & inductance - keep everything small & short.
A SO8 MOSFET driver, ceramic decoupling cap, SMD current-limiting resistor right on the back of the LED should be a good start.

[LaserSteve]

There are some tricks to sweeping the charge carriers out of the LED to speed up turnoff...

take a look at :

High-Speed LED Driver for ns-Pulse
Switching of High-Current LEDs
Hubert Halbritter, Claus Jäger, Rolf Weber, Michael Schwind, and Frank Möllmer

IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 26, NO. 18, SEPTEMBER 15, 2014

and do some Googling for LED Pulse PIV.. and you should be happy..

There are three schools of thought.. Avalanche Pulse, Bipolar Driver to sweep out charge by reversing polarity, and brute force high speed switching with a small series resistor. The above paper uses brute force, a fast mosfet driver, and adds an inductor which speeds up turn on and turn off... There is demand for this, with at least one research group developing a die with seven mosfets to do the current switching, charge sweeping, and shunting the led to correct for issues in turn on-turn off time.

If you want a nice sharp square wave with rise and fall times in nanoseconds, this becomes a LED chip and packaging modeling exercise, ..

This one is good, as well...

Pulsed operation of high-power light
emitting diodes for imaging flow
velocimetry

C. Willert1, B. Stasicki2, J. Klinner1 and S. Moessner1

2010 Meas. Sci. Technol. 21 075402


Avalanche Pulsing  is easy, until you need to clean up the trailing edge of the pulse...

Both of the above papers use available parts, no special magic...  Digikey sells some low ESR, medium voltage,  moderate capacity, lytic caps for pulse duty, and one paper makes mention of them. Board layout and material choices are crucial, with low impedance, high current, low inductance leads needed, as well as a "stiff" low ESR power supply.

Steve

[Kleinstein]

The speed of LEDs can be very different. Some are fast to follow sub ns pulses and some are rather slow and might produce light even a few 100 ns after the current stopped. So the choice of LED can be important. Ir and red LEDs tend to be relatively slow, blue LEDs can be rather fast. But it is not only the color but also model of LED that can make a difference.

To get a rough idea on how fast the LED is, one can look at the LEDs reverse recovery.

[Marco]

On second thought, I'd use a D flipflop to build the one shot. Because that's the only thing Fairchild has available in it's UHS family, which is the fastest cheap logic available. The NC7SZ74 has complementary outputs, which would be nice for driving a transistor pair to direct a current source through the LED.

My first thought was one of the cheap ST LDMOS devices such as the PD85004 (though they seem to be ditching them, the PD8xxxx series has been purged from ST's website). For some reason the datasheet says it only allows -0.5V Vgs though. NXP has the MRF1513NT1, slower but probably fast enough, and a lot more expensive. The -0.5 Vgs on the ST devices might be because of an internal diode used for gate protection.

PS. oops, extreme stability of the current is not a requirement ... in that case current steering is a bit wasteful. Still, LDMOS will give you a nice large voltage range to fight inductances with regardless.

[D_Money22]

There is a ton of awesome stuff in here, thanks. Just to clarify, the LED we purchased for this task is definitely able to operate at the speeds we need (that was a key buying criterion) and we have a setup which is fully functional (the purpose of this is to provide a backlight for a high speed camera, however we need the light to be able to fire two quick pulses in rapid succession so that we are left with two distinct frames with a specific amount of time between). The purpose of this circuit is to make the testing apparatus substantilly cheaper, as our current setup uses multiple lab grade devices with far more functionality than is necessary that each run in the tens of thousands of dollars. I will certainly look into all these ideas here, thanks for all the great advice.

[vealmike]

To clarify my post, I didn't mean to state that the goal was impossible. It was more of a "start thinking about.."

Really interesting stuff followed on switching diodes. I'm going to do some reading up.

[Kleinstein]

For a more or less constant pulse width, one can use he charge stored in a length of coaxial cable to create the pulse. As a very simple version: have a rather small FET (e.g. 2N7000 class) in series with the LED and a resistor in the 40 Ohms range followed by about 50 cm of 50 Ohms of coax, charges to about 50 V. The only thing left is to turn on the FET fast.

[Marco]

Using coax for timing doesn't really make sense if you're not using an avalanche transistor, so it really doesn't make sense. A one shot made from a UHS D flipflop will be more than fast enough and can vary the pulse length with a pot. Unfortunately not strong enough to directly drive a PD85004.