Ultrafast (10ns) high current (7A) LED pulser for Physics experiment

[DanInvents]

Hi!

For a Physics experiment I'm designing a fast (10ns) LED pulser (see attached image). The LED driver is a copy from the Art of Electronics where they explain that they used the circuit to drive an LED and send short light pulses. A gate drive controls the LED by sending a pulse of up to 7A of current. The circuit is carefully tuned so that the time constant of the RC snubber is the same as the 10ns long pulse. By doing so the LED is forward driven momentarily after which the conduction is quenched by the negative voltage imposed by the capacitor in the RC snubber.

My addition to this design is a CMOS oscillator and a fast (1GHz) buffer. To minimize LED heating I would like to drive the whole thing with a slow (1-10kHz) PWM signal. This way I can adjust the time that the LED is on through the duty cycle and easily modulate the repetition frequency of the bursts with a signal generator.

Any thoughts? Do you think that this approach will work?

Best,

Daniel

[Marco]

Why not have a free running oscillator and pull the output down with the PWM?

[CaptDon]

Somehow that RC circuit seems a flawed way to do the high current pulse. I think I would look into a laser diode driver optomized for LIDAR or the distance measuring gear like surveyors use.

[CaptDon]

10ns is a very fast pulse. It would be equal to light traveling about 15 feet from the start to the finish of the pulse. The old dinosaur way of doing this was with a light source and a spinning disk with an aperature hole or slot in the disk. The duty cycle would be fixed and never vary but changing the disk speed would change the repetition rate as well as the pulse width. I guess with a laser and spinning disk you could only get into the microsecond range and surely not nanoseconds.

[T3sl4co1l]

Emphasis on "old dinosaur"...?  I don't think anyone's used a spinning disk seriously since Fizeau. https://en.wikipedia.org/wiki/Fizeau_experiment Nipkow maybe, but not so much for sake of pulse generation.

Modern methods are fairly trivial, commercially speaking.  Direct drive laser and AOM (acousto-optical modulator) are typical methods to get into the 100s of MHz, and GHz.  Devices are readily available, including convenient modular formats like SFP.

But the quoted purpose is LED drive, so lasers are N/A here.

As for the circuit, gate drivers should be fine.  You might want individual resistors from each output, to avoid circulating currents between the two in case the propagation delays don't match.  I wouldn't think a capacitor is needed, it should suffice to discharge the LED to 0V.  Perhaps it's worth testing whether reverse bias clears charges from the junction faster -- in either case, the capacitor value can be adjusted freely, including infinity (replaced with short), so that still covers everything.

For still-faster drive, probably the next option is GaN transistors, but pretty soon at these time and impedance scales, you'll be hard limited by stray inductance on the PCB and into the LED chip itself (they're mostly/all wire bonded?).  If you want to get really fancy, you can consider peaking circuits (some series L and shunt C to somewhat compensate for the bondwire + pad + trace L), but that's going to get very tedious very quickly.

As for signal generation, buffering doesn't seem important, do the gate drivers have anomalously high input capacitance?

You say pulse is 10ns, but clock is labeled 100MHz which is a 5ns pulse.

How the oscillator works is important: if gated, you'll get a pulse train more or less as expected, give or take runt pulses at the start/end as the gate is asynchronous to the clock.  If powered down, expect long startup time before the output is stable and consistent.  It may be stuck high or low for some ~ms if it's a quartz type, and, I don't know about silicon oscillator types.

Using a latched clock gate, as might be constructed from some 74LVC or faster logic, would be fine, and then the oscillator can be free-running.

Doing both (a prompt start and accurate timing from an asynchronous trigger), is another matter.  There are methods such as triggering an inaccurate but fast RC oscillator, then disciplining it (give or take acceptable phase/frequency error during the burst), or measuring the phase and disciplining to that fixed phase offset, etc. You might be better off buying test equipment if it needs to be this accurate.

Tim

[Purduephotog]

This is the way. Until I was laid off, we were operating in about the same realm but with a few more zeros after the amps. No femtos.
The biggest question is 'experiment' vs 'production'. How often do you have to be right vs how often can you never be wrong? For experiments there are far smarter people than me that will offer you 'downgraded' production gear- still at a price.
My work was with laser output,  but they were pumped with LED, so similar time frames.

Do you have an error budget? If you don't, consider going through every component, getting the tolerances and acceptable deviation, and listing them. It'll take a bit of Zen but you may see something you can cheat from, or you may see other options you can't compromise on.

I do remember one team for (some application I can't remember) ended up using $$$$$$ worth of tantalum capacitors vs the regular because of the reliability and time- far exceeded cost but it hit (I think density and some other metric).

7A isn't a lot of current and technology has advanced significantly in the last 3 years since I last dove into it.

Good luck, and if you ever publish the results- if you'd care to send me a note I'd be flattered and pleased to read the paper you're involved with.

Good Science.

[DanInvents]

Thank you all for sharing your insight. I have used the rotating disk technique for laser experiments in the past. In this case however, I need very short light pulses that the rotating disk technique could not provide. In addition, I need very high brightness for a short time, as a matter of fact I have been considering multiplying the power stage 4 times to be able to drive 4 LEDs.

I'll look into 74LVC-based oscillators and get to work on the circuit design. I could keep in touch with those interested about future developments.

Thanks!

Daniel

[RoGeorge]

If rotating disks are too slow, maybe use rotating mirror(s) instead, placed at a distance, so to make the laser spot sweep its target faster.

[jonpaul]

Bravo for the hand drawn schematic.

Art of electronics is not the lastword....

Suggest to try discrete BJT, cascode or  an emitter follower.

Suggest current drive not voltage + resistor

Jon

[Phil1977]

Can you further specify the required pulse parameters?

Especially repetition rate, rise- and fall-time? Is a square-pulse-shape preferred or is a logarithmic decay also admittable? How is the ratio of max allowed LED current to max pulse current?

[nimish]

There are a bunch of LIDAR optimized gan transistor gate drivers, some fully integrated, that could do this.

[Marco]

If you want to be lazy, just buy LMG1020EVM-006.

[DanInvents]

As an answer to Phil1977, the most important parameter is the pulse duration, it should be approximately 10 ns. The pulse shape, rise, and fall times are not critical. Of course, in a 10ns pulse the rise and the fall times would have to be in the nanoseconds range. I don't know what the ratio of LED current to max pulse current should be. In practice I'm going to be abusing the LEDs and hoping that because I'm sending such short pulses the heat capacity of the die can absorb the heat before permanent damage occurs.

[Phil1977]

I don't say it isn't worth a try, but be aware that there are 2 heat capacities in the LED that can be harmed:

- The emitter die: That´s usually quite robust against current pulses
- The bond wire/and or conductive traces: In efficiency-optimated LEDs these parts are as thin as possible. That means they act as a fuse when treated with high current pulses and are destroyed.

You can easily test this if you connect the LED directly to e.g. a 220uF-low-ESR-cap and measure the current with a shunt and scope. But be prepared to kill some of the LEDs.


(I once got a very special cyan 480nm high power LED as a present. Together with the colleagues we wanted to see the colour and connected it to a usual lab power supply, slowly increasing voltage and current. The colour was really beautiful until we had one loose contact in the test strips. So the PS charged up its output cap to the set CV-value when the connection was established again. We couldn't believe that the LED was so easily broken so we put it under a microscope, and the bond wire looked exactly like a blown fuse. Luckily this LED was not resin filled, so we could lift up the protective glas and spot-weld a new wire to the residual bond wire. That was by far the smallest manual craftsmanship I´ve ever experienced.)

[T3sl4co1l]

Thermal failure is a non-issue for such short pulses, individually; electromigration is another possibility however.

Thermal may be an issue again for the tone-burst scheme; a few 10s µs is enough for such structures to heat up.  Heat also exacerbates electromigration.

Tim

[Conrad Hoffman]

If you can get the article listed here, they make some pretty good suggestions on circuitry. https://www.researchgate.net/publication/321415710_Low-Cost_High-Speed_In-Plane_Stroboscopic_Micro-Motion_Analyzer

[Positron Enthusiast]

The art of electronics circuit will probably work, but it's not what I usually go for. (just that I haven't tried it)

AC logic bus drivers can have their inputs/outputs ganged together. RC filter that into an nmos gate on the low side of your led. That will work down to around 1-10ns pulses. Lasers and LEDs behave very similarly below around 100ns pulsed in terms of driving circuitry. If you need faster pulses it gets very tweaky. Old NPN RF BJTs work to if you need even shorter/ gaussian shaped pulses. NMOS tends to give more of a square wave response, but the RC filtering offers a low of control. You will want fast zeners in parallel to protect the diode from inductive transients (and maybe some for the drive circuitry too).

Your high-side supply will need to be tweaked to get the best response. AC logic also performs well when run at the high side of the voltage range.

Any old logic family will do to generate pulses to feed the bus driver chips. clock goes into two logic elements A and B. Logic element A feeds B. That results in a pulse with fairly stable width of the gate delay. You can tune it somewhat with RC.

I'm assuming you already have an integrating or synchronous detector that you can couple the LED into and test the circuit? If not, cameras make decent integrators to get relative light levels, assuming you're in the wavelength range. For cameras you'll want a lens setup ideally that will let you see the LED element itself when it's off/ lit externally.

I looked at laser drivers a couple years back but they all had pretty big issues at this timescale. Most were intended for NRZ modulation with tight duty cycle restrictions and/ or very expensive.

AOM and EOM devices are an expensive way to turn CW light into pulsed last I checked, but might be easier to use. I think thorlabs carries some if you have a crazy budget haha. on a related note - if you haven't already checked with them they may have a pulsed laser/led circuit tutorial somewhere that would be helpful.

Mechanical handling of the LED matters a lot for pulsed applications. mechanical stress in the leads will cause early failures.

Not sure I understand your PWM modulation scheme - could you add more details? The clock sounds like a reasonable place to start but you might want to think about turning the rep rate of the pulses way down to start with. That will help with power/ thermal control.

Once you get a circuit up, start everything and low voltages and use high resistance values to keep currents low. Turn the resistances down slowly if you need more power or aren't hitting the threshold current. Don't go below 20-40 ohms current limiting resistance in series with the LED unless you're willing to cook it. Charge resistance for the anode capacitors should be on the order of kOhms.


Expect to see average pulsed power outputs less than 1% of the CW rating up to the CW rating. Most of the datasheet optical or electrical characteristics won't apply to a pulsed output. Wavelength especially tends to be distorted below 10ns.