How to synchronize a large number of fast, high power IR LEDs?

[Proxy]

I have a proof of concept board that triggers twelve LEDs (four parallel strings of three) matched to ~3ns with total propagation delay of ~15ns. They are triggered at 1MHz for up to 33% duty cycle.

I need to trigger 225 LEDs (45 parallel strings of five) matched to <5ns with a total propagation delay of <30ns.

Currently the trigger comes into the board and goes
1. to a transceiver (SN74AVC1T45) that level shifts to 3.3V, then
2. to a quad XOR (74LVC86A) that splits the signal, then
3. to four low side gate drivers (UCC27524) that switches the four strings of LEDs directly.

Scaling this up while keeping 225 LEDs matched to within 5ns is my biggest concern. What would be the best way to keep such tight timing?
1. Split the signal a couple more times and use 45 drivers, one for each string? The current XOR only imparts about 5ns delay so a couple more layers might be tolerable.
2. Add a few large power FETs after the drivers. Each FET could switch the ground for multiple parallel strings.
3. Something else?

[Ice-Tea]

Correct me if I'm wong, but the delay matching for the driver you are using is already 1-4ns... So, you may think you have a viable setup but the specifications say otherwise

How much current doe you need/LED?

[Proxy]

Each driver has two channels that are internally matched to 1-4ns, but how do I know I didn't just get lucky that the four drivers on my board are matched to each other so well? Total propagation delay of each chip can be 6-23ns. I really don't want to have to bin hundreds of chips just get 45 that match each other to within 5ns.

Currently (no pun intended) each string gets ~1A during the pulse. The new design will have slightly different LEDs requiring 4-5A per string.

[David Hess]

One way is to use a phase locked loop to compare each output to a reference and adjust the timing.  Some "zero latency" clock drivers work like this.

Unless there is a strict drift specification, I would add a short adjustable RC delay to each channel and calibrate them using an oscilloscope.

[Ice-Tea]

Part of the solution could be to use an FPGA. Delays are well caracterised and you can syncronise to the same clock. You'll still have the interesting task of getting 223 1A drivers to turn on within a few ns of each other 

[mikeselectricstuff]

If there is any distance involved I'd put the power drivers very close to the LEDs to avoid inductance issues, and have a sync signal distributed over LVDS to the drivers.
The drivers can have an adjustable delay if necessary to sync them.
It may or may not be worth having a PLL - if it has a long time constant it could maybe reduce the effect of any jitter. 

[Proxy]

The longest distance on board is about 4", is that enough to warrant LVDS?

[mikeselectricstuff]

The longest distance on board is about 4", is that enough to warrant LVDS?

Probably not.

[d-smes]

Be sure to take propagation delay into account if there is much distance between LEDs  1 ns per 6-inch on PCB IIRC.

[dmills]

Short lengths of RG316, carefully cut after measuring the driver delays to compensate the skew? You are looking at about 1ns per foot give or take so adjustment to within a few hundred ps is not hard, put the delay lines behind the drivers, and terminate into 50 ohms to keep reflections under control.

I would actually be very tempted by a fastish RF power mosfet, and just length match the nets to the LEDS, an unmatched 500MHz LDMOS part (Freescale make some) will switch in low single digit ns if you push the gate reasonably hard, even an old MRF150 or 151 would likely work.

Regards, Dan.

[mikeselectricstuff]

Might be worth looking for appnotes on laser diode drivers, as some of these will be in a similar range of speeds and powers

[Poe]

Do the LEDs have to go fully dark or just modestly reduce intensity?  If the latter, keeping them at 10% current (when 'off'/'low'/'dark') will minimize line voltage swings, effectively reducing response time.

This sounds like VCSEL array territory rather than LEDs though. 

[Proxy]

Short lengths of RG316, carefully cut after measuring the driver delays to compensate the skew? You are looking at about 1ns per foot give or take so adjustment to within a few hundred ps is not hard, put the delay lines behind the drivers, and terminate into 50 ohms to keep reflections under control.

I would actually be very tempted by a fastish RF power mosfet, and just length match the nets to the LEDS, an unmatched 500MHz LDMOS part (Freescale make some) will switch in low single digit ns if you push the gate reasonably hard, even an old MRF150 or 151 would likely work.

Regards, Dan.

RF MOSFETs never occurred to me, I will take a look at them.

Might be worth looking for appnotes on laser diode drivers, as some of these will be in a similar range of speeds and powers

Thanks I'll take a look

Do the LEDs have to go fully dark or just modestly reduce intensity?  If the latter, keeping them at 10% current (when 'off'/'low'/'dark') will minimize line voltage swings, effectively reducing response time.

This sounds like VCSEL array territory rather than LEDs though.

Ideally they would go fully dark so we can measure an accurate black point between each flash.