LED limit to short pulse and high power

Question

I’m studying the viability of a small footprint distance detector that relies on a LED, a photodetector and some lenses. It’s meant for a distance above 150m, where the light beam travel time is higher than 1us. Although the 1us time is within normal processor handling capabilities, the range is an obstacle of LED luminance and detection. However, if LEDs could handle very high power (but low energy) luminance output for a short period of time the detection could be made viable. The question is, what are the common LED physical limitations to handle such small energy but powerful output?

Answer 1

Generally speaking, there are two limits for semiconductor pulsed operation:

1. Junction temperature
2. Electromigration (and other spooky / ill-defined failure mechanisms)

Temperature is the easier one. You may even be able to measure it, because it takes on the order of microseconds for heat to spread out from the junction region to the rest of the die, and substrate, and so on; in which time, a small test current can be applied to measure V_F, which, knowing its curve vs. temperature (a calibration for the particular emitter may be required, as the manufacturing tolerance for this parameter is wide). Thus you can map out the SOA (safe operating area) of the device, determining temp rise for a given pulse power and duration.

Typically, heat flows by diffusive transport, and diffusion has a \$\propto \sqrt{t}\$ power law. Thus, as pulse width goes down, peak power goes up, and pulse energy goes down, roughly as the square root of the ratio in time. Exact values depend on how the device is constructed -- as heat flows through different width materials, with different conductivities, the actual scaling rate varies. Hence why transistors generally include a transient thermal impedance plot (showing the actual rate vs. time for the part).

The other effects are harder to predict. Electromigration is a current and temperature dependent phenomenon, which leads to erosion and eventual failure of interconnects (the metallization carrying current across the die). And perhaps semiconductor material itself as well. LEDs are doubly challenging here, because 1. they can't afford much metallization area because any metal directly blocks paths for light to exit the die, and 2. the thinner grid has higher resistance, increasing current density, and peak power dissipation, directly in the metal itself.

Preferably, you'll find LEDs rated for pulsed operation, respect their ratings, and that's that.

If you're having trouble finding such, you might rely on qualifying testing instead: run a few thousand pulses (or million, or however many you need over the expected product lifetime, or service interval if the light module should be replaceable), at worst-case ambient temperature, and see when they fail. If that failure rate is just acceptable, there you have it, that's your maximum operating level.

Offhand, I would not expect currents much higher than 10x DC rated, even for very short (~µs) pulses where temperature rise is negligible. A typical example is numeric display LEDs, which might be rated 20mA DC (at room temperature) but merely 100mA peak.

LED efficiency also declines at high level, making the better solution generally to use more -- assuming the optics work out, of course.

There's no way to combine individual dies into a single tight beam (or at least, nowhere near an easy one), so a multi-die solution is really only viable for broad illumination. Hence the minor adoption of LEDs for photoflash applications, but you'll have a damn hard time doing it for a narrow beam (e.g. point distance measurement LIDAR).

A related note: white LED phosphors have a decay time in the fractional µs, making them unsuitable for, say, data transmission applications above some Mbit. This is probably relevant here. Just stick with the bare kind (white LEDs are just blues with a yellow phosphor on top).

If wavelength is a free choice, probably blue is best, being the most efficient. (Hm, I forget offhand if red is similar efficiency but less efficacy because red; or IR for that matter. It may be that red is less efficient, enough to the point that, despite high-efficiency green having the InGaN drop of 3V for ~2eV emission, green also outshines red. It very obviously does by efficacy, but for green, I mean, that's practically cheating...) A single wavelength permits significant filtering of ambient light, which should be helpful.

Speaking of bitrate, you might consider a phased approach instead. This is the basis of typical ToF or distance cameras, as I understand it. By modulating the LEDs at a modest frequency (say, 10MHz, sinusoidally), and comparing the phase shift between a given point in the scene and the reference oscillator, its relative distance (within one cycle) can be determined. To determine absolute distance, repeat the measurement for slower and slower rates, until the phase shift stops decreasing (the target is less than one cycle away). Other modulations of course are feasible, from any general RADAR topics: pulsed, phased, chirped, wavelet, etc.

Answer 2

Beam divergence is much more important than high power for this kind of thing.

LEDs have a fairly wide cone of radiant emission and their light is incoherent (i.e. the light waves leaving the LED aren't in phase with each other). So, while you can collimate LED light fairly well, it's a pretty lossy and difficult process. This isn't helped by the fact that bright LEDs typically have large emitter die sizes, or come as COB arrays of emitters, resulting in the light acting like an area source rather than a point source. This all results in nonuniformity, penumbra effects, dispersion issues, and other complications.

Lasers, on the other hand, emit coherent light in a collimated beam. A perfectly collimated beam of light with zero divergence would leave the laser aperture with a given beam diameter (known as the "waist") and remain at exactly that diameter regardless of the distance. However, zero divergence is not achievable in practice (diffraction renders it a physical impossibility) and there is always some beam divergence. We typically measure beam divergence in milliradians (mrad).

The lower your beam divergence, the smaller cross-sectional area it will have after travelling a long distance. The lower the cross-sectional area, the higher the optical power density will be when it hits your photodetector (or at least the lens in front of it).

Solid state (diode) laser modules can easily achieve beam divergences of 0.5mrad or lower. With a low divergence laser at an optimal wavelength for your photodetector, you can do time-of-flight detection over pretty far distances without a lot of power. Exactly how far depends on your photodetector sensitivity and noise floor, the reflectivity of the distant surface, and the diffusion that occurs due to surface roughness.

If you want high power, it's available. Laser modules used for lighting effects typically have less than 1mrad of divergence and are available in powers from 100mW to tens of watts. You can electrically control the laser output to use it in a pulsed or modulate fashion. Typically, though, much lower powers are needed when using lenses, bandpass filters, and other tricks to improve sensor performance.

It is worth noting, though, that laser distance meters typically do not just use time of flight as a measurement technique. A laser source and line-shaped photodetector array sensor are positioned at a known distance away from each other within the device. The beam comes out of the front of the device, bounces off the remote surface, and back into the detector. This beam path forms a right angled triangle. By measuring the intensity pattern across the photodetector, the device can calculate the angle at which the beam entered the receiver lens. Basic trigonometry (SOH CAH TOA) then allows it to solve for the lengths of the sides of the triangle, and therefore the distance. By using a narrow bandpass filter tuned to the laser wavelength, and by modulating the laser light with a known pattern, external light can be rejected and filtered out, resulting in an extremely high signal to noise ratio. This allows the laser power to be reduced significantly - typically <1mW for distances up to 200m. All of this can be implemented in a device no bigger than an apple and costing under $30.