The end is nigh... for lighting efficiency

I was looking into lighting efficiency recently, wondering just how far things can go before we hit the absolute limit. And from what I found, it looks like there’s not a lot of room left to grow.

More specifically, I found these numbers…

  • Incandescent lights: ~5 to 15 lm/W
  • Current LED lights: ~50 to ~150 lm/W
  • True blackbody radiation source like the sun: ~93 lm/W
  • Ideal pure white (blackbody radiator minus invisible frequencies, visible light only): 251 lm/W
  • Most efficient light possible, narrow-band green at 555nm: 683 lm/W

So, for a high-quality pure white light with 100 CRI, it seems the limit is 251 lm/W. Going higher than that means sacrificing quality, mostly by decreasing CRI and increasing the green component compared to other frequencies.

The upper limit for “white” LEDs is estimated at 260 to 300 lm/W, depending on how much green tint people are willing to tolerate. Meanwhile, Cree already reached 303 lm/W back in 2014, at a power level of 350mA and CCT of 5150K.

In the past couple decades, there has been a huge and exciting explosion of LED lighting tech, dramatically improving lights by a factor of 10X or even 20X in a relatively short amount of time, mostly by switching from incandescent to LED. It has been a good time for everyone involved.

However, it looks like the party is coming to an end.

There are definitely still significant gains left to be accomplished. We could still double the efficiency of average lights. However, that’s about it. There will probably never again be a rapid explosion of progress like what happened in the past couple decades. We could see a 2X or maybe even 3X improvement over time, but nothing like the 10X or 20X increase we already experienced. What’s left is a matter of diminishing returns.

I think many of us have probably already noticed this, given how many lights don’t use the latest tech. Instead of XP-L2, we’re using the older XP-L HI. Instead of XP-G3, XP-G2 looks better. People happily forego new LEDs like SST-20 and SST-40 in favor of old “obsolete” items like 219B, because the new stuff looks too green.

And then of course there are also losses in the lens, the reflector/optic, the driver, the wires, the springs, etc. Each one is small, but it adds up. So instead of 251 lm/W, the practical limit for a pure-white flashlight may be more like 200 lm/W.

That’s a big improvement over my 219B lights with their measly ~70 lm/W. And a big improvement over my XP-L HI lights at ~100 lm/W. I don’t really need my lights to be brighter, but I like the thought of getting 2X or 3X as much runtime per battery charge… even though I rarely ever need to charge anyway.

So… that’s what appears to be ahead in the future. We’ll keep seeing improvements for a while, but each year will probably bring less and less exciting changes until it all seems fairly mundane… and we’ll move on to other exciting things.

This rapid improvement, followed by diminishing returns, is a pretty common pattern.

In a lot of fields, things suck for a long time, and then breakthroughs happen and tech advances at an exponential speed for a while, and it’s an exciting time to be alive… but eventually the tech gets about as good as it needs to be, and it slows down or stops.

For example, it happened with air travel. It’s how the Boeing 747 was made. Air travel was a hot industry with everyone competing to make faster, fancier planes, and Boeing dedicated their best people to engineering the best plane ever made. It was a supersonic sky limousine, the future of air travel, capable of going three times faster than sound. It was a crazy time full of hype and glamour, and one airline even started taking reservations for commercial flights to the moon.

Boeing also put some of their less-talented engineers on another project, an unexciting plane called a 747, meant to haul cargo and generally be a boring workhorse. But the 747 became the world’s most popular method of air travel… while the supersonic sky limo faded into obscurity. Because the sky limo was incredibly expensive and offered little benefit over more mundane options, while the 747 basically hit the most optimal point between cost and value. So 50 years later we’re still using it, and air travel hasn’t really improved since then. Like, why bother cutting a 4-hour flight in half (at triple the cost) when it takes another 8 hours to get to/from/through the airports? Air speed is the wrong part of the process to optimize.

It looks like the flashlight industry is following a similar pattern… and we’re approaching the end of the really exciting part — moving into the long phase of slow, incremental improvements.

Fun read, TK! :+1:

I gladly sacrifice efficiency for the rosy tint of the 219b sw45k.

On an average day, my EDC lights get used several times for a few minutes at moderate levels. I can usually go weeks before needing to top off their cells.

On a non-average day, I might use a headlamp for a few hours and top off the cell afterward. I rarely need to swap cells mid-task.

EDIT: I posted before you filled in your second post. Great stuff!

Maybe someone will make a new type of emitter with 99% efficiency :open_mouth:

This is an interesting topic. Could it be that there is actually still a lot of room for improvement? I think so, but I wonder if I am missing something that makes further improvement impossible.

Using lm/W as the one measurement to optimize for could lead us to the wrong conclusion. This value only tells you if “humans can see more light”, and not “is this LED better at converting electricity into light”. A hypothetical perfect LED, where 100% of electrical power gets converted into photons, would produce zero waste heat. Yet a green perfect LED would have a much higher lm/W measurement than a red perfect LED.

Is wall-plug efficiency not what we want to measure and improve instead? This measurement tells you what percentage of electricity is converted into light instead of heat. Small hotrods could be a lot more powerful if the amount of heat could be reduced, even if they didn’t produce much more light.

Of course, it is unlikely that we will ever have an LED with 100% wall-plug efficiency. But let’s consider a real-world example to see what wall-plug efficiency LEDs have today, and if we would see any benefits from improving it.

I found these AMAZING slides about LEDs and their use in horticulture. They are very technical, but I think many will enjoy reading them.

Here are some WPE values from the slides. WPE varies depending on how hard the LED is driven, among other things, but let’s ignore that for this example, and just consider the current at which WPE was measured.
SST-20-B 72% WPE (Blue)
SST-20-DR 59% WPE (Deep Red)
SST-20 2700K 35% WPE
SST-20 4000K 43% WPE

For SST-20-DR, 59% of electricity gets converted into light, and 41% into heat. 100% WPE might be impossible, but what if we could improve WPE to 80%? We would still be using the same amount of electricity, so runtime would be the same. There would be a small increase in the amount of light produced, but it would not be noticeable. On the other hand, heat would decrease by half! That would be noticeable.

What if we consider SST-20 2700K? With 35% WPE, about 2/3 of electricity is being wasted as heat. If we could improve WPE to 70, we would get twice as much light, and half as much heat! That would definitely be noticeable. What if we could improve WPE all the way to 95? There would be almost no heat produced, and small flashlights could become a lot more powerful while being comfortable to hold. It would be a game changer.

I don’t know much about physics, perhaps there is a reason that 35% WPE is approaching some kind of a physical limit for LEDs like SST-20 2700K, and it cannot be improved further. Does anyone know?

Keep in mind that when LEDs are overdriven, WPE suffers, so in practice a hotrod has a much worse WPE than the measurements shown above. Even if those measurements stay the same, we could see improvements that would significantly increase WPE when LEDs are overdriven, while keeping the measurements above approximately the same. There are also inefficiencies in the driver, and there is heat produced by the driver and the battery, so even a flashlight with a hypothetical perfect LED would still produce some heat.

There may still be significant room for improvements in efficiency at higher intensities.

Oh, great. Way to pop my balloon…

I will now sit and await the heat-death of the universe.

As said, as efficiency nears maximum theoretical, heat will stop being a limiting factor, and battery power will be the bigger limitation.

Therefore, if battery technology sees more rapid imprpvements, flashlights will follow.

Having said this, I don’t think there will be another huge jump in technology like there was in the first decade of the 2000s, more evolutionary and slow steps, as it has been pointed out.

The MPG illusion is interesting, and it is distantly related to one of the issues with using lm/W to measure efficiency.

Fuel consumption in the United States is measured in miles per gallon (MPG), while most of the world uses liters per 100 km. Most people will only be familiar with one or the other. You can use both to measure fuel consumption, but they are not the same. MPG naturally makes people underestimate the efficiency improvements on the low end and overestimate efficiency improvements on the high end.

And so it is.

I agree with that, and with most of whatever is below. :-)

That’s what I was thinking. At high output levels the efficiency of an LED drops a lot. If they can figure out a way to not let that happen, we could have lights with 2-3x the output whilst using the same power. Imagine a 25000 lumen EDC! :open_mouth:

Strange to see these comments about LED’s limiting out on output and efficiency. I have been thinking about this too from a noob’s point of view. I look and try to understand Lumens, KCD, CRI, etc… and will continue doing this with 3 lumen tubes, 2 light meters, 3 dmm, etc… because I love this hobby and everyone here at BLF.
But one thing I use to judge is the simple eye test. Thanks to Matt’s video, I came to a decision not to mod a light to the new 90.2 led because I like the 70.2 better. I know the numbers say different, but pictures seem to show different too. So I am most likely wrong about the 90.2 and I am going to get a kit from TA and put it in a light, this will most likely be the only one. So it does appear to me also that leds are near their max performance. You have to admit they are quite impressive and the variety suits many different needs.

XHP 70.2



The lights I use most often only clock in at about 100 lm/W (XP-L HI) or 70 lm/W (219B) with a linear driver. And, of course, that number drops when the lights are in really bright modes. It’d be really cool to have lights which get 200 lm/W instead, without sacrificing beam quality.

For me, the biggest win there is the increased runtime. Lights could run for a really long time between charges, and this would be especially important for small lights with 14500 / 16340 / 18350 cells. It’d be nice to get output and runtime on those similar to what we get from 18650 lights today.

Higher turbo modes would be fun too, though I feel like lights have already gotten brighter than I really need. I use a D4 a lot, and it can do over 4000 lumens… but I rarely ever want more than 100. So instead of a high-power quad, I’d mostly be happier with a more compact light with a single LED and a driver which runs very efficiently at low modes.

That might all be limited to 2X or 3X remaining improvement, but battery tech could go quite a bit farther. They’ve been working on a couple new types of cells, each one a huge upgrade from what we’re using today:

  • One type changes the internal structure of the cell. Instead of a coiled roll of materials, it changes to something like a 3D checkerboard pattern. This increases the charge and discharge rate dramatically… like, by a factor of 1000+. Today’s high-amp cells may be able to start fires in under a second, but that pales in comparison. Instead of 20 Amps output, it could do 20,000 Amps. It’s not an exaggeration to call it an explosive level of power, considering it works in the same way as gunpowder. I don’t need or want that much power in a flashlight, but it does have a really nice benefit — extremely fast charge times. Instead of hours, it could charge in a minute or two.
  • The new battery tech I’m most excited about is … wood. They’ve been developing cells which are, basically, rechargeable wood. The benefit of this is that it increases energy density by about 7X. So, instead of a 3000 mAh 18650 cell, it could be a 21000 mAh cell. Great for extending runtime and/or greatly decreasing the size of a light without decreasing the runtime.

Combine that wood battery with a 2X more efficient light, and it could be a 14X increase. That would mean a tiny 16340 or 14500 light in the future could compete with today’s 3x18650 lights. A little S-Mini or AA-sized light could make as many lumen-hours as a SP36, D18, or ROT66. Basically, look at the runtime of any current light, and add a digit… and that’s what we might be carrying around someday.

So… even though the LEDs themselves are running out of room to improve, the future of torches still looks pretty bright for another decade or two.

UVC LEDs have recently progressed quite a lot, moving from fancy pricey curiosities to possibility of practical (but still quite pricey) applications. They made major strides in both mW/W efficiency and total mW per LED that will sooner or later allow them to displace existing UV lamps.

I think that slow progress in Led industry is good for flashlights. Without new led and
without the ability to write larger numbers on boxes flashlight manufacturers would have to think about convenient UI, quality of spot, quality of light and so on.

I’m not following the latest developement but last time I checked, searching info about Soraa Vivid technology, it seems to me that the researches were gradually switching to laser based lighting for better efficiency that is more free of current droop. OLED panel might improve for diffuse lighting sources and laser for concentrate lighting sources.

In the future, your flashlight will be linked to eye-tracking. So you can light up a huge space with a narrow beam that rapidly scans across the scene to match where your eyes point. To the operator it will seem as though he has an infinitely powerful flooder.

May be you not know about this Field of view - Wikipedia

You only see sharply in a small area of your eye called the fovea centralis. To take in a scene you scan your eyes around rapidly, 3-5 times a second.

In VR there is the concept of foveated rendering. The eyes are tracked and only a small portion of the scene - central in your gaze - is rendered sharply. Everything else is rendered at low-quality.

It can work with a flashlight also.

It is intended merely as an example of how there are many inefficiencies in the field of flashlights. So even when we hit the limit of LED efficiency, there will be other things to work on, such as how efficiently we direct photons towards what we are looking at.

In automotive there’s scene detection already.