For a while now i have been working on what would become on of the brightest LED flashlights available. Unusably large and bulky but at least its bright π
This post however is not about the finished product. Because well…. I am not quite that far yet. Instead i will talk a bit about high power LEDs in general and my prove of concept: An LED upgrade for a theater light, that runs from a LiPo battery.
First I had to select a LED module. For this there are basically three options, either a high current module that gets a few LED chips very close together and connects them in parallel (high current low voltage), the more common COB arrays which have many chips in series-parallel arrangements on them or a module containing multiple led chips connected like the cob, but instead of using chips, they use normal LEDs.
The Zoom level is identical for all three pictures
The high current array:
These have the advantage of having a small LES (Light Emitting Surface) for the amount of light they output, making them common in theater spotlights among other things, because the beam is more easily collimated than with larges LES. They are also available with a different arrangement of chips, making the LES effectively round, which is even better for spots.
However these LEDs typically run at around 25-30A and i have driven them at as much as 40A without damage (even if this is not recommended), which results in difficulty with the driver design.

At this current even very small resistances from things like traces become a serious consideration and I had my fair share of magic smoke releases testing my drivers as you can see.
And because the forward voltage is only around 4-5V at 40A you don’t tend to get over 200W out of this type (at least i didn’t manage without damaging the LED)
So good for beam quality but not as powerful as other options.
The COB Array:
COB stands for Chip On Board, the modules are made of a substrate with the LED dies directly attached which are then bonded (wired) together and the phosphor loaded silicone gel is poured on top of them. A very similar construction as the high current module but here the LEDs are connected in series in very long strings (>=10 LEDs most of the time) of which many are connected in parallel, so each LED chip only has to handle a fairly small current and the overall Voltage of the module is much higher. It will be hard to find COBs that can handle more than 3A, with a forward voltage of usually ~30V. So around the same power of the high current modules, but with currents an order of magnitude smaller, which is much more forgiving in terms of driver layout etc. And with 50V COBs becoming more common, even higher power levels are possible (300W and beyond for ONE module).
Of course at these power levels cooling becomes a significant concern, and a normal CPU cooler might not suffice, especially since the COBs usually have a slightly higher thermal resistance from chip to cooler, as the substrate is ceramic/aluminium as opposed to the solid copper in the high current modules.
So COBs are cheaper to make, more powerful, easier to drive but also harder to cool and require better designed optics.
The array of individual LEDs:
This is a technique becoming more common recently; the module consists of a number of individual white LEDs, in this example with different white temperatures. This is being used in some of the Warm-Dimming LEDs that are becoming available now (and technically the way almost all low power lamps are made). The controller can vary the current flow through the warm/cold white LEDs individually and thus control the color temperature of the light. However the LEDs have an even worse thermal resistance to the cooler and thus power levels are much lower (<30W mostly).
So quite good for photography and common in the household use but less powerful.
A quick note on CRI:
CRI (Color Rendering Index) describes how accurately the spectrum, that LEDs output matches that of classic tungsten lamps.
In LEDs the spectrum is dictated my phosphor formulation (unlike filament temperature in old school lamps), which greatly influences how good we perceive the light to be. A good example are old CCFL lamps, the phosphor formulations in these were quite primitive compared to those in use today and are the reason why many people don’t like them and why we still used tungsten until very recently. Modern LEDs can get pretty much to 100 (wich means exactly like tungsten) but there is a tradeoff between efficiency and CRI (higher CRI usually means less efficient).
Cooler selection:
This is probably the most important thing after the LED.
Without a cooler the LED would die in seconds and you would be sad.
Most DIY projects use some sort of a CPU cooler, which is a very good choice as CPUs have a similar size, similar power output and similar temperature requirements to LEDs and the coolers are very cheap due to market volume.
So how should you choose which one to buy? This is unfortunately not so simple. Most websites just tell you to use the input power of the LED as the thermal power it outputs, but especially at high power levels the discrepancy due power radiated in the form of light becomes significant enough to reduce cooler requirements.
I am still working on the formula and a tool to calculate this so check back later Sorry :/
For small LEDs the input power should be ok to use for heatsink calculation though.
A look at the optics:
This is more important than you might think. Most lamps use a reflector to focus the light forward, but as the LES increases the projected beam becomes more unfocused requiring larger reflectors, which in a flashlight is obviously quite problematic.
So i used just a glass lens, though a combination of lens and a small reflector is also very common.
The issue with using just a lens is that there will be a lot of light lost, that just shines around the outside of the lens if it is far away from the LED, but if the lens needs to be close up a low focal distance lens is required, which can be harder to find. There are lens kits for LEDs available from china though and for most applications they should be fine.
The problems start to occur when you increase the power level. I’ve had plastic lenses crack and/or warp due to overheating when used with very powerful LEDs (>250W). So you need to make sure that the lens you get is made from glass if you want to go beyond maybe 150W (at least with the ones I tested).

This is the primary lens and cooler in my lamp. You can see that there is < 10mm of space between lens and LED. Even though this is technically still too much, closer or using a reflector between lens and LED would be better, I did not have a large enough lens, or reflector when I build it though.
The driver:
This is probably the hardest part of building a lamp if you design it yourself, as you are dealing with a lot of power. A >100W converter is not easy to build.
Luckily we have the recommended circuit diagrams in the datasheets so at least the circuit layout should be easys enough to get right. The hardest part by far however is the thermal design. You need to get rid of potentially >25W in you converter (250W output with ~90% Ξ·), so you will need a decent heat sink. And also lots of via stitching if you go with an SMD switching transistor (which is really your only option, advancements in power density don’t tend to make their way to THT transistors anymore).
You might be inclined to increase switching frequency to reduce inductor size and increase efficiency, however larger mosfets tend to have a larger gate capacitance and (for me at least) efficiency went down again as I increased switching frequency above ~125kHz.

I used an LT3795 step up constant current regulator, it supports analog dimming, PWM and input current limit.
The inductor is the only one i had rated for this current but was too low inductance (I initially planned for 750kHz operation but reduced it to 125kHz later so I needed a larger one than I bought) so i had to put two in series.
Power source selection:
Since we are living in the 21st century the only real option for power on the go is lithium, especially at this power level. (stationary you could just use any old CC regulated power supply)
Personally i would always go for the highest series cell count that fits in your space, as the reduction in required boost factor will reduce headache while designing the driver. For my lamp I used a 6s LiPo intended for model flying. One thing to keep in mind here is overdischarge protection. Depending on your application you might have ot sense every individual cell. If you are using a pre-made protected battery pack, or designing your own, thats not a problem but when you are using a standard LiPo you will need to design a custom solution. I didn’t here (I only have overall undervoltage protection) and managed to popcorn 2 batteries because I overdischarged one of the cells in them.
Back to my lamp
I initially chose to go with a high current module, as i wanted good beam quality but later switched to COB as the driver was causing too much trouble and because of the power increase from using a COB. I went with a CXM22 Gen 4 from Luminus, CRI of 80 and a rated maximum power of 190W, because it was cheap (~20β¬) and readily available.
For cooling i used an old server CPU cooler but upgraded the fan to some crazy powerful delta one (just the fan draws >10W π ).
The optics are a lens from an old image slide projector, collimating the beam to about what the original tungsten lamp in the light would have been like, and the old front lens.

However because the Cooler assembly is A LOT heavier than the old lamp holder I couldn’t just use the old slider to adjust beam angle and needed to put in some guide rods and a connection plate.
They also limit travel, as the small lens would hit the large one before the slider bottoms out, so I put some pipes onto the sliders to prevent this (still marked STOP on the slider though just to be sure π )
The driver needed two iterations, as the first version had a too tight track spacing. The back of the circuit board is partially not covered by solder mask, to improve thermal transfer, but there were still control/feedback traces going through that area, including ones carrying up to 80V (the open circuit voltage of the driver). This was fine until I put some heatsink compound onto the insulating sheet, as it apparently had a sufficiently low breakdown voltage to nuke basically every signal in the circuit as i screwed it back together (seriously there was not one chip alive after this happened). So in V2.0 i increased track spacing and added a guard ground in between low and high voltage signals.
Another trap for young players that you get in COBs is that you have to keep the phosphor gel on these LEDs perfectly clean. No Dust, no fingerprints nothing! The optical power density (radiant flux density) close to the LED is so high that any dark object will start to burn, and once the phosphor gel is heated enough it chars and goes dark, making it get even hotter and causing a chain reaction that, if not caught in time, could burn out the entire phosphor, making the led lamp useless. This happened to me because some sud from me burning wood in front of it got deposited on the gel and it immediately overheated and even caught fire. I realised it just in time though and managed to clear of the burned gel carefully with a cloth. But this still changed the color temperature in the burned area and noticeably reduced CRI of the light coming from the LED.
So how bright is it?
Unfortunately I can’t objectively tell. The datasheet for the LED sais that at 3.5A you should get ~25k lm from it, I was running the LED at 4.5A because I am a very naughty boy so the LED probably output >30k lm (and was probably not very happy even though temps were < 90Β° on the substrate) how much of that came out of the front i don’t know but it was brighter than a Olight x7r marauder with 12k lm even at some distance. Without any collimation (so just the LED on its own) it was about as bright as a old school 2kW Halogen filming lamp (though the hal. lamp had a wider lit up area at the same brightness). The lamp is also visible in cloud free noon daylight at 3m if you know about where to look. And it very visibly lights up clouds at night (verified by a friend who lives about 1km away from me).
And probably the best (but also most disturbing) thing is: it has enough radiant flux to light a black piece of paper on fire when it is held in front of the lens. Halogen was able to do this too but it had a ton of infra-red helping it along. The LED does this with just visible light flux.
So it probably is not such a good thing to stare into for too long if you still need you eyes, and I was actually wearing my welding mask when playing with objects in front of the lens, just because they were too bright to look at diretly.
Conclusions?
Well, i am happy with the performance, but would like to improve it some more. There is a more powerful series of COBs available from luminus (CVM32 Gen 4) that officially goes to 300W (so can I push 400W into it without it dying? π ) though at this power level even more significant cooling is in order (maybe water?) and i will need a better driver/converter.
The other direction I want to go in is portability. If i ever find somebody who is willing to let me machine something on his lathe (or make it for me for cheap) I will play around with a few passive heatsinks and see how much power they can dissipate and from there continue designing the portable version.

I made a CAD of what i imagined it too look like as my first CAD project, but got stuck not knowing if it would be able to dissipate the amount of power I want it to. Maybe I will need an axial fan in a heatsink maybe I can get by without. I can really only find this out if I get a few test heatsinks machined. Also I have doubled the cell count over what i had in this design, so i need to make the model longer. And maybe reduce wall thickness and,and,and… Honestly this is such a large project that it will take me years to complete so don’t expect a market release anytime soon π
Also I have to say that the only disadvantage I see with LED technology at the moment is the heavy and bulky cooling required for the high power ones. Except for that we are at a point where we do not need tungsten lamps for anything anymore.