.
I’ve had this happen wit SST-40s while trying to chemically de-dome and presumably on an sst-20 (but that one may have been burnt bond wires…).
Don’t know what could be happening besides bad heatsinking and thermal cycling fatigue.
Did you consider to change the mcpcb? And it might happen - I had a crappy batch of XP-G3’s a few years ago, they were dying on me after 20-30 hours in low current applications.
regards
Dominik
Could the driver be overvolting it?
I killed several of their xp-g3 emitters around this time last year. For some reason they don’t work with the AMC7135 drivers.
They’ll start flickering and eventually burn out or develop brown spots, even with 3,6V input at the driver.
They work fine with a Meanwell LCM60 or the DQG AAA driver.
Could be failing 7135s, especially since the G3 has very low Vf @1.8A, those poor guys are gonna be getting HOT
I don’t know the cause. I was trying to use them in several IKEA Janjsö with two AMC7135 stacked each.
To make things clear for those who are not familiar with this problem: The second you connect the xpg3 led to the AMC driver, no matter what input voltage or current limit you set on your power supply, they will get damaged.
Forward voltage will rise and output suffers until the led develops brown spots and fails. No overvoltage or overheating necessary. This happens even at the moon mode levels.
Sounds like black magic 
Regarding the xpg3 chip debonding, this also happened with the XHP50.2 when it first came out. It sounds like the bonding process is not as robust, and there are just some bad batches. Very frustrating.
G2 has much higher Vf than G3 and other flip chips
Could be worse, you could be stuck with just a dashboard light to experience paradise with.
I had one die after 30 minutes of use, too.
Since you’ve had this happen with several different emitters from different manufacturers, and on both copper and aluminum boards, do you think it could be something in your reflow technique that causes damage to the emitters? You mentioned that you tin the LEDs before reflowing…possible that you’re inadvertently overheating them in that step? Most of these are pretty robust really, compared to some like Nichia and various UV emitters, but it still doesn’t take much to damage them before they are mounted to a sink. With good solder (paste or wire) and a little extra flux, no tinning is necessary here (or helpful, imho). I don’t think rapid cooling would hurt so much but I just remove them from heat asap and let them air cool. That generally seems to come close to the cool-down ramp shown in the data sheets. With UV I’ll put those boards on a non-heated aluminum plate to let them cool just a bit faster. The 7135s did cross my mind, too, but I’m thinking it might be in your process.
Hmm...
who's the troll? 
@rayfish
thanks for all your testing.
I got mine from led-tech on stars. I doubt they are out of spec.
They even did some testing themselves and their leds ran overnight without problems.
It has to be the driver.
At 4,7khz you could be causing a resonance cascade, so be careful.
Rayfish…no worries using alcohol for cleaning the domes - assuming you mean isopropyl (or whatever it’s called in Europe) and not some other special/weird alcohol. In fact, isopropyl is the one and only (to my knowledge) solvent that is actually recommended universally by LED manufacturers. Best is to not need to clean at all, of course, but if necessary then iso is the stuff. Some emitter domes are silicone and some use other materials but iso is safe for all (maybe not for immersion but for wiping, yes). That said, I would use a less diluted version. Here in the US 91% isopropyl is excellent and quick drying. There is also a 70% which is usually sufficient, bit slower drying and easier to leave marks behind sometimes. We also have labels that say “rubbing” alcohol, and those are a bad choice because they most often contain additional oil and/or fragrance so they aren’t a clean solvent. We can buy 95% or 99% as well, but those are much more expensive and unnecessary for this kind of thing. 91% is great. I like to use electronics scrubbing applicators - like cotton buds except they use a polyester mesh for a tip instead of a wad of cotton. They’re cheap enough, never leave behind any fibers, and you can scrub harder with them if need be, like if some stubborn burned flux isn’t giving up easily. Other solvents like acetone or anything nastier are a poor choice for emitters but if those are ever used in the light or circuit boards, be sure to let them completely air out so their solvents can’t potentially damage the emitter dome (same is true for normal silicone adhesives and even many silver-bearing thermal pastes).
Sounds like you made a pretty fancy heater! A lot of people use a simple small kitchen electric hot plate (one that has good heat control is preferred) and maybe a small additional plate of metal to put in there as a buffer sometimes. These little PTC element heaters are very popular with the television/monitor repair crowd and they also work well…this is the first one I found on an amazon.de search and it looks like the typical product…has the right temperature on that model. There are lots out there and they can also be bought for much less money on Aliexpress (maybe Banggood too). I got one and have used it a few times…quick and convenient although since it heats up so fast I do use a small buffer plate…turn it on and off a couple times to allow for a more gentle ramp-up in heat…maybe not necessary.
Rayfish, great! And that’s a neat setup with the pumice stone. Those PTC heaters are nifty…here on US Amazon and on Aliexpress I find them as complete units with power cords, standoff legs/feet, and some with temp controls, but I couldn’t see much like that on the German Amazon listings. I may put a rheostat on my unit to have some ramp control and fine tuning of the temperature. I still use a hot plate sometimes but the PTC is so handy and compact that I’ve enjoyed using it a lot.
I think thermal shock may not be a huge concern. The bases of many emitters are ceramic, so metal tweezers or hemostats/forceps won’t hurt in that way. Avoiding the extra high temperatures is more important. I use some of the cheap ceramic-tip tweezers now mostly (tips are a little fat compared to electronics metal tweezers, but they work ok). A good solder interface is very important though, of course. Be careful when pressing down on an emitter to seat it…too much pressure at high heat and it’s possible to damage a dome or even the bond wires, and of course you don’t want to smoosh out too much solder from underneath or create bridges.
These emitters are pretty tough mostly. The way we use them in lights gives a lot of thermal shock but they hold up well. When I get a new light, I do like to run it on moonlight and low levels for a little while before I let it go to high or turbo, just to help remove any humidity (China is often very humid of course, during manufacturing sometimes but also in storage warehouses). I had a couple lights arrive with a lot of humidity inside to the point where I could see a little condensation/vapor residue, so I just do this all the time now. I would like to think that a gentle “burn in” with low heat is nice for the emitter, too, but that probably doesn’t matter. 
Glad you go things working well again!
I played around with every kind of home made device for years — then one day I bought one of these and the rest is history — once you use one you’ll never go back
https://www.amazon.com/gp/product/B07FP5988T/ref=ppx_yo_dt_b_search_asin_title?ie=UTF8&psc=1
I used amtech from ali. Highly doubt it’s your reflow. I use a jet flame lighter to heat the star and rubbing alcohol to clean.
My reflows and the store bought from led-tech.de (on stars) run just fine, just not on the 7135.