As I know LED is based on electrons and holes recombining in a piece of semiconductor. Emitted heat is very low. So why aren't LED's almost eternal? Why do they degenerate with time?
My guess is that maybe material reacts with eg. atmospheric water and it decreases crystalline quality and recombination possibility. Is my guess right?
The reason is Arrhenius Law of chemistry affects most things in life. It applies to batteries capacitors, semiconductors, magnetics, as well as non-electronic materials. The added thermal energy accelerates a chemical reaction of molecular leaching or separation or contamination. For LED's the industry standard is called LM70 where the "Lumen Maintenance" threshold for 30% reduction or 70% of initial radiated light in lumens where degradation occurs due to thermal sensitivity of chemical reactions. Not only the phosphor degrades but the thick blue LED substrate, but not necessarily at the same rate and thus shift towards blue also occurs.
This Law has been verified with field failure rates and used in MTBF calculations defined by (US) MIL-STD-HDBK 217 (also screening to Std 883) which gets revised at least every decade.
This degradation in MTBF or failure rate or life span is true for all semiconductors, caps, batteries and LED's and why accelerated life tests are done on Capacitors for 1000 hrs at 85'C or 105'C, and why car batteries sulphate faster in Arizona and have short life, and why LED's get dimmer and eventually fail.
Although the failure mechanism differ in each chemistry, the general law has process variables such that the reduction in MTBF of 50% for every 10'C rise in junction temperature ( the (semi- or )conductor <> dielectric interface ) So the Rule of Thumb often has a 20% range in results for one material vs another.
Some chemicals also has laws of physics where the thermal activation energy can cause secondary catastrophic failure and not simply a logarithmic degradation.
There are many causes of failure not included such as wire-bond weakness and sheared by CTE (coeff.thermal expan.) thermal stress plus vibration after testing and ESD from designs which do not have protection zeners included. (many)
The above examples of failure are normal lumped into infant mortality and depend on vendor quality/design and user abuse in manufacturing for ESD awareness where parts may get wounded and fail early from heat or some other energy.
