Its a firmware thing, built into the logic that runs the display board. As I noted, one could simulate varying levels of brightness in software, if a faster rate of interrupts were available (the kernel scheduler only guarantees 100 times/sec).

The idea is that LEDs are binary, they are either ON (very bright) or OFF. In-between levels of brightness are achieved by switching the LED on/off very quickly, far faster than the 100 times/sec that is easily available in the kernel.

Cheers