This is more interesting than it first appears. WS2815 appears to have been made in more than one revision but the latest design uses a clever constant current series trick.
The lower half is a constant current (say 16ma) regulator. The top rail is at 12v.
- If you want no leds, you turn off the regulator.
- If you want 0xFF0000 (red) you turn PWM3 off and PWM1 & 2 ON; the green and blue LEDs are actually shorted out.
- 16ma then flows through the red LED only, dropping about 1.2v in the process. The rest (about 10.5v) is dissipated as heat in the current regulator. This is quite inefficient (most of the 192mw goes as heat). However....
- Interestingly, the 16ma current consumption of the chip remains the same when you turn on Red+Green or even Red+Green+Blue - the wall-power consumed is the same (assuming 0xff intensity, is more complex otherwise), the difference is that less power is being dumped as heat in the CC regulator, and more is going through the LEDs. Sum is the same.
- Hence the efficiency of this is interestingly different to a 5v WS2812B; when driving 0xFFFFFF, it's actually slightly more efficient, however when displaying 0xFF0000 (or any other R/G/B single color) it's much less.
- I like it as a design; it's very cheap, it's not as awfully inefficient as
I first assumed. It'll work well on longer LED strips for several reasons; not least because a 300 LED strip on 0xffffff now only pulls about 5A @12v not 15A@5v, so less losses and much less heating of the strip due to copper trace resistance. Also, more tolerance of rail drooping.
- Adding an actual buck converter inside the LED looks to be impossible (cheaply) at the required power levels.
In short; power consumption on WS2815 is determined by MAX(r,g,b) NOT SUM(r,g,b) as it was on 5v LEDs.
Here's a detailed shot - note that not all WS2815's look like this; it appears that early versions may not have used the series trick
Note the Green and Blue LED dies are insulated from the pads they sit on, and have two wirebonds.
The Red only has one because it's connected to the 12v pad it's sitting on.
"NC" pad is I believe ~5v derived from the CC regulator, and is used to run the chip logic. The component (0201) visible inside the LED is I believe a decoupling cap for that supply
Update: This is confirmed by someone measuring the power consumption of a strip of 75 LEDs;