100W Addressable LED Panel

The adventure of getting 25¢, 9W RGB LEDs to run

https://github.com/wszeto9/chandelierLEDBoard

A spark of interest from mysterious Aliexpress listings

Aliexpress scrolling leads to very interesting finds of "I didn't know these things existed! I would love to try this out!". This part is no exception. I've gotten bored of the typical WS2812B 5x5mm LEDs. They're bright when looking at them directly, but they aren't very good at diffusing over a large piece of plastic for instance. After all, it's only 0.3W total. 9W on the other hand... now we're talking. One of these is just as bright as a cheap RGB lightbulb. Now imagine what a bunch of them could do...

Planning Drive Circuitries

The question after impulse buying these LEDs is... how does one drive them? They're at a weird voltage/current level. 700mA at 3.6V means a boost converter is out of the picture for individual channel control. A few other options that are possible are:

Constant-current buck converters
Open Drain LED drivers
The proven LED and resistor

When keeping in mind the cost of these solutions, very few options remain. After all, a 25¢ LED is not very cheap if it needs 3x $1 driver circuits. A constant current buck converter is around $0.50 to $1 per driver, so that's out of the picture. Most open-drain LED drivers aren't meant for LEDs that draw 700mA, so those are also out of the picture. What's left is a simple resistor to limit current to the LED. Those can be obtained for pennies. Is it efficient? Absolutely not. Perhaps one should calculate a theoretical efficiency before proceeding to create a power sinking monster.

Assuming a 5V supply and a 3.6V drop across the LED, a LED + resistor setup would be 72% efficient (3.6V/5V) at getting electrical power into the LED. This also means a 9W LED needs 9W / 0.72 = 12.5W of power input at full brightness.
Looking at a typical LED, the efficiency is around 40%. The total efficiency from LED + resistor input to LED light output is therefore 72% * 40% = 29%. A single LED and resistor would therefore draw 12.5W, and need to dissipate 12.5W * (1-29%) = 8.875W.
8 LEDs is probably a good number. Total heat dissipation is around 72W, which is manageable. Total power draw would be 12.5W * 8 = 100W.

The resulting circuit is surprisingly simple! All it has is a WS2811 to handle the data conversion from the standard WS28XX data protocol to RGB PWM, and a few P-fets to switch P5V to the LEDs.

Cooling the Monstrosity

How does one get rid of 72W of heat, and for cheap? Looking at industry uses for heatsinks, one large need for heatsinks is the desktop PC market. There's a 50-100W CPU that is essentially 100% efficient at converting power draw to heat, and the heatsink is expected to keep the die temp below 100C-- perfect for this project! Even more perfect, MIT has decided to throw out ~ 16 Optiplex computers around when this project was designed. I guess I am now in possession of 16 heatsinks that will perfectly fit a 4" board.

Board layout and routing

Rotationally symmetric, single layer aluminum board routing is quite a challenge! The Place footprints tool turned out to be very helpful in this journey. The sheet based schematic layout helps the software tool recognize which components should be selected as a group. Lots of shuffling was done to ensure that everything could be routed on a single layer. In addition, circular copper pours had to be created. KiCAD doesn't have a great way of officially creating a circular copper pour, so I thought of a workaround: pour a square of edge length L, and edit the copper pour to have a fillet edge smoothing of radius L/2. What results is a circle that has been fairly elegantly defined in KiCAD. Another challenge is sourcing a connector to power this board. I could not source a connector that could supply 16A+ in the space constraints of this board. So, I did the lazy thing: what if I just solder some thick wires to a SMT pad? Can't be that bad... right? Well... that might end up being a pain point.

5V Plane

The 5V routing is a bit tricky-- it needs to run under a lot of passives. The 2512 resistors are handy for this since they have a large area for the 5V current to run under. The jumpers are also sized to allow the 5V trace to run under it.

GND plane

Note the 8 jumpers colored in red. They jump GNDPWR (colored in green) to GND (colored in red). This solution is good enough in grounding, since GND is not expected to handle much return current. The components that use GND are the P-fet gates in order to turn on, and the WS2811 ICs in order to handle PWM logic.

A bit of CFD

Fusion 360 happens to have a handy tool for electronics cooling! The PCA was simulated in Fusion 360 in order to understand the expected temperatures on the board when the lights are at full power. The resistors are expected to output (0.7A * R) of power, and the LEDs are expected to output (9W * 0.4) of power. These values were input into the simulation parameters as the expected heat inputs of the system. For the heat removal, a solid metal heatsink was modeled to be roughly the size of the expected heatsink. Airflow was modeled as an average airflow of a 80mm fan of () CFM. The result was that no component exceeded 45C, which is an excellent result. The conductivity from junction to EP on the LED is likely high enough to keep the LED's die temperature under 80C.

Board bringup

As it turns out, aluminum is extremely hard to heat up. Who would have thought? The LEDs also pose a challenge-- the component can't be hand soldered due to its EP, and it can't be IR heated from the top since the plastic lens cap could start melting. The options left are IR bottom heating, or hotplate reflow. I actually requested that our EE makerspace at MIT obtain a hotplate specifically for board reflow, which, conveniently, works great for this board.

Soldering the wires onto the SMT pads turns out to be extremely difficult. Even a top of the line Hakko soldering station with a chisel tip could not melt the solder. The aluminum board simply sinked too much heat. The solution? Preheat the board on the hotplate to 150C, just below the junction temperature of the board. This provided just enough heat such that the wires cold be soldered with a soldering iron.

Once everything is soldered, time to power it on! And... it blinks. Hmm. Probing the 5V input of the WS2811 shows that it's dropping significantly once the LEDs turn on. Oops... forgot to add bulk capacitance. Guess I'll solder it next to the wires. Also, why is red not turning on?

Multimeter sanity check. Red has a 0R short to ground. Why is that? I assumed that the EP of a mystery aliexpress part would be connected to the negative pin of the LED. It is not. It is in fact, connected to the positive end of the red LED. WHY. Well... nothing a bit of kapton can't fix. I guess Rev A01 will have poor thermal performance. Time to fix it in Rev B01.

The LED legs also have trouble soldering properly. Looking under the microscope, some legs are not forming a proper heel fillet. Unfortunate, guess it needs to be reworked too.

Once that's all ironed out, the boards are working! Now... you may be wondering... how do you power this? Well, check out the 24 -> 5V buck converter that was designed and validated for this light!