Slim RGB Matrix Puts LEDs Inside The PCB

Sometimes all that’s required to build something interesting is to put the same old pieces together differently. [Sayantan Pal] did this for the humble RGB LED matrix, creating an extra-thin version by recessing WS2812b NeoPixel LEDs inside a PCB.

The popular WS2812B is 1.6 mm in height, which happens to be the most commonly used PCB thickness. Using EasyEDA, [Sayantan] designed a 8×8 matrix with modified WS2812B footprints. A slightly undersized cutout was added to create a friction-fit for the LEDs, and the pads were moved to the back side of the panel just outside the cutout, and their assignment were flipped. The PCB is assembled face down, and all the pads are soldered by hand. Unfortunately this creates rather large solder bridges which slightly increases the overall thickness of the panel, and is probably also unsuitable for production with conventional pick-and-place assembly.

We’ve seen some similar methods with PCB assemblies that use layered PCBs. Manufacturers are starting to even embed components inside multilayer PCBs.

16 thoughts on “Slim RGB Matrix Puts LEDs Inside The PCB

  1. This should be the new standard for how things are packaged! With cheap four layer boards we don’t need quite as much routing area, and it could be easy enough to socket or hand solder to replace DIP. You could surface mount an inductor directly on top of it’s in-PCB chip with all of its passives. The friction would presumably give a bit of mechanical support.

    Cuts could be slightly angled or funnel shaped and done by laser cutter, so not much precision would be needed to get a part wedged in, and things could be reworked with some effort by heating and pushing out from the other side.

    Maybe we could even use more bismuth low temp solder, with less concern about mechanical strength.

    1. For a board like the photo in the article, I don’t think more than 2L is needed. And if you could get “gull wing” packaged LEDs, it would be pretty easy to get a flat and thin assembly.

  2. I wonder if it would be possible to use the internal layers to prevent soldering on the external layer (by doing a small cutout to access these layers and thus the solder will be more flush.

    1. Or use solder paste and a toaster oven.
      Use 2mm FR4, make the pocket 1.6mm deep, put the pads on the inside bottom, coat with solder paste and stick in your toaster oven. Bob’s your dad’s brother, and the LEDs are flush.

    1. Before reading the whole article I thought better heat transfer would be the point of this hack.
      Skip the copper of an n-layer board and just put any kind of heatsink at the back with some thermal conductive pads (dunno the correct term).

  3. Instead of hand-soldering all those connections on the back, you could reflow the LEDs onto a polyimide (Kapton) film-type printed circuit: just 10 mils thick, probably thinner than the hand-soldered bumps.

    1. Isn’t the common construction for these pannels to use a flexable substrate? The ones I have are that way. Two layers so there’s some heat sinking–which you need badly for these larger arrays. I’ve got a 16×16 one and it can suck a lot of current.

      I’d rather see someone design one on Al-core PCB–an adhesive layer of amide board stuck to a piece of aluminum.

      1. The linear (1-D) strips are commonly found on flexible substrate. I have not seen a 2-D panel with that construction. Got a link for the one you mention?

        A thin Al-core PCB is helpful as a heat spreader, but it still gets hot: you still ultimately need to dissipate the heat somewhere. For my higher power arrays I laminate the flexible polyimide (not amide!) substrate directly to big finned heatsinks with thermal epoxy. I don’t use the pressure-sensitive adhesive types for that. Pretty easy to dump >1W/cm^2 even with just convection. I’ll run up to 4W/cm^2 for minutes at a time, but that gets pretty toasty, even with 3 cm deep fins.

      2. PCB laminated onto copper or aluminium sheet is pretty common these days. For working with the things yourself I would recommend copper – much easier to bond to than aluminuim.

        1. Unless you are soldering the device to the copper (which works a treat, BTW, if appropriate), I’ve found thermal epoxy bonding to aluminum to be far superior than copper. I etch the aluminum first with a 1N NaOH solution for about 30 seconds, then rinse with DI water and dry thoroughly. Bond it within a few minutes, before the oxide grows back. Darned near indestructible bond.

Leave a Reply to Fred Cancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.