PSA: WS2812B LEDs may not be what you think they are!

Hey guys,

So I’ve just spent the last week investigating this topic as I’ve recently learned that a couple thousand dollars worth of our inventory is defective because of bad WS2812B LEDs.

Story Time: Discovering the Issue

(Feel free to skip to How to Identify Smart LEDs)

For a while we’ve been sourcing Smart LED flexible strips from various sources: eBay, Amazon, Aliexpress, Alibaba with varying results. All searches were done using the term WS2812B which is a 4 pin integrated smart RGB LED. It only needs V+, GND rails and Data in/out using a daisy chain structure and sending data down the line sequentially.

We now rely on a very stable supplier who I’ve now purchased about 800m worth of LEDs without issue.

However, we also needed to make some custom double sided LED PCB bars so I went and sourced WS2812B LEDs directly from WorldSemi. These we spent about $800 worth on discrete LEDs, ~$100 on pcbs, and then another $1200 to have them assembled by a local fab. After everything was made we did some QC tests and found that the LEDs were failing in a weird way. Basically 1/10 LEDs had one or more internal diodes failing. Typically the blue diode, but occasionally red or green and in rare cases multiple or all three diodes failed.

Here are a couple videos of the behavior we were seeing:
Test 1:
Test 2:

The failure seems to be temperature sensitive, and in extreme cases pressure sensitive too. Being temperature sensitive means that from a cold start running full brightness white, in about 30 seconds you’ll start to see issues, but after it’s warmed up (~1-5min) those issues will disappear. If poked (pressure) it’s very likely you can permanently damage it. In the second video that one LED I poked never worked again and in fact the data line has been severed so subsequent LEDs will not work either.

This raised a lot of questions as to why our flexible strip LED supply was so reliable, but the LEDs bought directly from WorldSemi were not.

After investigating bad WS2812B chips and variations of the chip, we discovered there is a clone called the SK6812. And I’ve since confirmed that our stable flexible strip supplier has been selling these since early 2016 without telling their clients. Since they’re directly clones they felt there was no need to change the name of their product since it uses the same WS2812B protocol. However the main reason they switched was because they’re way more reliable than the WS2812B. There is also some slight improvements in the electrical characteristics, namely double the PWM frequency on the diodes (making them better for persistence of vision projects).

###How to Identify (4 pin) Smart LEDs

I’ve compiled a imgur album, but I will summarize the main points:

  • WS2812(S) is the 6 pin grandfather of the 4 pin smart LED, which follows a stupid naming convention. WS2812 is the name of the Smart LED with RBG diodes and an integrated IC. The name of that IC is WS2811. The reason this IC has 6 pins is because the WS2811 has separate power rails which is rather pointless. 6 Pins makes this easy to identify on it’s own.
  • WS2812B is the next evolution with 4 pins. From the images you’ll see I’ve found at least 3 different variations of it. To identify this IC you’ll need a microscope to look at the bonding wires. I’m mainly focusing on the IC itself because of the bonding wire pad placement. You’ll see that it has 3 closely place pads along the short edge of the IC and on the opposite side 3 spaced out pads. Other indicators for variations of the LED package are the placement of the blue diode and shape of the pads inside the package.
  • SK6812 is the clone Smart LED and is most easily identified by the IC being rotated 90°. Under a microscope you’ll also see that the 3 closely placed bonding wire pads are now along the long edge of the IC.

One other way to differentiate the ICs is the size:

  • WS2811 is 0.8x1.05mm
  • WS2812B is 0.7x0.9mm
  • SK6812 is 0.6x0.8mm


If you’re buying this type of smart LED, make sure you get the good stuff. Verify with your supplier which IC is being used in the product because it seems like most manufacturers aren’t bothering to tell people they’ve switch to clone.

I’ve done most of this investigation through discussion on a few blogs. For more information on bad WS2812B LEDs and other failure modes check out:

For more information on the SK6812 go here:

And for a summary on the differences between WS2811 and WS2812B, go here:

###After Credits

WorldSemi has yet to respond to my inquiry of the failed ICs but I expect with a small order (on their scale) I expect they’re probably not going to bother compensating me in any way.
Shiji has been very helpful and offered to supply me with the SK6812 discrete LEDs in tape/reel. I’m not sure if we’ll do that though because the cost of rebuilding the inventory is far higher than falling back to an old design using old inventory we already have.

Other things to note:
There are also a new generation of smart LEDs if you’re interested in looking them up. Some have RGBW diodes with varying white color temperatures. The next generation of ICs WS2813 and SK6822 which have dual data lines, this is a redundancy so that if one dataline breaks, you can still have signal going through. Interestingly WS2813 also incorporates the decoupling capacitor inside the LED package too allowing you to cram more LEDs closer together.
Another few other names to throw out: APA102 and SK9812. I haven’t looked into these in detail but I believe they’re also 6 pin packages with different protocols or data line redundancy.


Wow, so the SK6812 clones are more reliable than the WS2812B originals? That’s not the story I expected when I started reading. Thanks for writing it up!

On the subject of serial RGB LEDs, I’d just like to point out Tim’s Blog, which I don’t think enough people know about. He’s done interesting write-ups on [APA102 and SK9822 semi-clones] (, as well as WS2812 and PD9823 clones, and a bunch of other useful investigations and interesting projects. Highly recommended.

Edit: Oops, I see you already have Tim’s Blog in your list of resources. Don’t mind me, I’ll see myself out…

1 Like

Gonazar, Please put your posting, public service announcement on our wiki. That is where it belongs.

Great analysis - its interesting to note that the dies (LEDs and data
chips) are sure to be coming from different fab houses and thus the end
user quality and reliability will often depend upon the chip+die+wire
bonding process.
Most field use failures will I am sure be due to the stresses introduced by
the moulded nature of the encapsulation which as far as I have seen does
not hermetically seal the chips or internal bonding wires.
I have already seen LED Street lamps and traffic light assemblies with
missing elements after only a few years since they have been in use.

I have switched to LED COBs which so far have excellent reliability, if I keep them cool. I made this switch after a bad batch of individual LED chips. I had a similar problem with chip die wires, with my non-address able LED chips.

Is there an addressable LED COB implementation?

Did the WS2812’s from WS come properly sealed with desiccant, and the usual moisture indicator? I’ve heard similar stories before when people have opened/unsealed LEDs too long before assembly, and the moisture that gets absorbed is enough to damage the LEDs as they go through the reflow process.

As far as I know they were kept sealed. We sent them unopened to our fab and they didn’t catch any issues on their end. They have an automated pick and place + industrial reflow oven. Aside from setup, I believe they assembled everything within a day.

Something like this?

I don’t think I have write access.

I have a few of @Jarrett WS2812 RGB boards and a stock of WS2812 to reflow onto them…
And the WS2812 definitely did not come with desiccant (in fact I have yet to see anything come from AliExpress that is sealed with moisture indicators like DigiKey/Mouser does…

To try to avoid any issue I’ll try heating up the chips to dry them out for a period of time before I paste and reflow them. The oven has a mode to do this. I just have to find some rough idea of a recommended temp and time…

It’s a bad enough issue that manufacturers pretty much all have recommendations technical documents somewhere. It looks like the datasheet on the WS Alibaba page has the baking directions, but curiously the datasheet I’m finding elsewhere (Adafruit, etc) is older and does not. Even more importantly, they say always pre-bake the LEDs.

What I need is more a recommended temp and time for drying rather than the reflow cycle…
Something to ensure most of the moisture is gone from the WS2812 before you put it through a reflow…

That’s what’s at the bottom of the image, “Reflow considerations”. :wink:

To translate from engrish:

  1. Bake at 65-70 C for 24 hours before using
  2. Assembly and reflow within 2 hours of removing from the oven (I expect this is easy to push out in our dry winter air)
  3. If not reflowed within the 2 hours, rebake before assembling
  4. ??? Not sure. Maybe it’s saying if you’ve placed the parts but not reflowed in time, toss it in the bin?

Missed that…

I have 12 of the WS2812 strips to do (each has 5 leds)…
Perhaps I’ll try half without the pre-bake and half with…
Just to get an idea if it makes a difference…

Awesome! You have saved me from more disgruntled clients.

I don’t have any association with VHS, but I saw this post from a link on and it resonated with me because I ran into the same issue. Good job on your research.

We were making a simple 4x4 matrix of WS2812B LEDs. We bought a few thousand dollars worth of reels from World Semi directly and ran into a few problems with the sample boards we made. I thought that maybe we’d just forgotten to test them right after making them since they were shipped around to a couple different factories before they made it back to me for testing. That should have been a red flag.

Then, we started running the smt line and immediately started noticing about 5% of our panels were failing testing. We checked the obvious problems: placement, reflow profile, etc. Everything followed their datasheet. We’re usually close to 99.9% yield for smt, so this was concerning. To make it even worse, the pick and place machine was configured with 4 feeders (one per row of our matrix) and after reloading it, one of the reels World Semi sent us was mislabeled (labeled as a WS2812B but it was a completely different chip). We caught this soon enough, but we had to rework the first few panels after reloading the feeder with that bad reel.

Ultimately, we were able to get an informal response from World Semi. The source there claimed they know ~0.5% of LEDs are defective on the reel. That’s ridiculously high and shows a complete lack of QC in my mind. I won’t be using their chips again. Given that experience, I wouldn’t recommend any one use these WS2812B devices (or any devices) from that manufacturer. It took hours of manual testing/rework to figure out which LEDs were defective and replace them. Ridiculous to have boards fail at the component level and the manufacturer response was completely unacceptable.

I started googling around like you, and found many of the same blog posts you cited.

As for root cause analysis, I think it may have to do with World Semi’s semiconductor packaging process. I am just guessing here, but I think they’re going with their current method to avoid any IP/patent issues. Their packing seems to me to be the reason for failures. The fact they aren’t able to test their own packaging process on their factory floor shows bad management (in my mind). I’m not making any future products with their chips.

The most frustrating part - I had asked our supplier about the SK6812 as we were getting quotes for them that were much cheaper than World Semi’s WS2812B. The supplier said they don’t know much about the SK6812 manufacturer and was thus worried about doing business with them because they were unsure of the reliability. Lesson learned there.


That’s a really good confirmation. 5% is huge, especially in the high
volumes you describe.


This thread was helpful. I may have decoded “Reflow consideration” #4:

You cannot send the completed PCB back through reflow a second time.

1 Like

i’ve had a bad story with worldsemi on end of 2014 ,my guys ordered approx 150 reels at approx 10 dollars per meter and we get all the decoder from ltech (china),artnet from Madrix, we spent 12 days on workshop to do installation and testing ,and we start the application /final installation…when my client open the light every day, the color start to die ,and we started the crazy time , what wired was that we found the died LED start working again after we turn on the whole light …finally my friend help me figured the problem out ,WS2812B too sensitive , the chip inside has bad quality,when the strip getting hot,it’s stop working proper and keep working again once it’s cooler, i’ve never had such experience before .but finally we have removed all but get nonthing replaced from semi
same as poster i start to search on aliexpress ,amazon made in china and alibaba ,found there are too many company after talk to the different supplier (i dont trust anymore), all the salesman even didn’t knew this bad issue with this chip and all of them promised it’s top quality 2 year warranty since then i get the answers from ipixel led company ,the guy seems had same bad experience and opened to me that worldsemi made bad chip during that time ,he showed me a photo comparing two ws2812b , old one used bigger chip and new one smaller
i really scared of this kind products but we must fix the problem for my client who paid me, i tried 10reels of sk6812 sample from ipixel led and testing in my workshop ,seems not bad and stable , it’s another chip but program absolutely compatible this make me easier as we spent a lots on it
finally we ordered 200 reels from this <REDACTED -SPAM> after approved my own non-strop testing ,my guys spent 2 weeks again to remove and fix by this new one fortunately
just attention with the WS2812, ALSO important thing is that talk to someone profesional instead of keep asking to pay the invoice