How LEDs Are Made

An anonymous reader writes "The SparkFun team took a tour of a factory in China that manufactures LEDs. They took lots of pictures showing the parts that go into the LEDs, the machines used to build them, and the people operating the machines. There's a surprising amount of manual labor involved with making LEDs. Quoting: 'As shipped on the paper sheets, the LED dies are too close together to manipulate. There is a mechanical machine ... that spreads the dies out and sticks them to a film of weak adhesive. This film is suspended above the lead frames ... Using a microscope, the worker manually aligns the die, and, with a pair of tweezers, pokes the die down into the lead frame. The adhesive in the lead frame wins (is more sticky), and the worker quickly moves to the next die. We were told they can align over 80 per minute or about 40,000 per day.'"

Comments

Cheap labor versus automation

[Ichijo]

There's a surprising amount of manual labor involved with making LEDs.

You have to expect that in a country where manual labor is cheap. In other countries, it makes more economic sense to automate or otherwise fix inefficiencies in the manufacturing process.

Re:Cheap labor versus automation

[DocSavage64109]

I'm suddenly thinking that my job isn't so bad... I would go insane after the first three minutes.

Re:Cheap labor versus automation

[TubeSteak]

--American unions are changing their priorities. Appliance Park's union was so fractious in the '70s and '80s that the place was known as "Strike City." That same union agreed to a two-tier wage scale in 2005--and today, 70 percent of the jobs there are on the lower tier, which starts at just over $13.50 an hour, almost $8 less than what the starting wage used to be.

--U.S. labor productivity has continued its long march upward, meaning that labor costs have become a smaller and smaller proportion of the total cost of finished goods. You simply can't save much money chasing wages anymore.

Your article glides over this very quickly, but it's worth discussing further.
Management has essentially halted the growth of wages for decades and this has allowed all the productivity gains to accrue to business profits.
The knock on effects have ripped through the economy, from skewed stock valuations to screwed workers' debt loads.

http://tcf.org/assets/images/blog_images/20120814-graph-of-the-day-does-productivity-growth-still-benefit-the-american-worker.png
You can find other graphs that break down the wage growth by percentile (20th, 50th, 95th) and it's pretty much exactly what you'd expect.

closed on saturday

[rogoshen1]

The most striking thing to me about that .. article is that the factory was actually closed on a Saturday.

Re:closed on saturday

[rogoshen1]

the god damn article.

Somewhat off-topic: why not uncut LED panels?

[raymorris]

I suspect this story may draw comments from people who know something about LED manufacturing. If so, I hope someone can answer this question. I noticed that panels of LEDs, such as used for traffic lights or stage lights, are composed of 200 individual LEDs. So the process is:

Cut one LED panel apart, into hundreds of LED cores.
Glue hundreds of leads to the hundreds of fresh cut cores.
Align hundreds of cores into hundreds of little molds.
Inject resin into hundreds of little modes.
Assemble all of the hundreds of resin-covered LEDs back into a panel again.

Why not this?:
Attach ONE set of leads to the LED silicon panel.
Dip the whole dang panel in resin.

Re:Somewhat off-topic: why not uncut LED panels?

[Obfuscant]

I'm guessing for the same reason that the hundred of op amps on a wafer aren't all wired up on the wafer: the manufacturing process isn't perfect and you'd have a lot of nonworking or low output LEDs in the middle of your traffic lights, with modern pick and place machines it is automated to put them back into one device, and if you ever need to fix a failed LED it is cheaper to replace one than a whole wafer of them.

Probably also for heat dissipation.

And that traffic lights are a niche market so you'd be limiting yourself to that market by making LEDs that way. And stage lighting needs at least three colors of LED in one package.

Re:Somewhat off-topic: why not uncut LED panels?

[Anonymous Coward]

Attach ONE set of leads to the LED silicon panel.

LEDs emit light from the diode junction, not the surface. If you wired up the entire wafer, light would just come out only from the edge of the wafer and the rest would be trapped inside and turn to heat.

Re:Somewhat off-topic: why not uncut LED panels?

[Dan East]

All the value and cost of the LEDs is in the silicon dies they get from Taiwan. That's why they cram 4,000 dies onto a silicon sheet the size of a postage stamp. The labor required to manually pull them off is cheaper than the material and manufacturing process of the die.
The surface area of a stoplight led panel would be at least 40 dies worth. That's 160,000 LEDs worth of silicon and production time in the fab plant. Just to get 200 LEDs worth of illumination using your technique. Plus, setting up the fab to produce the silicon would be a large initial investment. There's no way it makes sense to make that manufacturing proprietary to something as specific as the stoplight application. It makes about as much sense as building a plant to make custom nails to build one house.

Re:Somewhat off-topic: why not uncut LED panels?

[b1t r0t]

I work in a part of the commercial lighting industry, and the current technology is indeed surface-mount LEDs, sort of like what you see on strip lighting, only on a metal PCB (for heat dissipation, naturally). Just run everything through a pick-and-place machine and roll it through the reflow oven. The lenses are clear plastic light pipes mounted through holes in a metal fascia. (And I was lucky enough to be able to pick up a bunch of LEDs that fell out of the PnP when we were making some a couple of years ago. Heavy parts do that.)

I doubt they're ever going to move to mounting bare dies (like in seven-segment displays) because they'd have to have their own wire bonder and epoxy, and I'm going to guess that white LEDs need the phosphor applied between mounting the die and molding the lens, so that's one more step. Some customers want Made in USA stuff, and that would be a true pain in the ass to do die-mount in the US vs. a tube of surface-mount LEDs and a normal pick-and-place machine.

I'm pretty sure just from the way they look that traffic lights are using regular 5mm or 8mm LEDs. Even if they used surface-mount LEDs, they'd still need a lens somewhere, and die-mounts would of course be even more trouble. You would have to buy the whole thing pre-made from China with die-mount. The only reason seven-segment (and presumably matrix) LEDs use die mounts is that they can be sold as a single pre-packaged part, and they're too small to even use 3mm molded LEDs.

Re:Somewhat off-topic: why not uncut LED panels?

[imsabbel]

They would be too bright, and too expensive. Those 5mm LEDs (outdated crap) have LED dies of a less than 1/20th of a mm^2 inside.

High-end power LEDs that put out the equivalent of a 100W incan bulb are 2-4mm^2.

A disk the size of a traffic light would be able to draw 10000s of Amps, and be bright enough to illuminate a stadium (if you could drain the waste-heat away). Driving it with lower currents would be a collosal waste of dies.

So the wavers are cut into conveniently sized pices that have reasonable power draw and thermal conductivity and then later are put together according to the requirements.

Re:Only 40,000 a day?

[K. S. Kyosuke]

They are also using some biological machines, apparently.

Re:Only 40,000 a day?

[PPH]

Understandable. They probably don't let their employees get on Sashdot. Or it would be 1200 per day.

Re:great photowork there

[50000BTU_barbecue]

My interpretation is as follows. The equipment shown is stuff that could have existed in the 1960s. In the West, that's pretty much how machines looked like in the 60s. The #1 company that made and still makes these machines is Kulicke & Soffa.

http://www.kns.com/en-us/Pages...

China basically scoured North America for all the old machines they could find. Ribbon machines that make incandescent lightbulbs. Pick and place machines. Board plating shops. Wire and ball bonders.

All this stuff that used to the define the West's technological prowess. K&S is now based in Singapore.

https://www.youtube.com/watch?...

Oh, and if you want to see something fast and automatic, look up chip shooter on youtube...

left out the most important steps

[zmender]

Note: I am speaking as a material engineer who spent about 6 years in R&D for the 65W LED bulbs you can now buy at HomeDepot. The articled failed to mention the most important aspects of the LED manufacturing: wafering and the MOCVD that deposits the light emitting materials (the PN junction) onto the wafer. In short, the steps would include: 1) Crystal growth / wafering / surface prep; (make the wafer) 2) Nitrite epitaxial growth; (grow the light emitting part) 3) Wafer fabrication (cut big wafers down to die-sized chunks) 4) Packaging and testing (encapsulating the die) -- what the article was describing The article only touched upon the 4th step of LED manufacturing, and concidently, the most automated aspect of manufacturing, as well as the part that contains the least amount of patents / trade secrets. The first 3 steps were marginalized as "This is a sheet of LED dies. YunSun buys their dies from a high quality Taiwanese company". To my knowledge, there is no high quality manufacturer in Asia outside of Japan. Samsung makes a great quantity of ok stuff, and China / Taiwan makes a great quantity of shitty stuff that is ruining the entire high profit margin products. Also, all of the major manufacturers of LED dies dare not introduce step (2) and (3) into China / Taiwan due to IP issues. Wafering is important because larger wafer sizes (2in to 4in to 6in) means more dies per area. However, crystal quality becomes harder to control as sizes go up, especially for US-based LED manufacturers that is based on silicon carbide instead of sapphire. The real issue is with the MOCVD, the deposition technique that grows the PN junction which actually emits light. In the world of deposition, MOCVD is archaic voodoo magic and we spent a lot of time praying to deities of deposition that our process would repeat for more than a day. Fab is more systematic than epitaxial growth, and the real science here has to do with light extraction. Again, big money is spent on R&D here, and we dare not bring the manufacturing process to Asia (except for Japan).

Re:left out the most important steps

[zmender]

Messed up my formatting.

Note: I am speaking as a material engineer who spent about 6 years in R&D for the 65W LED bulbs you can now buy at HomeDepot. The articled failed to mention the most important aspects of the LED manufacturing: wafering and the MOCVD that deposits the light emitting materials (the PN junction) onto the wafer. In short, the steps would include:

1) Crystal growth / wafering / surface prep; (make the wafer)
2) Nitrite epitaxial growth; (grow the light emitting part)
3) Wafer fabrication (cut big wafers down to die-sized chunks)
4) Packaging and testing (encapsulating the die) -- what the article was describing

The article only touched upon the 4th step of LED manufacturing, and concidently, the most automated aspect of manufacturing, as well as the part that contains the least amount of patents / trade secrets. The first 3 steps were marginalized as "This is a sheet of LED dies. YunSun buys their dies from a high quality Taiwanese company". To my knowledge, there is no high quality manufacturer in Asia outside of Japan. Samsung makes a great quantity of ok stuff, and China / Taiwan makes a great quantity of shitty stuff that is ruining the entire high profit margin products. Also, all of the major manufacturers of LED dies dare not introduce step (2) and (3) into China / Taiwan due to IP issues.

Wafering is important because larger wafer sizes (2in to 4in to 6in) means more dies per area. However, crystal quality becomes harder to control as sizes go up, especially for US-based LED manufacturers that is based on silicon carbide instead of sapphire. The real issue is with the MOCVD, the deposition technique that grows the PN junction which actually emits light. In the world of deposition, MOCVD is archaic voodoo magic and we spent a lot of time praying to deities of deposition that our process would repeat for more than a day. Fab is more systematic than epitaxial growth, and the real science here has to do with light extraction. Again, big money is spent on R&D here, and we dare not bring the manufacturing process to Asia (except for Japan).

Re:left out the most important steps

[b1t r0t]

The articled failed to mention the most important aspects of the LED manufacturing: wafering and the MOCVD that deposits the light emitting materials

It didn't "fail to mention" that, it started with the pre-cut dies. This place turns cut dies into finished LEDs, and they documented what they saw. This was in China, and the dies are apparently made in Taiwan (so maybe they are crap), and they weren't anywhere near there.

And yes, fabs are serious alchemy. Like the time when Fairchild was a new company, and they found out that their transistor yields were being affected by workers not washing their hands after taking a leak. The chemicals in urine were fucking up the process.

liberal propganda

[s122604]

More liberal propaganda to justify our loss of lightbulb freedom

Re:Lead is mentioned some 16 times

[MattskEE]

The lead frames are not typically made with any lead, it is just a case of easily confused homonyms.

In "lead frames", "lead" refers to the metal pins coming out of the packages, which are connected to the LED die. It not typically made with any lead content (Pb, element 82) due to RoHS restrictions. It could be made of tin-plated copper, or various alloys of tin, copper, and silver. Older ones would likely have been Pb-plated copper.

Re:Only 40,000 a day?

[50000BTU_barbecue]

State.