This is not a vintage teardown at all. This is a very modern LED module that just needs to be mounted and directly connected to electrical mains lead and you have a powerful floodlight. They are extraordinarily inexpensive, I bought mine on eBay for just $1.65 each (Free shipping). Less than the price of a cup of coffee.
I was inspired to get these from a bigclivedotcom youtube teardown where he has reverse engineered them. I wanted to get one to make a cheap workshop inspection lamp, and got a couple of spares to take a closer look at what a modern LED driver chip looks like.
The die are all covered in a soft silicone with the LED’s covered with a yellow phosphor containing material, all of which I laboriously scraped off. This reveals an array of 148 LEDs (74 pairs of LED) connected in series to the output of the bridge rectifier (Just above the neutral terminal) with the other end connected to 5 BP5132H driver chips (That are all in parallel each one providing 10W).
I verified that my circuit schematic is the same as what BigClive derived with one difference in that my device is using 148 LEDs in 74 pairs, presumably my board is using 148 x 0.33W LEDs (vs. 74 x 0.66W LEDs).
Getting the chips off the board is normally pretty straight forward. I just blast the board with a hot air gun until the solder joints melt and the IC’s can be lifted out. In this case the IC’s are solder mounted to the Aluminum substrate which is a heatsink. Getting them off the board took a lot of energy, and if you look closely you can see that for 4 /5 drivers the ic’s actually delaminated from the leadframe/heatspreader rather than the board. Fortunately I was still able to depot them successfully.
The BP5132H is an LED driver from a Chinese company Bright Power Semiconductor, the datasheet can be found online here. Rather oddly the datasheet that is freely downloadable on the internet from their corporate website has a big BPS Confidential watermark?!
I don’t think it works like that, if you’re going to publish a document on the internet then you really cannot mark it up as confidential!
The device is a pair of precision constant current regulators where the current output is set by the resistor on the current sense pin (In our case 6.8Ω). I was a little surprised when after depot I found two identical tiny die (BP5131HAA) – so the two current regulators are totally independent die. The die are very small – just 0.92mm x 0.52mm if you put that into one of the several die per wafer calculators online I get over 46000 die are on an 8″ wafer (Which is very probably the wafer size that is used to make these.)
What you can see is over half the chip is a large HV (500V) MOS transistor that is providing the precision current from Drain to Source (CurrentSense) by the control circuitry between Ground and CS pads. There appears to be only two metal layers of large geometry (Over 1μm) that is pretty densely packed and covers most of the transistors and other devices in the control block.
If we look closely at the HV transistor, you can see a 7μm wide polysilicon gate that is nearly complete buried under the source contact metal. In places there are notches where a metal ring makes contact to the polysilicon layer. I believe this is done to provide a low resistance path a long the polysilicon gate (To shunt the polysilicon resistance) to ensure that the voltage drop along the gate length is negligible. The device is asymmetrical with the gate-to-drain spacing of ~7μm and the gate-to-source spacing of ~30μm which needs to be so large to support the 500V breakdown that the transistor is specified for.
Turning to the LEDs. This is what the LED’s look like after I had scraped away the yellow phosphor containing material. They are soldered to the board with a band across the middle separating the anode and cathode regions.
I tried unsuccessfully to remove some LEDs by heating the substrate on a hotplate until the solder mounts. The tiny LED’s seem to disappear with the remains of the phosphor material? I resorted to brute force and soaked a board in hot sulphuric acid and managed to successfully depot a few LEDs.
This is what the LEDs look like after the acid depot (I believe their was a thin metal layer on the surface that has been etched away during the depot.) The die are tiny just 0.2mm x 0.5mm. They are made on Sapphire substrates depositing very thin (MQW multiple quantum well) layers of AlGaN/GaN (AluminumGallium Nitride/Gallium Nitride) using a rather marvelous process called MOCVD (Metal Organic Chemical Vapor Deposition) where some nasty precursor gases are passed over the heated sapphire substrates in a ultra-high vacuum chamber.
You can see the two anode connections at the top and four cathode connections at the bottom. I am not able to figure out exactly how the rod shapes connect to the anode cathode layers. However we can measure that the strip where light emits below the lower anode annulus and the upper cathode annuli is just 150μm, and the rod in the centre is where I believe the light emits is just 15μm wide so the light emitting area is just 2250μm2. These are 0.33W LEDs so I calculate they are running at ~105mA which makes the current density 47μA/μm2 (Or 47 x 107A/m2)! I am aware that LEDs (And solid state lasers) run at very high current density but it is still quite staggering how high it is when you do the calculation.
The price of these things at $1.65 seems unbelievable given they contain 148 LEDs 5 driver ics, 6 resistors, a capacitor and a bridge rectifier on a patterned aluminum substrate. I thought I would try with some WAGs (Wild Arse Guesses) to break down the costs to see if it is, or how it is, possible.
LEDs: These are made on Sapphire substrates, from what I can read most Chinese LED production is on 4″ substrates as this is still lower cost than 6″ substrates. I was not able to find any references to processed LED wafer costs but did find a link indicating that a 4″ Sapphire substrate was $20. The MOCVD is an expensive process with expensive pre-cursors and long process times, in addition metal contact layers need deposition and patterning. I am going to take a WAG at the wafer cost is ~$200. Using the die per wafer calculator a 4″ wafer will hold ~43000 die, so lets use 40000 and we get a die cost of $0.005 (0.5 cents)
We need 148 so LED cost = $0.74
BP5132H: I believe processed cost for a simple process 8″ wafer will be depending on foundry and volume somewhere between $500-$1000 so lets use $750. As I noted earlier there are 46000 die per wafer (Incidentally wafers are processed in lots of 24 so each wafer fab lot is ~1.1 million die!). Lets assume a 95% yield so each die cost is $0.017 (1.7 cents) there are two die per part so $0.034. I am not sure how much an ESOP8 package costs but will take a stab and say in volume it will be about the same, so lets say the packaged chip cost is $0.07
We need 5 of these so BP3152H cost = $0.35
Looking up the the passive costs on Digikey and it seems in very high volume they can be purchased for ~$0.002 (0.2 cents). There are 7 passives.
Passive device cost = $0.014
Bridge Recitifier, again looking at Digikey in very high volume I can find a 400V rectifier for $0.0754
Substrate: Really don’t know what price a 9cm x 4cm x 1mm thick polished Aluminum substrate with overlay, patterned and plated copper trace and top overlay, cut and drilled would cost. I am going to put a finger in the air and pick $0.15
Assembly pick and place and testing – Again don’t have a good guess but pick and place 161 components, solder reflow and deposit silicone and phosphor, then test – WAG of $0.15
So adding them up I get a grand cost of $1.47. So I estimate it as less than $1.65 but it only gives $0.18 or ~11% as a gross profit margin which seems way too low to me (But at this price the profit margin cannot be very much.)