Recently I have enjoyed email conversations with Mark Zinky (of Mark Zinky Design in San Francisco, California) after he posted a comment here. We discussed LED lamp design issues, and we revisited some of the problems I had when building the dodecahedron-shaped ‘Bright Light’ lamp. In particular, he’s helped me understand a fun-DUH-mental flaw in the way I was loading the CAT4101 driver. It’s given me an idea for another LED lamp, and it’s slick.

This is a picture of a ZeroTherm BTF95 fanless CPU cooler:

Off the shelf CPU cooler: ~ $50

So what happens if I strap a small PC board loaded with Cree XP-G high-brightness LEDs on the bottom of it?

CPU coolers are specifically made to pull heat away from a computer CPU and dissipate it into the air.  This particular one doesn’t have it’s own fan, but that’s not usually a problem, computers require a certain amount of internal ambient air flow, so there’s a breeze flowing through it when it’s hidden away inside your computer case. It can pull 80 watts of heat away from a LGA 775 (Intel) or an AM2/939 (AMD) CPU through a 40 millimeter square copper block on the bottom:

(Thanks to ZeroTherm for the sniped pics)

16 Cree XP-G LEDs might draw 17 watts, and the XP-G can be overloaded from nominal amperage by quite a bit.  Just on the face of it, this should work great.

I learned a lot from the “Bright Lights” project. The PWM frequency was just a little low for my own tastes, I vapidly overloaded the CAT4101 driver, and I totally nOObed the thermal management. Heat management in electronics is something I learned from a 1971 version of the Radio Shack 100-in-1 electronics experimenter kit when I touched a transistor during the process of burning it out. Using this CPU cooler, and how cool it is, I should be able to make a fairly bright dimmable LED lamp that could be mounted as a sconce. It could be mounted to shine down, or upside down and reflect off the ceiling, or tilted to highlight works of art, or mounted on a free-standing faux-goth candelabra thingy. The power supply is separate and will need mounting elsewhere, but one power supply could power many of these devices in a room if they are all tuned to require the same voltage.

The XP-G’s also come in a variety of white ‘colors’, as the wife doesn’t care much for the cooler white 6000K tones.

Re-purposefully engineering an LED circuit board

I use Ubuntu Linux.  It doesn’t cost anything.  On Ubuntu Linux I use Eagle electronic CAD software.  This board is within the restrictions of the ‘free’ license, and I’m sharing the design I make with it, so I think the back-end licensing is cool.  I tend to draw my own Eagle library components, it makes design a lot easier when trying to match up the math and the physics.

Here are the Gerber files for this project.  These files have had some mistakes corrected from the original design as prototyped, but there weren’t any particularly bad errors.

HouseLEDs-04

The schematic was actually pretty simple, since it was just a few variations from the basic ‘Bright Lights’ circuit. As usual, I’ll make it small so as not to bore my less-technical visitors:

For manufacturing, I’ve sent the design to Silver Circuits at http://www.custompcb.com . Their quoted price is very competitive, and they feature some advance processing that I considered necessary for this project… specifically double (2 ounce) copper, white soldermask paint, and ENIG gold finishing. Here is a generated representation of the expected PC board:

This is what the top should look like *
This is what the back should look like *

* these are not photographs, they are generated images.  See below for pictures of the manufactured boards.

The back of the board has a 48mm soldermask hole where the cooler bolts on, and I’ve used 55mm corner mounting holes to mount this specific cooler to the board. A more generic mounting would use 72mm corners and fit any LGA 775 capable cooler, but I was able to slim the board down a bit and simplify the mounting by making this design specific to the ZeroTherm BTF95 cooler. The copper face for the BTF95 is actually 36mm x 48mm, and can be mounted in any of 4 orientations. I don’t anticipate the CAT4101 drivers to generate very much heat due to the optimizations to the circuit that Mark made me aware of, but there is room for a couple of small chip heat sinks to be thermal taped (with Seksui 5760) onto the back side of the board. My wife’s VGA card recently needed replacing, so I have a few of those left over and ready to go.

The circuit features a potentiometer dimmer control, I’ve used a sealed Cermet ‘rock and roll’ quality Bourne pot that will be mounted directly on the board, although it could be moved off board if I think of a better way to mount it. The power supply will be a CUI Inc V-Infinity 50 watt 24 Volt supply, it can be adjusted +/- 10%, this is critical as the circuit needs about 25.1 volts and can be tuned to reduce the heat and power consumption. I can also use an off-board dimmer like the ‘Bright Lights’ wall-switch dimmer module.

I need a small metal shadow box, like an over-sized Altoids tin, that will fit over the heat-sink copper face and hold a Fresnel lens about 1/2 – 1 inch away from the LEDs (Update: trigonometry says an LED with a 125 degree shine mounted 20mm from the edge of the board will need 13mm of clearance right at the edge). This will also provide a way to strain-relief the power cord and maybe provide a mounting for the dimmer knob, and maybe a real on/off switch. The lamp can be turned off by turning the dimmer all the way down, or cutting the power to the supply. The power supply is ‘green’ (Energy Star compliant), 87% efficient per specs, and uses only 0.5 watts when idle, so I don’t really mind leaving the power on.

Here’s where the parts go:

Here is the parts list:

Digi-Key Part Number Manufacturer Manufacturer Part Number Quantity Description What it is Do I have to have it?
MCP6L94T-E/SLCT-ND
MICROCHIP TECHNOLOGY (VA)
MCP6L94T-E/SL
1 – IC4
IC OPAMP 10MHZ 2.4V 14-SOIC
PWM Generator Quad Op-Amp
Needed for on-board PWM
XPGWHT-L1-0000-00CE7CT-ND
CREE INC (VA)
XPGWHT-L1-0000-00CE7
16 – LED1 – LED16
LED WARM WHITE 3000K SMD
Cree XP-G LEDs (bin Q4)
Yes, can be any colour you like, warm white 3000K suggested
CAT4101TV-T75CT-ND
ON SEMICONDUCTOR (VA)
CAT4101TV-T75
2 – IC2, IC5
IC LED DRVR HP CONST CURR D2PAK
LED drivers
Yes
RGH16P1.2KCT-ND
SUSUMU (VA)
RGH1608-2C-P-122-B
1 – R9
RES 1.2K OHM 1/6W 0.1% 0603 SMD
PWM timing resistor
Needed for on-board PWM, high-temp stability suggested
RGH16P680CT-ND
SUSUMU (VA)
RGH1608-2C-P-681-B
2 – R2, R5
RES 680 OHM 1/6W 0.1% 0603 SMD
Maximum LED amperage set
Yes, high-temp stability required
RGH16P1.5KCT-ND
SUSUMU (VA)
RGH1608-2C-P-152-B
2 – R7, R8
RES 1.5K OHM 1/6W 0.1% 0603 SMD
PWM timing resistor
Needed for on-board PWM, high-temp stability suggested
RGH16P10.0KCT-ND
SUSUMU (VA)
RGH1608-2C-P-103-B
3 – R3, R4, R6
RES 10.0K OHM 1/6W 0.1% 0603 SMD
PWM pull-ups
Needed for on-board PWM, can be any value from 6K to 47K
MF-SM100CT-ND
BOURNS INC (VA)
MF-SM100-2
2 – PTC1, PTC2
FUSE RESETTABLE 1.0A 30V SMD
PTC, over-current over-temp protection
Highly recommended, but can be omitted (shorted)
3M9467-ND
3M
961102-5604-AR
2 – AMP-TEST1, AMP-TEST2
CONN HEADER R/A SGL 2POS GOLD
jumper for amperage testing
Highly recommended, but can be omitted (shorted)
ST32ETB202CT-ND
COPAL ELECTRONICS INC (VA)
ST32ETB202
2 – R1, R11
POT 2.0K OHM 3MM CERM SQ TOP SMD
LED amperage adjustment
Yes, high-temp stability required
ST32ETB501CT-ND
COPAL ELECTRONICS INC (VA)
ST32ETB501
1 – R10
POT 500 OHM 3MM CERM SQ TOP SMD
PWM hysteresis adjustment
Needed for on-board PWM, high-temp stability suggested
51AAD-B24-A15L-ND
BOURNS INC
51AAD-B24-A15L
1 – ‘GAS’
POT 10K OHM CERMET 1W
Brightness control
Needed for on-board PWM
497-1258-1-ND
STMICROELECTRONICS (VA)
74LX1G70STR
1 – IC3
IC BUFFER NON-INVERTING SOT235
logic voltage translator
Needed for on-board PWM
SBR2M30P1DICT-ND
DIODES INC (VA)
SBR2M30P1-7
1 – D1
RECTIFIER SBR 2A 30V POWERDI123
safety barrier power supply diode
Highly recommended, but can be omitted (shorted)
338-1841-1-ND
CORNELL DUBILIER ELECTRONICS (CDE) (VA)
AVEK107M35G24T-F
1 – C2
CAP ALUM 100UF 35V ELECT SMD
LED voltage power conditioning
Yes, LED voltage conditioning
338-1821-1-ND
CORNELL DUBILIER ELECTRONICS (CDE) (VA)
AVRF108M06F24T-F
1 – C4
CAP ALUM 1000UF 6.3V ELECT SMD
+5V power conditioning
Yes, +5V conditioning
641-1088-1-ND
COMCHIP TECHNOLOGY (VA)
CPDU5V0
2 – ESD2, ESD3
TVS ESD BIDIR 5V 0603
5V bi-directional zener diode
Recommended for ESD protection
641-1240-1-ND
COMCHIP TECHNOLOGY (VA)
TV04A240JB-G
1 – ESD1
TVS 400W 24V BIDIRECTIONAL SMA
24V bi-directional zener diode
Recommended for ESD protection
445-2537-1-ND
TDK CORPORATION (VA)
AVR-M1608C080MTAAB
2 – VAR2, VAR3
VARISTOR 8V 30A 650PF 0603 SMD
5V varistor ESD protection
Recommended for ESD protection
445-2549-1-ND
TDK CORPORATION (VA)
AVR-M2012C390KT6AB
1 – VAR1
VARISTOR 39V 100A 430PF 0805 SMD
24V varistor ESD protection
Recommended for ESD protection
568-3310-1-ND
NXP SEMICONDUCTORS (VA)
TDA3664AT/N1,118
1 – IC1
IC VOLT REG 5V 100MA 8-SOIC
+5V regulator
Yes, any physical form
445-2514-1-ND
TDK CORPORATION (VA)
C2012X8R1H104K
5 – C1, C3, C5, C6, C7
CAP CER 0.10UF 50V X8R 10% 0805
power supply hiss filters, PWM
Highly recommended, but some can be omitted (open)

Updates:

7-NOV-2010 The cooler is here, the parts are here, and Silver Circuits seems to have accepted my design, as they have billed me for it, and it is paid. Now just have to wait 2 weeks for the boards to arrive. I have time to recalibrate my toaster reflow oven.

9-NOV-2010 I’m seeing talk on the innerwebs about Cree XP-G bin R5, which is a particularly cool (5700 – 8000K) bluish white color, fine for flashlights.  On an all-copper heat sink of a lamp designed for indoor use, I don’t think it would look as good as warmer tones.  On the other hand, it does put out significantly more usable light, so if I can find 16 of these discrete (not mounted), I might have to try it just to glean on the hype.

15-NOV-2010 All the parts are here for two ZeroTherm BTF95 coppers and a stainless steel model with the Prolimatech Armageddon.  The boards are lovely, exactly what was expected.  There’s more wiggle room than I anticipated, I overshot the mounting screw measurement by a whole millimeter.  The test mounting still looks good, I’ll get pictures up soon.

The PC Boards arrive:

-= MANUFACTURING PHASE =-

It will take a whole day to paint the lead on a board with a toothpick and cook it, and I have a bit of prep work to do before that.  I can also do the physical stuff, easier to cut the sheet metal to size now that I have a template.  The leads on the brightness potentiometer are gonna take some stress if I mount it directly on the board and then screwed into the sheet metal casing, I’m considering mounting it with wires so they can bend a bit.  Prototype design phase complete, now on to manufacturing.

Compare this photograph to the generated image above.  What I saw is what I got!

20-NOV-2010 I cooked up one of the boards today, it took a long time to paint the lead on the pads, but after I dropped all the parts into place, I cooked it in the reflow toaster and it looks really good.  Only one 0805 SMDs tombstoned a little bit, and all of the LEDs look like they did just fine, there’s nice round melted lead in the holes on the back to indicate that it probably bonded the thermal center pad on all of them.

It works! I used the Prolimatech heat grease and mounted the board on a BTF95.  The dimmer isn’t linear, the sawtooth isn’t going rail to rail.  The quad op-amp is a slightly different chip than I used before, it may not be as good, but I suspect the components are just a bit aggressive and I need to slow it down a little with a bigger charge cap or resistors.  The frequency seems really high, like I was expecting about 4400 Hz, but on the scope it’s showing a waveform taking 2.4 divisions at 1uSec/div, which would be 416666 Hz.  What?  It sure isn’t strobing like the half dodecahedron at 330Hz, is it really going that fast?

It can’t be.  I suspect I’ve got some funky button set on the oscilloscope or I’ve stayed up too late and can’t do math right now.  I replaced the PWM charge capacitor from 47nF to 0.1uF and now the dimmer is working just fine.  There is a limit to the base speed of this style of PWM, but it’s working well for me so far.

To adjust: I set the amperage on each of the two arrays, then measure the voltage at the driver LED pins.  Using the greater value of the two voltages, I adjust the power supply voltage till the measurement is just over 0.5 volts, then I adjust the other driver’s amperage until the voltage on that LED pin matches.  This should give the most efficient use of the drivers, and they seem to barely generate any of their own heat.

These measurements should probably be repeated when the device comes up to full heat, and adjusted to split the difference.

The ‘max amperage’ resistor on each driver is 680 ohms, this lets me set a maximum LED amperage of around 450mA (or higher as the parts begin to derate with the heat).  I cranked the amperage (set to about 450mA but was over 500mA after it got hot), balanced the voltage drop on the CAT drivers, and the temperature seemed to stabilize a little over 75 degrees C, as measured with a Fluke 62 Mini infrared thermometer pointed at the LEDs in the center of the board.  The room is ~27C.  At 350mA (100% nominal), the center of the circuit board goes to a very reasonable 47-52C (toasty to the touch), and at 400mA, around 57-62C.  I’m sticking with 400mA for now, at that amperage the whole circuit draws 20 watts at the outlet, the heat is manageable, and the light output is 115%.   Cree bin Q4 warm white XP-G’s put out a minimum of 100 luminous flux at 350mA, so all 16 LEDs together at 0.4 amps should be generating about 1840 lumens calculated.

Laying on it’s side, but on my desk, so taking a slight advantage of the copper fin vertical convection, 400mA stabilizes at 57C.  It’s obvious a fan would really help, but this project won’t use any moving parts, we’ll save that for the grow light.

I’ve accidentally glanced at the array a few too many times tonight (16 evenly spaced little blurs on my retina), so I promise pictures tomorrow.  The “Sino Snow Goggles” from American Science and Surplus help quite a bit slightly, and really add to the ‘Mad Scientist’ image.

22-NOV-2010 I worked the sheet metal today.  I do NOT recommend hand working 40-mil hard copper, it was a bit rough but I got away with only a few minor cuts and scrapes.  The final assembly went okay, and it’s quite striking.  It’s somewhat smaller than I originally envisioned, but that really shouldn’t be a surprise for me.  Yes, yes I will take pictures tomorrow.

23-NOV-2010 Pictures!

view of the lamp with the butterfly shape of the ZeroTherm BTF95 from above and behind

POWER SUPPLY will be HIDDEN

inside the wall or mounting.

Here is the prototype lamp with power supply

view from front

detail on the gas lever

gas lever and power cord strain relief

removing the fresnel lens so we can see inside

the internal electronics.  The small yellow dots in the center are the Cree XP-G LEDs.  The grey ooze is thermal grease that has squeezed through the 24-mil thermal vias.  This is okay, it’s supposed to do that.

with the ‘gas’ lever all the way open, the lamp uses 20 watts of AC power as measured at the wall (The Kill-a-watt shows 19 watts in this picture as the lamp isn’t fully up to temperature).

This lamp runs at 24 volts DC (the white cable) with 400mA, 115% of nominal amperage.  The ability to adjust the power supply +/- 10 % is key to optimizing the thermal and electrical economy of the lamp.  The amperage directly affects the brightness, and is only limited by the efficiency of the heat sink, but lower temperatures and amperage allow long LED life.  I’m setting these lamps to run at 100% PWM continuously for years.  I’m thinking that the big capacitors will burn out before the LEDs, but I also used long life caps.  I could run this lamp at nearly twice this amperage in this circuit, but I have no desire to sacrifice the life of the components to increase the brightness.  If I really needed that much more light, I would have used a different design (see the Grow Light).

Here’s one of the improvements over the half dodecahedron model…. In each of the two pictures, some of the MX-6 LEDs are ‘out’ on the dodecahedron because the camera frame speed was competitive with the base dimmer frequency (~330 Hz).  On the new lamp, the base frequency is from four to seven times faster (1200~2200 Hz), so even a shutter speed up towards 1/1000th second should work just fine (note: I *will* be fixing this on the old model sometime, just need to replace one of the resistors).  Note also the color differences between the cool white 6500K MX-6’s and the warm white 3000K XP-G’s.

I’ve got the parts for two more of these, so I’ll make the stainless steel and aluminium one next.

24-NOV-2010 I dislike working with stainless steel sheet.  I botched the sheet metal pretty badly, but my manual pick and place was *perfect* for the cool white lamp.  The Prolimatech Armageddon mount was a little weird, but one quick trip to my favorite hardware store procured adequate bolt and spring mounting parts, the mount went fine, but the overall look to the shadow box is not good.  It will WORK, but I wouldn’t call it perfect.  My hands are a bit cut up, but I built the second one in a day from parts, and it works.  I’ve got it adjusted to nominal 350mA and it’s heating up nicely.  I’m pretty sure I’ll have plenty of room to overload this one as well, but I may just leave it as it is.

25-NOV-2010 I can really turn up the amperage.  It gets hot, but not unreasonable.  If I run it around 770mA, it’s supposed to put out 200% nominal light, but the board gets up to 72 C and the heat sink around 57 C.  It seems the outside of the sink about halfway up is usually 10 – 15 degrees cooler than the board (the board hot spot is right in the center and half an inch upwards, whichever way upwards happens to be).  If I run it at nominal 350mA, it never gets above 50 C.  This heat sink is doing very well!  I’ll run it at 550mA, which should give me 150% light output from nominal and run at a very reasonable temp.  If I let it run a while, the board hot spot goes to 58C (ambient is 27C), and the heat sink is 47 C.  All of this assumes my IR temp gun works well with the reflectivity of the surfaces, I really should work out a thermocouple to measure this stuff.

The hottest recorded temperature on earth is just under 55 C.  A 55 C ambient temperature might cause the board hot spot to go to 90 C, and I’m pretty sure it will either take that or blow off the self-resetting PTC fuses.

Here’s a pic in the middle of me painting the LED pads with lead solder paste.  Note that I haven’t painted the CAT4101 drivers yet, you can still see the gold plating.

Here I’ve placed all the parts.  This is ready for the reflow toaster oven.

Engineering for 16 CREE XP-G bin R4 LEDs…

Out of the oven and voltage applied.  It works!  This is tuned to minimal amperage at lowest dimming, so it won’t generate much heat just sitting on my desk.

…strapped to a Prolimatech Armageddon CPU cooler

assembled, the Prolimatech Armageddon LED sconce, from the back.  Here is the front:

and a close up on the back:

OOH, Shiny! Close up on the extremely shoddy shadow box, inventive spring mounting, power cable strain relief, and the gas knob.

27-DEC-2010

I’m 3 for 3 on melting the boards, I made a follower lamp for another BTF95 copper heat sink, it’s missing some of the PWM circuit parts as it takes the PWM line from the other lamp.  I send this back through the white cable on the white lead, black is hot, shield is negative and grounded at the power supply.

An important step on this lamp was that I used Kester NXG1 no-lead solder paste.  It definitely takes a hotter temperature to melt it, and I no longer fear that unknown.  I used regular lead solder to attach the wires, I think I might have some very thick gauge no-lead wire solder, I’ll have to get some thin stuff.

I’ve had the Armageddon on my desk and been using it for a few weeks now.  It gets hot, but it continued to work very well, and I’m very pleased and confident in this basic design.  I think the thermal compound (Artic Silver 5) cured, it seemed like the sink was drawing the heat away better after a while.  I’ve now switched it out (of my 50 watt test power supply) for these two others, I still have to mount the follower board on a heat sink.  I’ll cut the copper shadow box when I can, but I can still run it with low PWM and with the amperage rheostats turned minimum just to light it up on my desk.


Creative Commons License

Copyright (c) 2010 MDVE.NET. Some rights reserved.
This design is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License .

One Response to “LED Sconce”
  1. This is going to look cool!

Leave a Reply

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>

Copyright © 2010, Multi-Dimensional Visual Echo. All rights reserved.
concern-kittycornered