So, I've been alternating between searching digikey and staring at datasheets for a week and think it's possible to DIY a driver with similar features as a buckpuck for about $8/unit. ($5.87 for components to be exact, plus a buck or two for odds and ends, solder, wire, etc.) Unless someone tells me I'm crazy, I think this will be the approach I take for the nano I'm building. Perhaps it's not "worth it" on a nano since I'll only need 4 or 5 of the units, but I'm trying to stay focused on using this nano to pilot LED so I can use it on a very large tank at some point down the road. If you're using 20 - 25 buckpucks, saving $10 - $15 each would be significant.
The design is based on the ON Semiconductors NCP3065:
http://www.onsemi.com/PowerSolutions/product.do?id=NCP3063
I'm thinking I'll build a boost driver, since in theory it'll be a little simpler and a little more efficient. The tradeoff seems to be higher ripple on the output, but I'll be underdriving the LEDs compared to most people, so a little variance hopefully won't hurt.
Two 3-up Endor Stars (using the Rebel LED) per unit means 6 LEDs at about 20v when driven around 400mA. In a boost configuration running on an 18v power supply, OnSemi's design tools are predicting like 93% efficiency, and only like 18 mW of loss in the chip! The most efficient buck designs I can come up with are in the 80's efficiency, and closer to 1w lost in the chip. I'm guessing the buckpuck products use a similar chip to the NCP3065 and probably have, at best, ~85% efficiency - assuming you are driving them to within inches of their lives. According to the calculations on OnSemi's tools, boost drivers are most efficient with really low duty cycles (which happens when the power supply voltage is only a few volts under the LED voltage drop), whereas drivers are most efficient with really high duty cycles (which happens when the power supply voltage is only a few volts under the LED voltage drop). This explains the better absolute efficiency of the boost driver, since the switch in the chip is off most of the time, right?
I suppose this is all theoretical to an extent and I will never see a big drop in my electrical bill by saving 3 - 5% efficiency per driver, but the geek in me is excited by having come up with a way to improve things.
I realize that this post probably has little or no bearing on anyone else's projects, since I seem to be one of the only people interested in DIY'ing a driver, but I'm throwing it out there anyways. Some of you can probably tell, but I'm making this up as I go along, so please comment or correct if any of you with more EE background have thoughts about this.
The NCP3065 chip has an input for digital or analog PWM dimming, but I'm thinking I'll build without that feature, then add it later if I want. The way I am mounting the LEDs above this nano will give me a ton of flexibility in terms of placement and height, so hopefully that'll give me freedom if I want to adjust things. Plus, if I really discover that I want one string to be less bright, I can just swap the current sense resistor on the driver, or add PWM dimming to that one driver.
