How does this high power LED driver work? How much heat will it generate?

This is an interesting looking chip. I would assume it works like a linear regulator, simply turning excess voltage into heat, except the diagram seems to indicate the only thing between the input and the output is a transistor, which presumably is toggled at some frequency to limit the current, as one would do in a buck regulator, except... there's no inductors on the output. There is however a thermal pad on the bottom which indicates it may get hot.

I was looking at this for driving something like a 3W RGB led. 1W per channel, around 350mA each.

So how would I go about calculating how much power this is going to dissipate? Excess voltage * current? Something else? What about the voltage drops they mention, how should I account for those in my calculations?

It clearly is a constant current driver, little more and nothing less. (I think "TL" stands for Texas Linear).

So it will dissipate according to the current you set and the voltage across it.

This is a nasty circuit, if you want to run high current LEDs from a high voltage. It will dissipate as much heat as a resistor sized to give the same current and voltage-drop, but managing the heat from this device is much harder.

So to calculate: multiple the current by the voltage drop to get the dissipated power. E.g. if you drive the blue LED segment at 350mA from a 12V supply, and the LED drops 3.2V at 350mA, then the power Pd = 0.35 x (12 - 3.2) = 3.08W.

Consider a buck-converter solution instead, which drops the voltage across an inductor without dissipating the sort of heat a linear solution does. For example, this part http://uk.farnell.com/diodes-inc/ap8801m8g-13/ic-led-driver-buck-0-5a-8msop/dp/1843875 is as simple to add as your TI part, allows PWM brightness modulation and just requires the addition of a suitable inductor. There are many more similar parts out there, try the parametric search offered by Farnell/Newark - I use it as a design tool daily in my work.

Bill:
I'm aware of the issues with using a linear regulator; that's why I asked what it was, it wasn't clear from the datasheet.

I'm also aware of buck converters. There's one on Sparkfun that uses a similar design to the one you suggest, with one chip and one inductor per LED, but they're $20 each and I was wondering if there was a way to make them more cheaply.

By the way, in my search, I came across this chip:

That chip can only supply 150mA per channel, but you only need the one chip to control three LEDs. You also need only a single inductor. It does seem to require four microcontroller pins though, unless you can drive a differential input with one pin tied to ground? I'm not sure.

At $0.25 per inductor and $1 per chip for those chips you suggested that's $3.75 for parts, plus the cost of placing three times as many components. With the 5970, the cost of the chip is $3.23, and the inductor for it is $0.20, so your total BOM is a little over $3.43. Not a huge difference, but remember, you're placing 3x fewer components and assembly costs go up with increased component count. Also, if you're looking to make your board as small as possible, 3x fewer components will go a long way towards achieving that, and as the 5970 operates at 1.5mhz the inductor it needs is much smaller.

The only downsides are that differential interface and the lower current handling capacity. If someone knows if you can drive differential pins like that with a single IO pin somehow that would be nice to know.

With the differential signal the TLC5970 is apparently designed for long strings or on long cables. It's also an HTSSOP with a thermal pad and would be crazy difficult to solder properly; 16 pins each side only .5mm apart.

You could use a MAX488 or similar to drive the differential inputs. You still need two pins though; one for clock and one for data.

Hm, the 488 isn't really an ideal solution at $1.68 a chip in quantity.

Is that really the only solution? A chip like that, or use four pins? That's disappointing.

I did say "or similar" ;). A TI UA9638CP is $.275 at Newark. You're looking for a RS-422 driver for two differential outputs.

You can still use four IO pins instead of the driver but you seemed to be looking to reduce pin count.

By the way, in my search, I came across this chip {TLC5970}

That's a nice chip, new one on me. It seems to be a combination of a buck converter to drop the bulk of the supply voltage at high efficiency and three linear current sinks to fine-tune each RGB channel current. Best of both worlds, plus the dot correction and (greyscale) level features, and expandable along a chain to other devices.

If you're concerned about BOM cost differences in cents, then I guess you're intending to build a lot of these, will be laying out a PCB and maybe reflow soldering screen pasted boards. If that's the case, then I'd have no worries about the HTSSOP package. But I agree with Chagrin,wouldn't want to hand solder one.