Got a couple of Gadgets in the queue that involve switch-mode voltage regulators great and small.
Small, as in nominally 5V in, 1.2 to nominally-5V out, and maybe 150mA maximum, and it's gotta be cheap. I find a variety of chips that'll do this nicely. I should be able to get a few output-voltage steps by the simple expedient of switching feedback resistors in and out, and a few steps is really all I need.
Great, as in basically the guts of a bench power supply, only with a moderate-voltage DC input instead of running directly off the mains. Yeah, some dude in China is selling modular regulators that more or less do this (and apparently do it well), but my Gadget has a different feature set. This calls for a switch-mode controller using external power FETs, and if I could control the setpoints and maybe even read back some measurements, perhaps via I2C, that'd be great. Oh, hello! Turns out Microchip has a line of regulator chips with built-in microcontrollers and some very interesting capabilities. Install suitable firmware, and it's all ready to control and monitor via I2C. Win! And suitable power FETs are getting better every year; there are many to choose from. This leaves magnetics and capacitors to be sorted out (those depend on switching frequency, which depends on the controller chip's gate drive capability and the selected FET's gate capacitance, so there's a chain of design decisions).
FETs getting better? Yeah, there are some darn cheap ones in flat little packages rated for 60V, with ON resistance in the single-digit milliOhms. The lower the ON resistance, the less heat I need to dispose of.
Additional: There's a promising-looking FET that the controller can certainly drive usefully at 500 kHz, and maybe at 1 MHz. Comes in a 3.1mm square flat package. But... it's not clear whether the number given for RθJC is for junction-to-top-of-package (where a heatsink could be applied) or junction-to-underside-thermal-pad (where heat is transferred to the board). I expect it's the latter, with the former not characterized. Which is a nuisance. Oh, well: part of the point of using a way-overspecified FET is to reduce the power dissipation and thus the need for a heatsink, fan, etc.
More: Looks like the only surface-mount FETs characterized for a heatsink on top are the DirectFET line. Why isn't anyone else putting FETs in flat little packages with a thermal slug on the top?
And, in updating my part database, I notice that rather a lot of jellybean parts for which I'm showing a budgetary price of tuppence should now be guesstimated at ha'penny, or maybe even a farthing each in modest production quantities. And that's not even budget no-name Chinese parts; that's Japanese parts of verifiable provenance. But, for budgetary pricing of prototype runs, I think showing the jellybeans at tuppence per is probably good enough. And updating the "database" (a CSV file) is a pain; I really should be using some combination of a proper RDBMS and templates for standard passives. Maybe augmented by automated part lookups on Digi-Key, Mouser, and such. (Which reminds me: I should take another look at Octopart one of these days.)