I tell you, every time I see those videos of people deftly dropping an SMT chip on glistening solder pads and instantly forming a perfect bond it reminds me of tasks like juggling or making a pizza crust by spinning and tossing it in the air. Someone who's good at it can make it look so easy, but when you try it yourself... balls flying everywhere, pizza dough on the ceiling, dogs and cats living together, mass hysteria, etc, etc. One of these days I suppose I need to man up and try it.

As to the practicalities of the project itself, if the goal were limited to plain old 5mb/sec "slow SCSI" it's probably *just possible* that you might be able to pull off a converter entirely in software using, say, one of the faster AVR or PIC variants. They can bitbang "pretty fast"... and, look, in fact, someone has actually implemented a SCSI device using them. It's a fairly high parts-count device because he's using discrete 74LS-series parts to buffer the SCSI connector, but other than that it looks pretty simple. It does split out the functions between *two* CPUs, basically one handling the SCSI port and the other acting as "the drive"... it might be worth looking at the source code, as I wouldn't rule out *possibly* being able to compress it down to one if you were using media that natively is "semi-intelligent" and supports LBA, like an SD card.

(Edit: The designer of the RAM disk I linked to has already expanded it into a device which takes a PC card. So basically the work is already done. Converting from PC card to IDE is trivial.)