... on how much (time/money/energy) you wish to invest in the host application software (and perhaps driver). If you have an application that already talks to a COM: port, then a reasonable (but possibly not a slam dunk!) choice would be to use an FTDI part or that "USB to UART" chip from SiLabs (yes, the one who's advert is blinking to the left of the box in which I type).

If you don't already have an application and are writing one from scratch, you should be able to use one of the SiLabs parts ('320 and '321, I think). Use the HID class (your device will use "vendor defined" usages and pages (read the USB spec to understand what I mean by all that)) and you won't have to write a driver. Instead, your host app uses userland calls to the HID driver.

I have an HID interface in an audio device that does just this -- the "usage" is vendor-defined and is used to set gain, volume and control some features. (My device has no front-panel controls -- everything's controlled from the computer.) I even implemented a firmware-update feature using HID (the USB DFU spec is rather inscrutable, and no generic DFU class driver exists and I wasn't about to write or buy one) that turns my device into an I2C EEPROM programmer. I've written little control interface applets for Windows and Mac OS X. I'd write one for Linux but I don't have a Linux box (don't need one with the Macs around).

-a

PS: I wish the SiLabs parts had an external memory bus. The couple of ideas I have require an interface between the micro and the FPGA, and I'd prefer that interface be faster than SPI or I2C. An SiLabs part like this could seriously compete with the Cypress EZ-USB series. Plus, I love the JTAG debugger. And I hate Cypress.

PPS: regarding the Maxim DS9490. I won't use a Maxim part in a design. They usually require you to buy a whole reel when you really need only a tray or so. And sometimes they'll introduce a part with a lot of variants, most of which are unavailable until some Big Customer orders enough for them to do a fab run.