System-on-chip cpu's are coming like a train. The first one I spent time with with Motorola's 8xx line, the 860, 850 and 823. These chips had all the peripherals on the chip on one big melange. Static memory interface for flash, SDRAM interface for memory, an MMU, serial ports and built in ethernet.
Needless to say it ran linux well, thanks to people like href="http://www.embeddededge.com">Dan Malek. About the same time two chip SOC sets like the Intel SA1110/SA1111 came about and helped in the low-power area. More about ARM and low-power later. Why can't MOT ever make a low-power cpu?
The 850/823's are great but I wish MOT would allow people to write microcode for the serial engine inside it. It's certainly running some sort of microcode and it would be nice to be able to fix bugs and do custom work with it.
I can't much too many really flattering things about Intel and it'a ARM work except, "how many times can you EOL the SA1110 before no one will believe you?". Sure seems like the marketing people keep trying to kill it when at the last second the sales people say "but hey, we're selling a *lot* of these". Maybe it's just me :-)
I next ran into the Alchemy (now AMD) Au1100 line. Nice SOC. Everything but the kitchen sink and all the ants (bugs) too. Still, they seem to have a nice road map and do seem to be fixing some of the bugs. Seems that all too often large companies by IP and slap it into their chips without bothering to test it in a rational way. Sure it passes someones vectors or testbench but the tests don't reflect what normal mortals do... I keep finding the same USB host and function bugs. Some day I'll meet the guy who wrote that verilog :-)
The AU1100 line is very fast (400Mhz) but also rather power hungry. And not very incline to sleep. On the other hand the ARM SOC's can't get past 200Mhz but have many low power features.
I'm looking forward to next year, with even more SOC's and more power conscious SOC's.