That's a blast from the past. Yes, Dan Holcomb used the term "FERNS" in his project to harvest true randomness and "ID" fingerprinting bits from uninitialized SRAM at power-up. In retrospect, it might have made more sense to use the term "PUF." Bygones. He's now out at Berkeley. The IEEE Transactions on Computers journal version of the FERNS paper goes into more detail on things like negative bias temperature instability (NBTI). And Dan's latest work on using SRAM for significantly more "ID" fingerprint bits appears at RFIDsec 2012.
The idea of using SRAM from graphics cards would be a nice twist. But SRAM isn't the only building block on the chip (note the reverse poetry)---could be other cleverness too! I look forward to reading about the work.
Oh. And one more thing. Bad Wolf.