Comment by userbinator
6 years ago
From the paper:
Our novel operations function at the circuit level by forcing the commodity DRAM chip to open multiple rows simultaneously by activating them in rapid succession.
That reminds me of this:
https://news.ycombinator.com/item?id=11845770 , but the critical part is this:
In short, one memory cell gets refreshed with the bit value of a different memory cell.
What I find more amazing is that this effect has probably been noticed for as long as DRAM existed and treated as a bug (like above), and it took several decades for someone to think "that could be useful!"
I seem to recall an unusual proof of work scheme (cuckoo iirc) which depended on some similar exploit of off-label DRAM timing behavior, but just as a means of assuring a standard measuring unit rather than to perform compute ops.
As the author of the Cuckoo Cycle PoW https://github.com/tromp/cuckoo I don't recall any such dependence. I don't see how the operations described in the paper would benefit a Cuckoo Cycle solver, which in essense looks like https://github.com/tromp/cuckoo/blob/master/doc/simplesolve