Hi everyone,


We've recently seen a surge of hashrate that is speculated to have come from FPGAs: Recent services have come up such as those offered by OVH (https://labs.ovh.com/fpga-accelerators-on-public-cloud
 ), Intel support guide (https://www.intel.com/content/dam/www/programmable/us/en/pdfs/literature/ug/ug-qs-ias-opencl-a10-v1-0.pdf), as well as peripheral work on libre software projects for ASICs and FPGAs (https://github.com/altASIC/Open-CryptoNight-ASIC ).

Due to this, We're considering a medium-term solution that will retain decentralization, where we are considering Boolberry's Wild Keccak (WK) PoW (https://boolberry.com/files/Block_Chain_Based_Proof_of_Work.pdf
 ). This design we have found is similar in many ways to ideas in our prior PoW design discussions, and is potentially something we can expand development on if integrated. Currently, only Boolberry and Purk (a fork of BBR) are under this PoW which gives it higher financial risk that permits us more time to research and develop a more permanent PoW.

There is a reasonable ecosystem already present that could aid in quicker deployment of such a change to address PoW centralization concerns (as opposed to purely from scratch). We have found existing CPU and OpenCL miners that we can derive from (https://github.com/mbkuperman/cpuminer-multi-opencl, https://github.com/purkproject/cpuminer-multi, https://github.com/mbkuperman/boolberry-opencl )

Some of the discussion and criticism found on WK includes shared cryptographic concerns (https://www.docdroid.net/GYEk2YB/eval-mtp-argon2d-final.pdf#page=21 ) with potential optimization exploits identified by @gnock about some of the operations in the PoW, comparisons between CryptoNote (CN) and WK (https://bitcointalk.org/index.php?topic=588421.0 ), and the lack of peer review much like CN (https://github.com/turtlecoin/meta/issues/74#issuecomment-373961063 ).

Discussions on alternative medium-term PoW solutions included ones with highly available ASICs (i.e. SHA, Scrypt), PoWs that don't seem much affected by ASICs (Ethash), consideration of all CryptoNote variants as non-viable short-term solutions due to FPGA programmability (we would end up back to square 1), as well as recent relevant research that would be of value (https://github.com/tromp/cuckoo/blob/master/doc/cuckoo.pdf, https://www.docdroid.net/GYEk2YB/eval-mtp-argon2d-final.pdf)

If we were to go through with this, we would likely modify some components of WK, including but not limited to changes in PoW operations, an increase of the iteration count to increase difficulty, and a starting chain scratchpad size that is higher than known FPGAs memory limits: we would call this Fat Keccak, or FatKek for short (https://i.redd.it/0mk07wzuahny.jpg ).

This is a summary initial report that was collaborated by members of the Core team as well as other community members such as @josef, @kacamak, and @dubz.


Further feedback is much appreciated and encouraged!


Best regards,
Thaer