AFAIK the paper is referring to thin clients, right? You can already run those with the stock Bitcoin Core client. There's a command line argument that tells

AFAIK the paper is referring to thin clients, right? You can already run those with the stock Bitcoin Core client. There's a command line argument that tells it to prune any blocks older than X. The disk space required by such clients is X times the maximum size of blocks + the size of the UTXO. But you still need full nodes who maintain all historical data. Because, AFAIK, there's no way other way to trustlessly establish a UTXO without rescanning the whole blockchain. Maybe there are ways; I'm not that deep into Bitcoin's internal at the moment.