Lately, there's been some interest in sparse Merkle trees (SMTs). I know Ethereum uses Patricia tries (PTs), but this thread seems to indicate that working with SMTs would yield similar performance, even better if constructed correctly.
I understand how generally SMTs work (one generates the hash of all possible inputs, and the tree is sparse because you don't actually store all empty keys and slots), but how is it going to work within the context of Ethereum and sharding?