Vitalik Buterin has proposed chain cross-linking to assist in the implementation of his minimal sharding specification published earlier this month.
That spec is described by Buterin as being very simple, stating: “for every 5-block main chain period, the start of that period is used as a source of randomness to select proposers and notary committees, and the committees need to publish their signatures into the main chain by the end of the period.”
In simple terms, this is trying to link the main-chain to the shard-chains, with shards being previously described as their own “universes” which process, say, x-percentage of transactions, then are all linked together.
In the basic shard specification, Buterin proposes that linkage can be done in a simple straight line dependency, but that would be too slow and there might be a gap in time periods. He thus proposes chain cross-links.

“The idea is that there are two types of cross-links, one going from the main chain to shards, and the other going from shards to the main chain,” Buterin says before adding:
“A shard-to-main-chain link must be signed off on by a committee, and the committee’s responsibility is to attest to the availability of all shard blocks since the last cross-link that was made for that shard (alternatively, in the meta-committee approach, a single meta-committee attests to the fact that for every shard, a committee has attested to the validity of some particular hash).
Once an S2MC link is made, the validity of the main chain from that point depends on the validity of that shard chain; if a main chain contains a link to an invalid shard chain block, then that entire main chain past that point is to be considered invalid.”
This is clearly work in progress and it was proposed on the research site of ethereum, but it is somewhat interesting because it seems to indicate that simple specification could in fact be somewhat holistic.
Like Casper’s initial proposal was then simplified into a transitory Hybrid Casper, sharding too might start with a very simple parallelization, to then gradually be developed until Visa levels scaling.
It is also interesting because it shows it is quite unlikely we are to expect any sharding this year, but a simple implementation for next year might not be far fetched.