Looks like academic institutional repositories and figshare are doing the heavy lifting here. It's still neat to see Google aggregate everything, but it's not that different from what they do with other services relying on these sources already, and is largely dependent upon how rich these upstream sources are in the first place.