Edge placement error is the deviation between where a feature edge is intended to land and where it actually prints after exposure, etch and any prior layer overlay error compound. It matters most at advanced nodes because multi-patterning and tight pitch designs leave almost no margin for the combined error budget across overlay, CD variation and etch bias. Patent activity in this dataset concentrates on correcting EPE computationally at the mask or exposure stage (G03F, 60.2% of the 83 records) and at the device layer (H01L, 48.2%), rather than through a single dominant fix.

The ranking covers 15 companies in total, with the leader holding 31 of the 83 records in scope and the top 5 combined accounting for 80.7% of all records. That concentration means a small group of equipment, metrology and design-software firms hold most of the documented claim space, while the tail of the ranking — down to a single record at tenth place — is thin. Anyone evaluating freedom to operate should focus clearance searches on the top 5 first, since they cover the large majority of filed art.

Filings peaked in 2020 at 13 records and the field has trended flat to declining since, with the 2022 midpoint at only 2 filings. Publication lag of roughly 18 months means the very latest years will always look artificially low, so this is not proof the field has stopped moving. But the multi-year decline predating that lag window suggests the foundational optical and computational correction methods are largely staked out, and new filings are more likely to be narrow refinements than fresh platform claims.

US9748148B2, assigned to Applied Materials and dated 2017-08-29, covers correcting overlay error or substrate distortion by measuring a distortion map and then applying a doping recipe through electron or ion implantation to physically relieve the underlying stress. It is specific to implantation as the correction actuator, not to overlay correction generally, so approaches that correct through exposure dose, mask bias or etch adjustment sit outside its likely scope. Anyone building a stress-modulation correction loop using doping or implantation should review this family closely before finalizing a design.

The thinnest classes in this dataset are G06V (image and video recognition) and H10P, each at 4.8% of the 83 records, well below the 60.2% held by the core photolithography class. That points to machine-vision-driven defect classification feeding directly into a closed-loop correction step as an under-claimed combination, rather than a fully open field. A first claim there would need to tie a trained recognition model directly to an automated correction action, not just describe the two functions separately, to clear the dense prior art sitting in the optical-correction core.