OPC
Optical Proximity Correction (OPC) is the computational technique that deliberately distorts a photomask's shapes so that, after diffraction blurs them on the way to the wafer, the printed pattern comes out looking like the designer's original intent — not like the mask. Below roughly 130nm, printing a feature as a simple faithful copy of its intended shape stops working at all; OPC is what makes sub-wavelength lithography possible in the first place.
The problem OPC solves is pure physics, not a manufacturing defect. Light diffracts around any aperture edge, and once a feature's size approaches the exposure wavelength, that diffraction blur is no longer a minor rounding error — it measurably eats into corners, shortens line ends, and narrows spaces compared to the mask's drawn geometry. A mask that's a perfect, faithful copy of the intended layout will reliably print an imperfect, systematically distorted result.
Rule-based OPC is the fast, simple first line of defense. It applies pre-computed geometric corrections — adding a small serif at a corner, extending a line end, biasing an edge by a fixed amount — pulled from a lookup table built from known diffraction behavior. It's quick to run, but it breaks down wherever two-dimensional pattern interactions get complex, because a lookup table can't anticipate every possible neighboring-shape combination.
Model-based OPC is what actually carries advanced nodes. Rather than looking up a canned correction, it simulates the optical and resist response of a candidate mask shape, compares the simulated print to the design target, adjusts the mask edges to close the gap, and repeats — an iterative loop converging toward a mask that, once run through the real physics of exposure, reproduces the design far more faithfully than any rule table could.
Inverse Lithography Technology (ILT) removes the constraint that the mask has to look like the design at all. Instead of correcting a rectilinear mask, ILT treats the entire mask shape as a mathematical optimization target: given the desired wafer print, what *any* mask shape — including smooth, curved, non-rectilinear geometry — would reproduce it most accurately. The resulting curvilinear masks achieve the best fidelity available, at the cost of needing multi-beam mask writers capable of actually manufacturing those curved shapes.
A few terms show up constantly once you're working with OPC in practice: Edge Placement Error (EPE) is simply how far the simulated printed edge lands from the design target — the number OPC is directly trying to minimize. The Mask Error Enhancement Factor (MEEF) describes how a small error on the mask gets amplified (or shrunk) on the wafer — a MEEF above 1 means mask imperfections matter more, not less, as nodes shrink. And fragmentation — dividing each mask edge into small movable segments — is the mechanical degree of freedom model-based OPC actually adjusts, segment by segment, during that simulate-compare-adjust loop. Industry-standard tools for this work include Synopsys Proteus, Siemens EDA Calibre, and ASML Tachyon.