THE COURSE / M3 DESIGN TO MASK / M3.5
Masks and OPC
After this lesson you can:
- Explain why a mask's shapes differ from the layout's, and name the corrections OPC adds.
- Say why OPC needs a compute farm, and what inverse lithography changes.
- Describe how a mask is written, inspected, repaired and protected, and what drives the cost of a set.
Loading the lesson…
Sources
- Photronics, Inc., annual report on Form 10-K for the fiscal year ended 31 October 2025, filed with the SEC 17 December 2025 (read 12 November 2026). Photomasks are "high precision photographic quartz or glass plates containing microscopic images of electronic circuits," used "as masters to transfer circuit patterns onto semiconductor wafers"; they "are typically sold in sets comprised of layers, with each layer having a distinct pattern that is etched onto a different photomask." Manufacture "involves the receipt and conversion of circuit design data to manufacturing pattern data"; masks are made "using electron beam or optical (laser-based) lithography systems"; "a lithography system then exposes the circuit pattern onto a photomask blank"; "the exposed areas are developed and etched"; "the photomask is then inspected for defects and conformity to the customer's design data. After the repair of any defects, the photomask is cleaned, any required pellicles (protective translucent membranes) are applied". "The first several layers of photomasks are sometimes required to be delivered to customers within twenty-four hours." Eleven plants: Taiwan 3, China 2, South Korea 1, the United States 3, Europe 2. Glossary: OPC is "a photolithography enhancement technique applied to compensate for the limitations of light to maintain the edge placement integrity of an original design."
- eBeam Initiative, "2019 Mask Makers' Survey" results, collected independently by David Powell, Inc., July 2018 to June 2019 (read 12 November 2026). 11 companies (AMTC, DNP, Intel, Micron, PDMC, Photronics, Samsung, SK Hynix, SMIC, TMC, Toppan) reported 599,536 masks, 2,789 of them EUV; multi-beam masks reported for the first time, 30% written by VSB e-beam. Average VSB write time 8.64 hours (range 2.8 to 16.7), laser 1.49; longest VSB write reported 80 hours. Mask yield 93.7% overall, EUV 74%. "No repair" for 70% of masks; repairs by laser 12%, e-beam 7%, FIB 7%, nanomachining 4%. Weighted average masks per mask set: 16 at 130 nm and above, 38 at 45 to 65 nm, 46 at 22 to 32 nm, 57 at 11 to 16 nm, 76 at 7 to 11 nm (too few answers below 7 nm). Average critical-layer turnaround at 7 to 11 nm 11 days.
- D.-F. S. Hsu and F.-L. Liu (ASML Netherlands B.V.), "Optimization of assist features and source," US Patent 10,310,386 B2, granted 4 June 2019, background (read 12 November 2026). OPC "addresses the fact that the final size and placement of an image of the design layout projected on the substrate will not be identical to, or simply depend only on the size and placement of the design layout on the patterning device." "One of the simplest forms of OPC is selective bias." "A two-dimensional proximity effect is line end shortening. Line ends have a tendency to 'pull back' from their desired end point location as a function of exposure and focus"; "the degree of end shortening of a long line end can be several times larger than the corresponding line narrowing." "Extra structures such as 'hammerheads' or 'serifs' may be added to line ends to effectively anchor them in place"; "inner or outer serifs can be applied to any edge, especially two dimensional edges, to reduce corner rounding." "A regular array of non-resolvable scattering bars is drawn on both sides of an isolated line feature," so that it behaves, in the aerial image, like a line in a dense array, with a process window much closer to the dense pattern's; such sub-resolution assist features (SRAF) are "features outside edges of the main features." Model-based OPC "involves good process models and considerable computational resources," given "the many millions of features typically present in a chip design."
- X. Meng, C. Chen and J. Ni (Tsinghua University; Advanced Manufacturing EDA), "Inverse Lithography Technology (ILT) Under Chip Manufacture Context," Micromachines 17(1):117, 16 January 2026, open access (read 12 November 2026). OPC passed through "roughly four distinct generations: rule-based OPC (RB-OPC), model-based OPC (MB-OPC), rule/model-based sub-resolution assist feature (SRAF) insertion, and model-driven pixel-level optimization employed in inverse lithography technology (ILT)," each with more computation. Inserting SRAFs "substantially expanded" the process window for semi-isolated and isolated patterns. ILT was first proposed by B. E. A. Saleh and colleagues at the University of Wisconsin-Madison in 1981; it was long limited to hotspots by runtime, curvilinear shapes and mask manufacturability. GPU computing and multi-beam mask writers changed that: "in 2019, D2S demonstrated a full-chip ILT solution with practical runtimes (around 48 h) and mask-write times (around 12 h)."
- NVIDIA, "TSMC and Synopsys Bring Breakthrough NVIDIA Computational Lithography Platform to Production," press release, 18 March 2024 (read 12 November 2026). "Computational lithography is the most compute-intensive workload in the semiconductor manufacturing process, consuming tens of billions of hours per year on CPUs. A typical mask set for a chip — a key step in its production — could take 30 million or more hours of CPU compute time, necessitating large data centers within semiconductor foundries." "350 NVIDIA H100 systems can now replace 40,000 CPU systems." TSMC's tests: "a 45x speedup of curvilinear flows and a nearly 60x improvement on more traditional Manhattan-style flows"; "with curvilinear the mask shapes are represented by curves," while "Manhattan mask shapes are constrained to be either horizontal or vertical." Synopsys' Proteus optical proximity correction software runs on the cuLitho library. A generative-AI workflow creates "a near-perfect inverse mask or inverse solution to account for diffraction of light"; "the final mask is then derived by traditional and physically rigorous methods."
- IMS Nanofabrication, "The history of IMS Nanofabrication" (company page, read 12 November 2026). The proof-of-concept multi-beam mask writer, assembled in Vienna in 2011, had "262,144 programmable electron beams on 6-inch photomasks." In 2014 JEOL supplied the platform for "the world's first commercial multi-beam mask writer"; data rate raised tenfold in 2016 to 120 Gbit/s; MBMW-101 tools for the "7-nanometer technology node" in 2016 and MBMW-201 for the "5-nanometer technology node" in 2019. A first Intel investment in 2009.
- KLA-Tencor, "KLA-Tencor Announces New Teron SL650 Reticle Inspection System," press release, 20 May 2014 (read 12 November 2026). "A new reticle quality control solution for IC fabs that supports 20nm design nodes and beyond"; with 193 nm illumination it assesses incoming reticle quality, tracks degradation and detects yield-critical defects such as haze growth and contamination; installed at foundry, logic and memory makers for "incoming quality check and re-qualification of reticles"; "changes in reticle quality can have a devastating impact on every wafer printed."
- D. Brouns (ASML), "Development and performance of EUV pellicles," Advanced Optical Technologies 6(3-4):221-227, 2017 (read 12 November 2026). An EUV pellicle "is a thin membrane that protects particles from landing on the reticle surface." Particles "land on the pellicle membrane (which is not in focus) instead of on the reticle (which is in a field plane, so any obscuration will print directly onto the wafer)." Typical thickness about 50 nm, about 10% EUV transmission loss per pass; target lifetime 10,000 wafers.
- C. Weber, C. N. Berglund (Portland State University) and P. Gabella, "Mask Cost and Profitability in Photomask Manufacturing: An Empirical Analysis," IEEE Transactions on Semiconductor Manufacturing, November 2006 (author's copy, read 12 November 2026). From interviews covering six mask shops' 1990 to 2005 operations and a survey of seven photomask makers: "the uncontrolled growth of optical proximity effect correction and resolution enhancement techniques is driving up the cost of pattern generation and mask inspection." Table 1: masks without OPC, laser-written, under 4 hours to write and to inspect, US$1k to 5k; with OPC on laser tools, $5k to 40k; with OPC on vector-scan e-beam tools, 6 to 36 hours to write and 6 to 36 to inspect, $20k to 120k. E-beam writers cost US$15 million or more; up to two inspection tools at about $20 million each per writer. A mask set "expected to increase from ~US$ 500k for the 130 nm technology node to ~US$ 1 million for the 90 nm node and to ~US$ 2 million for the 65 nm node." Leading-edge layers are the primary drivers of a set's cost.