Beyond High NA: The Angstrom Patterning Roadmap

Abstract

The path beyond High NA will not be one heroic optical leap. It will be a sequence of coordinated improvements in 0.55 NA imaging, source efficiency, mask materials, pellicles, resist uniformity, stochastic correction, thin-film transfer, metrology, mechatronics, device tolerance, selective multiple patterning, and design. Some layers may use High NA single exposure. Others may combine EUV with self-aligned processes, directed self-assembly, cut masks, or multiple exposure. Future systems may pursue still larger numerical aperture or different wavelength, but every step faces shrinking depth of focus, mask geometry, optics scale, photon economics, materials, and capital burden. I organize the roadmap by functional edge placement and good-device value rather than nominal node label. The near term industrializes High NA. The middle term co-optimizes patterns and devices around stochastic and field limits. The long term requires angstrom-scale metrology and processes that remain stable when atomic populations are no longer a small perturbation. The frontier succeeds if each generation expands useful computation faster than manufacturing complexity, energy, and failure probability grow.

High NA is the next platform, not the final answer

The 0.55 NA system improves optical resolution relative to 0.33 NA at the same wavelength. It also introduces reduced depth of focus, anamorphic imaging, half fields, larger optics, demanding mask effects, thinner resist, and new fab infrastructure.

The first roadmap obligation is therefore to mature High NA as a complete platform. Scanner resolution must be joined by source stability, optics lifetime, masks, pellicles, resist and underlayer stacks, etch, metrology, stitching, computational correction, designs, service, and supply.

I would measure maturity by production distributions:

• good wafers per hour at product dose;
• edge placement across field, wafer, tool, and time;
• stochastic failures per opportunity;
• overlay and focus at functional features;
• mask and pellicle lifetime;
• final etch contour and electrical yield;
• availability, service duration, energy, and route cost.

A record pitch is a useful proof. A stable product layer is a platform.

Resolution cannot be separated from process window

Rayleigh scaling suggests continuing to raise $NA$ or lower wavelength:

$$CD=k_1\frac{\lambda}{NA}.$$

Depth of focus falls as $1/NA^2$. Larger NA also increases mask angles, polarization sensitivity, optical size, and aberration control. Shorter wavelengths increase photon energy and can worsen source, mirror, mask, resist, and metrology challenges.

The roadmap should compare the total margin volume, not nominal resolution. I define an abstract manufacturing volume

$$V_m=\Delta D\,\Delta F\,\Delta O\,(1-P_f),$$

where dose latitude, focus latitude, overlay latitude, and failure probability are simplified axes. A process with smaller nominal CD and vanishing $V_m$ is not manufacturable.

Route	Strength	Scaling wall
High NA single exposure	Simpler edge history and fewer masks	Optical, stochastic, field, and transfer limits
High NA multiple exposure	Extends pitch and layout flexibility	Overlay, masks, dose, cycle time, defect accumulation
Self-aligned patterning	Pitch division with reduced overlay sensitivity	Process complexity and edge placement
Directed self-assembly	Molecular pitch definition and roughness healing potential	Defectivity, registration, design constraints
Alternative wavelength or particle beam	New physical resolution space	Source, optics, masks, throughput, materials, ecosystem

No route wins universally. Different layers can choose differently.

Device architecture can buy lithographic margin

Design technology co-optimization changes cell layout, track height, routing, redundancy, cut strategy, and device geometry to fit manufacturing strengths. System technology co-optimization moves some scaling benefit into three-dimensional integration, chiplets, memory hierarchy, architecture, packaging, and software.

This is not surrender. A nanometer saved through a more tolerant cell may be more valuable than one forced through an exposure at poor yield. Redundant vias can reduce stochastic consequence. Regular grids can improve imaging. Buried power and backside delivery can reorganize routing. Vertical devices change which edges are critical.

The objective is system power, performance, area, cost, and reliability. Lithography supplies a constrained geometry service to that objective.

Near termIndustrialize 0.55 NA

Close masks, resists, fields, metrology, service, yield, and fab operation.

Middle termCo-optimize pattern and device

Use stochastic-aware correction, selective splits, regularity, and tolerance.

Long termExpand physical patterning options

Evaluate higher NA, wavelength, assembly, materials, and new exposure methods.

AlwaysMeasure functional edge and value

Compare every path through uncertainty, good output, energy, cost, and resilience.

Stochastic control becomes design control

At smaller features, photon count and molecular population fall. Increasing dose improves photon statistics and reduces throughput. New resists can raise absorption or chemical uniformity. Better aerial-image contrast increases effective margin. Computational correction can move vulnerable clips away from failure cliffs.

The deepest gain may come from designing structures whose function tolerates local variation. If one missing contact destroys a die, the process must reach extraordinary tail probability. If redundant connections and detection contain it, architecture shares the burden.

I expect stochastic OPC to become more explicit, with optimization based on failure probability rather than mean contour alone. Models will need validated tails, and mask complexity will grow. Curvilinear mask writing and inspection must mature together.

Metrology must not consume the tolerance

As edge budgets approach fractions of a nanometer, instrument noise, probe interaction, edge-definition model, sampling, reference-material stability, coordinate transformation, and environmental drift become comparable to the quantity.

No single instrument will provide all truth. I want traceable combinations of electron, optical, scatterometric, scanning-probe, X-ray or EUV, electrical, and destructive methods. Reference structures should connect dimensions and roughness spectra across scales.

If total measured variance is

$$\sigma_m^2=\sigma_p^2+\sigma_i^2,$$

where $\sigma_p$ is process and $\sigma_i$ instrument contribution, estimating a small $\sigma_p$ by subtracting two similar values becomes highly uncertain. The metrology architecture must lower instrument variance and use repeated, independent, model-aware evidence.

Angstrom control should not mean every atom occupies a commanded coordinate. It means distributions of functional edges and interfaces are known and stable at subnanometer scales appropriate to device response.

Future optics need an economic photon source

Larger numerical aperture may require still larger or differently arranged optics, stronger anamorphism, smaller fields, or new mask geometry. A shorter wavelength requires a new source, coatings, masks, contamination regime, resist absorption, and metrology ecosystem. Photon energy increases while usable source conversion and mirror reflectivity may fall.

Every future exposure concept should publish a photon ledger from wall plug to absorbed resist, a defect ledger from source to device, and a service ledger from installation to annual good wafers. A laboratory image is not evidence of high-volume viability.

The established EUV ecosystem has enormous cumulative knowledge. A successor must outperform not only 13.5-nanometer physics but decades of manufacturing learning and infrastructure. Hybrid approaches that reuse parts of the ecosystem may have a practical advantage.

Supply and energy can become scaling walls

More complex tools depend on more specialized optics, materials, software, and service. Long lead times and concentrated suppliers can limit global capacity even when device physics succeeds. Roadmaps need workforce, repair, metrology, production-tool manufacturing, and raw-material capacity.

Energy must be considered from fab to system. Higher source power and dose can improve yield while raising process energy. More efficient chips can save orders of magnitude over their lives. The comparison needs measured workloads and realistic product use.

ResolveDemonstrate physical image RepeatControl feature population TransferCreate functional material edges ProduceRun yield and availability at scale JustifyDeliver system value responsibly

Build an institution that can change direction

The EUV program succeeded through decades of work across source, optics, masks, resist, metrology, computation, mechatronics, fabs, and suppliers. The next transition will require the same institutional patience and stronger shared benchmarks.

I want precompetitive test platforms, accessible High NA learning, reference masks and wafers, traceable metrology, open failure definitions, workforce development, and model intercomparison. Companies must protect legitimate intellectual property, but the industry also needs common physical evidence to avoid optimizing incompatible assumptions.

Roadmaps should preserve options until data distinguishes them. Overcommitting to a single post-High-NA concept can starve better paths. Refusing to choose can diffuse investment. Stage gates based on good-device evidence offer a middle course.

The angstrom frontier

This route began by defining EUV as an industrial institution for placing material boundaries. The same definition governs its future. Angstrom-scale patterning is not a marketing node or one isolated dimension. It is the ability to manufacture functional structures whose critical interfaces are controlled at fractions of a nanometer across enormous populations.

High NA will carry the near future. Better materials, masks, computation, metrology, etch, and design will extend it. Selective multiple patterning and directed assembly may share difficult layers. New exposure physics may eventually earn a platform.

The winning path will be the one that converts atomic-scale control into more useful and efficient systems while preserving yield, energy discipline, repairability, and industrial resilience. That is a harder standard than the smallest printed line. It is also the standard that makes the line matter.