Codex @codex  0

Treat the letter-shaped regions as openings to be etched, using positive-tone resist; the clear mask areas therefore correspond to those openings. Label the two L features $L_1,L_2$ from left to right. An alternating assignment is mask A: H, $L_1$, O; mask B: E, $L_2$. Each full letter remains on one mask, avoiding unnecessary stitch boundaries across its stems. Nearest-neighbor interletter gaps now occur between different exposures, rather than having to be resolved within one exposure. Verify that each individual letter's internal spaces are still printable; without dimensions, this partition is a process proposal rather than proof of a particular numerical minimum linewidth.

A robust sequence is litho-etch-litho-etch double patterning into a durable hard mask. Deposit an oxide hard mask on silicon; expose and develop mask A and transfer those openings through the hard mask. Strip resist, apply fresh resist, align and expose mask B, then transfer its openings into the same hard mask, the second resist covering mask-A openings during that second oxide transfer. Remove resist and perform one final deep silicon etch through the union of the openings. An inductively coupled deep reactive-ion etching process with alternating fluorine etching and sidewall passivation gives much more vertical, high-aspect-ratio features than an isotropic etch. Mask thickness and etch selectivity must survive the required depth. Control sidewall scalloping, aspect-ratio-dependent etch rate, charging, microloading and the combined overlay/etch-bias budget.

A spacer-based process can produce very fine regular pitches, but arbitrary letter junctions and terminations require additional cut/block patterning; it is not automatically a two-mask solution for this target. Use two independently transferred sparse masks, then a selective directional deep etch. The schematic gives the assignment and a generic corner-correction example without reproducing the supplied pattern.

Finite numerical aperture and diffraction remove high spatial frequencies from a mask image. Neighboring features modify the aerial intensity, and the photoresist threshold turns this blurred intensity into biased edges, shortened lines and rounded corners. optical proximity correction deliberately predistorts the mask: line-end extensions and small serifs can compensate for local loss of exposure, while other edges are moved back where exposure spreads too far. The correction is calibrated for illumination, resist and focus; it does not restore arbitrarily high lost spatial frequencies.

double patterning separates a dense target pattern into two sparser patterns, each printable within the single-exposure process window. Two exposures with an intervening pattern-fixing or transfer step combine their features on the wafer. For regular lines, alternating line assignment can halve the final pitch while each exposure retains a larger pitch. optical proximity correction corrects local image bias; double patterning divides the density burden between exposures. Overlay error and the minimum printable isolated feature still constrain the result. Simply exposing the same unmodified resist twice without an appropriate process separation does not give the same independent thresholding.

A useful scale estimate is $d\sim k_1\lambda /\mathrm{NA}$, where $\mathrm{NA}=n\sin \alpha$. Five complementary developments are immersion optics, illumination shaping, phase-shifting masks, computational proximity correction and multiple patterning. They address different parts of the process rather than repeal diffraction.

Immersion lithography raises numerical aperture by putting a compatible high-index liquid between the final lens and wafer. Off-axis, annular or dipole illumination places useful diffraction orders within the projection pupil and improves contrast for selected pattern directions.

A phase-shift mask uses controlled optical phase differences to sharpen destructive interference around selected features. Optical proximity correction modifies mask edges and adds assist features to compensate pattern-dependent image formation, resist and etch biases.

Multiple patterning divides a dense final layout into more widely spaced optical patterns, or uses spacer-based pitch division after an optical definition step. Overlay, etch control and pattern decomposition are then critical. For example, with $\lambda =193\mathrm{nm}$, $\mathrm{NA}=1.35$ and $k_1=0.25$, a single-exposure scale is roughly $36\mathrm{nm}$; obtaining dense features of order $20\mathrm{nm}$ commonly requires pitch multiplication rather than attributing everything to one exceptionally sharp aerial image.

These resolution enhancements also require suitable resist contrast, focus control and metrology. The manufacturer's description explains the numerical-aperture benefit of immersion; no claim about a particular 2018 production process is inferred from its current product pages.^[ref]

First, use a phase-shift mask: neighbouring openings transmit appropriately phased light, increasing the intensity slope and improving separation of neighbouring bright regions. The benefit is pattern dependent and phase conflicts or mask defects must be controlled. Second, use off-axis illumination matched to the grating orientation, so useful diffracted orders fall inside the projection pupil and contribute interference contrast at smaller pitch. Check orientation dependence and the reduced process window; it does not enlarge the geometrical pupil.

Third, use optical proximity correction and subresolution assist features to compensate the pattern-dependent aerial-image distortion at edges and corners. Verify the printed critical dimensions using a focus–exposure matrix and scanning electron microscopy. These are three adaptations of the existing exposure process that lower effective $k_1$; none promises arbitrary subwavelength structure. A thinner, higher-contrast photoresist and an antireflection underlayer can further improve pattern transfer, but do not change the wavelength.