Can in-situ metrology preserve two-photon lithography accuracy over long runs? A single-cavity dual-comb laser supplies writing light and phase measurements; a dynamic model adjusts exposure from measured height. Fabricated diffractive optical elements and four-focus lenses showed improved contrast and focus uniformity. A 12-hour cuboid experiment reported an 84 nm mean height error for a 4.8 µm target. The demonstrated feedback controls height, requires negligible or consistent development shrinkage, and did not demonstrate lateral feedback.

TWO-PHOTON LITHOGRAPHY · NATURE COMMUNICATIONS · 2026

Printing tiny structures is one challenge. Keeping them accurate for hours is another.

A dual-comb laser writes into resin and measures the structure as it grows. Feedback turns those measurements into a more precise next dose of light.

mean height error 4.8 µm target · 12-hour test

higher measured image SNR closed loop versus open loop

Print through the drift.

Set an exposure loss, advance the clock, then try a fresh print with feedback. Can the columns still reach the same height?

Advance the clock to see how exposure drift changes the printed height.

Illustrative model · height exaggerated · drag to rotate

YOUR PRINT SETTINGS

Open loop keeps the original dose. Closed loop measures height and adds exposure toward the target.

This teaching model applies gradual exposure loss and simplified incremental correction. Changing strategy or drift starts a new simulated print. Curing is irreversible.

The 4.8 µm cuboids had a mean height error of 84 nm with feedback over a 12-hour test. The 1.0 µm target gave a mean error of 130 nm.

ONE LIGHT SOURCE. TWO JOBS.

How does light become a ruler?

Write at 800 nm

Focused femtosecond pulses trigger two-photon polymerization in liquid photoresist. A scanner and moving stage build the designed structure point by point.

Measure at 1052 nm

Curing changes the resin’s refractive index. A spectrally encoded line probes the growing structure; interference with a reference comb reveals its phase and, through calibration, its height.

Correct the next dose

A dynamic model connects the measured phase to the required exposure. The controller starts below the target, then adds light in small steps until the height approaches its setpoint.

The comb repetition rates differ by 356 Hz. The reported 4.8 nm height sensitivity uses 2 seconds of averaging ; it is a sensing result, separate from printing accuracy.

TRY THE FEEDBACK STEP

Approach. Don’t overshoot.

Polymerization is irreversible. You can add exposure, but you cannot un-cure a column that has grown too tall.

Start at half the predicted dose. Try fixed additions, or let each measured correction close half of the remaining gap.

Teaching model inspired by Methods, Eq. 4. The paper uses an approximation ratio of 0.5 or less.

Height 2.400 µm · 2400 nm below target.

Tiny height errors. A different image.

A phase plate controls how light waves interfere. Its pixels have different designed heights: even a small height error can send light away from the intended pattern.

Turn up the error.

Phase error 0.00 rad · equivalent height error ≈ 0 nm

Six published simulation cases: 0, 0.02, 0.05, 0.10, 0.20 and 0.30 rad. Height equivalents use the transmission-mode calibration below.

Compare the real optics.

Open loop · measured signal-to-noise ratio 4.94

The authors printed an 11-level, 100 × 100-pixel phase plate, 790 × 790 µm across. Illuminated at 800 nm, its computed phase pattern forms the letter “A.” Here, SNR is mean gray level divided by the background standard deviation.

A phase-to-height conversion. In transmission, a thicker cured region produces a larger phase change relative to uncured resin. With Δn ≈ 0.035 and λ0 = 1.052 µm, one micrometer of height gives about 0.21 radians of phase. Reflection doubles the phase change.

A SECOND DEVICE · FIGURE 6

Four foci. One precise print.

An eight-level Fresnel zone plate splits 800 nm light into four focal spots, 4 mm from the lens. Feedback makes their measured widths much more consistent.

Reported focus-width uniformity

F1 · Width: 7.93 → 7.80 µm · Peak-to-background ratio: 564 → 1820

Width is the full width at half maximum. Peak-to-background ratio compares mean intensity at a focus with mean intensity outside it. Values reproduced from the paper; chart redrawn.

From a fine feature to a reliable device.

The contribution is sustained height control during fabrication. It connects nanometer-scale metrology to a printing decision while the part is still being made.

The paper demonstrates optical elements and a 14-hour continuous fabrication process. More reproducible printing could support larger, more consistent optical devices for communication, computing and displays—applications the authors propose.

The experiments use 800 nm femtosecond pulses focused into IP-L 780 or IP-Dip photoresist on glass or silicon. The demonstrated cuboids are raster-scanned at 30 mm/s with 100 nm hatching distance and 45 mW laser power. Development removes uncured resin using propylene glycol monomethyl ether acetate, followed by isopropanol.

The phase-to-height calibration assumes negligible or consistent shrinkage after development; the authors checked this under their processing conditions using atomic force microscopy. The demonstrated feedback controls height. The dual-comb system’s lateral imaging resolution is 1.10 µm, and lateral feedback was not demonstrated.

The 12-hour cuboid comparison in Figure 4, the separate 10-hour comparison in Figure 1, and the 14-hour fabrication demonstration described in the abstract are distinct experiments.

THE RESEARCH BEHIND THIS EXPERIENCE

Closed-loop high-precision two-photon lithography based on a multiplexed single-cavity dual-comb laser

Yalan Yu · Zhiwei Zhu · Benjamin Willenberg · Justinas Pupeikis · Christopher R. Phillips · Shih-Chi Chen

Nature Communications 17 , 7135 (2026)

Source images: Yu et al., © The Author(s) 2026, licensed under CC BY 4.0. Figure 5d/e and Supplementary Figure 2a–f are cropped and resized; focal-width data are redrawn. The graphical abstract is an AI-generated conceptual illustration informed by Figures 1 and 5.

The virtual printing and dose experiments are illustrative teaching models. Source simulations and experimental results are labelled separately.