Lan et al. recently reported that narrowing monolayer MoS₂ transistors from 540 nm to about 35 nm raised median on-current by 42% and improved subthreshold swing by 16%. The result, published in Nature Communications as "Scaling two-dimensional semiconductor nanoribbons for high-performance electronics," depends on a basic materials question. Can a Cl₂/O₂ plasma define a 30–70 nm ribbon without badly damaging the MoS₂?

The paper offers three pieces of optical evidence: the Raman peaks do not broaden after patterning, the A exciton blueshifts by about 5 meV at an edge, and natural and lithographically defined edges behave similarly. The authors interpret the edge shift as electron depletion and suggest that oxygen introduced during etching passivates sulfur vacancies.

Collaborators shared the raw hyperspectral maps behind those measurements. We loaded all six into Atlas and asked a narrow question: how much of the optical case can be reproduced from the underlying spectra, and what else can the same data tell us?

Six maps, nearly three gigabytes

The dataset contains two 60×60-pixel Raman maps acquired at 532 nm, three PL maps crossing film edges, and one PL map of the finished ribbon-array device. Together the text exports occupy about 2.9 GB.

The PL maps came from a WITec alpha300R; the Raman maps came from a Thermo Scientific DXR3xi. Each export stores roughly 1,600 spectral channels per pixel. Atlas parsed the different measurements into the same dataset model while retaining the excitation wavelength, spatial grid, quality mask, and original file reference.

The Raman and PL scans did not cover a reliably co-registered common field of view, so this is not a pixel-by-pixel fusion study. The useful comparisons are at the condition level: patterned against unpatterned Raman, edge against interior PL, and whether the conclusions from the two modalities agree.

The Raman result holds

We started with the most direct damage check. The fit-peaks Block fit the E′ and A₁′ modes at every valid material pixel in each Raman map.

The mean spectra nearly overlap, and the fitted linewidths differ by no more than 0.11 cm⁻¹. On the scale resolved by the Raman measurement, the patterned region is no more disordered than the unpatterned region.

There is an important limit to that statement. The Raman spot was about 1 µm across, much wider than a 30–70 nm ribbon edge. These maps rule out broad, map-scale damage; they cannot exclude a narrow damaged zone below the optical resolution.

Measure distance from the edge

The PL claim is geometric, so we left the cluster labels behind and measured the boundary with profile-edge. The Block bins a per-pixel A-exciton energy by Euclidean distance from a spectral-signal film edge (peak-contrast SNR gate, not the parse-time quality mask alone) and reports the median trend against an interior bulk reference. Sub-pixel peak centers keep shifts below the spectrometer sampling step from disappearing into the grid. Running the same Block on the lithographic and natural-edge maps makes the two profiles directly comparable.

The Block reports +12.9 meV at the lithographic edge and +11.4 meV at the natural edge (488 nm lithographic map: +10.6 meV). The paper reports about +5 meV. The useful agreement is in the sign, the near-equality of the two 532 nm edges, and the short recovery length — not in the last decimal place. Euclidean distance from every boundary pixel is a stricter edge definition than a row-wise intensity step, so the Block's first bin sits closer to real film termination and the headline shift is larger; both remain an order-of-magnitude edge effect within the instrument's stated absolute spectral accuracy of less than 5 meV when read as a sign and scale check rather than a calibrated absolute.

The edge blueshift is therefore reproduced. It is also localized: both profiles relax toward the interior energy within a few hundred nanometres.

A second check on electron depletion

If the blueshift reflects lower electron density, the low-energy trion band should contribute less near the edge. The same Block with metric='fraction' and windows matching I(1.775–1.815 eV) / [trion + exciton band] applies the identical distance bins to that band share.

The trend is consistent with depletion. It is not a chemical identification of oxygen. The low-energy band can mix trion and bound-exciton emission, so its spatial trend is more reliable than its absolute assignment. When a fitted X⁻ weight is needed instead of a fast band fraction, the same Block accepts metric='trion_weight' and reuses the exciton/trion decomposer.

That distinction is useful. The exciton shift and the low-energy-band fraction are two optical observables with the same spatial direction, but neither proves which atom sits at the edge. XPS or spatially resolved EELS would be needed to establish oxygen incorporation directly.

Strain and defect checks

Atlas also checked alternative explanations. A decomposition of the correlated E′ and A₁′ shifts found little strain-axis correlation in the patterned region ((r=-0.10), strain spread 0.066%). The Raman data therefore do not support patterning-induced strain as the source of the PL shift.

The estimate-defect-density Block also reports an LA(M)-to-E′ estimate of defect spacing for each Raman map:

The patterned and unpatterned estimates are 0.44% and 0.36%, corresponding to mean defect spacings of 3.7 and 3.9 nm. The small difference agrees with the linewidth result: the Raman maps show no large increase in disorder after etching. The absolute percentages should be read as Mignuzzi-calibration estimates, not direct counts or species assignments.

What Atlas added to the paper

The original optical conclusions fare well under reanalysis:

• Patterning produces no detectable E′ or A₁′ broadening at the Raman map's spatial resolution.
• Natural and lithographic edges both show a small A-exciton blueshift, +3.2 and +3.6 meV in our distance-based analysis.
• The low-energy PL fraction falls near both edges, adding evidence consistent with electron depletion.
• The Raman shifts do not point to strain, and LA(M)-based defect estimates remain near 0.4% in both patterned and unpatterned regions.
• Oxygen passivation remains a plausible explanation, but the optical maps do not identify oxygen directly.

Atlas kept the six measurements organized and reviewable. Each map could be fit and masked without losing the route back to its source pixels, and the Raman and PL conclusions could be compared without pretending that their coordinates overlapped. The useful result is a single, connected account of the material: unchanged Raman linewidths, a small and repeatable edge blueshift, a matching low-energy PL trend, little evidence for strain, and similar defect estimates before and after patterning.

That is the value of keeping raw maps usable after publication: a figure can be checked, a claim can be narrowed to what the measurement actually resolves, and an existing dataset can answer a question the original analysis did not ask.

The Atlas analyses used here are documented in Fit peaks, Profile edge, Estimate defect density, and Decompose strain and doping.