GaAs and its alloys are the workhorse of photonics: lasers, VCSELs, detectors, solar cells, and an increasing number of quantum-dot and nanowire emitters. Almost all of that is heterostructure — quantum wells, tubes, dots — where the number that matters is not the bulk band gap but where the confined state sits and how sharp it is. Photoluminescence mapping is the standard way to ask, and the maps get large fast: a 100 µm field at 300 nm pitch is a hundred thousand spectra.

This post takes one such map, a published micro-PL dataset of GaAs/GaAsP quantum-well-tube nanowire heterostructures, and shows what Atlas does with it using the same Blocks that run in the Matter42 app. It is a shorter post than our WS₂ or GaN ones, because a band-edge PL map asks fewer distinct questions — but each one is asked a hundred thousand times.

The dataset

Patel and co-workers grew GaAs/GaAsP core–shell nanowires with a radial quantum well, embedded the arrays, and "nanoskived" them into thin sections laid flat on a substrate. Each section — they call them flakes — was then mapped by confocal micro-PL at 532 nm excitation, 334 × 334 pixels over 100 × 100 µm, with a 1,024-channel spectrum from 649 to 956 nm at every pixel. The raw archive is on Figshare; the paper is in ACS Applied Materials & Interfaces.

Atlas read the flakes into its PL map schema and was told material=GaAs, which selects the band-edge model family and the GaAs reference at 873 nm. That reference is bulk GaAs; the whole point of this sample is that it does not emit there.

One Block, a hundred thousand fits

Characterize PL fits the primary emission line on the map mean and then in every pixel, and reports position, width, intensity, and a deep-level defect-emission ratio as maps and distributions.

The headline number is the +138 meV. Bulk GaAs at room temperature emits at 1.42 eV; this sample emits at 1.56 eV. That is quantum confinement in the radial well, and it is the number a grower is targeting when they set the well thickness. Atlas reports it as a shift against the bulk reference and lists the three generic causes, because the Block does not know it is looking at a quantum well. In a workflow, you would tell it — or compare against the design value directly.

The map is where the sample's character shows. The nanowires are discrete emitters on a dark background, so the intensity map is speckled: bright wires, dim gaps, and a distribution with a 5th percentile fifteen times below the 95th. That is the sample, not the measurement, and the 96.8% emitting fraction says that almost every pixel in this flake still has a fittable line — the wires are dense enough to fill the field at 300 nm pitch.

Position and width as maps

These two maps are the pass/fail view for a heterostructure. The energy map tells you whether the well came out at the same thickness everywhere; here it did not, to about ±11 meV (one standard deviation), which for a GaAs well of this energy corresponds to a fraction of a monolayer of thickness variation between wires. The linewidth map tells you whether the interfaces are sharp; a 68 meV room-temperature line in a nanowire quantum well is respectable, and the fact that it is narrowest where the signal is strongest says the broadening is not an artifact of low counts.

A second flake from the same archive, S4F1, ran through the same Block with the same parameter and returned a narrower line — 56.0 meV at 1.5576 eV — but a 64% emitting fraction, meaning a third of its field is background with no fittable line. Same growth, different section, and the two numbers together say: sharper wires, sparser flake. That is a comparison the Block makes trivially and a person makes only if they remember to look at both.

What Atlas did not find, and why

The Block's deep-level defect-emission ratio came out at 0.0005 on the map mean and 0.003 on average per pixel — essentially zero. It is tempting to read that as "no deep-level defects". Do not. The GaAs deep-level window Atlas catalogues runs from 950 to 1080 nm, and this measurement stops at 956 nm. The low ratio is a statement about the spectral range, not the material. Atlas reports the window it integrated; a reader who checks it against the axis sees the gap. A measurement of the same flakes extended to 1.1 µm is what would actually answer the question.

Likewise, the linewidth-based defect score is low (p95 = 0.11), and that is correct as far as it goes: nothing in the line shape says "disordered". For a nanowire heterostructure, the failure modes that matter — stacking faults, non-uniform shell growth, surface recombination on an unpassivated sidewall — show up first as intensity and energy heterogeneity between wires, which is exactly what the maps above resolve.

What this gives a III-V team

From one 105 MB hyperspectral flake:

• the confined-state emission energy, +138 meV above bulk GaAs, fit at every one of 111,556 pixels,
• a per-wire energy spread of ±11 meV and a linewidth spread of 51–80 meV, both as maps you can point at,
• an emitting fraction that distinguishes a dense section (97%) from a sparse one (64%) across flakes from the same growth,
• and a deep-level defect ratio that Atlas reports honestly as near zero, together with the spectral window that makes it uninformative on this dataset.

Where this extends: the same Block reads GaAs, InP, GaP, InGaAs, AlGaAs, and their common substrates from a single material argument and carries reference energies and deep-level windows for each, so a multi-material photonics line can run one workflow. Excitation-dependent series, temperature series, and cross-modal registration against Raman or AFM maps of the same field are handled by the same Correlate modalities and Profile edge Blocks we use on 2D materials.

The Block used here is documented at Characterize PL. Every figure above was generated from the published archive through the same Atlas analysis functions the Block runs.

If you grow III-V heterostructures and have PL maps — of wafers, wires, or dots — that deserve more than a mean spectrum, request access or get in touch.