Additive manufacturing builds parts by adding material under digital control rather than cutting it away, and the current research frontier is less about the printers than about beating the resolution-versus-speed trade-off and making print fidelity predictable enough for real engineering parts.
Summary
Additive manufacturing (AM) is the family of processes that construct an object by depositing or solidifying material where a digital model says material should be, instead of removing it from a blank. The supplied evidence covers three distinct physical families rather than one technology. Light-based photopolymerisation cures liquid resin: two-photon polymerisation (2PP, also called direct laser writing) reaches sub-micrometre features but only over sub-millimetre volumes, while single-photon tomographic volumetric additive manufacturing (TVAM) forms centimetre-scale objects within seconds at feature sizes of tens of micrometres. Metal powder bed fusion (L-PBF) melts metal powder with a laser and is already used in production, but suffers from unstable melt-pool keyholes that generate porosity and spatter. At the smallest scale, focused electron beam induced deposition (FEBID) writes freestanding 3D nanowires directly.
The parameters that decide the field are resolution, build volume, throughput and fidelity, and they trade against each other. Photopolymerisation is explicitly described as constrained by a resolution/speed trade-off. Two lines of attack appear in the sources. The first is optical and computational: differentiable ray-optics and wave-optics models that compute projection patterns accounting for occlusions, refraction, scattering and even chemical blur, enabling overprinting onto pre-existing components and single-view holographic printing with roughly 10 micrometre lateral features. The second is hybridisation: running TVAM and 2PP on one platform in the same resin so a millimetre-scale body can carry sub-micrometre detail without stitching, platform changes or intermediate processing.
A parallel and equally large theme is process informatics. Prints deviate from the target because of physico-chemical effects, and neural networks are being trained to predict and pre-correct those deviations rather than relying on iterative trial prints. Related work covers automated defect classification that can absorb new defect classes and new part geometries from scarce data, machine health monitoring to replace experience-based maintenance, and physics-informed prediction of metal AM temperature fields that generalises to unseen materials without retraining. Read together, the sources describe a technology that already works and is now being industrialised through metrology, modelling and control.
Evidence base
- 19 June 2026: a single-platform hybrid printer combined single-photon TVAM for millimetre-scale volumes with 2PP for sub-micrometre detail in the same photoresin, without intermediate post-processing or stitching.
- 16 June 2026: TVAM overprinting onto absorbing, refracting, reflecting and scattering pre-existing components was demonstrated in round and square vials, with the open-source Dr.TVAM differentiable ray-optics framework producing high-quality projections within minutes.
- 17 June 2026: single-view holographic volumetric printing using time-multiplexed phase-only holograms achieved roughly 10 micrometre lateral features and higher fidelity than optical-only optimisation by pre-compensating for chemical blur.
- 16 June 2026: the MOSAIC keyhole segmentation algorithm averaged an F1 of 0.894 and precision of 0.953 across 12 L-PBF samples at 19.9 ms per frame on CPU, versus 54 ms and 5284 ms for the SAM and YOLO baselines.
- 15 May 2026: 41 cobalt and iron nanowires grown by FEBID at angles from 0 to 90 degrees showed measurable composition and magnetic induction variation, quantifying a fidelity limit for 3D nanoscale printing.
- 2 July 2026: two smartphones 60 cm away in a non-line-of-sight setup reconstructed a 3D printer’s G-code commands at 98.89% command-level accuracy, with transferability shown to a second printer.
- 21 July 2026: permittivity of commercial 3D printing materials was characterised from 70 to 110 GHz by open-waveguide extraction, targeted at 6G antennas and metasurfaces.
Open questions
- Does the unified TVAM plus 2PP platform hold its resolution and adhesion across many materials and repeated builds, or only in the demonstrated glass-rod, single-resin configuration?
- Do neural pre-correction models trained on one machine and resin transfer to other printers and chemistries, as the authors expect, or must they be retrained per installation?
- Can composition non-uniformity in beam-induced deposition of overhanging 3D nanostructures be compensated in-process, rather than only measured after the fact?
- Will acoustic and magnetic side-channel leakage be treated as a qualification requirement for distributed AM, given a 98.89% G-code reconstruction rate from consumer smartphones?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.