High-resolution tomographic volumetric additive manufacturing

Additive manufacturing (AM) is increasingly used across aerospace, healthcare, and bioprinting, demanding high throughput, high resolution, and material versatility. Conventional layer-by-layer AM (e.g., DLP and SLA) faces constraints including viscosity-limited throughput, support-structure requirements, anisotropy in part properties, and limited feedback control. Volumetric, multi-beam approaches that irradiate a transparent resin bath from multiple angles can achieve higher throughput and handle more viscous materials, but demonstrated feature sizes have been limited to ~300 μm. Unlike layerwise methods that rely on highly absorbing resins, tomographic volumetric AM requires transparent resins and precise spatio-temporal control of dose deposition to achieve high resolution. This work investigates how to push resolution by engineering the illumination étendue and by integrating feedback to control photopolymerization kinetics, aiming to produce centimeter-scale parts with sub-100 μm features within seconds.

Prior work highlights limitations of layer-by-layer AM: throughput constrained by resin viscosity and support structure requirements, leading to post-processing and anisotropy. Feedback control has been explored primarily in metal AM to improve quality. Volumetric AM methods using multi-angle irradiation and tomographic reconstruction can exceed the throughput of DLP/SLA and process viscous resins and gels, but feature size has remained ~300 μm. In polymer photochemistry, oxygen inhibition and radical diffusion can affect polymerization; DLP/SLA often use absorbing resins to confine cure depth, which is incompatible with volumetric approaches that require transparency. The present study builds on tomographic reconstruction concepts (Radon transform, filtered back-projection) and addresses optical and physico-chemical limits—specifically étendue-limited resolution and transport phenomena—to advance volumetric AM resolution and fidelity.

Tomographic AM process: A cylindrical transparent resin container rotates while being illuminated from the side by a sequence of 2D light patterns synchronized to the rotation. Patterns are tomographic projections (Radon transform) of the target 3D object over 360°; cumulative dose across angles locally exceeds the gelation threshold to solidify the desired voxels. Resins are chosen to be transparent with high viscosity to mitigate sedimentation and diffusion.

Physico-chemical control: High-viscosity formulations (μ > 10 Pa·s) minimize sedimentation during ~20 s builds, and reduce diffusion of oxygen/radicals. With a diffusion coefficient for oxygen of 1.2×10^−13 m^2 s^−1 in a 10 Pa·s resin, diffusion blur over 20 s is <2 μm, deemed negligible.

Optical design and étendue: Resolution is determined by DLP pixel size and magnification at the center, and degrades off-axis due to beam divergence. To limit pixel overlap at the build volume edge, a low source étendue L_s NA_s is required (condition L_s NA_s = n ρ L_vox). The system couples six 405 nm laser diodes (total nominal 6.4 W) into a 70×70 μm square-core fiber (NA 0.22) to create a low-étendue, homogenized source. The fiber output is relayed and shaped with an asphere and orthogonal cylindrical lenses onto a DMD (Vialux V-7000), then imaged via a 4f system into a cylindrical vial. Unwanted DMD diffraction orders are spatially filtered in the Fourier plane. Addressable volume ~17.5×17.5×23 mm; delivered power to the vial ~1.6 W with all pixels ON.

Projection algorithm: 3D STL models are voxelized. For each 2D slice, projections over 360° (0.6° steps) are computed using the Radon transform and filtered (Ram-Lak) in the Fourier domain. Negative projection values, not physically projectable, are thresholded to zero, producing approximate reconstructions where dose still concentrates in the object. Gelation thresholding ensures only voxels exceeding dose solidify.

Feedback algorithm: A camera at 90° to the illumination axis records 2D transmission images during rotation, with separate back-illumination at 671 nm. Solidification appears darker due to refraction/scattering. Reference images (pre-exposure) are subtracted from in-exposure frames; thresholding detects solid regions and timestamps their appearance. Solidification times are back-projected into a 3D grid across angles, yielding a 3D map of time-to-solidify. Because dose D = I·t, regions requiring longer time are assigned proportionally higher intensity for the next print. Corrected Radon projections are then generated for subsequent builds.

Materials and characterization: Acrylic resin prepared from di-pentaerythritol pentaacrylate (SR399) with 0.6 mM phenylbis(2,4,6-trimethylbenzoyl)phosphine oxide; heated to 100 °C for 1 h, degassed by brief centrifugation. Silicone elastomer was also printed for soft models. Micro-CT scanning: SkyScan 1076 at 9 μm pixel for selected parts; EasyTom S for artery/hearing-aid with registration and dimensional error analysis in VGStudio/myVGL.

Optical performance measurement: Modulation transfer function (MTF) mapped at on-axis, mid-field, and edge of field, at focus and ±8 mm, across the 16×16×20 mm build volume. Theoretical optical cutoff 2NA/λ corresponds to ~422 cycles/mm (1.2 μm diffraction limit), but practical resolution is limited by source étendue and DLP spatial frequency capacity (v_DLP = 21.9 cycles/mm).

• High-resolution features: Achieved 80 μm positive features (e.g., suspended buttresses on Notre-Dame model printed in 19.5 s) and 500 μm negative features (e.g., 3DBenchy chimney aperture printed in 25 s) on centimeter-scale parts in <30 s.
• Étendue-limited resolution: Experimental MTF confirms that resolution is limited by illumination étendue rather than diffraction. Center of build volume reaches DLP limit v_DLP = 21.9 cycles/mm; at edge, MTF ~0.4 at 11 cycles/mm (≈27 μm optical resolution spread). Low-étendue source (70 μm square-core fiber, NA 0.22) enables near-uniform high spatial frequency support across the field (theoretical optical spreads: 23 μm center, 33 μm edge).
• Physico-chemical transport: With μ > 10 Pa·s resins, sedimentation over ~20 s is negligible; oxygen diffusion blur <2 μm over build time, thus not limiting resolution.
• Feedback improves fidelity: Sequential feedback using transmission imaging and 3D back-projected solidification times led to synchronized solidification across complex geometries. In a mouse pulmonary artery, feedback eliminated occlusions and preserved open branches; without feedback, central Y-branch lagged and other vessels clogged. For a hearing-aid shell, feedback corrected delayed bottom formation and restored openings and dimensions to match the model more closely; results unattainable by simple exposure reduction.
• Demonstrated material range: Both hard acrylic and soft silicone parts were fabricated via tomographic AM.
• Throughput/time: Centimeter-scale objects produced in 19–25 s exposures, demonstrating potential for ultrafast volumetric fabrication.

The study addresses the central challenge of improving resolution and geometric fidelity in tomographic volumetric AM beyond prior ~300 μm limits. By engineering a low-étendue illumination system, the process maintains high spatial frequency content across the build field, confirming that volumetric AM resolution is constrained by source étendue rather than diffraction. Concurrently, selecting high-viscosity, transparent resins minimizes sedimentation and diffusive blurring within the short build times, ensuring that optical dose shaping directly translates to feature formation. The integrated (sequential) feedback leverages the transparency of volumetric AM to monitor and correct spatial dose non-uniformities by mapping solidification times and adjusting projection intensities. This improves the simultaneity of voxel gelation, preserves hollow channels, and enhances fidelity in complex, hollow geometries. Collectively, these advances demonstrate that tomographic AM can rapidly produce centimeter-scale objects with sub-100 μm positive features and sub-millimeter negative features, expanding its applicability to precise, functional constructs such as anatomical models and potentially bioprinted tissues.

This work demonstrates high-resolution tomographic volumetric additive manufacturing by combining a low-étendue illumination system with an integrated feedback strategy to control photopolymerization kinetics. The approach yields 80 μm positive and 500 μm negative features on centimeter-scale parts within ~20–30 s and verifies that system étendue, not diffraction, limits optical resolution in this modality. High-viscosity, transparent resins mitigate sedimentation and diffusion effects over build times. Feedback based on transmission imaging and 3D back-projection of solidification times significantly improves geometric fidelity in complex, hollow structures. These results pave the way for ultrafast fabrication of precise, functional constructs, including biomedical models. Potential future directions include implementing real-time, on-the-fly pattern corrections with dedicated electronics/FPGA, improving projection/dose deposition algorithms to better capture sharp corners and minimize negative-value truncation artifacts, and broadening the range of printable materials and geometries.

• Dose deposition algorithm limitations: Sharp features (e.g., Notre-Dame tower corners) were not faithfully reproduced, attributed to current projection/dose algorithm constraints (e.g., Ram-Lak filtered projections with negative values thresholded to zero).
• Feedback implementation: On-the-fly pattern correction was not possible with the existing DMD module; feedback was applied sequentially across prints, requiring additional electronics/FPGA for real-time correction.
• Field-dependent MTF: While improved by low étendue, MTF decreases toward the build volume edges (e.g., ~0.4 at 11 cycles/mm), indicating residual field non-uniformity.
• Material/process scope: Results are demonstrated on specific acrylic and silicone formulations and a defined build volume; generalization to other chemistries/geometries may require further optimization.
• Unwanted background dose: Thresholding negative projection values to zero yields approximate reconstructions, potentially depositing sub-threshold dose outside the object, though typically not enough to cause unwanted solidification.