Monday, October 19, 2026

3DP Daily

Today's digest has two papers on turning two-photon-printed polymer into carbon, an open-source volumetric printer from Berkeley, and two older papers worth reading: one on holographic volumetric printing and one on non-planar slicing.

Deep Pick

Today's one thing to read

Two-photon-printed resist, baked into carbon, becomes a reusable grayscale photomask

makes smooth 3D surface reliefs, such as microlenses, parts and templates for mass replication, in a single exposure. To do that it needs a mask whose transmission varies continuously from point to point, and such masks are expensive, complex to make and limited in resolution. This paper writes the grayscale pattern with a standard system instead. The printer writes a relief, where height encodes the desired gray level, and the relief is then (heated without oxygen) until it turns into . In the furnace the pattern collapses from micrometers thick to nanometers thick, so the original height map becomes a thickness map of a thin carbon film. Thicker carbon blocks more light, which gives a continuous-tone mask. The film stays stuck to the substrate, so lateral fidelity is preserved, but the pattern grows slightly in-plane in a systematic way that has to be compensated in the design. Because amorphous carbon is chemically inert and mechanically tough, the mask can be exposed many times and cleaned in between, which turns a slow, single-use direct-write print into a tool for fast, repeated . The abstract gives no numbers for the thickness-to-transmission curve, minimum feature size or how many exposures a mask survives, and the full text was not accessible here, so how well it compares with commercial grayscale masks is still open.

Why it matters. It treats a two-photon printer as a mask writer for a parallel process rather than as the final fabrication step, and getting the target transmission map means predicting how the print shrinks and expands in the furnace, which is an inverse-design problem.

Advanced Science · 2026-10-18

Read on doi.org ›

New research

Local Joule heating carbonizes two-photon-printed electrodes without damaging CMOS chips

Polymer nanostructures printed by can be turned into carbon electrodes by , but the usual furnace step at about 950 °C destroys the sensor circuits underneath. The authors instead pass current through the structures so they heat themselves (), concentrating the high temperature in the electrode while the chip as a whole stays cooler. They used analytical models and simulations to choose process conditions, then tested on samples on silica wafers using . Heating the whole sample to 500-600 °C gave only partial carbonization, and adding localized Joule heating completed it, as confirmed by and . This shows the conversion can be done within a temperature window that CMOS chips tolerate.

Why it matters. It uses the same print-then-pyrolyze route as today's deep pick and addresses its main limit when the substrate is an electronic chip: the substrate cannot take the furnace temperature.

arXiv: additive manufacturing · 2026-10-11

Read on arxiv.org ›

Laser powder bed fusion prints braided nitinol stent-like structures in one step

is a nickel-titanium alloy that can deform a lot and spring back, which is why it is used in stents and other devices delivered through a catheter. Those devices are usually made from braided wire, which takes many steps, while 3D-printed nitinol lattices tend to be stiff because their struts are fused together at the joints. The IMDEA and UPM team, with MIT and the University of Galway, designed braided patterns algorithmically so that the fibers pass over each other without joints, then printed them by and finished them by and . The finishing reduced fiber thickness by 47.7% and cross-sectional area by 71.1%. The braids survived 50% compression with more than 90% elastic recovery, and their (the steady push a stent exerts on a vessel wall) was up to five times lower than that of conventional printed lattices. The work is published in Virtual and Physical Prototyping, and this is a later write-up of research first reported in early 2026.

Why it matters. The gain comes from algorithmically designing a joint-free topology rather than from a new material, which is a computational design idea that could apply to other printed metal lattices.

VoxelMatters · 2026-10-19

Read on voxelmatters.com › Original: 3dprintingindustry.com ›

Three neural implicit fields produce topology-optimized parts that are fast to mill

produces organic, doubly curved shapes that are easy to 3D print but slow to machine, because a ball-end tool has to trace the curved surfaces point by point. (ones with zero ) can instead be cut in long passes with the side of the tool (), but requiring zero curvature everywhere oversimplifies the design. The authors represent the part as the product of three implicit neural fields. This gives mostly developable patches, with the curvature concentrated along seamlines where the fields intersect, and they optimize topology and curvature together. Extra constraints align surfaces with the cutter for 3-axis machines. Compared with earlier machinability-aware methods, the designs consistently take less time to machine and give competitive structural performance.

Why it matters. The process here is subtractive, but representing a part as a product of several neural fields so that manufacturing constraints are built into the representation could carry over to printing constraints such as overhangs or layers.

arXiv: additive manufacturing · 2026-10-11

Read on arxiv.org ›

Slicers & toolpaths

Nothing worth your time here today.

Industry

Berkeley releases OpenCAL, a roughly $1,638 build-it-yourself volumetric printer

, first published in Science in 2019, prints a whole object at once by projecting a sequence of computed images into a rotating vial of resin. The light dose builds up until it crosses the only where the part should be. OpenCAL packages this as an open kit. Parts cost about $1,638 (June 2026 pricing), one person can assemble it in five to six hours, and the hardware is released under the Prusa Open Community License. The software includes Tomo, a free Windows application, and VAMToolbox, an open-source Python library for computing the projections. A dedicated OpenCAL Resin is made by Formlabs and sold in the US through Prusa's channel (Printed Solid) for $119.99 per liter. Prints typically take one to two minutes; a copy of Rodin's Thinker formed in 90 seconds. The article does not give resolution or build volume. The project was led by Taylor Waddell with more than 40 contributors, mostly Berkeley students.

Why it matters. has mostly required a custom lab rig, so a cheap kit with an off-the-shelf resin and an open projection toolbox lets a CS-side researcher test their own or dose-optimization algorithms on real prints.

VoxelMatters · 2026-10-19

Read on voxelmatters.com ›

Rowan University runs image correlation backward in time to measure printed-concrete deformation

measures how a surface deforms by comparing camera images with an undeformed reference image. In 3D concrete printing there is no such reference, because the object does not exist until it is printed and each new layer loads the layers below while they are still curing. The Rowan team adapted DIC to work in reverse time, which lets them measure how lower layers settle and deform as new material is deposited and as the concrete cures. The article summary gives no accuracy figures or scale.

Why it matters. Any layer-by-layer process has the same problem of a reference that keeps changing, so the reverse-time trick may apply to in-situ monitoring of other printing methods too.

VoxelMatters · 2026-10-19

Read on voxelmatters.com ›

Worth reading from the past year

DISH prints millimeter-scale objects in 0.6 seconds with rotating holographic projections

Published in Nature in February 2026 and already cited 18 times. usually trades resolution against speed, and methods like CAL rotate the resin vial, which limits speed and rules out flowing resin. DISH keeps the sample still and spins a high-speed so that light enters from many angles. A is computed for each angle by iterative optimization so that the overlapping patterns add up to the target 3D dose. It keeps 19 µm resolution over a 1 cm depth, far beyond the objective's , and prints millimeter-scale objects in 0.6 s in resins across a range of . Adding a fluid channel allowed continuous mass production of varied parts.

Why it matters. Read next to today's OpenCAL release, it shows where goes when computation replaces sample rotation, and the core of the method is a multi-view hologram optimization problem.

Nature · 2026-02-11

Read on doi.org ›

AtomSlicer: non-planar FFF layers of near-constant thickness, with continuous toolpaths

Published in ACM Transactions on Graphics in July 2026. printers can tilt the nozzle to avoid supports and , but the curved layers they need must follow a desired orientation field, keep the thickness constant, and avoid collisions and frequent stop-and-restart travel. AtomSlicer encodes the part with three orthogonal sampled into oriented 'atoms', groups them into printable layers of near-constant thickness, and draws toolpaths aligned to a per-layer direction field. The input fields are treated as hard constraints: the method either returns a feasible toolpath or reports that the part cannot be fabricated, rather than quietly changing the specification. Across a large study and 13 printed models, non-extruding travel dropped to a few percent, travel moves fell by one to two orders of magnitude compared with planar slicing and the earlier Atomizer method, and it ran 9-60 times faster than Atomizer.

Why it matters. It shows a clear separation between choosing the design fields and realizing them, so other optimizers can produce the fields and AtomSlicer can check and print them.

ACM Transactions on Graphics · 2026-07-03

Read on doi.org ›

Also today

37 more stories worth checking

55 off-topic items skipped.

That's it for today. See you tomorrow.