Make your paper interactiveCreate

Text from the interactive explanation

Segmented projection image translation for large-area, high-resolution vat photopolymerization

Jize Dai, Jason M. Kronenfeld, Dan Ilyn, Christopher Linhardt, Philip R. Onffroy, Ian A. Coates, Colin Crown, Michelle Chen, Jacob Dobson, Max A. Saccone, John R. Tumbleston, Maria T. Dulay, Joseph M. DeSimone

Original paper

Science Advances · 2026 · Dai, Kronenfeld, Ilyn et al.

Picture a lattice with struts thinner than a hair, yet wide enough to hold between two fingers. The sharpest light-based 3D printers have had to choose between fine detail and size. A Stanford team found a way out: put one small lens on a turntable.

Simulated top view, drawn to scale from the paper’s geometry: a 2560 × 1600-pixel image of 3.86 µm pixels, a 4.8 mm lens offset and 16 stops 22.5° apart. The amber pattern is a 15 mm octet-lattice disk like the one the team printed. Slowed down.

A fixed budget of four million mirrors

These printers work over a vat of liquid resin. They project a picture of one thin slice of the object into it, and wherever light lands, the resin hardens. Slice by slice, the part grows. This is called vat photopolymerization .

The picture comes from a digital micromirror device (DMD): a chip with 2560 × 1600 tiny mirrors. Each one tilts to send its speck of ultraviolet light toward the resin or away from it, so each mirror is one pixel. Lenses decide how big each pixel lands on the resin. But however they are set up, the mirror count never changes.

Halve the size of every projected pixel. The printable area becomes…

Choose a pixel size

Small pixels draw smooth struts but cover a field smaller than a fingernail. Large pixels cover the disk but make every edge coarser. To cover the expanded 290.5 mm² area at 3.86 µm per pixel, a chip would need about 19.5 million mirrors , nearly five times what it has. Designers have been stuck choosing between resolution and area. That is the trade-off this paper attacks.

Go bigger by moving something

A printer can cover more ground by exposing neighbouring patches one after another. The question is what has to move, and what that costs.

Schematic animations of the approaches described in the paper’s introduction.

The secret is in the parallel light

Between the mirror chip and the resin sit two lenses. The tube lens turns the light from each mirror into a parallel bundle, and each mirror’s bundle travels at its own slight angle. Opticians call this stretch infinity space . The small objective lens then focuses each bundle back down to a sharp point on the resin.

Slide the objective lens 5 mm sideways. The picture on the resin will…

Slide the objective

Drag the lens in the picture, or use the slider. Each coloured beam comes from a different spot on the chip.

Architecture photographers use the same principle in tilt-shift lenses , sliding the lens to reframe a building without distorting it. There is a limit, though. The tube lens sends out a beam about 22 mm wide, and the objective accepts only 12 mm. That leaves roughly 5 mm of sliding before the beam starts to miss the lens. The team used an average offset of 4.8 mm . Push further and part of the light is clipped. One side of the image then grows dimmer, an effect called vignetting , which the brightness curve shows.

In the paper’s optical simulation, a focused spot measured 4.104 µm across with the lens centred and 4.102 µm with it shifted 5 mm: practically identical. It just becomes slightly oval along the shift.

Schematic thin-lens model, not to scale. Brightness follows how much of each beam the objective catches; spot sizes are quoted from the paper.

Put the lens on a turntable

Now mount the shifted lens on a gear and turn it around the optical axis. In the team’s printer a stepper motor, run by an Arduino board, turns a 3D-printed gear. The image follows the lens around a circle and always faces the same way. At each stop the projector flashes one segment of the layer. Then it projects a dark frame while the lens moves on. Choose how many stops to make, and how far off-centre the lens sits.

Challenge: push the offset until a hole opens in the middle. Where does it happen?

An image that orbits without turning sweeps a rounded rectangle: the image grown by the offset d on every side. Its area is the image area A (61.0 mm²), plus its perimeter P (32.1 mm) times d, plus a disk of radius d:

With a finite number of stops, small gaps remain near the rounded corners. The paper reports 75%, 91%, 96% and 98% coverage for 4, 8, 16 and 24 stops at 4.8 mm, and the stop buttons above reproduce those numbers. The model uses the nominal image (2560 × 1600 pixels of 3.86 µm, 61.0 mm²) and gives 287.6 mm² at 4.8 mm. The paper reports 60.4 → 290.5 mm² (4.8×) from its measured geometry.

Evening out the light

Shifting the lens has a side effect. The light is no longer brightest in the middle of each image. In the team’s measurements the brightest region sits about two-thirds of the way across. Uneven light means uneven curing: lines come out too thin in dim corners and too fat in bright ones.

The fix is a correction mask . The team photographed a plain white projection with an ultraviolet camera, then computed a grayscale mask that dims the bright areas to match the dimmest. The printer applies it to every image it projects. They made 32 masks, one per stop position, across two sets of stops 22.5° and 30° apart.

Apply the mask

(1) scales the camera image Ic so its brightest spot reads 255. (2) gives each pixel a gray level that brings it down to the dimmest value, Imin.

The price is subtlety. Where the mask dims a lot, fewer of the projector’s 256 gray levels remain, but exposure time, intensity and the resin’s response can compensate. The team also corrected barrel distortion , a slight bulging of the image.

Illustrative brightness map shaped like the paper’s measurement; the curing response of the test lines is schematic.

Slicing for a printer that orbits

An ordinary slicer cuts a 3D model into layers. The team’s slicer also cuts each layer into the segments every stop will project, and decides their order. The paper describes four ways to sequence them. Pick one and watch it print.

Which mode will finish the tapering tower first?

Race the rotary modes

Illustrative simulation of the sequencing logic, using 16 stops at 4.8 mm. Exposure, orbit and stage times are round numbers chosen for illustration, not the paper’s. The paper picks dynamic-mode stops with an integer linear programming optimizer; this simulation uses a simpler greedy search.

What came out of the printer

All of this was built and tested. Everything in this chapter is measured.

How thin can it print?

A 2-pixel line is designed to be 7.7 µm wide. How wide does it come out?

Selected

Line patterns printed in a single segment and measured by scanning electron microscopy. In camera images, 2-pixel lines stayed distinguishable at every tested lens angle from 0° to 315°.

alignment error where neighbouring segments overlap (camera and electron-microscope checks).

repeatability of the lens’s optical homing: misalignment stayed within one camera pixel.

in tensile strength or stiffness across stitched boundaries: no statistically significant change.

sharpness (MTF50) across one field, about 31–15 µm features; 40–100 lp/mm at MTF10 (12.5–5 µm). It is weakest near 135°–225°, which the authors attribute to minor misalignment.

Measured results reported in the paper.

Big, and still microscopic

We began by asking whether a printer could stay microscopic and still print big. SPIN’s answer takes remarkably little hardware: an off-the-shelf objective lens, a 3D-printed gear, a stepper motor and an Arduino. With them the team covered 4.8 times more area without giving up its 3.86 µm pixels .

Their printer also uses CLIP, an oxygen-permeable window that keeps the part from sticking. The authors stress that CLIP is not required, so the idea should carry over to other projection printers.

Proposed next steps, not yet demonstrated. Spin the lens continuously in step with the projector’s 120 frames per second. That gives 120 segments, 3° apart, in a one-second sweep (0.75° apart in four seconds). The authors also propose continuous and dynamic spirals, and moving the lens outward as well as around, for up to 14.7× the native area.

Fine-featured parts over larger areas for next-generation microelectronics cooling systems.

High-surface-area battery architectures. Printed lattices can be baked into carbon (pyrolysis), shrinking features to 10 µm or less.

Networks of tiny channels to supply blood to engineered organs.

Applications named by the authors as possibilities this approach could serve.

Read the paper →

Jize Dai, Jason M. Kronenfeld, Dan Ilyn, Christopher Linhardt, Philip R. Onffroy, Ian A. Coates, Colin Crown, Michelle Chen, Jacob Dobson, Max A. Saccone, John R. Tumbleston, Maria T. Dulay, Joseph M. DeSimone. “Segmented projection image translation for large-area, high-resolution vat photopolymerization.” Science Advances 12, eaef7779 (2026), published 11 November 2026. doi:10.1126/sciadv.aef7779.

The article is distributed under a Creative Commons Attribution 4.0 (CC BY) license. Photographs and figure panels are from Dai et al., Figs. 1C–E, 2C and 5, cropped from the published figures. Data and code: Dryad, doi:10.5061/dryad.fn2z34v90 (CC0).

The interactive scenes on this page are illustrations built from the paper’s reported geometry, equations and descriptions. The build-area and coverage calculations follow the paper’s geometry, while timings, brightness maps and curing responses are schematic. Measured results are labelled as such.

Topics in this iPaper

Engineering
we demonstrate a projection-based VP printer implemented with SPIN that expands the effective build area by 4.8-fold

Title page header and Abstract, page 1

Physics
The projection optics of this system integrate the principles of tilt-shift lens design into conventional projection-based VP

Results, Optomechanical simulation and characterization, page 2

vat photopolymerization
High-resolution projection-based vat photopolymerization (VP) 3D printing enables rapid fabrication of highly detailed polymeric structures.

Abstract, page 1

additive manufacturing
Additive manufacturing (AM) has redefined how engineering devices are fabricated

Introduction, page 1

continuous liquid interface production
The system additionally incorporates CLIP by replacing the standard vat window with a tensioned oxygen-permeable Teflon film

Results, Implementation and mechanical design of SPIN, page 2

digital micromirror device
a fundamental trade-off between projected pixel size and printable area due to fixed digital micromirror device (DMD) pixel counts

Abstract, page 1

modulation transfer function
the MTF was experimentally characterized using a slanted-edge MTF method

Discussion, Alignment and resolution characteristics, page 9

Segmented projection image translation for large-area, high-resolution vat photopolymerization — Interactive explanation | iPaper.ai