In the weeks after Carbon3D's announcement, several other projects made claims about high speed (>100mm/hr) stereolithography systems, some of them using top-down DLP. Given the magnitude of some of these claims (up to 500mm/hr), if true, it would appear that others had made substantial advances in the stereolithography process that would be worthy of investigation.

Ember team was curious about these claims, so we built a top down DLP SLA 3D printer (Fusion 360 and STEP files attached) and attempted to reproduce the results depicted here.

Step 1: Methods and Experiments

Our initial experiments using fully manual control of the hardware and an array of circles for an exposure pattern yielded surprisingly promising results (see above). By using Ember's Z-stage to lower a build platform into a pool of resin 100um at a time (with variable pauses of a second or less between motions due to manual control), we were able to build a somewhat solid geometry at a build rate of approximately 120-150mm/hr.

After these initial results, Cappie Pomeroy (cappiep) developed a workflow and a python script (attached) for automating the control of the testing hardware. With this hardware and control platform, we investigated the following process parameters and their effects on the following resins:

Resins:

PR48 (Autodesk standard prototyping resin, viscosity of 183 mPa·s)
Spot A Materials' Spot-GP Resin (63 mPa·s)
Fun To Do Industrial Blend Red Resin (viscosity not measured, but between that of PR48 and Spot-GP)
Homemade low viscosity resin formulated by Ember team polymer chemist Brian Adzima (viscosity not measured, but near 60 mPa·s)

Process parameters:

Continuous build speeds of between 100mm/hr and 500mm/hr
Discrete, layered build speeds of between 100mm/hr and 350mm/hr
Continuous exposures
Discrete, shuttered exposures
Liquid-liquid incident light interfaces
Sonication of the resin
Oscillatory resin vibration
Dipping cycles
Inclined build platforms
Mesh build platforms

Among the geometries we tested were a series of thin walls measuring between 0.5mm and 3.0mm thick (attached as "thinFeatureProbe.stl"), an array of tubes, and a skull geometry close to that used in the Gizmo3D video (shelled to 2mm, attached as "FTDskull.stl").

Step 2: Results and Conclusions

While we were able to replicate speeds close to those claimed in the Gizmo3D video, we were unable to do so with any meaningful object coherence or build quality. We were, however, able to gain a better understanding of the problem-space for top-down DLP SLA in its continuous and discrete implementations.

Effects that disrupt print quality and object continuity:

Rapid curing of resin inflow
(See illustration of process above) During continuous exposure and Z-direction motion, as the platform descends, new resin begins to flow into the area of exposure, but is immediately cured before it can flow past the boundary of the exposure. This process (illustrated above) continues until the pressure of the resin overcomes the thin, cured sidewalls and the resin rapidly flows into the cavity defined by the exposure. The top surface immediately cures, the sidewalls begin to form again, and the process repeats. This results in weakly connected "layers" and an overall very porous object.
Lateral migration of cured material
If curing is disrupted in the Z-direction, creating a gap of uncured resin in an object-in-progress, the surface material that cures occasionally begins to drift laterally, first likely due to fluid flow. As the amount of cured material continues to grow, its proportions and increased density cause it to sink, further driving the lateral motion and ultimately leading to the formation of a scroll-like film of cured material, the "tail" of which sometimes breaks the surface of the resin as pictured above.
Viscous flow effects
(See illustration of solid vs. mesh build platform above) Resins with higher viscosities required more time to flow laterally over the build platform to supply material for the first and all subsequent layers, both in continuous and discrete operation.
Surface tension effects
(See images above) The flow and distribution of resin depends on some degree on the resin's surface tension. We observed process disruption during early stages due to a combination of surface tension and viscous flow.

Strategies that improve process efficacy:

Porous build platform
(See illustration of solid vs. mesh build platform above) A porous build platform allows even resin inflow at the beginning of a print, leading to more successful first and early layers.
Inclined build platform
A build surface that is non plane-parallel combats, to some degree, the viscous flow problems for the first layers, but results in significant geometry deviations for early layers unless a rafting technique is employed.
Low viscosity resins
Low viscosity resins were able to more quickly replenish new "layers" in the continuous top down SLA process better than more viscous resins, but the above factors still proved too disruptive for viable printing at acceptable quality standards.
"Dipping" for resin replenishment
Discretely moving the cured resin well below the surface of the liquid resin while shuttering the exposure is an effective and demonstrated method of implementing the top down DLP SLA process, but is obviously not a continuous process. Dipping avoids a mechanical wiper for resin redistribution, but also potentially results in inconsistent layer heights.

Strategies that seemed to have little effect, but may merit further investigation:

Fluid-fluid light interface for viscous flow control
We attempted exposing through a low density, optically clear layer of solvent (pain thinner) floating on top of the resin in an attempt to lessen the differences in viscosity between the the two fluids composing the light interface in traditional top down SLA: a gas and the resin. A series of quick experiments yielded messy, but cured results that did not appear to be particularly more cohesive than air-resin interface prints.
Sonication of the resin
Using ultrasound to agitate the resin had no measurable effect on print quality.
Mechanical vibration of the resin
Agitating the resin with an oscillatory vibrating motor may be worth further investigation. We observed standing waves and minor increases in print quality that we suspect to be due to improved resin inflow to curing areas.