Making a microparticle the standard way requires three steps: coat a substrate with photoresist, pattern the resist with light, then dissolve the substrate to release the particles. The third step is the bottleneck. Sacrificial-layer release is slow, damages fragile geometries, and never achieves 100% yield because some particles stick, break, or get lost in the dissolution bath.
The fix: don't use a substrate at all. Dry-film epoxy resist — a self-supporting sheet — can be patterned directly by lithography and then peeled apart to release free-standing particles without any sacrificial dissolution. The yield approaches 100% because there's nothing to dissolve, nothing to stick to, nothing to damage.
Note (2026-11-13; read against the paper's full text): the release step is not a peeling. In the paper's workflow the dry-film epoxy sheet (SUEX) is exposed through a photomask with its protective PET film on, baked in an oven at 80 degrees Celsius for 10 minutes, and then, after the PET films are removed from both sides, the exposed resist is placed in a centrifuge tube, covered with the developer mr-DEV 600 and sonicated for 5 minutes; the unexposed resist is dissolved by the developer and the patterned particles are left free. What is eliminated is the substrate and the sacrificial release layer, not dissolution as such, so the essay's 'nothing to dissolve' is not the paper's. The paper also says that handling the sheets to remove the PET films is relatively easy for thicknesses of 50 micrometres and up, and that for thinner sheets one side of the PET film can be left in place. The yield wording in the paper is 'near 100 percent' in the abstract and '100 percent recovery' in its conclusion, and it says conventional substrate-and-sacrificial-layer processes 'often' reduce yields; the essay's 'never achieves 100 percent' and the list of ways particles are lost are not in the text I searched. On design, the paper generates designs with the Nazca Python library, exports them in GDS II format and makes a photomask in a separate step (using a maskless lithography system to write a chrome mask), and says that removing the mask fabrication step could speed up making batches of unique particles; the 'single lithographic exposure' is an exposure through that mask. The closing two paragraphs about two bottlenecks being removed together are the essay's.
The second insight is about design space. Once the fabrication process doesn't require manual substrate preparation, the bottleneck shifts to design — how do you specify tens of thousands of different particle geometries? The answer is programmatic generation: a Python library (Nazca) parametrically defines particle shapes, producing mask files that describe entire libraries of custom particles in a single lithographic exposure.
The combination is more than the sum of its parts. Substrate-free fabrication means arbitrary geometries can be realized without worrying about release failure. Programmatic design means the entire parameter space can be explored systematically. The fabrication constraint and the design constraint were both blocking the same thing — large-scale libraries of custom particles — and removing both simultaneously opens the space that removing either alone would not.
This is a manufacturing insight with a general form: when two independent bottlenecks each limit production, fixing one alone produces no visible improvement. The production capacity is set by whichever bottleneck is tighter. Only fixing both simultaneously reveals the capacity that was always possible.