In short: Maskless lithography writes a pattern directly onto resist with a steered laser or electron beam guided by GDSII/OASIS layout data, trading the throughput of mask-based projection for the agility of needing no physical reticle.
Conventional optical lithography is a photographic process: a master pattern etched into a quartz-and-chrome reticle is projected, reduced 4×, onto a resist-coated wafer, and the whole field prints in one flash. That reticle is the bottleneck. A single leading-node EUV mask set can cost well north of a million US dollars and take weeks to fabricate. For prototypes, research wafers, photomask writing itself, or any run of a few dozen die, paying for a mask makes no economic sense. Maskless, or direct-write, lithography removes the reticle entirely and paints the pattern pixel by pixel.
Why eliminate the mask at all?
The motivation is almost entirely economic and iterative. A mask is a fixed, non-recurring cost that only amortises across high volume. If you are a university group taping out one MPW shuttle slot, a photomask shop writing the reticles that everyone else uses, or an R&D team spinning a new test chip every fortnight, the mask is pure overhead and a multi-week delay. Direct-write collapses that: you send layout data and the tool starts exposing. Change a metal layer in Cadence Virtuoso or Synopsys Custom Compiler, re-stream the GDSII, and the next wafer carries the new pattern with zero tooling lead time.
Key fact: Essentially every photomask in the industry is itself made by a maskless tool — an electron-beam or laser pattern generator writing directly into resist on the blank. Maskless writing underpins the mask-based ecosystem.
Two families: photons versus electrons
Direct-write splits into optical (laser) and charged-particle (electron, or occasionally ion) approaches, and the choice is driven by the resolution you need.
Laser direct imaging (LDI / DMD-based pattern generators) uses one or more modulated laser beams, often fanned out through a Digital Micromirror Device from Texas Instruments, to expose features down to a couple of microns. Heidelberg Instruments and Durham Magneto Optics build maskless aligners on this principle. The wavelength (often 355–405 nm) sets the diffraction limit, so LDI dominates in MEMS, printed-circuit and photomask prototyping rather than sub-100 nm logic.
Electron-beam lithography (EBL) replaces photons with a focused beam of electrons whose de Broglie wavelength is picometres, so the resolution limit is no longer diffraction but the electron scattering and the resist. Commercial Gaussian-beam and shaped-beam writers (Raith, JEOL, NuFlare) routinely resolve sub-10 nm features. The physics, the proximity effect, and the resist chemistry deserve their own treatment — see our deep dive on electron-beam lithography.
The datapath: turning GDSII into a dose map
A maskless tool is as much a data-processing machine as an optical one. The layout, exported as GDSII or the denser OASIS format from your P&R or custom-layout flow, must be fractured into primitive shapes the writer understands, then dose-corrected. Two corrections dominate:
- Proximity-effect correction (PEC). In EBL, electrons backscatter from the substrate and expose neighbouring resist, so dense regions receive unwanted extra dose. PEC solves the inverse problem, assigning each shape a dose so the developed pattern matches intent. Tools such as Synopsys Proteus and Siemens Calibre perform analogous optical-proximity correction for the mask-based world; EBL uses dedicated PEC engines like GenISys BEAMER.
- Fogging and loading correction for long-range scattering and etch effects.
The sheer volume is the challenge: a full-reticle pattern can be terabytes after fracturing, which is exactly why single-beam EBL is slow and why multi-beam architectures exist.
Throughput: the fundamental trade
Projection lithography prints an entire ~26×33 mm field in one exposure. A single Gaussian e-beam writes one pixel at a time. The arithmetic is brutal: writing a dense full wafer with one beam can take many hours to days. That is why maskless has historically lived in prototyping and mask-making, not volume production.
| Attribute | Mask-based projection | Maskless direct-write |
|---|---|---|
| Reticle cost | High, fixed (NRE) | None |
| Design iteration time | Weeks | Hours |
| Throughput | >100 wafers/hr | Low (single-beam) to moderate (multi-beam) |
| Best resolution | Sub-10 nm (EUV) | Sub-10 nm (EBL) |
| Ideal use | High-volume production | Prototypes, masks, MEMS, research |
Key fact: Multi-beam mask writers such as IMS Nanofabrication's MBMW platform fire on the order of a quarter-million electron beamlets in parallel, decoupling write time from pattern complexity — the breakthrough that made EUV mask writing practical.
Where it fits in a real flow
In practice, direct-write serves three steady roles. First, mask fabrication: the reticles for DUV and EUV are themselves written maskless. Second, rapid prototyping and MPW, where a design team validates a layout before committing to a mask set. Third, MEMS, photonics and nano-device research, where geometries are custom and volumes tiny. Teams building these flows often pair the write step with the same sign-off decks (Siemens Calibre, Synopsys IC Validator) used for production tape-outs, so the layout is DRC-clean before a single electron lands. If you need production-grade layout and physical-design services behind such a flow, specialist houses like Avecas design services bridge research prototypes to manufacturable GDSII.
For a broader comparison of patterning strategies, our nanoimprint versus optical lithography article maps where each technique wins.
Want to go from GDSII to a working wafer? CourseTron's Nano Fabrication track walks through lithography, resist chemistry and PEC with hands-on layout labs. Explore the courses →
Frequently asked questions
Is maskless lithography faster than using a mask?
For time-to-first-wafer, yes — there is no multi-week mask build. For wafers-per-hour on a mature design, no: projection prints a whole field per flash, while a beam writes serially, so volume production still favours masks.
Can direct-write reach the same resolution as EUV?
Electron-beam direct-write can resolve features below 10 nm, comparable to or finer than EUV, because electron wavelengths are far shorter than any photon used in lithography. The limiting factors become electron scattering, resist resolution and throughput rather than diffraction.
What layout format does a maskless tool consume?
The same hierarchical formats as mask-making: GDSII or OASIS exported from EDA tools such as Cadence Virtuoso or Synopsys Custom Compiler, then fractured and proximity-corrected into the writer's native shape and dose instructions.
Why is almost every photomask made with a maskless tool?
Because the mask itself is the first physical copy of a design — there is no earlier master to project from. A laser or electron-beam pattern generator writes the pattern directly into resist on the mask blank, making maskless writing the foundation of the entire mask-based ecosystem.