July 2025 · pre-alpha
The first control program
Windows Forms: a table of fields, each with its own position and exposure time — dose sweeps set up by hand.
DESKTOP · MASKLESS · DIRECT FROM GDSII
Desktop Maskless Lithography
Photolithography straight from a layout file. No photomasks, no mask aligner, no waiting — a design change is a file change.
Two machines, one line of work. Prototype 1.0 is built and working: it exposes real wafers today and runs software written in-house. Prototype 2.0 turns what it taught me into a modular sub-micron motion platform, with lithography as its first instrument.
Prototype 1.0 · a full exposure job, ×15
GDSII
native input · no masks
1920 × 1080
DMD micromirrors
440 nm
LED source · no laser
Full closed-loop
servo axes on optical scales
Built
Prototype 1.0
Built and working
The machine that proved the idea. Enclosed desktop unit: DMD projection head, 440 nm LED, closed-loop XY stage on servo drives and optical scales, a camera looking through the same objective. It exposes resist on real wafers.
How it works
A TI DLP4710 DMD — 1920 × 1080 micromirrors — is a programmable photomask. Each frame of the layout is projected through a microscope objective onto the resist; the stage steps and stitches frames into one continuous pattern.
Results
Made on this machine
Photographs, not simulations: exposed and developed on Prototype 1.0.
Precision from calibration
Sub-micron precision here does not come from granite and air bearings. The machine measures its own geometry and illumination with a camera in the wafer plane and cancels them on every field; the software refuses to expose with an uncalibrated objective.
One engineer
Mechanics, optics, electronics and software — designed, built and brought to working condition by one person.
In design
Prototype 2.0
Modular platform · CAD stage
A precision 3-axis motion platform first; lithography is one configuration of it. Three identical linear modules make the XYZ gantry, and everything above the stage bolts to a standard optical-breadboard grid.
Renders with dots are interactive: hover over or tap a part to see what it does.
Optical tableA standard breadboard with M6 threads on a 25 mm grid, so ordinary optomechanics mounts right next to the machine.
Gantry frameTwo columns on a base plate carry the gantry axis and bolt down to the optical table.
Stage axisMoves the substrate. The same linear module as above: servo drive, ball screw, crossed-roller guides and an optical scale.
Gantry axisThe second horizontal axis, on the gantry. Together with the stage axis it gives at least 100 × 100 mm of XY travel.
Z axisThe same module mounted vertically: 40 mm of travel for focusing, with the exposure head on its carriage.
Servo drivesOne industrial servo drive per axis, all on a single CANopen bus.
Chuck and objective rackA vacuum chuck on a rotary axis — in development — next to a rack for spare objectives.
Exposure headDMD projection optics with a microscope objective; the camera looks through the same path. The whole head unbolts and can be replaced by other optics.
Sub-micron linear module
Every axis is the same block: servo drive, ball screw, crossed-roller guides and an optical scale, closed-loop and complete in itself — 100 nm as standard, 2.5 nm as an option. A product line of its own.
Base plateMachined base plate that carries the guides and bolts to a table, a frame or another module.
CarriageThe moving table. A 25 mm grid of M6×1 threads takes fixtures, optics or another module directly.
Optical linear scaleA glass scale with a read head measures where the carriage actually is, and the drive closes its loop on that reading — 100 nm as standard, 2.5 nm as an option.
Servo driveAC servo motor on an industrial CANopen bus. It turns the ball screw through a coupling.
Base plateMachined base plate that carries the guides and bolts to a table, a frame or another module.
Crossed-roller guidesTwo pairs of crossed-roller rails carry the carriage: stiff, smooth and low in friction.
Ball screw and nutTurns the motor’s rotation into straight travel. The nut sits in a bracket under the carriage; the screw end runs in a fixed bearing support.
Optical linear scaleA glass scale with a read head measures where the carriage actually is, and the drive closes its loop on that reading — 100 nm as standard, 2.5 nm as an option.
Servo driveAC servo motor on an industrial CANopen bus. It turns the ball screw through a coupling.
Details
Substrate end: chuck and objectives
What changes against Prototype 1.0: an open gantry instead of a box, and every part of the optical path made swappable.
Stage axisMoves the substrate. The same linear module as above: servo drive, ball screw, crossed-roller guides and an optical scale.
Rotary base and objective magazineA compact motor under the chuck will turn the substrate for alignment — in development. The same housing holds the spare objectives.
Vacuum chuckHolds the substrate flat during exposure: silicon up to 4 inches, glass or PCB.
5× objectiveA wide field for finding the substrate and navigating over it.
10× objective10× / 0.5 NA: a larger exposure field per frame.
20× objective20× / 0.75 NA: the finest features, with a smaller field per frame.
Datasheet
Target specifications
Prototype 2.0. The optical core and software are carried over from Prototype 1.0.
- Feature size
- 0.7–1 µm typicalmin 0.4 µm, optics & resist dependent
- Exposure area
- 100 × 100 mmscalable to 200 × 200 mm
- Light source
- LED, 440 nm405 nm option · no laser
- XY repeatability
- 0.2 µmclosed loop on optical scales
- Position feedback
- 100 nm scalesdown to 2.5 nm option
- Throughput
- ~2.8 mm²/min @ 1 µm10×; ~0.9 mm²/min @ 0.5 µm (20×)
- Light modulator
- TI DLP4710 DMD1920 × 1080 micromirrors
- Objectives
- 10× / 0.5 · 20× / 0.75interchangeable
- Substrates
- Si up to 4″ · glass · PCB
- Travel X / Y / Z
- ≥100 / ≥100 / 40 mm
- Design input
- GDSII nativeDXF — tuning
- Mounting
- 25 mm grid, M6×1optical-breadboard standard
In development
- Layer-to-layer alignmentrotary vacuum chuck + alignment workflow in software
- Automatic objective changemotorized swap between 10× and 20×
- 405 nm LED optionfor sub-0.5 µm work and broader resist compatibility
- 2.5 nm optical scalesfor repeatability in the 25–50 nm class
Today it is a single-layer tool: expose, develop, done. If your process needs aligned multilayer now, ask me about the timeline.
GDSII layoutIn resist
5 mm
Layout → resist
Design against result
Drag the divider. On the left is the GDSII file that went to the machine; on the right, a photograph of what came out, corrected for perspective so the two line up.
The silicon chip is 25 × 25 mm; the layout on it, a logo with captions and four dithered pictures, takes 20 × 20 mm. One exposure job.
The machine exposes one DMD frame at a time: this chip is 1 869 fields of 0.25 × 0.38 mm, placed side by side by the stage. Switch the picture to Exposure fields and back, and look for the seams.
What ends up in resist depends on development as much as on exposure: developer, time and temperature all change the result. These are samples from process tuning. Works by Magritte and Escher are blurred on both sides: they are still under copyright.
Applications
Where it fits
One base, different instruments
Maskless lithographyDMD head, vacuum chuck, exposure software
Inspection microscopeimaging optics on the same stage
Fluorescence imagingfilters, dichroic, excitation source
Your instrumentprobe stations, laser processing, AOI
Electronics & PCB
Fine-line boards and flex circuits, straight from the layout — same-day iteration.
Microfluidics
Channel networks, master molds for PDMS, on-chip electrodes.
MEMS & sensors
Resist patterns for etching, lift-off and plating on Si up to 4″.
Photonics
Gratings, diffractive elements, waveguide test structures.
Battery R&D
On-chip test cells, microelectrode platforms, microbatteries.
Education & labs
No laser class, compact, enclosed — real workflows from GDSII to wafer.
Software
Written in-house, runs the job from layout to wafer
No third-party CAM chain, no licence dependencies. MVP level: everything below runs on the tool today.
What it does
- Native GDSII import, layers, automatic tiling
- Die arrays / step-and-repeat
- OPC assist and focus-exposure matrix
- Two-phase contrast autofocus under red light
- Resist-safe illumination: only blue exposes
- Per-tile exposure time and dose-sweep wafers
- Substrate height map, per-field Z
- CANopen servo motion, exposure interlocks
- Scientific camera pipeline, scale bar per objective
- Project files with stale-tile detection
Calibration · both prototypes
The machine measures itself
A camera goes on the stage with its sensor facing up, in the plane where the wafer would be. The projector then draws straight onto the sensor, so the software sees exactly what the resist would get — and measures geometry, dose and stitching from it. The same procedure carries over to Prototype 2.0.
The payoff: tolerances move from hardware into software. Parts, rails and optics do not have to be perfect, and the machine does not have to be assembled perfectly — whatever is left is measured and cancelled. That keeps the parts cheaper and the build simpler.
BeforeAfter
500 µm
Before → after
What calibration changes
Drag the divider. Measured sides are camera frames taken in the wafer plane (one sensor pixel is 3.75 µm on the substrate). The compensated side of the white field illustrates the correction algorithm on that same frame.
One square frame exposed at 25 stage positions and stacked into one image. Before: projector rotation against the stage and the rails’ shear show as tilted squares with uneven gaps. After: every move goes through the measured 2 × 2 correction.
Geometry: one 2 × 2 matrix
Projector rotation against the stage, the shear between the X and Y rails and the scale of each axis are measured together and applied to every move. Sub-micron placement at the edge of a 100 mm wafer needs the angle to about 4 arcseconds.
Dose: a per-pixel mask
The camera maps illumination across the field; the software turns it into an attenuation mask on the DMD and adjusts exposure time per tile. The camera then measures the result again, and the mask is refined until the field is flat to within 2–3 %.
Final result · closed loop
Illumination flat to within 2–3 %
| Objective | Before | After | Darkest vs brightest |
|---|---|---|---|
| 10× | 48.7 % | 97.8 % | −2.2 % |
| 20× | 63.0 % | 97.9 % | −2.1 % |
| 40× | 70.0 % | 97.2 % | −2.8 % |
Uniformity here is the dimmest 2 % of the field — or the darkest corner, whichever is worse — over the brightest 2 %, measured through the objective and averaged over 256 frames. Each objective gets its own mask, refined pass by pass until a pass stops improving it. July–August 2026.
Progress
How it got here
A year of iterations. The pre-alpha ran on a different projector with a first program written in Windows Forms; Prototype 1.0 got its own stage, its own software and a stitched exposure process.
Microscope, ×2
July 2025 · pre-alpha
First developed well arrays
Test arrays in resist, each exposed in a single projected field with no stitching, labelled with its own exposure settings.
Stage rework
Into the milling shop
The XY stage on a milling machine: the rotary platform fitted and the mounting plane machined level.
2026
Prototype 1.0
Its own stage on servo drives and optical scales, the QtLithoCAN software, autofocus and a height map, jobs stitched across whole wafers — everything above on this page.
A test chip made of 1 869 exposure fields shows no visible seams.
Partnership
Looking for a manufacturing partner and investor
On the table: a working Prototype 1.0, the Prototype 2.0 design, full documentation and in-house software. A partner brings production, supply chain and market access; deal structure and investment size are open. Demonstrations and sample exposures by arrangement.