XY Lab.
Mask projection lithography
on a printed stage
- Resolved
- <20nm
- Repeatable
- 210nm
- Noise floor
- 0.6nm
- Travel
- 3×3mm
- Materials
- ₹1,500
See the measurements Build it or buy it Buy one built · ₹2,500
Same idea. Stiffer frame.
The second body keeps the one-piece print and gives the drive something solid to push against.
XY One proved that a printed flexure can move in nanometres. This body is what happened next: a cross-shaped base that bolts flat to a table, a closed frame around the slide holder, and printed rods running either side of each lead screw so the drive block cannot yaw as it travels.
It still comes off the bed in one piece. You still bolt on two motors, and it still moves.
- Print the body: one piece, flexures included
- Drop an M3 nut into each housing
- Lead screw into the coupler cap, washer on
- Thread the screw into the captured nut
- Push the coupler cup onto the motor shaft
- Slide the motor on, two M3 screws each
Fig. 5 — The body prints as a single piece, then the two drives go on in the order of the assembly guide, both axes at once. The body turns translucent while parts go inside it. About thirty minutes on the bench.
It bends instead of sliding.
The slide holder hangs on folded flexure beams and nothing else. Each drive block is tied to the frame by thin blades that bend, so it travels in a straight line along its own lead screw and resists every other direction. No friction to overcome, no backlash to unwind, nothing to shim into alignment by hand.
That is the trade the stage makes. Give up long travel, and precision stops being a machining problem. It becomes a printing problem — and printing is cheap.
Detail B — The top plate is cut away to show what is underneath. Each lead screw pushes in turn, first X, then Y, then both: the yellow blades bend, the light platform shifts against its datum outline, and the dark frame does not move. Deflection is exaggerated by roughly four thousand times.
Four bolts to an optical table.
The base is a flat cross with a clearance hole at each arm, so the stage drops straight onto a standard breadboard on the 25 mm grid. Nothing is kinematic, nothing is bonded: bolt it down, and the frame borrows the table's mass.
Above it sits whatever writes. Here that is a mask and a projection lens pointed straight down at a post standing on the slide holder, which carries the sample clear of the frame. The beam never moves. The stage does all of the work.
Fig. 6 — The stage bolted to a 254 × 218 mm breadboard, with the sample post on the slide holder. Board and post follow the setup CAD; the stage itself is the source geometry.
The beam stays still. The wafer moves.
A 5 × 5 mm diced wafer sits on a post on the slide holder, lifted clear of the frame. The optics are fixed overhead. Each field is printed in a single shot through a mask, so the die is patterned a field at a time, and the shutter stays closed while the stage travels to the next one.
The mask carries one test cell — two bond pads, a supply rail, a single transistor with its gate finger, and a pair of meander test structures — imaged into a 900 µm field, the whole cell at once. Then the stage steps 900 µm, settles, and the shutter opens again. Nine fields in a 3 × 3 grid cover 2.7 × 2.7 mm of the die, inside the 3 mm envelope, and the whole thing repeats until the die is full.
Fig. 7 — Live simulation of a step-and-repeat exposure. Stage travel is drawn one-to-one; die, fields and pattern are to scale. Feature size is set by the mask and the projection optics above, not by the stage — the stage decides only where each field lands.
Lithography for the rest of us.
Patterning is how you turn a material into a device. Almost nobody gets to try.
Every microfluidic chip, every photonic waveguide, every MEMS sensor starts the same way: something has to put a pattern on a substrate with sub-micron placement. The optics for that are increasingly within reach — a UV source, a mask, a projection lens, a shutter. The stage under them has not been. A positioner good enough to place one field next to the next has cost more than most departments spend on a whole experiment, and that single line item is where a great many good ideas quietly stop.
This is the part we can fix. A stage that prints in one piece for about ₹1,500 turns the expensive half of a lithography tool into a consumable. A student can print two, break one, and keep going. A school can bolt one to a table beside the microscope it already owns. A lab in a place with no cleanroom can write real patterns on real substrates and find out whether the idea works before anyone writes a grant.
We are not claiming a fab on a bench. We are claiming that the positioning stage should stop being the reason you never tried.
What this is not
- Not a resolution claim. The smallest feature you can write is set by the wavelength and the optics overhead, not by the stage.
- Not a cleanroom. No filtered air, no temperature control, no vibration isolation beyond the table you bolt it to.
- Overlay between fields is bounded by the stage's repeatable increment, currently 210 nm, and by whatever the frame drifts through a long write.
- The exposure shown here is a simulation built from the stage's measured numbers. Patterned substrates from this setup are the next thing we publish.
The fibre stays still. The chip moves.
Swap the projection optics for a fibre array and the same stage becomes the moving half of a photonics test bench. Eight fibres at 127 µm pitch sit in a glass V-groove block, polished so the cores meet the chip at 8°. Under them, a silicon die carries rows of grating couplers: the outer two channels joined by a loopback waveguide, the inner pairs feeding ring resonators.
Nobody can see a ten-micron coupler through a block of glass, so alignment is done by listening to light. The stage brings the chip within a few tens of microns by eye, then spirals outward until the power meter registers first light. From there it hill-climbs in shrinking steps, the last of them 210 nm — the stage's repeatable increment — until the loopback peaks. Then it steps to the next block and does it again.
Fig. 8 — Live simulation of active alignment. The stage carries the chip under a fixed fibre array; the map follows fibre channel 1 over its input coupler, in micrometres, coloured by loopback power wherever it has been sampled. Coupling is the overlap of two 10.4 µm Gaussian modes with a typical 4.5 dB coupler loss — textbook inputs, not our measurements. Array and chip are drawn to scale.
Light in, light out.
A photonic chip is useless until you can get light into it. The last ten microns are where the money goes.
Silicon photonics has quietly become something you can design on a laptop. Multi-project wafer runs share one mask between many groups, so a die full of waveguides, rings and interferometers no longer needs a fab of your own. Getting light onto it is another matter. A grating coupler accepts a fibre over a few micrometres; miss by five and more than half your signal is gone. So every photonics lab has the same instrument at its centre: a fibre held still, a chip that moves under it, and something that walks the two together until the power meter peaks.
That something is usually a motorised or piezo alignment stage, and it is usually the reason a group owns one test setup instead of four. XY Lab repeats 210 nm steps across a 3 mm envelope — an order of magnitude finer than the coupler can tell apart, with enough travel to step along a row of test structures. Bolt one under a fibre array, and the expensive half of the bench becomes a print.
We are not claiming a probe station. We are claiming that the alignment stage should stop being the reason a chip sits untested in a drawer.
What this is not — yet
- Not a measurement. Alignment and coupling on this page are simulated from the stage's measured step and textbook coupler figures. Coupled-light data from our own bench is what we are working towards.
- Two axes, not six. The stage moves the chip in the plane; fibre height and the 8° tilt are set once on a fixed mount. A grating coupler is far more forgiving of height than of in-plane position, which is exactly where the stage earns its place.
- Surface couplers only. Butting a fibre against a polished chip facet needs vertical alignment this stage does not have.
- Plastic moves with temperature. A printed frame grows several times more per degree than steel, so expect to re-peak before each measurement rather than align once and walk away.
Moves in Nanometers
Under optical tracking the stage resolves individual increments below 20 nanometres — roughly one four-thousandth the width of a human hair — against a measurement noise floor of 0.6 nanometres. The smallest increment it repeats dependably, move after move, is 210 nanometres. Both numbers belong to a machine built for the price of a good pair of shoes.
Scale — Every length quoted on this page on one logarithmic ruler, a decade per division. From 3 mm of travel down to a 0.6 nm noise floor is a span of nearly seven orders of magnitude. Heavy marks are this stage; light marks are there for reference.
Fig. 1 — Live simulation of the slide holder seeking random targets inside its travel envelope. The holder moves on both axes; each drive block and its hardware travel only along their own lead screw, and the blades bend to take up the difference. Frame, base and motors are rigid. Displacement against the datum outline is magnified roughly 4,000×.
Measured, not modelled.
Every number on this page came off the bench. A marker on the slide holder, filmed at 20 frames per second and tracked frame by frame against a calibrated scale. Two runs from that work are below.
nine commands
of variation
moves
linear fit
increment
1.36 nm on X
pixel
coupling
| 25 microsteps | 1 microstep | |
|---|---|---|
| Mean move | 210 nm | 6.8 nm |
| Scatter | CV 0.13 | CV 1.60 |
| Run backwards | 0 % | 25 % |
| Land dead | — | 15 % |
Notes
- Sub-20 nm individual increments optically resolved, against a 0.6 nm measurement noise floor.
- Minimum reliably repeatable increment is currently 210 nm, at 25 microsteps per command.
- Single-microstep commands resolve below 20 nm but do not yet repeat: 15 per cent land dead, and a quarter reverse.
- All figures are from video tracking. Independent interferometric validation is the next step.
Not the finest. The cheapest by three orders of magnitude.
A piezo stage will out-resolve us by a factor of fifty. A bearing stage will out-travel us by a factor of four. Neither of them comes off a print bed for the price of lunch, and that is the whole argument.
Range — Each bar runs from a stage's finest quoted motion to its full travel, on a logarithmic axis. The tick on our bar is the 210 nm we repeat; the tick on the NPXY200SG is its 4 nm closed-loop figure. The piezo stages reach further down, the bearing stage further out, and the printed one covers the same middle ground as the Thorlabs.
| Stage | Travel | Fine motion | Price |
|---|---|---|---|
| F2R-XY-01 |
3 × 3 mm | <20 nmresolved; 210 nm repeatable (per 25 microsteps) | ₹1,500≈ $15.60 |
| Thorlabs NFL5DP20 |
5 mm20 µm piezo | 20 nmopen loop; 10 nm on the −S | ≈ $945+ driver |
| Newport 9062-XY-NP-M |
12.7 mm | <30 nmmin. incremental motion, Picomotor axis | On request |
| Newport NPXY100 |
100 µm | 0.2 nmopen loop | ≈ £3,854+ amplifier |
| Newport NPXY200SG |
200 µm160 µm closed | 4 nmclosed loop; 0.4 nm open | ≈ £6,577+ amplifier |
in materials
The XY stage costs about ₹1,500 to build. That is $15.60 in materials.
In less time than it takes to fill in a quotation form for a commercial stage, you could be doing real positioning work on an F²R.
Writing patterns. Scanning samples under a microscope. Aligning optics. Building instruments nobody has thought of yet.
The hard part of nanopositioning was never the physics. The hard part was affording it.
Ratio — Sixty squares, one per XY One. Together they cost in materials what a single Thorlabs NFL5DP20 lists for, about $945 before its driver.
The things people ask.
How can a 3D print position to 20 nanometres?
Because nothing slides. The stage moves by bending 0.5 mm blades of plastic, and a blade under load deflects by an amount set by its geometry, not by its surface finish. Layer lines and printer tolerance would wreck a sliding bearing; they barely matter to a flexure. Precision moves from the machine shop into the CAD file, and CAD is free.
What is the difference between resolved and repeatable?
Resolved means we can see the stage make a move that small: individual increments below 20 nm show up cleanly above a 0.6 nm measurement noise floor. Repeatable means you can ask for that increment and get it every time. Today that number is 210 nm. The gap between the two is honest work still to be done, and it is why both numbers are on this page.
So what resolution can it write?
That is the wrong question to ask the stage. Feature size comes from the wavelength, the numerical aperture and the resist. The stage decides where a feature lands, not how small it is. Ours will place one field next to the next to within its repeatable increment, currently 210 nm, which is what sets overlay across a step-and-repeat job. Feature size is the mask's problem; placement is ours.
What would I need on top of the stage to actually write?
A source, a mask carrying your pattern, a lens to project it onto the substrate, a shutter, a coated substrate, and controls that can sequence them together. None of that is in the ₹1,500. What the ₹1,500 buys you is the part that used to cost more than all of it put together.
Can it align fibres to a photonic chip?
For surface grating couplers, that is what we are exploring now. A standard coupler gives up 1 dB at about 2.5 µm of lateral error, and the stage repeats 210 nm steps, so the resolution is there with room to spare. Fibre height and tilt go on a fixed mount; the first coupled-light measurements are on our bench. If you have a chip that needs light in it, write to us.
How far does it travel?
About 3 mm on each axis. That is the trade a flexure makes: the blades have a finite elastic range, and pushing past it costs you linearity long before it costs you the part. A 5 mm die needs two mountings to cover; if you need centimetres in one go, you want a bearing stage.
How was it measured?
Video tracking. A marker on the slide holder, filmed at 20 fps and tracked frame by frame against a calibrated scale, with sub-pixel centroid fitting. Independent interferometric validation is the next step, and until that is done every figure here should be read as tracker-derived.
What do I need to build one?
A printer that can hold a 0.5 mm wall cleanly, two small stepper motors with drivers, a microcontroller, and the screws, rods and lead screws in the bill of materials. Print, bolt on the motors, flash the firmware. There is no assembly step in between, because there is nothing to assemble.
Can I buy one instead of printing it?
Yes. We build XY Lab to order with its stepper drivers and your choice of Arduino Nano or ESP32, for ₹2,500 including shipping within India — order one here. Pilot builds, lab sets and collaborations go to the same inbox, and a human answers.
Something not answered here? Send it over — this list grows from real questions.
Build it, or buy it built.
The stage is open source. The repository is where the project lives, and it fills out as each part of the build is cleaned up and documented. Clone it, print it, drive it past where we stopped — and tell us what breaks. No printer, or no time? We will print, assemble and check one for you.
github.com/atharvdubey22ug-glitch/XY-Stage-files — and if you write something with one, we very much want to see it.
Contact us
Collaborations, pre-orders, questions about the characterisation, or an application we haven't thought of. A human reads every message, and answers quickly.
atharvdubey.nj@gmail.com
+91 81092 85020
github.com/atharvdubey22ug-glitch/XY-Stage-files
Supported by an equity-free grant from
,
the Frugal Founder’s Fund.
© 2026 F²R — Fiction to Reality. Application notes AN-01 · Step-and-repeat lithography, AN-02 · Fibre-to-chip alignment, AN-03 · Free-space laser link / F2R-XY-01. Stage figures obtained by video-tracking characterisation and may vary between builds. The exposure sequence and the fibre alignment on this page are simulations driven by those figures.