Photolithography in a university cleanroom: what students can be trained on and what research needs

Spin coating, mask aligners, honest resolution limits and resist handling in a humid climate, plus how to size a lithography bay so a class of thirty and a PhD student can both use it without fighting.

Two requirements arrive on the same purchase order. An undergraduate class of thirty has to produce a patterned wafer each semester, reliably, in a three hour session. A doctoral student needs 1 micron features aligned to a previous layer on a diced 10 mm chip. Departments usually specify for the second requirement and then discover they cannot deliver the first, because the bottleneck in a teaching laboratory is never the exposure tool.

This article covers the lithography sequence and where it fails, what a mask aligner can and cannot resolve, resist handling in Dhaka's humidity, where photomasks come from, and how to size a bay so that teaching and research do not compete for the same hour.

The sequence, and where it actually goes wrong

Dehydration bake, adhesion promoter, dispense, spin, soft bake, align, expose, post-exposure bake, develop, rinse and dry, inspect, hard bake, then etch or lift-off, then strip. Eleven steps before the pattern means anything.

Almost every failure in a university lab traces back to three of them: adhesion, which is a dehydration bake and priming problem, bake control, which is a hotplate calibration problem, and develop time, which is a discipline problem. The mask aligner is rarely the culprit, which is inconvenient, because the aligner is the expensive item everyone inspects first when results go wrong.

Spin coating

Resist thickness falls roughly with the inverse square root of spin speed and rises with resist viscosity. Start from the manufacturer's spin curve, then build your own on your own hotplates with your own room conditions, because the published curve was measured somewhere else.

• Dispense method changes the result. A static dispense onto a stationary wafer then ramping up, and a dynamic dispense onto a slowly turning wafer, give different coverage and different defects. Pick one, write it into the process sheet, and stop letting each student decide.
• Small pieces are a different process. On a 10 mm die the edge bead occupies a serious fraction of the usable area, and surface tension pulls resist toward the corners. Buy the small chuck adapters with the spinner, and consider mounting chips on a carrier wafer.
• Comets and striations come from particles on the substrate and from dispensing onto a poorly wetted surface. Filter the resist at the point of dispense and clean the chuck between users.
• Hotplates, not ovens, for soft bake. Contact hotplates give repeatable heat transfer; a convection oven gives a different bake and a different final thickness. Do not mix the two in one process, and calibrate the plate with an independent probe rather than trusting the display.
• Exhaust the spinner bowl. Solvent vapour has to be extracted at source, and the bowl needs cleaning on a schedule or it becomes the particle source that contaminates every subsequent coat.

Mask aligners and exposure modes

A mask aligner brings a photomask and a substrate into a defined geometric relationship and floods them with ultraviolet light. The mode of contact sets the resolution and the mask life.

Mode	Mask to wafer	Resolution	Mask wear	Where it fits
Proximity	Held apart by a set gap	Coarsest	Minimal	Teaching, forgiving feature sizes, protecting expensive masks
Soft contact	Touching, no vacuum	Better	Moderate	General purpose work
Hard contact	Pressed together	Better still	Significant	Research layers needing tighter features
Vacuum contact	Evacuated between mask and wafer	Best available on an aligner	Highest	Finest features, flat substrates, experienced users

• Masks are consumables. Thirty students in hard contact will damage a mask in a term. Run teaching layers in proximity or soft contact, keep a separate research mask set, and budget for replacement rather than treating a scratched mask as a fault.
• Expose by dose, not by time. A mercury lamp loses output as it ages, so a fixed exposure time slowly under-exposes as the term progresses. Measure intensity with the aligner's probe and calculate the time from the required dose. This one habit prevents more failed student batches than any other, and it is the step most often skipped.
• Buy the chucks and mask holders you will actually use, at the time of purchase, covering every substrate size including small pieces. Retrofitting a chuck later is slow and disproportionately expensive.
• Decide about backside alignment early. If MEMS or through-wafer work is on the roadmap, that option has to be specified with the machine. It is not a field upgrade on every platform.
• Chuck cleanliness is an alignment issue. One particle trapped between wafer and chuck tilts the substrate and destroys contact across a whole quadrant. The symptom looks like a resolution fault and is housekeeping.

SUSS MicroTec is the reference manufacturer for mask aligners in research cleanrooms, and Vvon supplies their lithography systems in Bangladesh alongside Oxford Instruments plasma and deposition tools, which matters when the etch step has to be developed against the resist the aligner is exposing.

Resolution: what an aligner can and cannot do

In proximity mode, the smallest usable feature scales with the square root of the wavelength multiplied by the gap between mask and wafer. In contact mode the gap term is replaced by the resist thickness, which is why contact does so much better. Working with the ultraviolet lines of a mercury lamp and a resist a micron or two thick, that arithmetic puts a mask aligner comfortably in the few micron regime, working hard at around a micron, and out of the running below that.

Which leads to the most useful sentence here for anyone drafting a tender: a resolution figure with no resist thickness, contact mode and substrate specification attached is not a specification. Writing "0.5 micron resolution" into a mask aligner tender sets the laboratory up to fail its own acceptance test, and the supplier will be right while the university is disappointed. State the feature size, the resist and its thickness, the contact mode and the substrate, then require a demonstration on your own material.

If the research genuinely needs sub-micron features, plan for a different technique: electron beam lithography, nanoimprint, or a projection stepper. Or, usually the better answer for a first facility, design devices that do not need it. A great deal of publishable MEMS, microfluidics, sensor, photovoltaic and photonic work sits at 2 to 5 microns, and a group chasing sub-micron capability before it has a reliable 2 micron process tends to end up with neither.

Resist handling in a humid climate

• Yellow light means all of it. Filter everything below roughly 500 nm in the bay, including corridor light through a vision panel, laptop screens and phone torches. A single unfiltered window has quietly ruined more resist than any chemical problem.
• Storage and warm-up. Store resist and developer at the temperature the supplier states, and let a bottle reach room temperature sealed before opening it. Open a cold bottle in a Dhaka laboratory and moisture condenses straight into the resist.
• Shelf life against import time. Resist and developer ship as flammable or corrosive goods, with transport restrictions and customs clearance in the middle, so a product with a limited shelf life can arrive with much of it gone. Order quantities you will use, insist on the batch date at delivery, and date every bottle when it is opened.
• Priming matters more here. Ambient humidity leaves adsorbed water on the substrate, which is the direct cause of resist lifting during develop and etch. A proper dehydration bake followed by vapour phase HMDS priming is worth more in Dhaka than in a dry climate, and vapour priming beats a puddle of promoter on a spinner.
• Developer discipline. Developers are toxic and their strength drifts with use and with carbon dioxide absorbed from the air. Use fresh developer, time it with a timer rather than by eye, agitate the same way each time, and control the temperature. Under-developed resist looks like a bad exposure and is not.
• Segregate the waste. Developer, solvent and rinse water go to separate collection, with a disposal route agreed before the laboratory opens rather than improvised when the first drum is full.

Photomasks: the line item nobody budgets

Masks are imported, have a lead time, and are consumed. Film or emulsion masks printed on polyester are inexpensive and adequate down to features of tens of microns, which makes them ideal for teaching. Chrome on soda lime glass is the workhorse for features of a few microns. Chrome on quartz costs more and earns it where deep ultraviolet exposure or tighter dimensional stability is needed.

1. Freeze the layout and have it checked by somebody who did not draw it. Alignment marks, layer polarity and the orientation of the whole die are the three things that come back wrong.
2. Order several revisions and several layers in one purchase to absorb the shipping time once instead of three times.
3. Log every mask in a register with its layer, polarity, date and photographs of any defect.
4. Store masks in dedicated boxes and clean them only by the prescribed procedure. An unlabelled scratched mask left in a drawer will be used by a student at two in the morning, and the failure will be blamed on the aligner.

Sizing the bay for teaching and research together

Requirement	Teaching class	Research group	What to buy
Feature size	10 to 20 microns	1 to 2 microns	One aligner with vacuum contact capability, film masks for teaching and chrome for research
Throughput	Thirty students in one session	A few chips a week	Two spin stations and two hotplate sets, not one of each
Substrates	Small wafers or glass slides	Diced chips and full wafers	A chuck set covering every size in use, bought with the tool
Alignment	Single layer, no overlay	Two to four layers with overlay	Split field microscope, and backside alignment if MEMS is planned
Process control	One fixed qualified recipe	Recipes under active development	A locked teaching module plus a separate development log

The insight in that table is in the throughput row. Thirty students each need to prime, spin, bake, expose, develop and inspect. Exposure takes seconds. Spinning and baking take minutes each and cannot be parallelised on one station. A bay with one spinner, one hotplate and one aligner runs a class of thirty as a queue that consumes an afternoon, with the aligner idle for most of it.

The second protection is procedural. Write a locked teaching module: a fixed resist, fixed spin speed, fixed bake, fixed dose and fixed develop time, qualified against a control wafer with a measured linewidth, that students run unchanged. Researchers develop new processes in their own logged space, on their own consumables. That separation protects the class from a half-finished experiment and protects the researcher from thirty people changing the hotplate set point.

A lithography bay is judged by how many people can complete a process in an afternoon, not by the smallest line it has ever printed.: Vvon engineering team, on university cleanroom planning

Back to all Insights