Mask cost and lead time are charged per design, laser writing time is charged per square millimetre. Which way that arithmetic falls depends on how often your designs change, how large the pattern is and how many students run the same exposure.
A doctoral project designs an electrode array, orders a chrome on quartz mask, waits for it to clear customs, exposes twelve wafers, and discovers that one dimension is wrong by five micrometres. The correction is trivial in the CAD file. Getting it onto glass and back into the building takes another six weeks, and the student has lost a term. That sequence, repeated across a department, is the strongest argument for direct write lithography that exists, and it has nothing to do with resolution.
Both routes end with patterned resist on a substrate. What differs is where the money and the waiting go.
A mask based process pays once per design. The mask is written by a service, shipped, inspected and then used as many times as it survives. Per exposure the marginal cost is close to zero, which is why every production line uses masks. In Bangladesh the mask is imported, so the cost per design carries shipping, customs clearance and the risk of damage in transit, and the delay is measured in weeks rather than days. A five layer process needs five masks and five of those delays, usually not in parallel because layer two often depends on how layer one turned out.
A direct write system pays per exposure instead. There is no mask and no lead time; the pattern goes from the CAD file to the substrate in one session. What you pay is machine time, which scales with the area written and the dose, plus the depreciation and service of a machine that has moving optics and a precision stage. Change the design at four in the afternoon and you can expose the corrected version the same evening.
So the comparison is not equipment cost against equipment cost. It is design iterations per year against exposures per design. A group that settles a design and then makes two hundred of them wants a mask. A group that changes the design every fortnight for eighteen months wants a writer. Most university research is firmly in the second category and most teaching is firmly in the first, which is why facilities that serve both keep arguing about it.
Count the calendar, not the process steps. With a mask, an iteration is: revise the layout, submit to the mask house, wait for writing and shipping, clear customs, inspect, then expose. With a writer, an iteration is: revise the layout, load the file, expose. In a research programme where the number of iterations is the thing that limits progress, that difference compounds. Three iterations a year against thirty is a different project.
There is a middle path that is often the correct answer and gets overlooked. A direct write system can write your own masks. Buy the writer, produce masks in house on blank plates for the designs that have stabilised, and expose those on an aligner for volume and for teaching. That removes the import delay from the mask route entirely and gives the facility both capabilities from one capital purchase plus a relatively cheap aligner.
Procurement committees compare headline resolution figures because they are the easiest number in the brochure to line up. It is close to the least useful.
A contact aligner with a good mask, a thin resist and genuine contact resolves finer features than most people assume, and its practical limit is set by contact quality, resist thickness and diffraction in any gap between mask and wafer. A direct write tool's limit is set by the write lens, the beam and the stage. In both cases the achievable line width in a real laboratory is dominated by process rather than by the tool: resist selection and thickness, soft bake temperature and time, exposure dose calibration, developer concentration and temperature, substrate flatness and cleanliness, and the humidity of the room.
That last one deserves emphasis in Bangladesh. Resist behaviour is sensitive to humidity, and a coating bench in a room whose relative humidity swings widely between the dry season and the monsoon will not give the same line width in June as in January. If the specification calls for sub micrometre features, the environmental control and the process development effort are larger line items than the difference between two exposure tools.
Specify the feature size you actually need across the whole device, not the smallest feature anyone has ever wanted. A great many microfluidic, sensor, electrode and MEMS designs live comfortably in the several micrometre range, where both routes work and the decision falls back to turnaround and throughput.
A mask aligner exposes the whole field at once. Whatever is on the mask lands on the wafer in a single flash, so a two centimetre long waveguide is one continuous feature with no internal boundaries.
A direct write tool builds a large pattern out of write fields, moving the stage between them. Wherever two fields meet there is a stitch, and a feature crossing that boundary can show a small discontinuity or lateral offset. For a pad, a via or a short electrode, nobody will ever see it. For a long optical waveguide, a precision resistor, a diffraction grating or an interdigitated structure whose period must not vary, a stitch is a defect. Interferometric stage feedback reduces stitching error substantially and is the feature to ask about if the group makes long continuous structures.
Overlay, meaning the alignment of each layer to the one below, is a separate matter and both routes need it. An aligner does it optically through the mask against marks on the wafer, with the operator or an automatic routine doing the alignment. A writer does it by locating marks with its own optics and then transforming the pattern before writing, which means it can correct for wafer rotation, offset and sometimes scale, and that is a genuine advantage on a warped or non standard substrate.
The common failure is neither tool's fault. Layer one is designed without global alignment marks, or with marks that are destroyed by the etch that follows, and layer three then cannot be aligned to anything. Put durable alignment marks in the first mask level, put them outside the die area, and make sure they survive every process step in the flow. Write that into the design rules for the facility and enforce it.
Direct writing is serial. Time on the machine scales with the area to be exposed and inversely with the writing speed, so a full wafer covered in dense pattern is a long session and a small die is quick. Mask exposure is parallel: the exposure time is set by the dose and the lamp, and it is the same whether the mask carries one die or four hundred.
That is why a teaching laboratory should not run thirty students through a direct write tool on the same pattern. One mask, one aligner and a well drilled sequence gets a class through in an afternoon and gives every student the same starting point. Reserve the writer for the research groups, where the pattern is different every time.
It is also why the mask wins for any small volume production the facility takes on for industry. If a local sensor company wants a hundred devices a month, the mask pays for itself in the first batch and the writer becomes the tool that made the mask.
Three patterns cover most Bangladeshi facilities.
Whichever route is chosen, the exposure tool is a minority of the spend. The supporting chain is what determines whether the facility produces devices: a spin coater with reproducible speed and acceleration, calibrated hotplates rather than an oven, a developer bench with temperature control, deionised water and a clean drying route, a wet bench with the right chemical handling, a metrology microscope with a calibrated stage for checking line width, and a resist store that is actually refrigerated.
Resist deserves its own sentence. It has a shelf life, it degrades with heat, and it is shipped from Europe or the United States. Resist that spent three weeks in an ambient container and a further two in a customs warehouse in July is not the material described on the datasheet, and the failed exposures that follow will be blamed on the exposure tool. Order in quantities you will use, insist on cold chain shipping where the supplier offers it, record receipt dates and store it properly.