SU-8 process recipe (spin-coating, soft bake, dose, PEB, development)

Two kinds of information are deliberately kept apart. Soft bake, PEB and development are **read** from a published table: they are the manufacturer’s values, not calculations. The corrected dose and the exposure time are **computed**, but only through exact operations, a multiplication by a tabulated factor, then a division by an irradiance you measure. No interpolation occurs in these recipe parameters or in these calculations, interpolation. Where it exists, concerns only the spin speed, and it is declared as such.

The dose recommended by the datasheets assumes a silicon substrate. On a reflective substrate, part of the UV crosses the resist, reflects off the interface and passes back through the film: the energy actually absorbed differs from the energy emitted by the lamp. The SU-8 2000 datasheets publish a multiplier to compensate, 1.5× for glass, Pyrex and ito, 1.5 to 2× for gold. Aluminium, copper, nickel and titanium. This tool applies that factor automatically and shows the dose before and after correction. When the datasheet gives a range rather than a value, it is because reflectivity depends on surface condition: start at the lower bound and refine with an exposure matrix.

The thickness obtained depends on spin speed, but the datasheets do not all present that relation the same way, and this tool does not treat the two cases alike. For the classic series, each datasheet publishes a **table** of three speeds per product: the values are exact, and an interpolation between two exact points is flagged as such. For the SU-8 2000 series, the relation exists only as a **figure**: the markers were read off, and the error of that reading was measured rather than assumed. Since the SU-8 2-25 datasheet publishes both a figure and a table for the same products, the reading procedure was applied there and compared against the exact values. Result: up to 13 % deviation above 10 µm, but up to 33 % below. Where the curve flattens against the axis. The declared uncertainty is therefore 15 %, rounded above the worst measured deviation, and no speed is proposed below 10 µm for that series. Every result states its provenance: tabulated value, interpolation between exact points, or figure reading.

A datasheet cannot give an exposure time, because time is not a property of the resist. What the resist imposes is a **dose**: an amount of energy per unit area, in mJ/cm². The time needed to deliver it depends entirely on your source intensity, the same 150 mJ/cm² takes 15 seconds under a lamp measured at 10 mW/cm², and half that under a lamp twice as intense. This is why the irradiance field is optional: if you have not measured it, the tool gives you the dose and stops there, rather than inventing a time. Measure it with a radiometer at i-line (365 nm), the reference wavelength of the SU-8 datasheets.

Soft bake

· Bake after coating, which evaporates solvent and densifies the film. Too short and the film stays soft and sticks to the mask; too long and it becomes brittle and adheres poorly.

Exposure

· UV exposure which, in a negative resist such as SU-8, forms a strong acid in the illuminated areas. The pattern is not yet fixed at this stage.

PEB (post-exposure bake)

· Bake that triggers crosslinking catalysed by the acid formed during exposure. It is what makes the exposed areas insoluble, an insufficient PEB produces patterns that lift off during development.

Development

· Dissolution of the un-crosslinked areas. Published times are approximate, since dissolution rate depends strongly on agitation and temperature.

Dose

· Energy per unit area, in mJ/cm². A property of the resist and thickness, independent of equipment.

Irradiance

· Power per unit area delivered by the lamp, in mW/cm². A property of your equipment, to be measured with a radiometer.

It does not cover the SU-8 3000, TF 6000, xft and dfr series. Nor SU-8 2100/2150, because their process datasheets could not be consulted from their primary source. Adding them from a secondary source would contradict the rule this project imposes on itself. It proposes no spin speed below 10 µm for the SU-8 2000 series: cross-validation measured up to 33 % reading deviation there, which makes the figure unusable. Finally, it does not replace an exposure matrix: the datasheets themselves present their values as a starting point to be optimised, not as a validated recipe for a particular process.

This is what sets this tool apart from existing SU-8 calculators, and it is not cosmetic. The 25–40 µm range is covered by **two different datasheets**: the one for the thin SU-8 2000 formulations (2000.5 to 2015) and the one for the thick formulations (2025 to 2075). They do not give the same soft bake, 4 to 5 minutes at 95 °C for the former, 5 to 6 minutes for the latter. A calculator asking only “what thickness do you want?” would have to arbitrate between the two, and would therefore arbitrate arbitrarily for half its users. By asking for the product first, the question disappears: the datasheet to consult is determined, and the recipe with it. The same logic applies to the classic series, whose tables are indexed product by product.

Datasheets are not extracts from a single database: they are distinct documents, published at different dates. For different formulations, and validated on different processes. Two datasheets overlapping in thickness have no reason to agree exactly: they describe resists whose viscosity, solids content and drying behaviour differ. When a discrepancy exists, this tool neither arbitrates nor hides it: it shows the value from the datasheet matching your product, cites that document’s exact revision, and flags that other products cover the same thickness. It is up to you to know which resist is actually in your bottle.