Lift-off lithography, step by step

1 · Why the undercut matters

Deposit a film over a resist edge that slopes *outward* toward the top — a positive, tapered wall — and the material coats the resist and the substrate as one continuous sheet, joined at the sidewall. When you then try to dissolve the resist, the solvent can't reach it through that continuous film, and the metal tears raggedly instead of lifting: the classic fencing or flagging defect, where a wall of unwanted material stands up at every feature edge.

An undercut — a resist profile that is *narrower at the top than at the bottom*, a re-entrant or overhanging edge — fixes this by geometry. The deposited film lands on the substrate at the base and on top of the resist, but the overhang shadows the sidewall so the two never connect. That break gives the solvent a path to the resist and a clean edge for the material to snap along. Everything else in lift-off is in service of producing and preserving that overhang.

2 · Three routes to the undercut

Single-layer negative resist

Some negative resists develop with a naturally re-entrant sidewall — the top of the exposed feature cross-links more than the base, so development leaves an overhang in one coat. The AZ nLOF series is the workhorse here: Kim et al. 2017 use AZ nLOF 2035 as the negative-tone lift-off stencil for a 100 nm gold-on-chromium metamaterial, and Huang et al. 2025 use a 7 µm AZ nLOF 2070 film, direct-write patterned, to lift off superconducting air bridges. The AZ nLOF 2020 recipe publishes 66 mJ/cm² at 365 nm for a 2.0 µm reference process, the AZ nLOF 2035 recipe 80 mJ/cm² at 365 nm for 3.5 µm, and the NR9-1500PY recipe 190 mJ/cm² at 365 nm normalized to a 1 µm film.

Bilayer (underlayer + imaging resist)

The most controllable undercut comes from a bilayer: a non-photoimageable underlayer beneath an ordinary imaging resist. The underlayer dissolves *laterally* in the developer at a rate set by its own chemistry, cutting a controlled overhang under the patterned imaging layer independently of the exposure. Chen et al. 2004 reach sub-100 nm lift-off with a PMMA/LOR bilayer this way, and Cord et al. 2006 use a PMMA/PMGI stack for the Dolan-bridge undercut of evaporated Josephson junctions, exploiting that PMGI develops in aqueous base independently of the imaging layer. The LOR 3A recipe (0.28–0.567 µm), LOR 5A recipe (0.47–0.98 µm) and PMGI SF6 recipe (0.26–0.485 µm) are the underlayers — all not photoimageable, chosen by thickness, as the lift-off underlayers type page explains.

Image reversal

A positive resist run through an image-reversal process develops with the overhanging sidewall lift-off wants, from a single coat. AZ 5214E is the classic example — the AZ 5214E recipe doubles as an image-reversal resist and a direct lift-off stencil, though it publishes no absolute dose (only a rule of thumb that the imagewise dose is roughly half a standard positive-process dose). The image-reversal resists type page covers the reversal-bake mechanism.

Sources: Kim et al. 2017; Huang et al. 2025; Chen et al. 2004; Cord et al. 2006

3 · What deposition the undercut can take

Lift-off only works with a line-of-sight (directional) deposition — thermal or e-beam evaporation, and sputtering only cautiously. The overhang has to *shadow* the sidewall, and it can only do that if the arriving material travels in nearly straight lines from a distant source. A conformal process — CVD, ALD, or heavily scattered sputtering — coats the sidewall and the underside of the overhang too, reconnecting the film and defeating the undercut. Keep the substrate roughly normal to the source and avoid rotation that would let material creep under the lip.

The film thickness must stay well under the resist thickness — a common rule of thumb is no more than about a third — so the deposit never climbs to the top of the resist wall and bridges the gap the overhang created. That is why lift-off resists are coated thick relative to the metal: a thin evaporated layer lifts cleanly under a micron-scale stencil, while trying to lift a film approaching the resist height fences badly. Keep the substrate cool, too; a hot evaporation can reflow or burn the resist and close the undercut.

4 · The process, step by step

The single-layer negative-resist flow is eight steps (a bilayer adds an underlayer coat + bake before the imaging resist; an image-reversal route adds the reversal bake and flood exposure before develop). Values for your resist live on its recipe page.

• —Clean and prime. Clean, dehydration-bake and (where called for) HMDS-prime the substrate. For a bilayer, coat and bake the underlayer now.
• —Coat. Spin-coat the imaging resist thick relative to the metal you'll deposit, read from the resist's spin curve.
• —Soft-bake. Soft-bake to drive off the casting solvent.
• —Expose. Deliver the resist's working dose; on a maskless tool this is a direct write from your layout — Bonafe et al. 2025 pattern S1818 by maskless direct-write on an ML3 Microwriter and use a second S1818 layer as the lift-off stencil.
• —Post-exposure bake. For a chemically amplified negative resist (the nLOF series) this bake is essential and drives the cross-link that forms the re-entrant edge; for an image-reversal route it is the reversal bake. Check whether your resist needs one.
• —Develop (form the undercut). Develop to clear the pattern and open the overhang; for a bilayer, this is where the underlayer's lateral develop rate sets the undercut depth. Don't over-develop past the intended undercut.
• —Deposit. Evaporate (line-of-sight) your material over the whole surface, film thickness kept well under the resist height, substrate cool.
• —Strip and lift off. Soak in the resist's solvent (acetone or a dedicated stripper, sometimes warmed or with gentle agitation) until the resist dissolves and carries away the material on top of it. Rinse and dry. The strip & lift-off section covers the solvents and agitation.

As a fully worked example, Zhang et al. 2025 publish the complete NR7-1500PY route — spin, bake, expose, post-exposure bake, a short TMAH develop, then an overnight Cr/Au lift-off — for ring electrodes on glass.

Sources: Bonafe et al. 2025; Zhang et al. 2025

5 · Failure modes

Lift-off failures nearly all trace back to the undercut or the deposition mismatch.

• —Fencing / flagging — walls of material standing at feature edges → insufficient or absent undercut, or a film too thick relative to the resist. Deepen the undercut (longer develop, thicker or dedicated underlayer) or thin the deposit.
• —Metal won't lift / lifts partially → the deposit bridged the resist top: reduce film thickness, increase resist thickness, or move to a bilayer for a cleaner overhang. Warming the solvent or longer soak time helps only if the resist is still reachable.
• —Ragged, torn edges → a marginal undercut tearing rather than snapping; increase the overhang, and avoid ultrasonic agitation that shreds fragile features (use gentle flow instead).
• —Resist reflowed / undercut closed during deposition → the substrate ran too hot; slow the evaporation or add cooling so the resist keeps its overhang.
• —Redeposition / metal debris settling back on the surface → catch lifted material in flowing solvent and rinse promptly rather than letting a still bath dry down.

Common questions

Where to next

1. Kim et al.. Electrical access to critical coupling of circularly polarized waves in graphene chiral metamaterials. Science Advances (2017). doi:10.1126/sciadv.1701377
2. Huang et al.. Fabrication of metal air bridges for superconducting circuits using two-photon lithography. Applied Physics Letters (2025). doi:10.1063/5.0271788
3. Chen et al.. A lift-off process for high resolution patterns using PMMA/LOR resist stack. Microelectronic Engineering (2004). doi:10.1016/j.mee.2004.02.053
4. Cord et al.. Robust shadow-mask evaporation via lithographically controlled undercut. Journal of Vacuum Science & Technology B (2006). doi:10.1116/1.2375090
5. Zhang et al.. Proteomics-Empowered Microfluidic-SERS Immunoassay for Identifying and Detecting Biomarkers of Micropapillary Lung Adenocarcinoma. Advanced Science (2025). doi:10.1002/advs.202501336
6. Bonafe et al.. Dissipative charge transport in organic mixed ionic-electronic conductor channels. Nature Communications (2025). doi:10.1038/s41467-025-57528-9