Plasma processing for MEMS: deep etching, release and the stiction problem

MEMS devices fail at the last step more often than the first. A practical walk through deep silicon etching, the release step, why released structures stick down, and a full process sequence for a simple cantilever.

A group runs its first MEMS batch. The lithography looks clean, the trenches are deep and vertical, the SEM images are good enough for a conference poster. Then the devices go through release and drying, and every cantilever comes out lying flat on the substrate, permanently stuck. Nothing in the etch step was wrong. The failure is in the last twenty minutes of a three week process, and it is the single most common reason a university MEMS project produces no working device in its first year.

This is a walk through the plasma steps that build a MEMS device and the wet and dry steps that decide whether it moves, written for groups setting up microfabrication in a Bangladeshi university.

Why MEMS breaks the habits of planar processing

Integrated circuit processing works with thin films and shallow features, and everything stays attached to the substrate for life. MEMS turns both assumptions over. Structures are tens of microns deep with aspect ratios in the tens to one. More importantly, part of the device has to be freed from the substrate so it can bend, resonate or rotate. That release step, and the drying that follows it, has no equivalent in IC work, and it is where the process sequence has to be designed rather than copied.

The practical consequence: you cannot judge a MEMS process by looking at the wafer halfway through. A batch that looks perfect under an optical microscope after etching can be entirely dead after release.

Deep etching: the Bosch process and its alternatives

Deep reactive ion etching gets vertical walls in silicon by alternating two steps in an inductively coupled plasma. An etch step, usually SF6, removes silicon almost isotropically for a few seconds. A passivation step, usually C4F8, deposits a thin fluorocarbon polymer over every surface. On the next etch step, directional ions strip that polymer from the trench floor while the sidewalls stay coated, so the etch advances downward and not sideways. Repeat several hundred times.

The signature of this process is scalloping: a ripple on the sidewall with a period set by the cycle time. Shorten the cycles and the scallops get smaller at the cost of etch rate. That trade matters if the sidewall is an optical surface or a capacitor plate, and matters much less if it is a structural wall.

Cryogenic etching is the alternative. SF6 and O2 at a deeply cooled substrate temperature grow a thin passivating layer in place, giving smooth walls with no scallops. It needs a liquid nitrogen cooled chuck, and it is less forgiving of mask adhesion and of the small pieces that university labs usually process.
Wafer cooling is not optional. Long deep etches deposit real power into the wafer. Without helium backside cooling and mechanical or electrostatic clamping, the resist heats, hardens, cracks and stops masking. Half the "the recipe stopped working" reports in a teaching lab are a clamping or helium leak problem.
Design for aspect ratio dependent etching. Narrow trenches etch slower than wide ones, so a layout that mixes 2 micron gaps and 100 micron open areas will not land at a single depth. Either keep trench widths similar across the die, or use a buried oxide as an etch stop and accept the consequences below.
Notching at the buried oxide. When the etch reaches an insulating layer, charge builds up and deflects incoming ions sideways, undercutting the silicon just above the oxide. It is controlled with a low frequency or pulsed bias and a reduced energy final step, and it is the reason silicon-on-insulator devices need a deliberate landing recipe rather than a longer version of the main etch.

The release step

Release means removing the sacrificial material underneath the moving structure. The two routes commonly used in a university lab:

Sacrificial layer	Release chemistry	Nature	Watch out for
Silicon dioxide	Hydrofluoric acid, liquid	Wet, isotropic	Timing, anchor undercut, and the drying step that follows
Silicon dioxide	Anhydrous vapour HF	Dry, isotropic	Water management, condensation, needs dedicated equipment
Silicon	Xenon difluoride vapour	Dry, isotropic, very selective to oxide, nitride and aluminium	Roughening of the etched silicon surface, pulsed dosing

Release is timed, and the timing is unforgiving in both directions. Too short and the structure is still tethered. Too long and the same isotropic etch that freed the beam eats into the anchor that was supposed to hold it. Put release test structures of graduated width on every mask so you can read the completed undercut optically before committing the real devices.

Stiction, and the ways out of it

When a wet released device is pulled out of a rinse and left to dry, the liquid between the structure and the substrate forms a meniscus. Surface tension in that meniscus pulls the beam down with a force that easily exceeds its restoring stiffness. The beam touches the substrate, and once in contact it is held by van der Waals forces, hydrogen bonding and any residue in the gap. In most cases it never comes back up. That is release stiction, and it is a physics problem, not a workmanship problem: careful handling does not avoid it.

There are four ways out, and a good process uses more than one.

Remove the liquid to vapour interface. Critical point drying is the standard answer: exchange water for isopropyl alcohol, exchange alcohol for liquid carbon dioxide in a pressure vessel, then take the carbon dioxide around its critical point so it leaves as a gas without ever forming a meniscus. A critical point dryer is an inexpensive item beside an etcher and it is the difference between working devices and a drawer of stuck ones.
Sublimation drying. Exchange into a solvent that freezes and sublimes, then let it leave the solid phase directly. Cheaper than critical point drying, slower, and workable for teaching devices with generous gaps.
Avoid the liquid entirely. Vapour HF or xenon difluoride release never puts liquid in the gap. This is the cleanest solution where the material stack allows it.
Change the surface and the design. A self-assembled monolayer or a thin fluorocarbon film deposited from a C4F8 plasma makes both surfaces water repellent, which reduces both release stiction and later in-use sticking. On the layout side: dimples or standoff bumps under the beam to cut contact area, release holes through wide plates so the etchant reaches the middle and the plate does not act as a squeeze film, and beams stiff enough that the restoring force wins.

In-use stiction is a separate and very local problem. Water adsorbed from humid air condenses in the narrow gap between a released structure and its substrate and pulls the two together long after fabrication. In Dhaka, that is not a rare failure mode, it is the default outcome for an unpackaged device left on a bench through a monsoon week. Store released dies in a nitrogen purged dry cabinet from the moment they leave the dryer, and treat that cabinet as process equipment rather than furniture.

A working sequence for a simple cantilever

This is the shape of a first process that a masters student can be expected to complete in a semester on a silicon-on-insulator wafer.

Clean and dehydration bake the substrate. Prime with HMDS vapour so the resist adheres through a long etch.
Spin, soft bake, expose and develop the mask layer. For a deep etch, use a thick resist or deposit and pattern a hard mask of oxide or nitride first, because ordinary thin resist will not survive several hundred Bosch cycles.
Measure the patterned linewidth and inspect for resist scumming before etching. Fixing a lithography problem costs an hour; discovering it after the deep etch costs the batch.
Deep etch the device layer in the ICP tool with helium backside cooling and clamping, stopping on the buried oxide, and finish with a reduced bias landing step to limit notching.
Strip the mask. Oxygen plasma for resist, a wet or dry chemical strip for a hard mask, then inspect in the SEM while the structure is still anchored and mechanically safe to handle.
Release: timed hydrofluoric acid on the buried oxide, or vapour HF if available. Read the release test structures and stop as soon as the design undercut is reached.
If wet released, exchange the rinse water for isopropyl alcohol without ever letting the die dry between steps, then transfer wet into the critical point dryer.
Dry through the critical point, or sublime, and move directly to a nitrogen cabinet.
Optional anti-stiction treatment: a self-assembled monolayer or a short fluorocarbon plasma deposition.
Inspect and characterise. Confirm in the SEM that the beam is free and check its resonant response, because an optical image of a released beam and a stuck one look almost identical from above.

Inspection deserves a note. A variable pressure SEM such as the Hitachi SU3500 lets you image insulating and lightly charging samples without sputtering a conductive coating on them. For released MEMS that matters, because a gold coating adds stress to a thin beam and changes the very thing you were trying to measure.

Where university MEMS projects usually stall

No dry release path. Wet HF and a hotplate is not a release process. If the budget cannot stretch to a critical point dryer at the same time as the etcher, design the first devices with large gaps and stiff beams, and buy the dryer in the same financial year.
No metrology between steps. Without a profilometer or an SEM cross-section, nobody knows the etch depth, so recipe development becomes guesswork spread over months.
Recipes developed on wafers, applied to chips. Loading changes the etch rate. A recipe qualified on a full wafer will over-etch or under-etch a diced piece, and students will report the tool as unstable.
Layouts with no release holes and no dimples. Design rules for release are the cheapest fix available and they are almost always missing from a first mask set.
Devices stored in room air. Working parts stop working over a weekend, and the cause is recorded as "device degradation" when it is humidity.

Back to all Insights