Deep reactive ion etching projects rarely fail at the equipment. They fail earlier — in the gap between what the device team assumed and what the equipment discussion actually covered. A platform is chosen against an incomplete picture, the demo tests the wrong thing, and the disagreement surfaces six months later during qualification, when it is most expensive.
The fix is unglamorous: define the application completely before the first platform conversation. These are the twelve inputs we ask for, and why each one matters.
Geometry
1. Target depth and tolerance. Not “deep” — a number, with the tolerance your device can accept. A 50 µm cavity at ±3 % and a through-wafer via at ±10 % are different projects.
2. Critical dimension and pitch. The mask opening sizes, the smallest and the largest, and how densely they are packed. Aspect-ratio-dependent etching (ARDE) means a chamber that is uniform on isolated 100 µm openings may not be on dense 5 µm trenches — the etch rate a vendor quotes was measured on some geometry, and it was probably not yours.
3. Profile requirement. Sidewall angle target and what failure looks like: bowing, tapering, undercut at the mask, notching at a buried interface. State which of these your device actually cares about; they trade against each other.
4. Sidewall texture. Scallop size matters if the sidewall becomes an optical surface, a bonding surface, or a liner interface. If it does not matter, say so — smoothness costs rate.
Materials
5. Substrate and stack. Bulk silicon, SOI with a buried oxide, bonded stacks, or something already carrying devices. SOI brings notching behaviour at the oxide; carrier-bonded wafers bring thermal and handling constraints.
6. Mask material and budget. Photoresist, oxide, or metal — with thickness. Selectivity determines whether your mask survives the full depth, and re-masking mid-project is not a plan.
7. What else is exposed. Metals, polymers, films on the backside. The etch chemistry and the plasma see everything, not just the pattern.
Production context
8. Wafer size and format. Including thickness, warp, and whether wafers arrive on carriers or tape frames. Handling capability is a configuration decision made at purchase, not a recipe parameter.
9. R&D or production — honestly. A tool bought “for research, but we will ramp” needs the production conversation now: automation level, chamber matching, uptime expectations.
10. Throughput at specification. Wafers per hour at your profile and uniformity requirement, not at the relaxed condition where headline rates are measured.
Acceptance
11. The metrics that decide acceptance. Depth uniformity across the wafer, CD loss, profile angle range, defect criteria — written down, with the measurement method for each. If you cannot measure it, it cannot be an acceptance criterion.
12. The failure mode you are escaping. If this project replaces or supplements an existing process, describe how the current one fails. That single input often does more to focus a platform discussion than everything else combined.
What this buys you
With these twelve inputs, an equipment manufacturer’s process team can give you a real answer: a platform, a configuration, and a credible statement of what a demo can prove. Without them, you get a brochure.
The list also protects you commercially. Every input above becomes a line in the demo success criteria and, later, in the acceptance test. Ambiguity at the start is a discount you give away for free.
Preparing a deep silicon etch evaluation? We help teams turn device requirements into equipment questions manufacturers can answer precisely — start with your application.
