Gate-All-Around GAA Nanosheet Channel Release Sige Selective Etch MBCFET Ribbonfet
Where finfet-fin-patterning-sadp-saqp-process starts from a single-crystal silicon fin etched straight out of the substrate, the gate-all-around nanosheet architecture starts from something that has to be built before it can be patterned at all: an epitaxial superlattice stack of alternating silicon and silicon-germanium layers, typically three to four silicon "sheets" each a few nanometers to around eight nanometers thick, separated by sacrificial silicon-germanium layers of similar thickness, grown one on top of the other on the substrate before any fin-like shape exists. This stack is the raw material the rest of the process operates on, and the two materials play opposite roles from the moment they are grown: the silicon layers are the future transistor channels, while the silicon-germanium layers exist purely to be removed later and never carry current themselves. That sacrificial role is worth stating plainly because this batch has already used silicon-germanium for a completely different job in source-drain-epitaxy-strained-channel-process, where silicon-germanium is grown directly in the source/drain regions specifically to stay in the finished device and mechanically strain the channel for mobility; here, inside the gate-all-around stack, silicon-germanium is grown specifically so that it can be etched away again, and confusing the two uses, strain-engineering silicon-germanium that stays versus superlattice silicon-germanium that is sacrificed, is one of the easiest ways to misread how this transistor is actually built.
Once the superlattice is grown, the stack is patterned much the way a fin is patterned, an anisotropic etch cutting vertical sidewalls down through the full alternating stack to define the lateral shape the eventual nanosheet channels will share, conceptually the same fin-patterning operation named in finfet-fin-patterning-sadp-saqp-process but now cutting through several alternating material layers instead of one uniform crystal. A dummy gate and spacer are then formed over the stack using the same replacement-gate strategy already established in high-k-metal-gate-hkmg-process, a sacrificial placeholder gate that protects the channel region through the high-temperature steps that follow and gets pulled out later, and source/drain epitaxy is grown at the exposed ends of the stack using the strain-engineering concept from source-drain-epitaxy-strained-channel-process, except now that epitaxial growth has to make electrical contact with every silicon sheet in the stack simultaneously rather than with a single fin, merging around the stack ends to form one continuous source or drain region that all the sheets terminate into.
The step that actually defines this architecture happens after the dummy gate is pulled out: with the Si/SiGe stack exposed inside the gate cavity, a highly selective lateral etch, typically a dry or wet chemistry chosen to attack silicon-germanium while leaving silicon essentially untouched, removes only the sacrificial silicon-germanium layers from between the silicon sheets. Because that selectivity has to hold laterally, reaching in from the gate cavity sidewalls to clear germanium out from between sheets that may only be a few nanometers apart, without nicking the silicon sheets themselves or undercutting the source/drain epitaxy at the stack ends, this release etch is one of the most selectivity-critical steps in the entire flow: insufficient selectivity leaves germanium residue that blocks gate material from fully wrapping the sheet, while excessive aggressiveness can thin or damage the suspended silicon itself. Once the silicon-germanium is cleared, what remains are silicon nanosheets suspended in the gate cavity, mechanically anchored only at their source and drain ends, with open space on all four sides, top, bottom, and both lateral faces, where the sacrificial layer used to be. The high-k dielectric and metal gate fill, deposited with the same atomic-layer-deposition and conformal-metal-fill concept already used in high-k-metal-gate-hkmg-process, now has to flow into those narrow inter-sheet gaps and coat every exposed surface, which is what "gate-all-around" literally means: the gate material fully encircles the channel's perimeter, a four-sided wrap, rather than draping over three sides of a standing fin the way a FinFET's gate does.
| Characteristic | FinFET (three-sided gate wrap) | GAA nanosheet (four-sided gate wrap) |
|---|---|---|
| Electrostatic control | Strong, but channel bottom remains gate-uncontrolled, limiting subthreshold swing and DIBL improvement versus planar | Full channel perimeter under gate control; further improves subthreshold swing and DIBL beyond FinFET |
| Effective channel width tunability | Fixed by fin height set during patterning; width scaling requires adding more fins | Nanosheet width and sheet count per stack are independent design knobs, not tied to a fixed vertical fin height |
| Process complexity | Mature flow: fin pattern, STI reveal, replacement gate, strain epi | Adds epitaxial superlattice growth and a highly selective lateral SiGe release etch on top of the same replacement-gate flow |
| Adoption node / company | Dominant from the 2010s FinFET generation through late sub-10nm nodes | Samsung MBCFET at 3nm, Intel RibbonFET at 20A/18A-class, TSMC's planned N2 nanosheet node |
That four-sided wrap is the direct answer to a limitation FinFETs could not design around: a fin's height is fixed once patterning locks it in, so increasing a FinFET's effective drive width means adding more fins side by side, while a gate-all-around nanosheet stack lets designers tune channel width and the number of stacked sheets independently of any fixed vertical dimension, giving the standard-cell and SRAM design teams a continuously adjustable sizing knob that FinFET's fin-counting approach never offered. That flexibility does not come for free: growing a multi-layer epitaxial superlattice before patterning even begins, and then performing a release etch selective enough to clear germanium from gaps only a few nanometers wide without damaging the silicon sheets it leaves behind, adds real process steps and tighter selectivity windows on top of everything FinFET already required, which is part of why the industry's move to gate-all-around tracks closely with the move to the most advanced nodes, where the electrostatic and width-tunability gains are judged worth that added complexity.
Gate-all-around nanosheet is the FinFET-era toolkit's next device-architecture generation, not a wholesale replacement of it: the replacement-gate concept from high-k-metal-gate-hkmg-process, the strained source/drain epitaxy concept from source-drain-epitaxy-strained-channel-process, and the anisotropic stack-patterning etch that echoes finfet-fin-patterning-sadp-saqp-process all carry forward essentially unchanged in purpose, they are simply applied around a fully suspended, multi-sheet silicon channel instead of around a single standing fin. What is genuinely new is the epitaxial superlattice growth step and the highly selective lateral silicon-germanium release etch that turns that stack into suspended sheets, a pair of steps with no FinFET equivalent, needed specifically because wrapping a gate around all four sides of a channel first requires suspending that channel in open space rather than leaving it anchored to the substrate the way a fin is. Extending the same gate-electrostatics logic that drove the planar-to-FinFET transition one generation further, from three sides of control to four, gate-all-around nanosheet closes out the fin-based architectural arc this batch has traced and opens the industry roadmap's current leading edge at Samsung's, Intel's, and TSMC's most advanced nodes.