Advanced Logic
Enabling the atomic-scale etch precision that makes today's most advanced AI processors manufacturable
- FinFET
- GAA
GAA changed everything.
The etch process has to keep up.
The processors that run AI are built at sub-3nm. The etch process that makes them possible demands atomic-scale precision.
As the industry moves from FinFET to Gate-All-Around (GAA), the process complexity at the etch level increases significantly. Conventional RIE cannot meet the precision these architectures demand – and the gap only widens as nodes shrink.
Where AAT Fits
Building GAA structures requires removing sacrificial silicon-germanium (SiGe) layers from between stacked silicon sheets with extremely high selectivity to silicon. The process window is too narrow, the surfaces too sensitive, and the tolerance for damage essentially zero. Isotropic ALE is the only process that makes nanosheet release possible.
Beyond nanosheet release, inner spacer formation, self-aligned contact trimming, and backside power delivery structures all introduce new etch steps where atomic-scale control is the requirement, not a preference.
The gap widens as nodes shrink
- AAT ALEMeets the precision requirement
- Conventional RIECannot keep up
AAT's platform handles these precision finishing steps – not replacing high-throughput RIE for bulk removal, but deployed at the steps where surface quality and dimensional accuracy are non-negotiable. Because the AI optimizer is running in real time, process engineers spend less time on recipe development and more time on device results. At a node where process windows are measured in angstroms, that speed of iteration is not a convenience – it's a competitive advantage.
Demonstrated Performance
- Synergy Factor>80%
- Linearity (R²)>0.9947
- Cycle Time~2 seconds
- Modification saturation~2.0 Å at 1 second
- Removal saturation~2.1 Å at 1 second
Source: SPIE Advanced Lithography + Patterning 2026, Paper 13984-24
The AAT Advantage
Purpose-built for precision, not retrofitted
AAT's decoupled chamber architecture independently controls the modification and removal steps – giving process engineers a level of step-level tunability that coupled-source tools cannot match.
AI-native process control
Machine learning models optimize gas flow, power, and timing parameters continuously – as an active part of how the tool runs. For GAA development, where the margin for error is essentially zero and manual recipe tuning can take weeks, real-time AI control means faster convergence to a production-worthy process window with tighter within-wafer uniformity.
Production-relevant cycle times
~2 second cycle time – approximately 10× faster than conventional ALE tools – means atomic precision no longer comes at a throughput penalty.