Etch Endpoint

An etch endpoint is the precise instant during a plasma or wet chemical removal process when the target thin-film material has cleared across the wafer and exposed the underlying boundary or etch stop layer. In advanced logic below 3nm, high-aspect-ratio 3D NAND memory holes, and gate-all-around nanosheets, etch endpoint detection is not a passive timer: it is a real-time, sensor-driven closed-loop trigger using optical emission spectroscopy (OES), laser interferometry, mass spectrometry, or plasma impedance monitoring to switch recipes from high-rate bulk clearing to high-selectivity soft-landing over-etch. At a high level, how an etch endpoint is identified, filtered, and acted upon in real time looks like this:

A few principles are essential for understanding how etch endpoint systems operate, why blind timed etches guarantee yield collapse, and how modern fabs detect endpoints even at sub-1% open area:

Endpoint marks the onset of clearing, not the end of the etch. Because thin films exhibit across-wafer thickness variations (typically 1% to 3% 3-sigma), plasma density gradients across the 300mm wafer, and micro-loading differences between dense and isolated layout structures, the target film never clears instantaneously everywhere. When an endpoint algorithm registers a signal inflection, the fastest-etching regions have cleared, but slower-etching die (typically at wafer edges or high-aspect-ratio dense trenches) still retain target material. Consequently, triggering an endpoint automatically initiates a controlled "over-etch" phase (typically specified as 5% to 30% of main etch duration) with reduced ion bias energy and elevated selectivity to clear residual stringers without eroding the underlying stop layer.

Reactant consumption vs. byproduct emission defines the optical signature. Optical Emission Spectroscopy (OES) monitors the light emitted by electronically excited neutral radicals and ions in the plasma glow. As the target film is consumed:
1. *Byproduct Emission Drops:* Chemical reaction byproducts escaping into the plasma diminish. For example, during silicon etching with halogen plasmas, the emission intensity of $SiF$ radicals (440 nm) or $SiCl$ (281 nm) drops sharply as silicon clears.
2. *Reactant Emission Rises:* Because the target film no longer consumes plasma radicals, excess unreacted etch gas accumulates in the chamber, causing reactant emission lines (such as atomic fluorine $F^*$ at 703.7 nm or bromine $Br^*$ at 734 nm) to rise precipitously.

Sub-1% open area demands multi-wavelength PCA algorithms. In advanced contact holes, FinFET cut masks, or nanosheet inner spacer etches, the exposed target film constitutes less than 1% of the total wafer surface area (low open area). Under these conditions, the reduction in byproduct emission or increase in reactant signal is buried beneath baseline plasma background noise and chamber wall recombination fluctuations. Modern etch tools employ multi-channel charge-coupled device (CCD) spectrometers monitoring hundreds of wavelengths simultaneously from 200 nm to 900 nm. Principal Component Analysis (PCA) and multivariate regression algorithms combine subtle, correlated intensity variations across multiple spectral lines to extract clear, high-confidence endpoint triggers where single-wavelength monitors fail.

In situ laser interferometry provides cycle-by-cycle depth tracking. For non-selective or blind trenches where no distinct chemical stop layer exists (such as shallow trench isolation silicon recess or recessed contact depth), optical emission cannot detect a material transition because the underlying material is identical to the target. In these applications, tools incorporate normal-incidence laser interferometry (e.g., 670 nm or 980 nm diode lasers directed at dedicated test scribelines). As the film etches, the interference between reflections from the receding surface and the bottom interface produces sinusoidal interference fringes ($2nd\cos\theta = m\lambda$). By counting completed fringes and measuring instantaneous phase shifts, the tool determines etch depth in real time and triggers endpoint precisely at the target depth with single-nanometer accuracy.

Algorithmic filters eliminate false triggers from plasma instability. A real-world plasma chamber is susceptible to instantaneous electrical disturbances: RF impedance matching transitions, electrostatic chuck helium micro-bursts, or chamber wall polymer flaking can cause transient dips in light emission. If an endpoint algorithm relied solely on an absolute intensity threshold ($I < I_{\text{threshold}}$), a 50-millisecond plasma flicker would trigger premature recipe termination, leaving micro-bridges and un-etched vias. Robust algorithms require a conjunction of criteria:
- *Windowing / Delay Time:* Suppressing endpoint evaluation during the initial 15–30 seconds of bulk removal.
- *Derivative Zero-Crossing:* Requiring the first derivative ($dI/dt$) to remain negative and the second derivative ($d^2I/dt^2$) to cross zero at the inflection point.
- *Persistence Time:* Demanding that the signal transition persist for a user-defined verification duration (e.g., 500 ms) before executing the over-etch switch.

Read etch endpoint through a *closed-loop-process-boundary* lens: the primary metric of endpoint capability is not whether it detects the arrival of an interface under ideal blanket test wafers, but how reliably it protects ultra-thin underlying gate oxides and interconnect barriers across wafer non-uniformities and low pattern densities without a single nanometer of over-erosion.