What Is an Etch HfO2 Process Oes Endpoint Algorithm
An etch hafnium oxide process Optical Emission Spectroscopy endpoint algorithm (etch $HfO_2$ process OES endpoint algorithm) is a specialized multi-spectral trace-element diagnostic and statistical change-point detection engine engineered to govern the atomic-layer clearance of high-$k$ gate dielectrics in advanced CMOS nodes, ferroelectric memories (FeRAM), and monolithic 3D NAND stacks. In cutting-edge sub-5nm FinFET, GAA nanosheet, and CFET architectures, the gate dielectric consists of an ultra-thin layer (typically 1.0 nm to 2.5 nm) of hafnium dioxide ($HfO_2$) or silicon-doped hafnium oxide ($Hf_{1-x}Si_xO_2$) deposited over an ultra-thin interfacial silicon dioxide ($SiO_x$) or silicon germanium ($SiGe$) channel. Removing $HfO_2$ from non-gate regions (such as I/O transistor zones or replacement metal gate dummy recess) requires inductively coupled plasmas (ICP) using boron trichloride and chlorine chemistries ($BCl_3/Cl_2/Ar$). Because the $HfO_2$ film is only 5 to 10 atomic monolayers thick, volatile hafnium halide reaction byproducts ($Hf^*$ and $HfCl^*$) generate exceptionally faint, transient optical emission lines that emerge and vanish in a matter of seconds. An advanced $HfO_2$ OES endpoint algorithm combines high-gain ultraviolet-visible spectrometer channels, multi-line hafnium-to-boron ratioing ($Hf^*/BCl^*$), background polynomial baseline subtraction, and real-time inflection detection to command an instantaneous plasma quench before the underlying 0.8 nm interfacial oxide is breached. At a high level, how an $HfO_2$ OES endpoint algorithm tracks hafnium depletion and protects the channel looks like this:
A few principles are essential for understanding how $HfO_2$ OES endpoint algorithms operate, why the thermodynamics of hafnium reduction dictate specific chemistry, and how trace atomic emission lines are reliably resolved against intense background plasma glow:
Thermodynamics and spectral signatures in $BCl_3/Cl_2$ high-$k$ etching. Hafnium dioxide possesses one of the highest metal-oxide bond enthalpies in materials science ($\Delta H_{\text{bond}} \approx 801\text{ kJ/mol}$ for $Hf-O$, compared to $799\text{ kJ/mol}$ for $Si-O$). Pure chlorine ($Cl_2$) or fluorine ($CF_4$) plasmas cannot spontaneously volatilize $HfO_2$ at low wafer temperatures. Therefore, boron trichloride ($BCl_3$) is required as an oxygen scavenger, chemically reducing the oxide while chlorinating hafnium:
In the ICP glow, electron-impact excitation dissociates these volatile products into distinct optical emitters:
- *Trace Hafnium Byproducts (Plummet at Endpoint):* Neutral hafnium atomic line $Hf^*$ emits at 343.82 nm and 368.22 nm ($5d^2 6s^2 \rightarrow 5d^2 6s 6p$). Singly ionized hafnium ($Hf^+$) emits at 339.98 nm. Hafnium monochloride ($HfCl^*$) emits faint bands around 385–405 nm. Because the film is only 1.5 nm thick, the peak signal is tiny ($<3\%$ above baseline).
- *Boron Halide Reactants (Surge at Endpoint):* Boron monochloride ($BCl^*$) radical emits a sharp, prominent molecular band system at 272.0 nm ($A\,^1\Pi \rightarrow X\,^1\Sigma^+$). Atomic boron ($B^*$) emits doublet lines at 249.7 nm and 249.8 nm. When the atomic layer of $HfO_2$ clears, boron trichloride consumption ceases abruptly, causing $BCl^*$ emission at 272.0 nm to surge upwards.
Cross-line ratioing ($Hf^*/BCl^*$) and real-time polynomial baseline subtraction. Because hafnium emission intensity is inherently small, any baseline shift caused by chamber wall heating, RF power fluctuations, or viewport clouding can obscure the clearing step. The algorithm combines dynamic baseline subtraction with cross-line ratioing:
1. *Local Polynomial Subtraction:* The algorithm samples two adjacent non-emitting spectral continuum windows ($\lambda_{\text{ref1}} = 341.5\text{ nm}$ and $\lambda_{\text{ref2}} = 346.0\text{ nm}$) surrounding the $Hf^*$ 343.8 nm peak, fitting a localized linear or polynomial baseline $B(\lambda, t)$ and subtracting it from the raw peak intensity:
2. *Dual-Line Ratio Formation:* The baseline-subtracted hafnium signal is divided by the surging boron monochloride reactant line:
This creates a sharp downward ratio cliff with an amplified signal-to-noise ratio ($SNR > 24\text{ dB}$), completely rejecting chamber seasoning artifacts.
Second-derivative inflection zero-crossing and sub-50ms recipe cutoff. The total etch duration of an ALD $HfO_2$ layer is often fewer than 10 to 15 seconds. If over-etch persists for even 500 ms, energetic $BCl_2^+$ ions punch through the 0.8 nm interfacial $SiO_2$ layer and cause catastrophic silicon or $SiGe$ channel recess. The algorithm implements real-time Savitzky-Golay filtering to monitor the second derivative $d^2 R_{Hf} / dt^2$:
Once zero-crossing is locked and verified across three consecutive sampling intervals (15 ms), the chamber controller triggers an emergency recipe abort: RF source power is dropped instantly to zero, bias power is deactivated, and chamber pressure is purged with inert helium, stopping the etch with atomic-scale precision.
Read an $HfO_2$ etch process OES endpoint algorithm through an *atomic-tripwire* lens: operating on the absolute frontier of high-$k$ metal oxide etching, it senses the fading whisper of hafnium atoms amid a blizzard of boron radicals to switch off the plasma furnace before a single monolayer of channel oxide is harmed.