Optical Emission Spectroscopy Endpoint (OES)

1. Theory

• Definition: Optical emission spectroscopy (OES) records light emitted by excited atoms and molecules in a processing plasma, and endpoint detection follows lines of etch products or etchants that change when a film clears (Tokyo Electron patents; Oxford page).
• Emission physics: Line intensity depends on emitter density, electron density and electron temperature, so raw intensities mix chemistry with plasma conditions (background knowledge, Lieberman and Lichtenberg).
• Actinometry: Ratioing a reactive-species line to a rare-gas line of similar excitation energy gives relative atom density (Donnelly abstract).
• Open area: The endpoint change scales with exposed area, so small open areas are hard (Oxford page; Tokyo Electron method patent).

2. Working Principle

• Collection: A viewport and fibre feed a spectrometer covering about 200 to 850 $\mathrm{nm}$ or a filtered photomultiplier at one line (Oxford page; Tokyo Electron detector patent).
• Channels: A product line and a reference band are recorded together and ratioed to remove drift (Tokyo Electron patents).
• Algorithms: Smoothed ratios, derivatives and noise-circle thresholds call slope start and end (Tokyo Electron method patent).
• Over-etch: A timed over-etch follows the detected endpoint (background knowledge).

3. Applications

• Oxide and nitride etch: CO* and N* lines in fluorocarbon plasmas (Tokyo Electron detector patent).
• Chamber cleaning: F* at 704 $\mathrm{nm}$ rising at clean endpoint (Oxford page).
• Radical monitoring: F-atom density by Ar actinometry (Donnelly abstract).
• Fault detection: Spectral fingerprints for chamber matching and drift (background knowledge).

Plasma Emission and Actinometry

Name	Formula	Related Formulae	Variables/Constants
Emission line intensity (corona model)	$I_{\lambda }=K_{opt}{n}_X{n}_e{k}_{exc}\left(T_e\right)\frac{A_{ul}}{\sum A_u}$	$E_{ph}=\frac{hc}{\lambda }$	$I_{\lambda }:lineintensity$ $K_{opt}:opticalcollectionfactor$ $n_X:ground-statedensityoftheemitter$ $n_e:electrondensity$ $\begin{array}{r}{k}_{exc}:electron-impactexcitationrate\\ coefficient\end{array}$ $T_e:electrontemperature$ $A_{ul}:transitionprobability$ $\begin{array}{r}{E}_{ph}:photonenergy;coronamodel\\ (backgroundknowledge,Liebermanand\\ Lichtenberg)\end{array}$ $\begin{array}{r}(computed:5.66\mathrm{eV}at219\mathrm{nm},2.56\mathrm{eV}at\\ 483.5\mathrm{nm},1.84\mathrm{eV}at674\mathrm{nm},1.76\mathrm{eV}at\\ 703.7\mathrm{nm},1.65\mathrm{eV}at750.4\mathrm{nm})\end{array}$
Rare-gas actinometry	$\frac{n_F}{n_{Ar}}=C_{act}\frac{I_{703.7}}{I_{750.4}}$	$C_{act}=\frac{k_{exc,Ar}}{k_{exc,F}}\approx \text{const}$	$n_F,n_{Ar}:fluorineandargondensities$ $I:lineintensities$ $\begin{array}{r}{C}_{act}:actinometricconstant;Donnelly\\ abstract:F703.7\mathrm{nm}andAr2p1750.4\mathrm{nm},\\ similarrelativeenergydependenceofthe\\ twoexcitationcrosssections,negligible\\ dissociativeexcitationinstrongly\\ dissociatedCF4/O2,SF6andNF3plasmas,\\ methodfirstpublishedbyCoburnandChen;\\ ratioform(backgroundknowledge,Coburn\\ andChen)\end{array}$ $\begin{array}{r}(myreading:becausebothexcitationrates\\ scalealikewith{T}_e,theratiocancels\\ changesinelectrondensityand\\ temperature)\end{array}$
Endpoint species and lines	$\mathrm{Si}{\mathrm{O}}_2:\mathrm{C}{\mathrm{O}}^{\ast }219\text{or}483.5\mathrm{nm};\mathrm{S}{\mathrm{i}}_3{\mathrm{N}}_4:{\mathrm{N}}^{\ast }674\mathrm{nm}$	${\text{F}}^{\ast }704\mathrm{nm}\text{for chamber clean}$	$\begin{array}{r}C{O}^{\ast },N^{\ast },F^{\ast }:excitedemitters;Tokyo\\ Electrondetectorpatent:CO\ast at219\mathrm{nm}\\ and483.5\mathrm{nm}foroxideinCF4,N\ast at674\\ \mathrm{nm}fornitride,referencechannelon\\ CF-seriesemission,0.1\mathrm{mTorr}toseveral\\ \mathrm{mTorr};TokyoElectronmethodpatent:COat\\ about219\mathrm{nm}andCF2\ast atabout260\mathrm{nm},\\ priorartwithHe706.5and667.8\mathrm{nm}\\ references;Oxfordpage:704\mathrm{nm}\\ fixed-wavelengthunitforchamber-clean\\ endpoint,fluorinerisingsharplyat\\ endpoint\end{array}$ $\begin{array}{r}(myreading:reactionproductssuchasCO\\ fallatendpointwhileetchantlinessuch\\ asFrise,sothetwomoveinopposite\\ directions)\end{array}$

Endpoint Signals

Name	Formula	Related Formulae	Variables/Constants
Open-area scaling	$\frac{\Delta I}{I}\approx {\left(\frac{\Delta I}{I}\right)}_{100}\frac{A_{open}}{A_{wafer}}$	$A_{open}\gtrsim \text{several}\mathrm{c}{\mathrm{m}}^2$	$\begin{array}{r}\Delta I/I:fractionallinechangeatendpoint,\\ (\Delta I/I{)}_{100}:changeforafullyopenwafer\end{array}$ $\begin{array}{r}{A}_{open}:exposedarea;Oxfordpage:OES\\ endpointtypicallyneedsseveralsquare\\ centimetres,dependingonetchrateand\\ emissionintensity;TokyoElectronmethod\\ patent:detectableapertureratioimproved\\ threefold(statedas1to3percent);\\ proportionalscaling(background\\ knowledge)\end{array}$ $\begin{array}{r}(computed:a300\mathrm{mm}waferis707square\\ cm,so1percentis7.1and3percent21\\ squarecm,and3.5squarecmis0.5per\\ cent)\end{array}$
Two-channel drift removal	$S=\frac{S_1}{S_2}$	$\text{product line}\downarrow ,\text{etchant line}\uparrow \Rightarrow S\downarrow \downarrow$	$S_1:product-linesignal$ $\begin{array}{r}{S}_2:reference-bandsignal;TokyoElectron\\ detectorpatent:secondchannelusedto\\ removedriftfromthefirst,CO\ast falling\\ andCF-seriesemissionrisingatendpoint,\\ photomultiplieronthefirstchannel;\\ ratioform(backgroundknowledge)\end{array}$ $\begin{array}{r}(myreading:common-modechangesinRF\\ powerorwindowcloudingcancelinthe\\ ratio,whiletheendpointchangeadds)\end{array}$
Linear-fit normalized ratio	$A^{\prime }=\frac{a_1}{a_2}\left(Y_2-b_2\right)+b_1,R_n=\frac{A}{A^{\prime }}$	$\frac{D}{C}>T_{start}\Rightarrow \text{slope start}$	$\begin{array}{r}{Y}_1=a_{1}X+b_1,Y_2=a_{2}X+b_2:least-squaresfitsof\\ thetwoemissionsoveranaveragingperiod\end{array}$ $R_n:normalizedratio$ $\begin{array}{r}D:distancefromtheoriginintheplane\\ of{R}_{n}andd{R}_n/dt\end{array}$ $\begin{array}{r}C:radiusofthevariationcirclefromthe\\ maximumdeviations\end{array}$ $\begin{array}{r}{T}_{start}:threshold;TokyoElectronmethod\\ patent:stepsaslisted,slopeendwhen\\ thenormalizedslopedeviationfallsbelow\\ athreshold,repeatableformulti-step\\ etches\end{array}$ $\begin{array}{r}(myreading:thefitremovestheslow\\ driftofeachchannelbeforetheratiois\\ formed,soonlytheendpointtransition\\ leavesthenoisecircle)\end{array}$

Signal Processing and Timing

Name	Formula	Related Formulae	Variables/Constants
Sampling and filtering	$f_s=\frac{1}{\Delta t_s},f_{Nyq}=\frac{f_s}{2}$	${\sigma }_{\bar{x}}=\frac{\sigma }{\sqrt{N}}$	$f_s:samplingrate$ $\Delta t_s:samplinginterval$ $\begin{array}{r}{f}_{Nyq}:Nyquistfrequency,{\sigma }_{\bar{x}}:noiseafter\\ averagingNsamples;TokyoElectronmethod\\ patent:samplingevery0.1s,removalof\\ noiseattwicethesamplingcycleor\\ higherstatedas20\mathrm{Hz}ormore,dynamic\\ averaging;relations(background\\ knowledge)\end{array}$ $\begin{array}{r}(computed:0.1sgives10\mathrm{Hz}anda5\mathrm{Hz}\\ Nyquistlimit;0.1percentnoiseaveraged\\ over10samplesfallsto0.032percent)\end{array}$
Endpoint signal-to-noise	$SN{R}_{EP}=\frac{\Delta I/I}{{\sigma }_{\bar{x}}}$	$\delta t_{EP}=\frac{{\sigma }_{\bar{x}}}{|dS/dt|}$	$SN{R}_{EP}:endpointsignal-to-noise$ $\begin{array}{r}\delta t_{EP}:timinguncertainty;relations\\ (backgroundknowledge)\end{array}$ $\begin{array}{r}(assumed:50percentchangeforafully\\ openwafer;computed:0.5percentat1\\ percentopenareaandSNRof16with\\ 0.032percentnoise)\end{array}$
Clearing spread and over-etch	$\Delta t_{clear}=\frac{\Delta d}{ER}$	$t_{OE}=f_{OE}{t}_{EP}$	$\Delta t_{clear}:spreadinclearingtime$ $\begin{array}{r}\Delta d:rangeoffilmthicknessand\\ etch-depthvariation\end{array}$ $ER:etchrate$ $t_{OE}:over-etchtime$ $f_{OE}:over-etchfraction$ $\begin{array}{r}{t}_{EP}:endpointtime;relations(background\\ knowledge,LiebermanandLichtenberg)\end{array}$ $\begin{array}{r}(assumed:500\mathrm{nm}filmwithplusorminus2\\ percent,500nm/min,20percent\\ over-etchof60s;computed:2.4sspread,\\ 12sover-etch)\end{array}$

Important Questions

Ques: How does OES detect the end of an oxide etch?

By watching a reaction-product line fall as the film clears. In a fluorocarbon plasma, oxygen from SiO2 forms CO, whose excited emission at 219 or 483.5 $\mathrm{nm}$ is strong while oxide is exposed and drops when the underlying layer appears (Tokyo Electron detector patent). The patent adds a second channel on CF-series emission, which rises at endpoint because fluorocarbon is no longer consumed, and uses it to remove drift from the CO signal. For nitride, N* at 674 $\mathrm{nm}$ plays the same role. Raw intensity follows $n_X{n}_e{k}_{exc}(T_e)$ (background knowledge, Lieberman and Lichtenberg), so ratioing two lines cancels much of the variation in RF power and window transmission (my reading). A timed over-etch then follows. Reference: S. Saito, ‘Endpoint detector for plasma etching’, US patent 5,888,337, Tokyo Electron (filed May 1996) (fetch-tool summary); M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges and Materials Processing, 2nd ed., Wiley 2005, and J. W. Coburn and M. Chen, J. Appl. Phys. 51, 3134 (1980) (background knowledge; not opened).

Ques: Which CO line is better for endpoint: 219 nm or 483.5 nm?

Both appear in the sources. The Tokyo Electron detector patent lists CO* at 219 $\mathrm{nm}$ and at 483.5 $\mathrm{nm}$, and the method patent uses about 219 $\mathrm{nm}$ in its embodiment while citing 483.5 $\mathrm{nm}$ in prior art. The photon energies are 5.66 and 2.56 $\mathrm{eV}$ (computed). The 483.5 $\mathrm{nm}$ band lies in the visible, where ordinary viewports, fibres and silicon detectors work well, while 219 $\mathrm{nm}$ needs UV-transmitting quartz and is more sensitive to window deposits (background knowledge). On the other hand, the 219 $\mathrm{nm}$ band sits away from strong visible CF and F emission. For production tools with fibre-coupled spectrometers, 483.5 $\mathrm{nm}$ is more likely the common choice, while the patents’ 219 $\mathrm{nm}$ suits filtered UV photomultipliers (my reading). Reference: S. Saito, ‘Endpoint detector for plasma etching’, US patent 5,888,337, Tokyo Electron (filed May 1996) (fetch-tool summary); C. Koshimizu and S. Saito, ‘Method of detecting end point of plasma processing and apparatus for the same’, US patent 5,928,532, Tokyo Electron (filed November 1997) (fetch-tool summary).

Ques: What is rare-gas actinometry?

A way to turn emission intensity into relative atom density. A small, known fraction of a rare gas is added, and the reactive-species line is divided by a rare-gas line whose excitation has a similar energy dependence, so changes in electron density and temperature cancel (Donnelly abstract; background knowledge, Coburn and Chen). For fluorine, the F line at 703.7 $\mathrm{nm}$ is ratioed with the Ar 2p1 line at 750.4 $\mathrm{nm}$; Donnelly calls this one of the most reliable cases because the cross sections match and dissociative excitation from other F-containing species is negligible in strongly dissociated CF4/O2, SF6 or NF3 plasmas. The photon energies are 1.76 and 1.65 $\mathrm{eV}$ (computed). The Oxford 704 $\mathrm{nm}$ chamber-clean unit monitors the same fluorine line. Reference: V. M. Donnelly, ‘Rare Gas Actinometry Turns Thirty Nine and is Still Finding Applications’, AVS 66th International Symposium, paper PS-WeA3 (2019) (fetch-tool summary of the abstract); M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges and Materials Processing, 2nd ed., Wiley 2005, and J. W. Coburn and M. Chen, J. Appl. Phys. 51, 3134 (1980) (background knowledge; not opened); Oxford Instruments Plasma Technology, ‘Optical end point detectors’ upgrade page (fetch-tool summary).

Ques: Why is endpoint hard for low open area etches?

Because the endpoint change scales with exposed area. The Oxford page states that OES endpoint typically needs several square centimetres, depending on etch rate and emission intensity. A 300 $\mathrm{mm}$ wafer is 707 square centimetres, so 3.5 square centimetres is only 0.5 per cent open area (computed). If a fully open wafer changed a product line by an assumed 50 per cent, 1 per cent open area gives 0.5 per cent, and with 0.1 per cent noise averaged over 10 samples, the signal-to-noise is about 16 (assumed; computed). The Tokyo Electron method patent claims a threefold improvement in detectable aperture ratio, stated as 1 to 3 per cent, which is ambiguous; it most likely means the minimum detectable open area falls by three times (my reading). Reference: Oxford Instruments Plasma Technology, ‘Optical end point detectors’ upgrade page (fetch-tool summary); C. Koshimizu and S. Saito, ‘Method of detecting end point of plasma processing and apparatus for the same’, US patent 5,928,532, Tokyo Electron (filed November 1997) (fetch-tool summary).

Ques: How does the noise-circle method call endpoint?

By detecting when a normalized ratio leaves its noise band. The Tokyo Electron method patent fits each channel with a least-squares line during an averaging period, forms an affine copy of the second channel matched to the first, and computes the ratio and its time derivative. Plotted against each other with the averaging-period means as origin, normal operation stays inside a circle whose radius is the root-sum-square of the maximum deviations; the slope start is called when the normalized distance exceeds a threshold, and slope end when the normalized slope returns near the mean. The patent samples every 0.1 s, a 10 $\mathrm{Hz}$ rate with a 5 $\mathrm{Hz}$ Nyquist limit (computed), so its stated 20 $\mathrm{Hz}$ filter likely acts before sampling (my reading). Reference: C. Koshimizu and S. Saito, ‘Method of detecting end point of plasma processing and apparatus for the same’, US patent 5,928,532, Tokyo Electron (filed November 1997) (fetch-tool summary).

Ques: How is over-etch time chosen after endpoint?

From the spread in clearing time across the wafer. Endpoint marks when most of the film clears, but thickness and etch-rate nonuniformity leave thinner and thicker regions clearing at different times (background knowledge, Lieberman and Lichtenberg). For an assumed 500 $\mathrm{nm}$ film with plus or minus 2 per cent variation etched at 500 $nm/min$, clearing spreads over 2.4 s, and a typical assumed 20 per cent over-etch of a 60 s main etch adds 12 s (assumed; computed). The endpoint transition in the ratio signal broadens with this spread, so the noise-circle start and end points in the Tokyo Electron method bracket the clearing interval. Selectivity to the underlying layer sets how much over-etch is acceptable (background knowledge). Reference: M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges and Materials Processing, 2nd ed., Wiley 2005, and J. W. Coburn and M. Chen, J. Appl. Phys. 51, 3134 (1980) (background knowledge; not opened); C. Koshimizu and S. Saito, ‘Method of detecting end point of plasma processing and apparatus for the same’, US patent 5,928,532, Tokyo Electron (filed November 1997) (fetch-tool summary).

References

1. S. Saito, “Endpoint detector for plasma etching,” US patent 5,888,337 A, Tokyo Electron, filed May 17, 1996. Available: https://patents.google.com/patent/US5888337
2. M. A. Lieberman and A. J. Lichtenberg, Principles of Plasma Discharges and Materials Processing, 2nd ed., Wiley, 2005; J. W. Coburn and M. Chen, “Optical emission spectroscopy of reactive plasmas: a method for correlating emission intensities to reactive particle density,” J. Appl. Phys. 51, 3134 (1980) (cited from background knowledge).
3. C. Koshimizu and S. Saito, “Method of detecting end point of plasma processing and apparatus for the same,” US patent 5,928,532 A, Tokyo Electron, filed November 3, 1997. Available: https://patents.google.com/patent/US5928532
4. V. M. Donnelly, “Rare Gas Actinometry Turns Thirty Nine and is Still Finding Applications,” AVS 66th International Symposium and Exhibition, paper PS-WeA3, 2019. Available: https://www2.avs.org/symposium2019/Papers/Paper_PS-WeA3.html
5. Oxford Instruments Plasma Technology, “Optical end point detectors.” Available: https://plasma.oxinst.jp/service-support/upgrades/optical-end-point-detectors

Related pages in this vault: UV-Vis Spectrophotometer, Photoluminescence Spectrometer, Spectral Reflectometer, Contacts and Vias, Optical CD Scatterometer (OCD), Patterned-Wafer Optical Inspection, Langmuir Probe, Residual Gas Analyzer (RGA), Quartz Crystal Microbalance (QCM), Laser Interferometric Endpoint, Vacuum Gauges, Plasma and Radical Oxidation Tool, DC and RF Magnetron Sputter System, PECVD Reactor, Plasma Asher (Barrel and Downstream), Parallel-Plate Reactive Ion Etcher (RIE), Inductively Coupled Plasma Etcher (ICP-RIE).