AVS2009 Session PS2-ThA: Plasma Diagnostics, Sensors, and Control II

Dual–frequency capacitively-coupled etch reactors using Ar/fluorocarbon/O₂ mixtures are widely employed for etching of dielectric films for integrated circuit manufacture. CF_x free radicals play an important role in the gas-phase and surface chemistry controlling etching and polymer deposition. The CF₂ radical is the most abundant, and its density can be measured by UV absorption via the A-X band (230-270 nm). Previously Xe arc lamps have been used for the absorption light source, but these sources are rather unstable, limiting the sensitivity of the technique, as well as being cumbersome and relatively expensive. We have successfully replaced the Xe arc with UV light-emitting diodes. The baseline stability is of the order 2x10^-4, compared to 1x10^-3 with an arc lamp. We determined the variation of the CF₂ density as a function of gas composition and power in a modified 2 + 27MHz commercial etch reactor operating in Ar/C₄F₈/O₂. As expected, the CF₂ density decreases rapidly as the O2/C4F8 ratio is increased. The CF₂ density increases with RF power at both frequencies, but is most affected by 27 MHz power. There is speculation that CF₂ may play an important role in either or both the creation and destruction of F- negative ions. However, we did not find any simple correlation between CF₂ density and electronegativity (as determined from electron density and ion flux measurements).

We also attempted to measure the F^- negative ion density by the continuum absorption below 365 nm. However in this case we observed a broad but structured absorption when the O2/C4F8 ratio is small. This absorption is too intense to be attributed to F^-, and we speculate that it is due to unsaturated CxFy oligomerisation products.

We wish to thank the Lam Foundation for financial support.

In low-pressure discharges commonly used for materials processing, neutral gas temperature are routinely determined from the rovibronic structure of N₂ inserted as a tracer. This is realized by comparing the measured emission intensities of the bandhead and the violet-degraded tail of the second positive system of N₂ (C³Π_u ν’ -> B³Π_g ν’’) to the prediction of a model with the rotational temperature, T_rot, as the adjustable parameter. Such measurements are usually performed using the most intense (ν’,ν’’)=(0,0) and (0,2) bands at 337.1 and 380.5 nm. However, in argon-containing plasmas, these bands typically yield temperatures larger than those obtained from other methods such as Doppler-shifted laser-induced fluorescence (D-LIF). Hypotheses for such discrepancy vary; either the emitting C³Π_un’=0 level (11.026 eV above the ground state) could be populated by the ³P₀ and ³P₂ argon metastables (11.723 and 11.548 eV above the ground state) or the spatial non-uniformity of the plasma could skew D-LIF measurements. In this work, we examined the influence of Ar metastables on the rotational temperature of N₂. We compared T_rot values obtained from different N₂ bands, notably (0,0), (0,2), (1,0), and (4,2) to a less conventional plasma sampling mass spectrometry (PSMS) technique in which the Ar plasma on-to-plasma off signal intensity ratio is linked to the gas temperature. These measurements were performed in an inductively coupled 98%Ar/2%N₂ plasma as a function of pressure and absorbed power. We show that N₂ bands with ν’≤2 generated much higher T_rot values than the (4,2) band or the PSMS. For example, for a 20 mTorr, 1000 W Ar plasma, the (0,0), (0,2), (1,0), (4,2), and PSMS yielded temperatures of 973, 715, 920, 485, and 415 K, respectively. We then computed the reaction rates for excitation of the C³Π_u ν’=0 level by collisions with Ar metastables, R-Ar^m, and by electron-impact, R-e, from the ground state. The densities required as inputs for those calculations were measured by Langmuir probe for electrons and, for the metastables, were determined from a global model. We found that that the ratio of the temperatures obtained from the (0,0) and (1,0) bands to that of the (4,2) band increased quasi-linearly with the R-Ar^m-R-e ratio, going from 1.5 to 2 as the Ar metastable-to-electronic excitation rate ratio increased from 0.01 to 0.1. Since Ar metastables can have a strong influence on rotational temperatures even for R-Ar^m-R-e ratio as low as 0.01, accurate gas temperature measurements in mostly Ar plasmas thus require analysis of bands for which ν’≥3 as these levels (e.g. 11.74 eV for C³Π_u ν’=4) are above the ³P₀ Ar metastable level.

Dual-frequency capacitively coupled plasmas (2f-CCP) used in the fabrication of modern integrated circuits may provide quasi-independent control of ion flux and energy. The accurate determinations of the electron temperature (T_e) and the electron energy distribution function (EEDF) are important for understanding plasma behavior and optimizing plasma processes in 2f-CCPs. In this study, measurements of T_es and EEDFs in CF₄/O₂ plasmas generated in a 2f-CCP etcher were performed as a function of pressure, applied RF power, and O₂ feed gas content by using trace rare gases-optical emission spectroscopy (TRG-OES). The parallel plate etcher was powered by a high frequency (60 MHz) “source” top electrode, and a low frequency (13.56 MHz) “substrate” bottom electrode. 80%CF₄+20%O₂ or 90%CF₄+10%O₂ plasmas were ignited at pressures ranging from 4 to 200 mTorr, top RF powers of 500 and 1000 W, four different bottom RF powers (0, 100, 300 and 500 W), and three different wafers (Si, Al and anodized Al). T_e was measured across the plasma at a height of 5 mm above the lower electrode. For Si substrates, T_e increased with increasing pressure between 4 and 20 mTorr (typically from 5 to 6.5 eV). The dependence of plasma electronegativity on pressure may be responsible for this behavior. T_e decreased rapidly with increasing pressure in the 20-60 mTorr range, and then slowly decreased with further increases in pressure to 200 mTorr, where T_e = 2.4 to 2.7 eV. Increasing the applied bottom RF power resulted in higher T_e, caused by enhanced stochastic heating of electrons with increasing low frequency voltage. Over the entire pressure range investigated, T_es in 90%CF₄+10%O₂ plasmas were similar to those in 80%CF₄+20%O₂ plasmas. The EEDFs exhibited bi-Maxwellian characteristics with an enhanced high energy tail, especially at pressures >20 mTorr. Different dependences of T_e on pressure and applied top and bottom RF powers were observed for Al and anodized Al wafers.

External cavity tunable diode lasers (DL) and Light Emitting Diodes (LEDs) are cost effective and easy to use tools that can be easily implemented for the diagnostics of process plasmas. We have used these techniques to characterize plasmas produced in several industrial etch reactors (Applied Materials and LAM Research) and to better understand the interaction mechanisms of the plasma with surfaces present: wafers or/and reactor walls.

Using near infrared DLs, we deduce the gas temperature in different silicon etch plasmas (Cl₂, CF₄, SF₆,..) from the Doppler width of absorption lines from argon metastable atoms when traces of argon is added to the process gas [1]. Absorption and Laser Induced Fluorescence experiments with a blue DL permits to map up the angular dependent velocity distribution function of aluminum atoms sputtered from a RF biased Aluminum wafer under argon ion bombardment [2].

We have shown that the sensitivity of Broad Band Absorption Spectroscopy (BBAS) can be enhanced by at least one order of magnitude when a LED is used as a light source [3]. Thanks to the high stability of the LEDs, it is also possible to perform time-resolved measurements of radicals densities in pulsed plasmas [4]. With a 350 nm LED, the time variation of Cl₂ density down to 3 mTorr can be measured with a time resolution of about 10 millisecond [3]. With a 275 nm LED, we have measured the decay time in the afterglow of the BCl radical produced in BCl₃/Cl₂ plasmas. It has been shown that the main loss mechanism of BCl radical is its gas phase reaction with Cl₂ to form BCl₂ molecule [4].

[1] G. Cunge, R. Ramos, D. Vempaire, M. Touzeau, M. Nejbauer, and N. Sadeghi, JVST A 27, 471 (2009)

[2] R. Ramos, G. Cunge, M. Touzeau, and N. Sadeghi, J. Phys. D: Appl. Phys. 41, 152003 (2008)

[3] G.Cunge, D.Vempaire, M.Touzeau and N.Sadeghi, Appl. Phys. Letters 91, 231503 (2007)

[4] D.Vempaire and G.Cunge, Appl. Phys. Letters 94, 21504 (2009)

Process yield in many plasma assisted processes may be improved significantly by real time, closed loop control of certain plasma species. This paper describes the closed loop control of a low pressure, inductively coupled plasma simulation. The plasma simulation consists of a global model of the plasma chemistry coupled to an equivalent circuit. The equivalent circuit incorporates an impedance matching box and an model of power coupling from the antenna into the plasma which has been derived from the wave equation and the two term solution to the Boltzmann equation. In addition, mass flow controller models and gas flow transport delays are included in the simulation. The design of effective, real time, closed loop control algorithms is facilitated by simple, control-oriented, dynamical models of the relationship between actuators (inputs) and the process quantities to be controlled. The paper will indicate how the parameters of a control algorithm may be determined from the process model (model-based control) in order to guarantee a robustly stable closed loop response. In general, process measurements are noisy and may not provide direct estimates of process quantities to be controlled. For example, estimates of atomic oxygen density obtained from optical emission spectroscopy are ambiguous due to dissociative excitation. Furthermore, many process parameters such as wall sticking coefficients are extremely difficult to estimate and may change due to chamber seasoning. The paper will indicate how an optimal state estimator may be used to improve estimates obtained from optical emission spectroscopy and how such estimates may be used to adapt the control algorithm in real time in order to guarantee process stability despite changes in process parameters.

During the last ten years there has been a growing awareness about the impact of the plasma etch chamber wall condition on the density of reactive species in the etch chamber. This is especially the case for ICP chambers at low pressure (5-80mTorr) where gas phase diffusion and recombination at the reactor wall can be a dominant loss mechanism for reactive species. The majority of the studies are carried out for reaction chambers, coated with SiOCl or CF_x -based polymers.

In this paper we expanded the study to metal deposits such as titanium and tantalum which are frequently used for metal gate application in front-end-of-line, or as hard mask material for low-k etching in back-end-of-line. Additionally we made an evaluation of the cleaning procedures to remove these metals from the reactor walls.

CF_x-based polymers where deposited by igniting a polymerizing plasma, while Si, Ti and Ta were deposited by etching Si, TiN and TaN, respectively, in Cl₂/HBr/O₂ plasmas. The composition of the chamber wall deposits was investigated by XPS analysis of so-called ‘floating samples’. Relative density variation of reactive species (Cl, Br, O and F) in “contaminated” chambers was analyzed by actinometry. For the actinometry experiments we added 5% Ar to respectively Cl₂, HBr, O₂ and SF₆ plasmas and we related the peak intensities to the Ar 750nm peak. Our results indicate that the effect of the metal deposits is very similar to the CF_x-based deposits. We can summarize this as a lowering of all tested species densities when the chamber walls contain metal(oxide) traces. By comparing our data with previously published results we can also estimate a recombination probability.

Next to the study on density variations of reactive species we investigated how these metals are best cleaned from the chamber walls. This is again done with actinometry, by comparing the data of ‘clean’ and ‘coated’ chamber walls. For CF_x based coatings and SiOCl coated chambers the cleaning procedure is well known and is mostly done with O₂ and SF₆ based plasmas respectively. For the metals however we found that titanium should be cleaned with Cl₂ plasma and for Ta we observed that none of the tested chemistries was able to quickly remove it. Since our XPS data indicate that the Ti and Ta on the walls is partially in the form of TiO₂ and Ta₂O₅ this is not a surprising result. Finally, we did observed that it’s possible to clean both metals with BCl₃ or SiCl_x containing plasmas but this is beyond the scope of this work.