AVS2012 Session PS1-TuM: Plasma Diagnostics, Sensors and Control 1

Several different phenomena were investigated. First, we investigated the changes in plasma electron density that occur when the rf bias is turned on or off. Analysis shows that such changes can be explained by the stochastic heating of electrons by rf bias power, the interaction of the rf bias and the inductive source, and perturbations in gas composition caused by etch products. Second, we investigated the changes in all plasma parameters that occur when etches of silicon dioxide films on silicon substrates reach endpoint. Analysis shows that the changes observed at etch endpoint were primarily due to changes in gas composition, rather than changes in electron emission from the wafer surface. Third, we also observed nonidealities in the inductive source that affect its ability to maintain constant plasma parameters.

Simultaneous with these measurements, the rf current and voltage were measured outside the reactor, in the rf bias and inductive source circuitry. These external measurements were strongly correlated to — and explained by — the wave cut-off and Langmuir probe data. These results strongly suggest that nearly all of the phenomena observed here can be monitored equally well by external rf measurements, instead of invasive probes.

Precise optical measuring techniques of spectral lines are necessary for low pressure plasma semiconductor manufacturing analysis. Problems which add to the inaccuracy of the optical measuring techniques include spectral line broadening mechanisms such as self-absorption.

Self-absorption has been widely neglected or its importance overlooked for plasma diagnostics.

In this study several different techniques are used to examine the impact of self-absorption on emission spectral lines from a reactive ion etch (RIE) plasma chamber, measured using a high resolution optical spectrometer. The experiments are performed in an Oxford Instruments plasma lab 100 RIE chamber which operates at a standard single frequency (GEC) 13.56 MHz with maximum RF power of 600 W. The OES spectrums were taken with a Horiba Jobin Yvon Czerny-Turner design Auto MicroHR spectrometer with focal length of 140 mm and spectral resolution of 0.25 nm at 400 nm wavelength.

One technique which is used is to check line intensity ratios within multiplets which abide by so-called LS-coupling rules. The time or spatial fluctuation in the observed intensity of the strongest or metastable line within the multiplet, in respect to the weakest spectral line in the same multiplet shows that self-absorption is present. Another technique which is examined is to change the optical path length by measuring the plasma from different points or different viewports on the plasma chamber. If the increase in signal intensity changes corresponding to optical path length, there is no self-absorption.

A Design of Experiments (DOE) has been used to cover the multidimensional external parameter space e.g. power, pressure, flow rate and chemistry. The results from the techniques are compared to each other for accuracy and the impact of self-absorption is then quantified. In this work the influence of self-absorption of neutral Argon and Oxygen spectral lines for a range of parameters in an RIE plasma chamber has been studied. Almost all the measured spectral lines have been affected, up to some level, by self-absorption. Some spectral lines shapes are changed by self-absorption up to 60%. One of the most widely used actinometry spectral lines for plasma diagnostics, Argon 750 nm, has its intensity affected by self-absorption by up to 40%.

Correction of self-absorption is a necessary step for OES based plasma diagnostics. In the case of Actinometry calculation correction of self-absorption could change the final result up to 20%.

Measurement of the thickness of a growing film as it is deposited can provide real advantages in a production environment. Consistent film thickness is achieved despite long term process drifts and machine to machine variations in deposition rate. Also the time lost due to system re-qualification after routine cleans and maintenance procedures can be reduced significantly.

In this work film thickness is measured in situ in a PECVD system by reflectance techniques, using the plasma emission as a light source (Optical Emission Interferometry, OEI). A parallel plate deposition system is described in which the plasma emission reflected from the substrate is monitored through one of the gas introduction holes in the upper electrode. During the deposition of silicon dioxide and silicon nitride films, the emission bands from molecular nitrogen in the 300 – 400nm region are used for measurement purposes. As the film thickness increases the reflected intensity undergoes a cyclical variation due to interference effects. The film thickness change for one complete cycle is known from the values of the wavelength and the refractive index of the film at that wavelength. The number of interference cycles, including fractional cycles is counted and the film thickness calculated in real time as the film is deposited. The process is terminated when the desired film thickness is reached. An example is shown where the long term variation in film thickness is significantly reduced as compared to running a process terminated by time alone.

For thin films (eg < 100nm) a series of interference cycles is not generated, so the “peak counting” approach cannot be used. Instead, multiple wavelengths (multiple nitrogen bands) are monitored, generating more data points which permits a more accurate determination of the film thickness. This approach is applied to the deposition of a 50nm silicon nitride film.

The above approaches strictly measure change in film thickness, not absolute film thickness. It is shown that by monitoring the complete spectrum and calculating the spectrum change with time, the result is equivalent to the differential of the film's reflectance spectrum. From this the absolute film thickness is calculated. This is particularly useful for the deposition of thicker films where a final film thickness is required, despite a variable or unknown starting film thickness. Although not discussed here, the same approach can be used when etching thick films down to a final required thickness.

A precise shape control at less than 1 nm scale is demanded in large-scaled-integrated-circuits (LSI) fabrication of the 10-nm half pitch and beyond. In realization of super-fine plasma etching process, surface reactions of ions and radicals play important role because they determine etched feature ^[1]. Thus control of actual plasma parameters such as substrate temperature, radical density, and electron density is required. It has been proved that the radical density is changed as a result of emission into bulk plasma both of etching products and adsorbed species on chamber wall. However, investigation of influence of inner wall condition on bulk radical density has not been studied yet. Thus, in this study we focused on gaseous radical density in H₂/N₂ plasmas, and investigated temporal changes subsequently after different kind of plasmas where inner wall condition could be changed.

Measurements of gaseous radical densities have been carried out by the Vacuum Ultra Violet Absorption Spectroscopy (VUVAS) equipped with a micro-discharge hollow cathode lamp (MHCL) ^[2]. In this case we measured transition lines for Lyman α at 121.6 nm for H atoms and ⁴P_5/2-⁴S⁰_3/2, ⁴P_3/2-⁴S⁰_3/2 and ⁴P_1/2-⁴S⁰_3/2 at 120.0 nm for N atoms. H₂/N₂ plasma was generated in a 100-MHz capacitively coupled plasma (CCP) reactor with H₂/N₂ gas mixture ratio of 75/25. Applied power was 400W (100 MHz) for upper electrode, and 200W (2 MHz) for bottom electrode respectively. The gaseous radical densities for the H₂/N₂ plasmas were monitored subsequently after different plasmas, (a) seasoning condition of H₂/N₂ plasma, (b) O₂ plasma, and (c) air exposure.

First, temporal changes for N radical density in the H₂/N₂ plasma subsequently, after the various plasma exposures were compared. In case of (a) seasoning, the N density was stable. On the other hand, the N radical densities were much varied after (b) O₂ plasma, and (c) air exposure. Notably, the density just after H₂/N₂ plasma ignition reached value which is more than twice higher when compared to stabilized value for (a) seasoning. Interpretation can be given as etched products or process gases were adsorbed on inner wall surface at the previous process, and then those species were desorbed from the wall into bulk plasmas in the subsequent processes. In accordance with organic low-k etching using H₂/N₂ plasmas, when a ratio of H radical and N radical (H/(H+N)) was changed more than 15% against an optimal original value, a shape of etched feature is resulted in modifications with scales of 10 nm ^[1].

[1] M. C. Sung et al, J. Appl. Phys. 107, 113310 (2010).

[2] S. Takashima et al, J. Vac. Sci. Technol. A 19, 599 (2001).

The ever-increasing demand for higher device density in integrated-circuit fabrication has resulted in stringent requirements in respect of quality, reliability and precision of all fabrication processes in semiconductor manufacturing. Reactive Ion Etching (RIE) is a critical technology used at many stages of the manufacturing process. At present, most semiconductor manufacturing equipment is operated in open loop mode. Consequently, key plasma parameters such as ion flux and radical densities at the substrate surface are sensitive to drift in tool subsystems, changes in wall condition and wafer loading, for example. Such disturbances may affect process metrics such as etch depth and anisotropy and result in a significant degradation in device yield and performance.

In this presentation, we describe the implementation of a multi-variable Linear Quadratic Gaussian/Loop Transfer Recovery (LQG/LTR) controller to regulate key plasma parameters in an updated Mini-lab RIE 80 capacitively coupled chamber. The custom designed, real-time data-acquisition and control system has been implemented in Labview on a National Instrument CompactRIO real-time controller, which has been interfaced to a hairpin resonance probe and an optical emission spectrometer, the Ocean Optics USB4000. Experimental results are presented which compare the performance of the multivariable LQG/LTR controller to that of a suite of decentralised single-input-single-output (SISO) controllers.

This research forms part of program to implement multivariable closed-loop control on an industrial Electron Cyclotron Resonance (ECR) plasma etch chamber. The presentation concludes by describing progress in this part of the program.

Present practice in reactive ion etching specifies etch recipes in terms of inputs such as gas flow rates, RF power and pressure. However, ostensibly identical chambers running identical recipes may produce very different results. Extensive ‘chamber matching’, i.e. initial iterative, empirical tuning of the process recipe, which entails time-consuming, ex situ statistical analysis of etch metrics, is required to ensure that an etch chamber produces acceptable results. Once matched, etchers are run ‘open loop’ and are thus sensitive to disturbances such as actuator drift, wall seasoning and substrate loading, which may have deleterious effects on process metrics such etch depth, uniformity, anisotropy and selectivity,. An alternative approach, which would reduce sensitivity to disturbances of the plasma generating process, would be to specify a recipe in terms of plasma quantities such as radical densities, and to regulate these in real time by adjusting the inputs with a suitable multivariable control algorithm.

Multivariable closed loop control of an SF₆/O₂/Ar plasma in an Electron Cyclon Resonance (ECR) etcher is the focus of a major research program in the National Centre for Plasma Science and Technology (NCPST) in Dublin City University (DCU). As an intermediate step, a multivariable LQG/LTR control algorithm has been implemented on a capacitively coupled plasma using the same gas mixture. This presentation describes the design of the algorithm, and its efficacy is demonstrated via simulation with a variety of disturbances. The performance of the multivariable algorithm is compared to that of a suite of decentralised single-input-single-output (SISO) controllers.