AVS2012 Session PS1-WeA: Plasma Diagnostics, Sensors and Control 2

Dry etching technology was introduced to integrated circuit manufacturing due to its unique ability to do anisotropic removal of material to create high aspect ratio structures. Dry etching tools in the semiconductor industry use plasma to generate electrons, bombarding ions and free radicals. These radicals are chemically reactive and highly efficient at removing material from substrate surfaces, so the etch profiles depend on ion density and ion energy.

Electron cyclotron resonance (ECR) plasma etching reactors are widely used in wafer production. They use a low pressure, high density plasma source that has resonant magnetic field coils and two independently controlled microwave and radio frequency (RF) sources. ECR plasma etching is a very complex process and most of the progress with ECR etching tools up to this time has been accomplished empirically. Introducing new and adapted plasma diagnostic techniques is essential for further characterization and better understanding of plasma and etching processes.

In previous work we presented electrical measurements obtained using a retarding field energy analyzer (RFEA) installed in an commercial ECR etching reactor. RFEAs are generally used for charged particle flux and energy distribution measurements at electrically grounded surfaces. We demonstrated that an electrically isolated RFEA can be successfully used for measuring ion energy distribution functions (IEDFs) at the surface of RF driven electrodes in the presence of strong DC magnetic fields.

In this work we present further achievements in plasma diagnostics and ECR etching characterization. The original experimental setup is expanded for additional diagnostic tools. Phase resolved optical emission spectroscopy (PROES) measurements are taken at three different discharge locations simultaneously with RFEA measurements. Electrical and optical measurements show strong correlation in comparable aspects which proves their validity and together they provide more information about plasma in complementary aspects. The RFEA has also been used to measure the electron energy distribution function (EEDF) with the aim of electron temperature and density estimation in the bulk plasma. For validating EEDFs obtained with the RFEA, we performed similar measurements of EEDFs in a capacitively coupled plasma (CCP) reactor using the RFEA and the Langmuir probe. Results of this validation and estimations made from EEDFs are presented and discussed. Experimental and theoretical plasma etching characterization has been done through the analysis of spatially resolved data sets obtained from RFEA and spectroscopic ellipsometry at various radial locations.

A new type of microwave resonator probe, curling probe, has recently been proposed [1] which employs mono-pole excitation of spiral slot antenna. This probe enables direct measurement of electron density in reactive plasmas, based on a quarter-wavelength resonance at the frequency f=b(c/4L)[2/(e_in+e_out)]^1/2. Here, b is the finite-size correction factor, L is the antenna length, e_in and e_out are the relative permittivities of two dielectrics inside and outside the slot antenna, respectively. When the probe is inserted into plasma, the outer dielectric constant e_out is expressed as e_p =1-(f_p/f)² with the electron plasma frequency f_p=(e²n/me₀)^1/2. Thus, measuring the resonance frequency of curling probe, one can determine f_p, and hence the electron density n. The finite-difference time-domain (FDTD) simulation of 11-mm-diam curling probe of 35-mm antenna length shows a sharp resonance at the frequency from 1 GHz to 6 GHz uniquely determined by the electron density. The sheath formed in front of the probe surface was modeled as a vacuum layer of ~5l_D in thickness. The resonance frequency was hardly influenced by the sheath at the density higher than 10¹¹ cm^-3. Basic experiments of electron density measurement by the curling probe were performed in an argon (1~20 Pa) low-power (<2 kW) discharge in ICP device for 300-mm-wafer process. The power dependence of the resonance frequency observed by the curling probe was explained well by the FDTD simulation result as well as the analytical formula. The radial distribution of electron density was measured by a movable curling probe.

When a polymer layer (permittivity e_d) is deposited on the probe surface, the resonance frequency decreases with the increasing layer thickness. Using this probe characteristic, one can in situ monitor the thickness of dielectric layer deposited onto a wall of plasma vessel, where the curling probe is positioned to just the same surface of inner wall. The FDTD simulation shows ~10 MHz shift in the resonance frequency for 15-mm-thick deposition of polymer (e_d =3.2), in good agreement with the experimental observation. Thus, the wall deposition layer during chamber cleaning or CVD process can be monitored by the curling probe. In fact, amorphous carbon layer formed on the vessel wall was in situ monitored by the curling probe in a 1-kW surface-wave plasma in CH₄/Ar deposition discharge and in O₂/Ar cleaning discharge. This model experiment successfully demonstrated an applicability of curling probe to monitoring the dielectric layer deposited on the plasma vessel wall.

[1] I. Liang, K. Nakamura, and H. Sugai, Appl. Phys. Express 4, 066101 (2011).

In this contribution we report a series of experiments based on time resolved laser induced fluorescence (tr-LIF) on low temperature discharges. This technique is used, in combination with Thomson scattering (TS) measurements, to get insight in the excitation kinetics of the discharge.

This method is used to pump different levels within the 4p and 5p groups of the Ar excitation space, from the Ar metastable level 1s₅. Measuring the tr-LIF decay times in combination with the TS measurements of electron density provides simultaneously total destruction rates by electron and heavy particle collision of the pumped levels. These effective excitation rates are of great importance for the development of models, and are scarce in the case of high excited argon levels.

Similarly, the time response of the collisional induced fluorescence signals (tr-LCIF) emitted by other excited levels is used to study the interaction between levels. In the particular case of Ar this method is used to obtain the “depopulation rates” of metastable 4s levels. When other gasses are introduced in the discharge tr-LCIF signals are obtained from the emission of different excited species of those gasses. These experiments enable us to probe in details the excitation exchange between different species. The use of a high rep-rate system is fundamental for the LCIF measurements, where the low intensity of the signals demands measurement times up to 30 minutes.

Deep silicon etching is a crucial step for several emerging and evolving technologies, including micro-electromechanical systems (MEMS) and CMOS image sensors. Among etching methods available, the Bosch process is widely used because it allows fast etching rates together with high aspect ratios. This process relies on rapidswitches between isotropic etching in SF₆-containing plasmas and polymer deposition for sidewall passivation in C₄F₈-containing plasmas. As a result of reactor wall conditioning, run-to-run variability in etching rates and uniformity is commonly observed. In order to understand the origin of such drifts and thus to develop new strategies to reduce these effects, we have started investigations of the temporal evolution of the properties of SF₆ and C₄F₈ plasmas in the Bosch process on an industrial, inductively-coupled plasma reactor (ICP) at Teledyne Dalsa. In this context, a rf probe was installed on the transmission line connecting the generator to the substrate holder. This system measures the amplitude of the current and voltage as well as the phase between them, allowing determination of the real and imaginary parts of the impedance. During one etch cycle in Ar/SF₆ (total duration of 10s), the absolute value of the reactance, |X|, decreased sharply within the first second and then reached a plateau. For one sample exposed to many etch and deposition cycles, the cycle-averaged value of |X| slowly decreased between the first and the 70^th cycle. For many samples etched in the same chamber with a 2 minutes O₂ plasma clean between each run, the run-averaged value of |X| decreased with the sample number. Over this whole range of experimental conditions, the resistance remained fairly constant. From optical profilometry measurements, the etching rate was found to decrease from 340 to 300 nm/s between run #1 and #5. This decrease matched relatively well the observed decrease of |X|, suggesting that the reactance determined from rf probe measurements is a good parameter to examine in a non-intrusive way run-to-run variability in deep silicon etching. More recently, preliminary analysis of the plasma emission in the Ar/SF₆ plasma revealed that the F (703.7 nm)-to-Ar (750.4 nm) line-intensity ratio (related to the F density) remained constant within the first few seconds and then decreased mid-way in the etch cycle due to strong F uptake. Future experiments will attempt to determine whether the observed decrease of the Si etching rate due to wall conditioning can be attributed to a reduction of the fluorine concentration during the etch cycle or to a higher density of fluorocarbon radicals during the passivation cycle.

Optical emission spectroscopy has evolved a mighty tool for evaluating the high-energy tail of the electron energy distribution function EEDF [1]. Since it is a non-evading technique, a spatially-resolved measurement cannot be performed, which limits its general applicability, especially for inhomogeneous plasmas, in particular capacitively-coupled discharges with their thick plasma sheaths and large rf electric fields, and microwave-driven plasmas which exhibit a very high plasma density in front of the microwave window which prevents the penetration of the microwave into deeper regions of the reactor. A spatially averaging technique therefore overdraws the region close to the window leading to erroneous results and conclusions.

To overcome these limitations, a cyclindrical parallel-plate reactor was equipped with a magnifying double lens system with different focal lengths halfway between the two “electrodes” to resolve radially dependent spectroscopic data obtained with the two actinometric gases Kr and Xe [2]. From these data, the electron temperature of the high-energy tail of the EEDF of capacitively coupled discharges through argon and the weakly electronegative gas mixture CF₄/O₂ can be calculated [3].

These data are modeled with ray tracing programs to evaluate the degree of magnification. Compared with an experimental setup with a plane window instead of a converging lens and only one focusing lens for the glass fiber, we can enhance the S/N ratio by a factor of at least 100 yielding a radial resolution of about 1 – 2 mm which is in the same range of Langmuir probes.

Compared with measurements obtained with Langmuir probes in the plasma bulk, electron temperatures show in fact the expected behaviour of definitely hotter temperatures.

[1] V.M. Donnelly, J. Phys. D: Appl. Phys. 37, R217 (2004)

[2] G. Franz, I. Krstev, F. Schamberger, Plasma Sci. Techn. 14(8), to be published 2012

[3] M.V. Malyshev and V.M. Donnelly, J. Vac. Sci. Technol. A15, 550 (1997)

In particular for III-V semiconductors, plasma processes close to active zones of surface-sensitive devices are critical, demanding low damage through ion bombardment and so an excellent process understanding and control. In order to get the ion energy into the right range, the bias power is switched off working then a downstream-like mode. A complete model of the entire system with this special setup is presented and validated. It demonstrates that the bias matchbox capacitances are excellent control elements for the ion energy. We can also show that the Vp measurement is not representing the ion energy here. The ion energy is estimated by the sheath voltage from the model and shows a reasonable correlation to the etch rate observed.

This paper focuses on a general method to characterize equipment and plasma by a combination of off-line and real-time measurements. The first major goal of this paper is to show how reduced models of plasma equipment can be used to estimated important plasma parameters as to understand how the plasma process works. The equivalent circuit i only a tools to explain the over all plasma physical picture, it does mean that an equivalent circuit is always the right tool to describe a plasma. This resulting, reduced and so also not self-consistent model must be verified and then used for the ion energy control in an ICP etcher.

The model is based on an combined plasma and RF model in the special case of an existing bias matchbox but zero bias power. The control elements are the capacitance in the bias matchbox. We show the tune capacitance to be an excellent control element, but we can also demonstrate that – at least in this case – an RF peak voltage measurement is not representing the ion energy and alone not very useful.

Trikon/Aviza Omega 201 ICP etcher was used. A 100 mm electrode, a cylindrical coil around a ceramic chamber wall, and a Al chamber lid are main parts of the chamber.

Usually two generators are running at 13.56 MHz as master and slave with an adjustable phase shift between source and bias generator. In our case the bias generator is switched off in order to achieve a very low ion energy and thus almost no damage at the wafer. First investigations have shown that the position of load and tune capacitance in the bias matchbox to be very important for etch rate and wafer damage. Therefore a model was built to understand and finally to utilize these effects for process control.

The measurement of ion energy at the wafer surface at conditions that are realistic for commercial equipment and process development without extensive modification of the reactor geometry has been an industry challenge. High energy, wide frequency range, process gases tolerant, minimally perturbing, contamination free and accurate ion energy measurements are the base requirements. In this work we will report on the complete system developed to achieve the base requirements, including the safe and easy use by engineers. The system includes: a reusable silicon ion energy analyzer (IEA) wafer, signal feed through, RF confinement, and high voltage measurement and control.