Monitoring Sheath Voltages and Ion Energies in High-Density Plasmas Using Noninvasive Radio-Frequency Current and Voltage Measurements Mark A. Sobolewski Process Measurements Division, National Institute of Standards and Technology, Gaithersburg, MD 20899 Abstract. To obtain optimal results from plasma processing, the energy of ions incident on substrate wafers must be carefully controlled. Such control has been difficult to achieve, however, because no practical method exists for monitoring the energy distributions of ions at a wafer surface during processing. To solve this problem, we have developed a noninvasive, model-based method for determining ion energy distributions (lEDs) that is suitable for use during actual processing in commercial plasma reactors. The method was validated by tests performed in argon and CF4 discharges at 1.3-3.1 Pa (10-23 mTorr) in an inductively coupled, high-density plasma reactor, with radio-frequency (rf) substrate bias at frequencies of 0.1 MHz to 20 MHz. Plasma potential waveforms and sheath voltages obtained from the noninvasive rf technique agreed well with independent measurements made using a capacitive probe. Ion energy distributions from the rf technique were in good agreement with distributions measured by ion energy analyzers. INTRODUCTION In plasma etching, substrate wafers are bombarded by reactive neutral species and energetic positive ions. These energetic ions are essential for etching. Etch rates, etch profiles, and damage rates all depend on the kinetic energy of the ions. To obtain optimal results, ion energies must therefore be carefully optimized and controlled. Unfortunately, no good method now exists for monitoring energy distributions of ions at a wafer surface during actual processing. Thus it is difficult or impossible to know what the ion energy distributions are, and whether they are stable or drifting with time. Ion energy analyzers, sometimes coupled with mass spectrometers for ion mass analysis, are used in plasma research studies, but they are not well-suited for process monitoring during manufacturing. When exposed to actual processing conditions, ion energy analyzers may fail due to deposition or etching at analyzer surfaces. Commercial plasma reactors often cannot accommodate an ion energy analyzer anywhere close to the wafer. Furthermore, materials etched or sputtered from the analyzer could possibly contaminate wafers. Ideally, one wants to know the lEDs at the wafer itself, but to directly measure these distributions would require fabrication of an ion energy analyzer on each wafer, which is impractical. Because of these difficulties in measuring lEDs directly, there is much interest in developing indirect methods, which would use models to deduce the ion energy distributions from other types of measurements. It would be best to have a completely noninvasive method, one that does not require anything be inserted into the plasma reactor. This paper describes one such noninvasive, model-based method. It relies solely on rf measurements of current and voltage made outside the reactor. The measurements are analyzed using electrical models of the plasma and its sheaths, the thin regions at the boundary of the plasma adjacent to the wafer or other electrode surface. The method was tested by comparisons performed in argon and CF4 discharges in an inductively coupled, high-density plasma reactor, equipped with rf substrate bias. Model results for the sheath voltage and plasma potential waveforms, which are important intermediate results needed to accurately determine the lEDs, were validated by comparisons to independent measurements made using an invasive, capacitive probe. lEDs from the rf technique were compared to distributions measured by ion energy analyzers at grounded [1] as well as rf-biased [2] surfaces, and found to be in good agreement. EXPERIMENT The plasma reactor used in this study, an inductively coupled Gaseous Electronics Conference Reference Cell [3], is shown in Fig. 1. Mounted on the top of the reactor is an inductive plasma source, consisting of a flat coil powered at 13.56 MHz and an electrostatic shield. Wafers to be processed are placed on the substrate electrode located below the inductive source. Using a sinusoidal signal generator and a wideband rf amplifier, the substrate electrode can be biased at frequencies from 10 kHz to 100 MHz. This rfbias is CP683, Characterization and Metrology for VLSI Technology: 2003 International Conference, edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula 2003 American Institute of Physics 0-7354-0152-7/03/$20.00 195 high-density, inductive plasma source Vb(t) - + VgeW ground —o grounded sheath surfaces Vps(t) powered sheath A Vpe(t) F>owered electrode (a) 1 ^rw 4 _Vge(t) Rs wall sheath Vps(t) 1 powered sheath A Vpe(t) powered electrode (b) FIGURE 1: Diagram of the plasma reactor and electrical measurement system. FIGURE 2: Circuit model of discharges in (a) argon and (b) CF4. designed to control ion energies at the wafer independent of the plasma density and ion flux, which are controlled by the inductive source. On the electrical connections between the rf amplifier and the electrode, a current probe and voltage probe are mounted. Signals from these probes are digitized by an oscilloscope and transferred to a computer. Using techniques described previously [4-5], the computer accounts for propagation delays and stray impedance, resulting in an accurate determination of the current and voltage signals Ipe(f) and Vpe(t) present at the substrate electrode or wafer surface, and the small voltage Vge(t) induced on grounded reactor surfaces. These signals are then further analyzed, as described below, using an electrical model of the discharge. Plasma potential waveforms and sheath voltages obtained from this analysis are compared with independent measurements made using a capacitive probe [1] inserted into the plasma. Ion energy distributions obtained using the model are compared with lEDs measured either by a mass spectrometer and ion energy analyzer system inserted from the side of the reactor [1], as shown in Fig. 1, or an analyzer incorporated into the rf-biased substrate electrode [2] (not shown). Additional plasma characterization was performed using a Langmuir probe. The Langmuir probe provides a measurement of the dc plasma potential which is necessary to calibrate the dc offset inherent in the capacitive probe. It also provides measurements of the radial variation of the dc plasma potential and values for the electron temperature. discharges is shown in Fig. 2(a). The model includes two sheaths: the powered sheath—that is, the sheath adjacent to the rf biased substrate electrode — and the opposing sheath adjacent to grounded reactor surfaces. There is no need to model the sheath adjacent to the inductive source, since the electrostatic shield minimizes any capacitive coupling between the source and the plasma. The Ipe(t) and Vpe(t) signals, and capacitive probe measurements of the bulk plasma potential, V^(t), show only very small Fourier components at the fundamental and harmonic frequencies of the inductive source. These small components need not be included in the analysis; they are removed from all measured signals by Fourier filtering. There is also no need to include in the model any impedance associated with the bulk plasma. For the high-density argon discharges studied here, the bulk plasma impedance is negligible compared to the impedance of the sheaths [6]. On the other hand, Langmuir probe studies [3,7] have detected a radial drop in dc potential across the bulk plasma. Because the rf biased electrode is located near the radial center of the reactor but grounded surfaces are located further out, this radial dc voltage drop, Vr, should be considered to be in series with the two sheaths, as shown in Fig. 2(a). The powered sheath and ground sheath are modeled using a 1-dimensional numerical sheath model described previously [8]. An iterative procedure solves for the rf current and voltage across each sheath, subject to the two constraints imposed by the circuit of Fig. 2(a). First, the sheath currents must be equal and, second, the sum of the sheath voltages, plus Vr, must equal the total voltage Vpe(f)—Vge(f). To solve the model, one METHOD The model used to analyze the rf waveforms of argon 196 must input the Vpe(t)-Vge(t) waveform, obtained from the rf measurements described above, as well as a measured or estimated value for Vr. Other model input parameters are the area of the powered electrode Ape, the total grounded area Age, the electron temperature Te, and the total time-averaged ion current across each sheath, IQPS and lQgs. The model also requires the mass mi, charge Z/, and relative flux FI of each species of positive ion. Several of the model input parameters can be considered to be constants, since they vary little or not at all over a wide range of plasma conditions. In the argon discharges studied here, Ar+ constitutes 98% or more of the total ion flux [9], so the model need only consider one ionic species with m/ = 40 amu, Z/ = +e, and JTJ = 100%. Langmuir probe measurements of argon discharges [8] yield an electron temperature given by kfiTe = 3 eV. This value is an effective temperature for higher energy electrons, which are the most relevant for the sheath model. Low energy electrons are less important, since they are reflected back into the plasma before reaching the electrode surface, and thus they make no net contribution to the sheath current. Langmuir probe measurements also yield a value for Vr = 6 V. The dimensions of the substrate electrode and the reactor give values of Ape = 81 cm^ and Age = 6000 cm^. Unlike the other parameters, IQPS and lQgS vary strongly with inductive source power, and even at fixed source power they tend to drift. Consequently, IQPS and lQgs are treated as fitting parameters. They are varied in iterative fashion until the best agreement is obtained between the measured rf current waveform Ipe(f) and the rf current waveform output by the model, If(t). Specifically, the fitting algorithm minimizes the sum of the squares of the difference, Ipe(t)-It(t). Typically, the fitting procedure yields excellent agreement between Ipe(f) and It(f). When the fit has converged, plasma potentials, sheath voltages, and lEDs are output by the model. For ease of comparison, model lEDs are then broadened by an amount (2 eV) comparable to the energy resolution of our ion energy analyzer. The procedure used for CF4 discharges is slightly different. It will be discussed below. First, however, results obtained in argon discharges are presented. ARGON RESULTS Figure 3 shows plasma potential and sheath voltage results for argon discharges at three different values of the rf bias frequency, ranging from 100 kHz in Fig. 3(a) to 10 MHz in Fig 3(c). In the middle of each figure is plotted Vpe(f), the voltage waveform on the surface of the rf biased substrate electrode. Near the top of each figure we see plots of the bulk plasma potential, Vfr(f). The solid plot is the Vb(t) waveform measured directly using the invasive, capacitive probe. The dotted plot is the Vfr(t) waveform obtained from non-invasive rf current and voltage measurements using the model-based 197 6 8 10 12 time (ps) CD D) §-50 (b) 1.0 1.5 2.0 2.5 3.0 time (ps) 10MHz -100 0.7 0.8 0.9 1.0 time ([is) FIGURE 3: For argon discharges, waveforms of the plasma potential V/,(0 and the powered sheath voltage Vps(t) obtained by the noninvasive, model-based method (dotted curves) fall nearly on top of V/,(0 and V p s ( t ) waveforms measured using an invasive, capacitive probe (solid). The powered electrode voltage Vpe(t) is also shown. Conditions are 1.33 Pa (10 mTorr), an inductive source power of 200 W, a peak-to-peak Vpe(t) of 100 V and a bias frequency of (a) 0.1 MHz, (b) 1 MHz, and (c) 10 MHz. procedure described above. Similarly, near the bottom of each figure we see two plots of the powered sheath voltage, Vps(f)=Vpe(t)-Vb(t). The solid curve is the VpsW waveform obtained from capacitive probe measurements of Vfr(t). The dotted curve is the Vps(t) waveform obtained by the noninvasive, model-based method. The agreement between the invasive and noninvasive measurements is striking. They agree within ±2 V, which is roughly the uncertainty of the capacitive probe. Thus, one can obtain plasma potentials and sheath voltages from the noninvasive rf technique with an accuracy comparable to the invasive measurements. Excellent agreement was also obtained at other bias frequencies ranging from 10 kHz up to 20 MHz. Our plasma reactor is not equipped to measure ion energy distributions at the rf biased electrode, but > 200 '(a) peak-to-peakVps(t) = 36V CM I 150 > 50 CD O) a 100 J 50 g o 6 10 8 12 time (us) > 50 0 10 20 30 40 50 60 ion energy (eV) FIGURE 4: (a) Ar+ ion energy distribution measured by Woodworth et al. [2] for argon discharges in an inductively coupled GEC Reference Cell at 3.1 Pa (23 mTorr), with an inductive source power of 100 W, rf bias at 13.56 MHz, and a peak-to-peak powered sheath voltage Vps(t) of 36 V. (b) Ar + ion energy distribution obtained from noninvasive rf measurements obtained for the same conditions in our GEC Reference Cell. 2.0 2.5 3.0 3.5 4.0 time (us) 50^ Woodworth et al. [2] have measured such distributions, for argon discharges, in a GEC Reference Cell that is ostensibly identical to ours. Figure 4(a) shows an ion energy distribution for Ar+ reported by Woodworth et al. Figure 4(b) shows an Ar+ distribution obtained in our cell under the same conditions, using the noninvasive rf method. Both distributions show the familiar two-peaked structure. The energies and relative amplitudes of the two peaks agree rather well. The peaks do appear broader in Fig. 4(a) than in Fig. 4(b), but this may simply be a result of broadening due to the resolution of the gridded energy analyzer used by Woodworth et al. 10MHz, -50 0.5 0.6 0.7 0.8 time (us) FIGURE 5: For CF4 discharges, waveforms of the plasma potential V^t) and the powered sheath voltage Vps(t) obtained by the noninvasive, model-based method (dotted curves) and measured using an invasive, capacitive probe (solid). The powered electrode voltage Vpe(t) is also shown. Conditions are 1.33 Pa (10 mTorr), an inductive source power of 200 W, a peak-to-peak Vpe(t) of 100 V and a bias frequency of (a) 0.1 MHz, (b) 1 MHz, (c) 10 MHz. split the ground sheath into two separate parts as shown in Fig. 2(b). The orifice and other grounded surfaces close to the radial center of the reactor are considered one sheath (called the mass spectrometer sheath, with area Ams and total ion current I$ms) and the more remote grounded surfaces such as the vacuum chamber wall are considered to be another sheath (the wall sheath, with area Aws and total ion current IQWS). The radial drop in dc plasma potential contributes a dc voltage in series with each sheath, denoted Vrm and Vrw. Also, to obtain good agreement with electrical waveforms measured at high bias frequencies, e.g., 10 MHz, it is necessary to include in the model a resistance Rs in series with the wall sheath. This represents the resistance of the diffuse plasma adjacent to the wall sheath as well as the stochastic resistance associated with the boundary between the plasma and the wall CF4 RESULTS The configuration of the plasma reactor used for CF4 studies was somewhat different than for argon. To extend the operating range, CF4 plasmas were partly confined by a quartz ring suspended from the inductive source and a steel plate placed on lower electrode. [1] The ion energy analyzer and mass spectrometer system used in the CF4 studies entered from the side of the reactor and protruded into the intense plasma inside the confinement ring. The entrance orifice of the mass spectrometer system is grounded, so the sheath in front of the orifice is part of the ground sheath. However, the plasma density and ion current density are much higher close to the orifice than at the more remote grounded surfaces. Because of this nonuniformity, it is useful to 198 sheath. It is not necessary to include any resistance in series with the mass spectrometer sheath. Apparently, the plasma adjacent to that sheath has a high enough density that its resistance is negligible. For CF4 discharges, the model of Fig. 2(b) is solved in a similar manner to the argon model in Fig. 2(a), except that we now solve three sheaths simultaneously. Measured values of constants input into the CF4 model were Ams = 13 cm2, Aws = 6000 cm2, Ape = 214 cm2, kpTe = 3 eV, Vrm = 2 V, and Vrw = 6 V. The relative ion fluxes were 39% CFs+, 27% CF2+, 23% CF+, and 11% F+. For simplicity, the total ion current density at the mass spectrometer sheath and the powered electrode sheath were assumed to be equal, that is, lQpS/ApS = lQmsIAms. The three remaining independent variables, /Ops, /Qw5» and RS* were treated as fitting parameters, again adjusted until the rf current waveform output by the model agreed with the measured Ipe(f) waveform. Figure 5 shows plasma potential and sheath voltage results for CF4 discharges, again at four different values of the rf bias frequency. At each frequency, the plasma potential Vb(t) as well as the sheath voltage VpS(t) obtained by the noninvasive, model-based method agree well with measurements made using the capacitive probe. The agreement is not quite as good as that seen for argon discharges in Fig. 3, perhaps because of the greater number of model input parameters for CF4, or greater uncertainty in their values. Figure 6 shows ion energy results for CF4. lEDs obtained from the noninvasive method are compared against lEDs measured by the mass spectrometer, at bias frequencies of 0.1 MHz [Fig. 6(a )and (b)] and 10 MHz [Fig. 6(c) and (d)]. Good agreement is obtained for the energies of all peaks in the distributions, both at the lower rf bias frequency, 0.1 MHz, where ion inertia is negligible, and at the higher rf bias frequency, 10 MHz, where ion inertia is beginning to be dominant. Like the mass spectrometer measurements, the model based results show the expected narrowing of peak separation with rf bias frequency and with increasing ion mass at 10 MHz. The relative amplitudes of the peaks are usually in fair agreement, but there are several disagreements, for example CF3+ at 10 MHz. The amplitude disagreement would appear to indicate a failure in the sheath model, perhaps because it uses a simplified treatment of the boundary conditions between the plasma and the sheath. Peak amplitudes are more sensitive than peak energies to the boundary conditions. CONCLUSIONS Using the model-based method described here, one can obtain accurate plasma potential waveforms, sheath voltages, and ion energy distributions solely from noninvasive rf current and voltage measurements. One potential disadvantage of the method is that it requires values for several model input parameters. A sensitivity analysis too lengthy to be included here identifies the electron temperature as a particularly important parameter. To obtain the most accurate results, the electron temperature should be known ahead of time, or monitored by some independent measurement technique. mass spectrometer data noninvasive I/V method 100kHz (c) REFERENCES mass spectrometer data 10MHz 1. M. A. Sobolewski, Y. Wang and A. Goyette, J. Appl. Phys. 91, 6303 (2002). 2. J. R. Woodworth, I. C. Abraham, M. E. Riley, P. A. Miller, T. W. Hamilton, B. P. Aragon, R. J. Shul and C. G. Willison, J. Vac. Sci. Technol. A 20, 873 (2002). 3. P. A. Miller, G. A. Hebner, K. E. Greenberg, P. D. Pochan and B. P. Aragon, J. Res. Natl. Inst. Stand. Technol. 100, 427 (1995). 4. M. A. Sobolewski, J. Vac. Sci. Technol. A 10, 3550 (1992). 5. M. A. Sobolewski, IEEE Trans. Plasma Sci. 23, 1006 (1995). 6. M. A. Sobolewski, Phys. Rev. E 59, 1059 (1999). 7. A. Schwabedissen, E. C. Benck and J. R. Roberts, Phys. Rev. E 55, 3450 (1997). 8. M. A. Sobolewski, Phys. Rev. E 62, 8540 (2000). 9. Y. Wang and J. K. Olthoff, J. Appl. Phys. 85, 6358 (1999). noninvasive I/V method 10MHz 20 40 60 80 100 ion energy (eV) FIGURE 6: Ion energy distributions in CF4 discharges obtained by a mass spectrometer system (a) and (c); and by rf measurements (b) and (d). The bias frequency is 0.1 MHz in (a) and (b) and 10 MHz in (c) and (d). Other conditions are the same as in Fig. 5. 199
© Copyright 2025 Paperzz