Experimental verification of physical models for defect states in. crystalline and amorphous ultrathin dielectric films

Transcription

1 Experimental verification of physical models for defect states in crystalline and amorphous ultrathin dielectric films H. Ren, 1 M. T. Nichols, 1 G. Jiang, 2 G. A. Antonelli, 3 Y. Nishi, 4 and J.L. Shohet 1 1 Plasma Processing & Technology Laboratory, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA 2 Novellus Systems, Tualatin, Oregon 97062, USA 3 Novellus Systems, Albany, New York 12203, USA 4 Stanford University, Stanford, California 94305, USA ABSTRACT Defect states in different dielectric materials with crystalline and amorphous structures were investigated with two models, the Broer model and the Bloch model. For crystalline structures as in hafnium oxide, defects were found to fit well with the Broer model due to strong local magnetic-field-orientation dependency. In amorphous materials, for example, organosilicate glass, defects were found to fit to the Bloch model due to field broadening effects. Using the two models, behaviors of defects can be described as a result of electron spins. Defect concentrations and locations are determined. 1

2 When a dielectric sample with certain defect states (unpaired electrons) is immersed in a radiofrequency-modulated magnetic field, the defect states will absorb energy through their magnetic susceptibility. 1 The susceptibilities of different defects are subject to different relaxation mechanisms as a function of frequency. 2 The relaxation may depend on local magnetic strength and direction. 3 In this Letter, we investigate the defects in the dielectric films that are popular in manufacture of microelectronic devices. These materials are divided into high- and low-k dielectrics, depending on the dielectric constant of the material compared with that of SiO 2 (k = 3.9). High-k dielectrics with crystalline structures, such as hafnium oxide and zirconium silicate, have received considerable attention. 4, 5 However, defect states in high-k dielectrics strongly influence the dielectric properties of these materials and are still bottlenecks in the fabrication process. 6 Low-k dielectrics with amorphous structures, on the other hand, can increase the speed of integrated circuits by providing low capacitance. Low-k dielectrics include silicon-based (such as SiCOH, 7 HSQ, 8 SiOF, 9 etc.) and nonsilicon-based (e.g. polymer 10 ) materials. However, for low-k dielectric materials, chemical and mechanical properties are often compromised in order to reach the lowest possible values of dielectric constant, resulting in the introduction of defects. The nature and concentration of these defects are important in determining the lifetime and potential applications of these materials. 11,12 Hence, for both high-k and low-k dielectric materials with different crystal structures, an investigation of their defect state concentrations and determination of their locations is extremely important. 2

3 Previous work has concentrated on the determining the defect concentrations in some dielectric materials, such as HfO 2, Al 2 O 3, ZrO 2, 13,14 etc. The results successfully found out point-defect concentrations in these materials. However, without accurate physical models that can describe and predict the behaviors of the defects that are detected with electron-spin resonance (ESR) measurements, the location of the defects can hardly be generalized and applied to other related materials. Furthermore, difficulties may occur during identification and measurement of the defects in dielectric materials with amorphous dielectric films such as organosilicate glass. 11 For example, different deconvolution methods of the ESR signal can result in different interpretations of the defect fingerprints and concentrations. Based on these problems, a solution is proposed here using two physical models. A systematic investigation of defects in dielectric materials with various crystal structures was done during ESR measurements. With this work, concentrations and locations of the defects in various dielectrics can then be correctly interpreted. We consider two physical models, the Broer model 15 and the Bloch model 16, to investigate the defect behavior of dielectrics with crystalline and amorphous structures during ESR measurements. These models were developed based on different assumptions of the response of the defects to the local magnetic field during ESR measurements. Each defect has its fingerprint, i.e., the g-factor, which depends on the microwave frequency and the magnetic field strength as (1) where h is Planck s constant, µ B is the Bohr magneton eh/(4πm e ), ν is the frequency of the 3

4 oscillating magnetic field, and B 0 is the magnetic field strength at the peak of ESR absorption for each defect. In the Broer model, the defect spins of crystalline dielectrics are assumed to be randomly oriented parallel or antiparallel to the magnetic field. 15 The absorption spectrum of the defects as a function of the sweeping magnetic field in the limit of large numbers of spins clustered around the nucleus, 3 shows a Gaussian shape described as (2) where A = A(B 0, χ 0 ) represents the amplitude of the distribution while χ 0 is the normalized susceptibility of the defects. B 0 can be used to determine the g-factor of the defect as in equation (1). σ is defined as the B-field width which represents the width of the ESR spectroscopic signal. Since this Gaussian distribution is a function of magnetic field strength, it can also be transformed into a function of frequency using equation (1). Experimentally, the absorption spectral signals, if they are written as a Gaussian derivative, are more helpful in determining the g-factors of the defects. The integral of the Gaussian distribution is used to determine the concentration of the defects. 17 They can be easily expressed as (3) and (4) where f Gauss (B) is the signal shape of Gaussian derivative and C Gauss represents the 4

5 concentration of the defect state in the Broer model. To further visualize the physical picture of this model, we consider the defects in a crystal structure or in the interfacial layer between the dielectric and silicon substrate. The spin directions of these defects are very likely to be uniformly parallel or antiparallel to the ESR magnetic field and therefore satisfy the assumption of this model. This is schematically shown in Figure 1 (a). In the Bloch model, defect spins of amorphous dielectrics are assumed to be randomly oriented with respect to the magnetic field. 16 With a single relaxation mechanism, 18 the defect distribution is supposed to be in Lorentzian form as (5) where A, B 0, and σ are the same as in equation (2). The corresponding experimental spin signal and defect concentration will then be the derivative of the Lorentzian and the integral of Lorentzian distribution, respectively, given by (6) and (7) From equations (6) and (7), we establish the expectation value of the measured signals and the calculation method for defect concentrations for the Bloch model. That is, if the 5

6 Bloch model applies, the signals should obey equations (6) and (7) while if the Broer model applies, the signals should obey equations (3) and (4). A physical picture of the Bloch model is shown in Figure 1 (b). When the bulk dielectric structure is amorphous, the spin orientations will be more random. In this case, it is likely that defect spins will form a Lorentzian distribution under the action of the magnetic field. In order to verify the two physical models, electron-spin resonance (ESR) spectroscopy was employed on two ultrathin dielectric samples: 1) atomic-layer deposited (ALD) 20 nm-thick 4000 Ω cm HfO 2 on (100)Si with a crystalline structure and 2) plasmaenhanced chemical vapor deposited (PECVD) 50 nm-thick deposited on 8000 Ω cm lowk organosilicate glass (SiCOH). This material has an amorphous structure. The ESR measurements were made with an X-band Bruker Elexsys 500E EPR spectrometer. In order to obtain a clear spectroscopic signal in a high Q-factor cavity, high resistivity of the substrate is required to obtain effective ESR measurements. 19 A % KCl weakpitch sample ( spins/cm) was used to calibrate the system. For crystalline HfO 2, the ESR signals are shown in Figure 2. The ESR signal as the black curves can be decomposed into three defect states as expected. They are P b -type states (P b0 and P b1 ) and E states with different g-factors. The P b -type states are silicon dangling bonds. 20,21 The two P b -type states on (100) silicon wafers are defined as P b0 and P b1. On (111) silicon wafers, there is only one P b state, which is called P b or P b0. 22 The E states 6

7 are positively charged oxygen vacancies. 23 Since the concentration of the E states is relatively small and is hard to fit, as shown in Figure 2, instead we examine the possibility of fitting the data to silicon dangling bonds. Both physical models were applied to fit ESR signals for HfO 2 using least-squares. It is seen that a Gaussian fitting matches very well with the signal while Lorentzian fitting introduces a large deviation. This indicates that these dangling bonds can be fit to the Broer model. The concentrations of the defects in this case were calculated to be approximately around spins/cm 2. Based on these concentrations and the fingerprint of these silicon dangling bonds, we believe the dangling bonds are located mainly at the interface between the dielectric layer and silicon substrate. There are few silicon atoms in the bulk dielectric so that it is unlikely that these defects are located in the bulk dielectric. Now we move on to examine the defects in SiCOH with amorphous structures and determine whether the defects in SiCOH match either of the two physical models. Low-k organosilicate glass (SiCOH) was deposited so that carbons were highly doped and porosity was introduced. 24 This makes the structure of SiCOH amorphous. 11 Hence, we expect the ESR signal of SiCOH to be Lorentzian rather than Gaussian. Figure 3 shows that the ESR signals of SiCOH have a Lorentzian-derivative shape. Thus, the Bloch model fits these very well. The defect concentrations in SiCOH were calculated to be on the order of spins/cm 2, which are two orders of magnitude larger than the number of potential interfacial defects. Because of the relatively large value of the defect 7

8 concentration, the defects are likely located in the bulk dielectric. The interfacial defects in SiCOH are negligible compared to the large number of bulk defects in the amorphous structure of SiCOH. The B-field widths as defined in equations (2) and (5) and defect concentrations as calculated using equations (4) and (7) are summarized in Table 1. It is seen that defect concentration in SiCOH is much higher than these in HfO 2. In conclusion, we applied physical models to the defects of dielectric films with crystalline and amorphous structures. We determined that defects are located at the interfacial layer for crystalline HfO 2 while for amorphous SiCOH, the defects are primarily located in the bulk dielectric. Electron-spin resonance measurements verified that the Broer model with Gaussian fitting can be applied to calculate interfacial defect states in HfO 2 while the Bloch model with Lorentzian fitting can be used to interpret bulk defects in SiCOH. These models can be further extended to other dielectric films so that the defect concentrations and locations can be determined. We thank M. Ivancic for helping set up the ESR experiments. This work is supported by the Semiconductor Research Corporation under Contract Number 2008-KJ-1781 and by the National Science Foundation under Grant CBET

9 References Cited 1 C. P. Poole and H. A. Farach, ASM Handbook, (ASM International, Materials Park, Ohio, 1986), Vol. 10, pp J. G. Castle, Jr., D. W. Feldman, P. G. Klemens, R. A. Weeks, Phys. Rev. 130, 577 (1963) 3 G. E. Pake and E. M. Purcell, Phys. Rev. 74, 1184 (1948) 4 J. Aarik, A. Aidla, A.-A. Kiisler, T. Uustare, V. Sammelselg, Thin Solid Films 340, 110 (1999) 5 M. Youm, H. S. Sim, H. Jeon, S. Kim, and Y. T. Kim, Jpn. J. Appl. Phys. 42, 5010 (2003) 6 W. S. Lau, L. Zhong, A. Lee, C. H. See, T. Han, N. P. Sandler, and T. C. Chong, Appl. Phys. Lett (1997) 7 A. Grill and D. A. Neumayer, J. Appl. Phys. 94, 6697 (2003) 8 P. T. Liu, T. C. Chang, S. M. Sze, F. M. Pan, Y. J. Mei, W. F. Wu, M. S. Tsai, B. T. Dai, C. Y. Chang, F. Y. Shih, H. D. Huang, Thin Solid Films 332, 345 (1998) 9 S. Lee and J. W. Park, J. Appl. Phys. 80, 5260 (1996) 10 C. V. Nguyen, K. R. Carter, C. J. Hawker, J. L. Hedrick, R. L. Jaffe, R. D. Miller, J. F. Remenar, H.-W. Rhee, P. M. Rice, M. F. toney, M. Trollsas, and D. Y. Yoon, Chem. Mater. 11, 3080 (1999) 11 A. Grill, J. Appl. Phys. 93, 1785 (2003) 12 A. A. Volinsky, J. B. Vella, and W. W. Gerberich, Thin Solid Films 429, 201 (2003) 13 B. B. Triplett, P. T. Chen, Y. Nishi, P. H. Kasai, J. J. Chambers, and L. Colombo, J. Appl. Phys. 101, (2007) 9

10 14 A. Stesmans and V. V. Afanas ev, M. Houssa, J. Non-Cryst. Solids 303, 162 (2002) 15 L. J. F. Broer, Physica 10, 810 (1943) 16 F. Bloch, Phys. Rev. 70, 460 (1946) 17 H. Ren, S.L. Cheng, Y. Nishi and J.L. Shohet, Appl. Phys. Lett. 96, (2010) 18 D. C. Koningsberger and T. De Neef, Chem. Phys. Lett. 4, 615 (1970) 19 M. Tabib-Azar, D. Akinwande, G. E. Ponchak, and S. R. LeClair, Rev. Sci. Instrum. 70, 3083 (1999) 20 Y. Nishi, Jap. J. Appl. Phys. 10, 52 (1971) 21 A. Stesmans and V. V. Afanas ev, J. Appl. Phys. 97, (2005) 22 E. H. Poindexter, P. J. Caplan, B. E. Deal, and R. R. Razouk, J. Appl. Phys. 52, 879 (1981) 23 E. P. O Reilly and J. Robertson, Phys. Rev. B 27, 3780 (1983) 24 K. Maex, M. R. Baklanov, D. Shamiryan, F. Lacopi, S. H. Brongersma, and Z. S. Yanovitskaya, J. Appl. Phys. 93, 8793 (2003) 10

11 Figure and Table Captions Figure 1. Physical interpretations of (a) Broer model and (b) Bloch model. Broer model shows the dielectric sample with spins parallel or antiparallel to the magnetic field strength while Bloch model shows dielectrics with randomly orientated spins of defects. Figure 2. ESR signals for HfO 2, defect states were fitted by (a) Broer model and (b) Bloch model. The g-factors for P b0, P b1, and E states in this case are , , and respectively. Figure 3. ESR signals for SiCOH, defect states were fitted by (a) Broer model and (b) Bloch model. The g-factor for bulk defect state in this case is Table 1. Magnetic field widths and defect concentrations for thin dielectric films. 11

12 He Ren-Figure 1 12

13 He Ren-Figure 2 13

14 He Ren-Figure 3 14

15 He Ren-Table 1 Dielectric Film and Defects B-field Width (gauss) Defect Concentrations (cm -2 ) HfO 2 (Broer Model) SiCOH (Bloch Model) P b ± P b ± E 2.8 ± Bulk Defect 4.05 ±