This is the accepted version of the article: Ruiz, Ana; Seoane, Natalia; Claramunt, Sergi; [et al.]. «Workfunction fluctua- tions in polycrystalline TiN observed with KPFM and their impact on MOSFETs variability». Applied Physics Letters, Vol. 114, issue 9 (March 2019), art. 93502. DOI 10.1063/1.5090855 This version is available at https://ddd.uab.cat/record/249164 under the terms of the license > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1 Workfunction fluctuations in polycrystalline TiN observed with KPFM and their impact on MOSFETs variability A. Ruiz1, N. Seoane2, S. Claramunt1, A. García-Loureiro2, M. Porti1, C. Couso1, J. Martin-Martinez1, M. Nafria1 1 Electronic Engineering Department, Universitat Autònoma de Barcelona, Barcelona 08193, Spain 2 CITIUS, Universidade de Santiago de Compostela, Santiago de Compostela 15782, Spain Electronic address: ana.ruiz@uab.cat Abstract— A more realistic approach to evaluate the impact of From KPFM measurements, the metal gate WF fluctuations polycrystalline metal gates on the MOSFET variability is are determined at the nanoscale. This information is then presented. 2D experimental workfunction maps of a introduced into a device simulator to analyze the impact of the polycrystalline TiN layer were obtained with Kelvin Probe Force metal granularity on the variability of the device electrical Miscroscopy with nanometer resolution. These data were the properties. input of a device simulator, which allowed to evaluate the effect of the workfunction fluctuations on MOSFET performance Experimental data were obtained from a sample containing a variability. We have demonstrated that in the modelling of TiN 100 nm thick TiN layer grown by continuous e-gun workfunction variability not only the different workfunctions of evaporation of metallic Ti over a HfO2/Si substrate. The the grains but also the grain boundaries should be included. formation of TiN was ensured by passing the Ti atoms through a reactive nitrogen-enriched atmosphere, at a nitrogen partial Index Terms— KPFM, polycrystalline metal gate, variability, MOSFET. pressure of 8ꞏ10-4 mbars. The structure of the TiN layer was determined by X-Ray Diffraction (XRD) using a Panalytical X’PERT PRO with the CuK line in Bragg-Brentano T HE continuous scaling of MOSFET dimensions has led to geometry. The morphological and electrical properties of the the introduction of high-k/metal gate stacks in recent polycrystalline metal layer have been measured with a Nano- technological nodes. However, high-k dielectric Observer AFM (from Concept Scientific Instruments), which allows bimodal single pass AM-KPFM measurements to polycrystallization has already been shown to be a variability obtain simultaneously topographical and sample-tip contact source in ultrascaled MOSFETs affecting its electrical potential difference (CPD) 2D maps with nanometer properties1–5. Metal gates were also introduced with high-k resolution12,13. Note that although FM-KPFM usually offers a dielectrics because they have several advantages compared to better resolution than AM-KPFM14–16, bimodal single pass polysilicon6,7. However, depending on the growth temperature, AM-KPFM measurements, thanks to the smaller tip sample metals become polycrystalline, leading to grains with different distance, can offer comparable resolution to lift mode FM- sizes and orientations. Since the grains workfunction (WF) KPFM14,17,18. The first flexural eigenmode resonance depends on their orientation8, the random distribution of grains frequency of the cantilever (f1 = 61.3 kHz) was used to track (with their corresponding WF) results also in variations in the the topography. The feedback to measure the CPD of the metal WF and threshold voltage variability (TVV)9–11. sample works at a frequency around the second eigenmode Therefore, it is critical to understand the origin of this resonance frequency of the cantilever15,19 (f2 = 380 kHz, high variability in order to select the best fabrication process and enough to avoid any KPFM signal dependence on the materials that can reduce it. frequency20). An internal algorithm selects the working Some works have already evaluated the impact of the grain frequency that maximizes the signals used to nullify the orientation induced workfunction variation (WFV) on the electrically driving oscillation, improving the sensitivity of the variability of MOSFETs. In Ref. 11, for example, it was measurements. Since the ratio f2/f1 is typically 6.2, self- experimentally demonstrated that the TVV of a MOSFET is excitation of second eigenmode through the 6th-harmonic of affected by the grain size. However, since this study was the first eigenmode is not present in the measurements, performed at device level, the individual impact of the WFV minimizing cross-talk between topography and CPD image21. on the device properties cannot be separated from other To obtain the CPD image, the tip is biased at VAC = 1 V while the TiN layer is directly connected to ground. Since the sources as random dopant fluctuation (RDF) or line edge CPD data correspond to the difference between the WF of the roughness (LER). Other works have evaluated the TVV by tip and the metal layer, assuming a constant value of the tip means of simulation and statistical models that make WF during the experiment, the measured CPD variations are assumptions that could be unrealistic9,10. related to the WF fluctuations of the metal layer. Therefore, In this work, a more realistic approach which relies on 2D the CPD image provides information of the local value of the experimental WF maps obtained with Kelvin Probe Force WF of the metal. In our case, KPFM images were obtained in Microscopy (KPFM) is proposed to study the impact of metal air and using a highly doped20 Si tip (radius <10 nm) from polycrystallization on the electrical parameters of MOSFETs. AppNano. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2 Fig. 1a presents a 850 nm x 290 nm topographical image of the TiN layer obtained with KPFM at a scanning rate of 0.2 Hz. Note that it shows a granular structure, which has been attributed to the polycrystallization of the metal layer: grains (Gs) are related to individual (or a cluster of) randomly oriented nanocrystals separated by grain boundaries (GBs, depressions in the topographical image)2. A statistical analysis of the Gs diameter has been done using the image analysis software Gwyddion22. Fig. 2a shows a histogram of the grain diameter suggesting an average diameter (Gdiameter) of ~25.4 nm. These results are compatible with values already reported, which point out a large range of diameters, depending on the growth conditions of the layer9,10. The effect of the metal polycrystallization on the nanoscale WF of the layer was investigated based on the CPD image. Fig. 1b shows the measured WF at the same surface region as in Fig. 1a, suggesting that the WF is not uniform. A granular pattern, as also observed in Ref. 23, overlaps with that of the topographical image. Since the tip WF is not known and Fig. 1. Topography (a) and WF (b) maps obtained with KPFM on a absolute values of WF are also very sensitive to ambient TiN layer (850 nm x 290 nm). (c) Topographical (continuous line) conditions20, it has been assumed that the average value of the and WF (dashed line) profile across the line plotted in Fig. 1.a-b. CPD image is 4.22 eV, which corresponds to the average WF of a TiN metal gate obtained from experimental devices24, and only relative variations will be considered. This assumption does not affect the conclusions of the work because we are interested on the impact of the WF fluctuations on the MOSFET variability. Fig. 1c shows a topographical (continuous line) and WF (dashed line) profile across the line plotted in Fig. 1a and 1b (inside the black square). Note that, in general, the positions with lower WF are located along the topographical depressions, though Gs with very low WF can also be found. This qualitative observation is verified statistically in Fig. 2b, where the dependence of the WF with the Z-axis relative position (Zrel, defined as the Z-position with respect to the Z- Fig. 2. (a) Histogram showing the distribution of nanocrystals diameter. The average diameter is 25.4 nm. (b) Relationship between axis mean value of the image, which has been arbitrarily the WF and topography (Z-axis relative position, Zrel) map pixel by considered to be the zero-reference level) of all the pixels in pixel. Color map shows the density of points for each region. Zrel is Fig. 1 is shown. The color scale of Fig. 2b indicates the defined as the Z-position with respect to Z-axis mean value of the density of sites with a given WF and Zrel. This figure suggests image, which has been arbitrarily considered to be the zero-reference that, instead of discrete WF values, there is continuous WF level. The general trend is that depressed areas (regions with low Zrel) distribution that spans from ~4.0 eV to ~4.5 eV. The show lower WF. depressed areas (low Zrel) tend to show lower WF than hillocks. Since depressed areas are associated to GBs (as shown in Fig. 1a and 1c)2,23, the results indicate that Gs tend to have a higher WF value. In addition to the WF difference between GBs and Gs, the WF of Gs show also a continuous distribution. This distribution can be associated to, on the one hand, a non- homogeneous WF in the G or intra-G variability (see for example the G located at ~200 nm in Fig. 1c) and, on the other, to differences in the WFs between Gs (inter-G variability). To evaluate this inter-G variability, we have considered the maximum value of WF as metric parameter (the same conclusions would have been drawn if the average WF would have been considered instead). Grains with different maximum WF values are measured, as in Ref. 23. As Fig. 3. XRD spectrum of the TiN layer, showing the [111] and [200] an example, Fig. 1c shows two Gs with a similar height and orientations of TiN. different maximum WF values (4.30 and 4.48 eV), whose > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3 difference is ~200 meV. This result is compatible with the maps (and, therefore, device configurations) were analyzed. presence of two grain orientations, [111] and [200], with 0.2 Fig. 4 shows 2 extreme examples of device configurations. eV9 WF difference. The presence of crystals with 2 Fig. 4a corresponds to a device with very low TiN WF values orientations is further supported by XRD analysis. The XRD (mainly occupied with GBs), while Fig. 4b corresponds to a spectrum of the TiN layer (Fig. 3) reveals the presence of two device mainly occupied with Gs with WF values ranging from peaks, corresponding to aforementioned orientations, being ~4.29 to ~4.5 eV. The corresponding statistical analysis of the the [111] orientation dominant over the other. However, the 100 MOSFETs electrical characteristics is presented in Fig. 5 KPFM map suggests a continuous distribution of values that for the off-current (IOFF), subthreshold slope (SS), threshold spans from ~4.29 eV for low WF values to ~4.5 eV for high voltage (VT) and on-current (ION) at a low drain bias of 50 mV. WF values23, within the WFs range of the two crystal The mean value and standard deviation (σ) of the distributions orientations. A continuous distribution of WFs is also are also indicated in this figure. The VT, IOFF and SS observed for GBs. In this case, however, WFs range from 4.0 histograms are clearly asymmetric with behaviors far from the to 4.1 eV, so the WF difference between grains and GBs commonly assumed Gaussian distributions, with skewness ranges between ~200 meV and ~500 meV. values ~ ±1, because of the unbalanced grain WF The data obtained from KPFM images (WF at the probabilities. A correlation between Gs/GBs distributions and nanoscale) adds new information on the properties of the electrical characteristics of the device can be found, showing, polycrystalline metal layer compared to the data used in for example, that higher values of WF (as in Fig. 4b) lead to previous works to simulate WFV. Besides the presence of MOSFETs with higher VT and lower IOFF. Then, from the IOFF different grain orientations, we have observed that they do not and VT histograms in Fig. 5, it can be concluded that the show discrete WF values, but a continuous range of them (due probability of occurrence of Gs with extreme high WF values to inter- and/or intra-G variability). Moreover, the presence of (4.5 eV), is extremely low. GBs with lower WF values is also observed (which were not considered in previous works). Such variations could also affect the electrical characteristics and variability of semiconductor devices and, therefore, they should be taken into account. In order to demonstrate this, the nanoscale information experimentally obtained with KPFM is introduced in a device simulator to study the variability of MOSFETs (due to the WF fluctuations in the metal gate). Fig. 5. Distributions of IOFF, SS, VT and ION due to the experimentally observed WFV for a 50 nm gate length Si MOSFET at a drain bias of 50 mV. The mean value and standard deviation (σ) of the statistical Fig. 4. TiN metal gate experimental profiles that produce the highest ensembles are indicated, together with the values obtained for a (a) and lowest (b) off-currents when considered as the metal gate of a device with a uniform gate WF (set to 4.22 eV). 50 nm gate length Si MOSFET. The electron concentration inside the device, at VG=0.0 V and VD=50 mV, is also shown. To perform a fair comparison between our WF variability To evaluate the impact of the nanoscale WF fluctuations on results and those previously published by Wang et. al.26, the device electrical characteristics, a 3D in-house built finite- corresponding to a 35 nm gate length (GL) MOSFET, the ratio element drift-diffusion device simulator25 was used. For Gdiameter /GL needs to be used27. When comparing results with simplicity, a WxL= 50x50 nm2 gate area n-type Si MOSFET a similar Gdiameter /GL value of 0.5, our predicted σ VT (~ 49 with a HfO2/SiO2 gate stack was considered. The gate WF mV) is around 10% lower than that reported by Wang. These value (4.22 eV) and Source/Drain doping were obtained from large variability values explain why WFV is currently the appropriate scaling and calibration of a 67 nm effective considered one of the major sources of variability impacting gate length MOSFET experimental device24. From maps as device´s performance. However, most of the previously TiN those shown in Fig. 1, 100 different 50 nm x 50 nm metal gate WFV published studies26–29 assume that TiN has only two maps (without overlapping) were obtained and introduced in possible WF values spanning 0.2 eV with probabilities of the device simulator, so that an ensemble of 100 different WF > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4 occurrence 60% and 40%30. Our results demonstrate that such Tsunashima, K. Suguro, and T. Arikado, IEEE Trans. Electron Devices 48, 1604 (2001). assumption is quite simplistic since in real materials GBs may 7 W.P. Bai, S.H. Bae, H.C. Wen, S. Mathew, L.K. Bera, N. Balasubramanian, be present. We have observed a maximum excursion in VT N. Yamada, M.F. Li, and D.L. Kwong, IEEE Electron Device Lett. 26, 231 values of 0.28 V (see Fig. 5), being the mean VT of the (2005). 8 H. Dadgour, K. Endo, D. Vivek, and K. Banerjee, Tech. Dig. - Int. Electron distribution, 0.68 V, a value lower than the one obtained for a Devices Meet. IEDM 3, 5 (2008). device with a gate only composed of Gs with extreme low WF 9 H.F. Dadgour, K. Endo, V.K. De, and K. Banerjee, IEEE Trans. Electron values of 4.29 eV (0.75 V). This indicates a heavy influence of Devices 57, 2515 (2010). 10 the GBs (with WF values ~ 4.0 eV) in the statistical N.M. Idris, A. Brown, J. Watling, and A. Asenov, Ultim. Integr. Silicon 165 (2010). distribution. Note that, the gate of the device that exhibits the 11 K. Ohmori, T. Matsuki, D. Ishikawa, T. Morooka, T. Aminaka, Y. Sugita, highest off-current (see Fig. 4a) is mainly occupied by GBs T. Chikyow, K. Shiraishi, Y. Nara, and K. Yamada, Tech. Dig. - Int. Electron and Gs with WF values ranging from ~4.0 to ~4.3 eV. Devices Meet. IEDM (2008). 12 V. Iglesias, M. Lanza, K. Zhang, A. Bayerl, M. Porti, M. Nafría, X. Therefore, the influence of the GBs should not be disregarded Aymerich, G. Benstteter, Z.Y. Shen, and G. Bersuker, Appl. Phys. Lett. 99, when modelling metal gate WFV. 103510 (2011). 13 T. Berthold, G. Benstetter, W. Frammelsberger, R. Rodríguez, and M. Nafría, Thin Solid Films 584, 310 (2015). In summary, the impact of the metal gate polycrystallization 14 J.L. Garrett, D. Somers, and J.N. Munday, J. Phys. Condens. Matter 27, on the variability of MOSFET devices has been studied using 214012 (2015). 15 a combination of 2D experimental WF maps obtained with S.A. Burke, J.M. Ledue, Y. Miyahara, J.M. Topple, S. Fostner and P. Grütter, Nanotechnology, 20, 264012 (2009). KPFM and a device simulator. The KPFM shows WF 16 T. Glatzel, S. Sadewasser, and M.C. Lux-Steiner, Appl. Surf. Sci. 210, 84 variations between Gs and GBs (being the WF lower in the (2003). GBs). Though two crystal orientations are observed 17 T. Ouisse, M. Stark, F. Rodrigues-Martins, B. Bercu, S. Huant, and J. Chevrier, Phys. Rev. B - Condens. Matter Mater. Phys. 71, 205404 (2005). (supported by XRD data), a continuous distribution of WFs is 18 T. Ouisse, F. Martins, M. Stark, S. Huant, and J. Chevrier, Appl. Phys. Lett. measured. Regions of the WF images have been mapped onto 88, 043102 (2006). the gate of a MOSFET, to evaluate the device ID-VG curves. 19 Y. Miyahara, J. Topple, Z. Schumacher, and P. Grutter, Phys. Rev. Appl. 4, This procedure is a more realistic approach for the analysis of 054011 (2015). 20 C. Baumgart, A. Müller, F. Müller, and H. Schmidt, Phys. Status Solidi A the device-to-device variability (due to the presence of 208, 777 (2011). different grain orientations) since nanoscale differences of the 21 G. Li, B. Mao, F. Lan, and L. Liu, Rev. Sci. Instrum. 83, 113701 (2012). 22 metal WF are considered from direct experimental data. D. Nečas and P. Klapetek, www.gwyddion.net. 23 C.I. Enriquez-Flores, E. Cruz-Valeriano, A. Gutierrez-Peralta, J.J. Though TiN layers have been considered as case study, the Gervacio-Arciniega, E. Ramírez-Álvarez, E. Leon-Sarabia, and J. Moreno- proposed methodology can be extended to any metal layer Palmerin, Surf. Eng. 34, 660 (2018). 24 (polycrystalline or not) and growth conditions, since the A.J. García-Loureiro, K. Kalna, and A. Asenov, AIP Conf. Proc. 780, 239 (2005). electrical properties only depend on the WF. 25 A.J. Garcia-Loureiro, N. Seoane, M. Aldegunde, R. Valin, A. Asenov, A. Martinez, and K. Kalna, IEEE Trans. Comput. Des. Integr. Circuits Syst. 30, 841 (2011). This work has been partially supported by the Spanish AEI 26 X. Wang, A.R. Brown, N. Idris, S. Markov, G. Roy, and A. Asenov, IEEE and ERDF (TEC2016-75151-C3-1-R, TEC2014-53909-REDT Trans. Electron Devices 58, 2293 (2011). 27 and RYC-2017-23312). D. Nagy, G. Indalecio, A.J. García-Loureiro, M.A. Elmessary, K. Kalna, and N. Seoane, IEEE Trans. Electron Devices 64, 5263 (2017). 28 S.M. Nawaz and A. Mallik, IEEE Electron Device Lett. 37, 958 (2016). 1 K. Shubhakar, K.L. Pey, S.S. Kushvaha, S.J. O’Shea, N. Raghavan, M. 29 R. Saha, B. Bhowmick, and S. Baishya, IEEE Trans. Electron Devices 64, Bosman, M. Kouda, K. Kakushima, and H. Iwai, Appl. Phys. Lett. 98, 072902 969 (2017). (2011). 30 H.F. Dadgour, K. Endo, V.K. De, and K. Banerjee, IEEE Trans. Electron 2 V. Iglesias, M. Porti, M. Nafría, X. Aymerich, P. Dudek, T. Schroeder, and Devices 57, 2504 (2010). G. Bersuker, Appl. Phys. Lett. 97, 262906 (2010). 3 C. Couso, M. Porti, J. Martin-Martinez, A.J. Garcia-Loureiro, N. Seoane, and M. Nafria, IEEE Electron Device Lett. 38, 637 (2017). 4 O. Pirrotta, L. Larcher, M. Lanza, A. Padovani, M. Porti, M. Nafría, and G. Bersuker, J. Appl. Phys. 114, 134503 (2013). 5 K. Murakami, M. Rommel, B. Hudec, A. Rosová, K. Hušeková, E. Dobročka, R. Rammula, A. Kasikov, J.H. Han, W. Lee, S.J. Song, A. Paskaleva, A.J. Bauer, L. Frey, K. Fröhlich, J. Aarik, and C.S. Hwang, ACS Appl. Mater. Interfaces 6, 2486 (2014). 6 A. Yagishita, T. Saito, K. Nakajima, S. Inumiya, K. Matsuo, T. Shibata, Y.