You are on page 1of 4

1112 IEEE ELECTRON DEVICE LETTERS, VOL. 42, NO.

8, AUGUST 2021

Improving Low-Frequency Noise in 14-nm


FinFET by Optimized High-k/Metal Gate
Thermal Processing
Hao Zhu , Bin Ye , Chengkang Tang, Xianghui Li , Qingqing Sun , and David Wei Zhang

Abstract — Low-frequency noise has become one of the experimentally identify the defect traps in a non-destructive
critical factors in ultra-scaled MOSFET devices, and is also manner. Various models have been theoretically proposed
effective as an evaluating tool in characterizing device seeking the origin of 1/ f noise in FinFET which provide
structure and reliability. Here, the low-frequency noise good guidelines to optimize the fabrication process [1]–[5].
in 14 nm-FinFET is studied, and its dependence on trap In the gate stack of FinFET devices, the SiO2 interface
defects of high-k dielectric and interface is further experi-
layer is typically prepared by wet chemistry. The surface
mentally investigated. By using NH3 /N2 thermal processing
in high-k/metal gate (HKMG) module, the 1/f noise char- of such SiO2 interlayer more favorable to the growth of
acteristics have been greatly improved. The dominating HfO2 high-k dielectric as compared to dry oxidation. The
mechanism is quantitatively analyzed focusing on the defect intrinsic defects in the wet SiO2 are suppressed by the thermal
trap density in bulk HfO2 and SiO2 interface based on processing after HfO2 deposition which is known as post-
the carrier number fluctuation model. Moreover, the carrier dielectric annealing (PDA), but with the penalty of increasing
mobility and the bias temperature instability property of the interlayer thickness degrading device reliability. Moreover, the
FinFET devices are also enhanced which further confirms HfO2 dielectric deposited by atomic layer deposition (ALD)
the improvement in low-frequency noise upon optimized is generally oxygen-deficient, and the defects of oxygen
thermal processing. vacancies can act as trapping centers deteriorating the 1/ f
Index Terms — Low-frequency noise, thermal annealing, noise characteristics. Additional thermal processing like post-
FinFET, gate dielectric, reliability. metal annealing (PMA) in the HKMG module can drive more
oxygen diffusing from HfO2 to the SiO2 interface generating
additional interface traps. This will not only further degrade
I. I NTRODUCTION the 1/ f noise performance but also lower the BTI reliability of
T HE continuous downscaling of MOS transistors is degrad-
ing the signal-to-noise ratio, and further device explo-
ration at advanced technology node is severely limited by the
the device. HfO2 -based dielectric nitridation by NH3 annealing
has been found with effective improvement in defect trap
density [6], [7]. However, there still lacks a comprehensive
low-frequency noise, which is inversely dependent on the gate analysis of the nitridation impact on the DC performance and
size. Moreover, the employment of high-k dielectric replac- reliability characteristics including low-frequency noise and
ing conventional SiO2 as the gate oxide introduces defects BTI property based on 14-nm FinFET technology.
between the gate and channel, and increases the low-frequency In this work, for the first time, we have optimized the
noise due to its high defect-sensitivity. The technical challenge thermal processing in the HKMG module to improve the 1/ f
to form thinner and taller fin structures in novel FinFET noise characteristics in 14-nm FinFET devices. By introducing
can also generate defective channel/dielectric interface which NH3 /N2 annealing after PDA, the oxygen vacancies in HfO2
further degrades the low-frequency noise performance. are passivated and SiON is formed in the SiO2 interface layer
As a typical dominating low-frequency noise, the 1/ f noise effectively prohibiting the oxygen diffusion. The carrier num-
has been widely studied in solid-state electronic devices. ber fluctuation mechanism is quantitatively analyzed, and the
In addition, 1/ f noise characterization also enables a metrol- 1/ f noise is significantly improved upon optimized processing
ogy approach to evaluate gate dielectric quality as well as to which enhances the carrier mobility and reliability of the
device at the same time.
Manuscript received May 22, 2021; revised June 11, 2021 and
June 17, 2021; accepted June 18, 2021. Date of publication June 22,
2021; date of current version July 26, 2021. This work was supported in
part by the National Key Research and Development Program of China II. D EVICE FABRICATION AND E XPERIMENTAL D ETAILS
under Grant 2018YFB2202800, in part by NSFC under Grant 61904033, The FinFET devices are fabricated based on the Semi-
and in part by the Science and Technology Commission of Shanghai
Municipality under Grant 20501130202. The review of this letter was conductor Manufacturing International Corporation (SMIC)
arranged by Editor K. J. Kuhn. (Corresponding authors: Hao Zhu; 14-nm FinFET CMOS process platform. The length of the
Qingqing Sun.) 20-fin channel is 240 nm for both nMOS and pMOS devices.
The authors are with the State Key Laboratory of ASIC and System, Fig. 1 shows the process flow of the HKMG module used
School of Microelectronics, Fudan University, Shanghai 200433, China
(e-mail: hao_zhu@fudan.edu.cn; qqsun@fudan.edu.cn).
in this work. Based on the baseline process, a millisecond
Color versions of one or more figures in this letter are available at annealing (MSA) in NH3 with 1.4 ms annealing time is
https://doi.org/10.1109/LED.2021.3091488. performed after PDA as an experimental split. 3.5% and 10%
Digital Object Identifier 10.1109/LED.2021.3091488 NH3 in N2 environment are tested and compared.
0741-3106 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: UNIVERSITY SABAH MALAYSIA. Downloaded on August 28,2021 at 04:15:02 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: IMPROVING LOW-FREQUENCY NOISE IN 14-nm FinFET 1113

Fig. 1. Process flow of the HKMG module. The millisecond thermal


processing in NH3 is performed after PDA as an experimental split.

Fig. 3. The Sid /I2d as a function of Id (blue) and the dependence


of (gm /Id )2 on Id (red) of (a) baseline nMOS, (b) baseline pMOS,
(c) 10% NH3 MSA nMOS and (d) 10% NH3 MSA pMOS.

Fig. 4. XPS spectra of HfO2 dielectric after different thermal processing.


Fig. 2. The normalized drain current PSD as vs. frequency of (a) nMOS
and (b) pMOS devices. Numerical ratio of normalized PSD between
different process and baseline of (c) nMOS and (d) pMOS devices.
As mentioned above, the bulk defects of oxygen vacancies
in HfO2 dielectric as well as the interface traps at the Si inter-
III. R ESULTS AND D ISCUSSION layer can trap and de-trap carriers degrading the low-frequency
Fig. 2(a) and 2(b) show the normalized drain current power noise property. X-ray photoelectron spectroscopy (XPS) is
spectral density (PSD) Sid /I2d of nMOS and pMOS devices used to analyze and compare the elemental bonding char-
based on the baseline and 3.5%/10% NH3 MSA treatment, acteristics in HfO2 with MSA processing in N2 and NH3 .
respectively. The spectra well follow the 1/ f characteristics As shown in Fig. 4, the presence of feature peaks at 395.7 eV
from 10 Hz to 100k Hz, and the PSD is clearly lowered observed from spectra of the NH3 MSA sample indicates
upon NH3 MSA processing. As shown in Fig. 2(c) and 2(d), formation of stable N-Hf bonds. The atomic concentration
a maximum reduction of 37% and 45% in normalized PSD of N in HfO2 is 2.0% and 2.7% with 3.5% and 10% NH3
has been achieved for nMOS and pMOS devices, respectively MSA, respectively. The introduction of N atoms in HfO2
by using 10% NH3 MSA processing. can compensate the oxygen vacancies and suppress the bulk
According to the analysis based on the Hooge’s mobility trap density in the gate dielectric, and thus, improve the 1/ f
fluctuation (HMF) model [8] and McWhoter’s carrier number noise performance. It should be noted that a shoulder peak at
fluctuation (CNF) model [9], the Sid /I2d can be described 396.7 eV is also observed which is ascribed to the of N-Si
as [10]: bonding at the SiO2 interlayer [11]. Such N atom diffusion
to the SiO2 interlayer is favorable for the formation of SiON
Sid /Id2 = (gm /Id )2 × (1 +  × Id /gm )2 × Sv f b (1) which not only increases the permittivity of SiO2 interlayer,
but also functions as a barrier prohibiting oxygen diffusion to
where Svfb is the PSD of flat-band voltage fluctuation,  is the the interlayer in subsequent thermal processing.
CNF factor and gm is the device transconductance. If the 1/ f We then characterize the influence of bulk oxide traps on
noise is dominated by CNF, Sid /I2d is proportional to (gm /Id )2 the 1/ f noise with different thermal processing based on CNF
with  = 0. Fig. 3 shows the dependence of Sid /I2d on Id for model. To extract the oxide trap density (Not ), we rewrite
the baseline and 10% NH3 MSA devices (nMOS and pMOS). Eq. (1) as:
Sid /I2d varies with Id as (gm /Id )2 for all samples, suggesting the  √ √
dominating mechanism of CNF model in our 14-nm FinFETs. (S id /Id2 ) = S v f b × gm /Id +  × S v f b (2)

Authorized licensed use limited to: UNIVERSITY SABAH MALAYSIA. Downloaded on August 28,2021 at 04:15:02 UTC from IEEE Xplore. Restrictions apply.
1114 IEEE ELECTRON DEVICE LETTERS, VOL. 42, NO. 8, AUGUST 2021

Fig. 6. (a) Extracted interface trap density and inversion layer thickness
 under different processing conditions. (b) Effective mobility of nMOS
Fig. 5. Linear fitting of (Sid /Id2 ) to gm /Id of (a) nMOS and (b) pMOS and pMOS FinFET devices with and without 10% NH3 MSA annealing.
devices. Frequency is 10k Hz. Extracted bulk trap density as a function (c) nMOS PBTI and (d) pMOS NBTI reliability of the devices at 125 ◦ C
of trap depth of (c) nMOS and (d) pMOS devices. Data obtained at using baseline and 10% NH3 MSA processing. Vg is 1.5 V.
frequency of 10 Hz to 100k Hz were used.
√ processing. This is consistent with reported results that gate
S v f b can be extracted as the slope from the linear fitting stack nitridation with small thermal budget (like MSA used
of (S id /Id2 ) to gm /Id under different gate voltage, which here) can lower the Nit and Tinv due to the formation of
SiON barrier at the SiO2 /HfO2 interface after NH3 thermal
is shown in Fig. 5(a) and 5(b). Thus, Not can be further processing [14], [15]. Based on the above results, we further
calculated by: characterized the carrier mobility which can be easily damp-
Sv f b = (q 2 kT λNot )/(W LC O
2 ened by Coulomb scattering at the oxygen vacancies. By using
X f) (3)
split C-V measurement, the calculated effective mobility is
where λ is the oxide tunneling attenuation factor, Cox is the plotted in Fig. 6(b) which demonstrates enhancement with
gate capacitance per unit area, f is the frequency, W and L are NH3 MSA processing due to the minimizing of defective
the channel width and length, respectively. At 10k Hz, Not is trapping centers. Fig. 6(c) and 6(d) show the BTI reliability
decreased from 1.0 × 1018 to 6.8 × 1017 for nMOS and from characterization results. Five devices located at different sites
7.0 × 1018 to 4.1 × 1018 for pMOS after 10% NH3 MSA, on the wafer are selected for each process condition. With
respectively. 10% NH3 MSA processing, it is observed that the interface
Furthermore, the depth distribution (xt ) of the bulk traps trap-dominated pMOS NBTI degradation [16] is suppressed
in HfO2 can be evaluated from the noise level at various with smaller threshold voltage Vt shift (Fig. 6(c)). On the
frequency according to xt = λln(1/2π f τ0 ), where τ0 is the other hand, the nMOS PBTI affected by the oxygen vacancies
Shockley-Read-Hall recombination time with a typical value in HfO2 dielectric is also improved with less electron trap-
of 10−10 s [12]. By assuming Not as the trapping centers based ping and de-trapping. Such results are well aligned with the
on the carrier number fluctuation mechanism, the distribution improvement in 1/ f noise illustrating effective optimization
of trap density in HfO2 dielectric can be estimated by plotting in device structure and electrical performance with engineered
the calculated Not versus xt for nMOS and pMOS, which thermal processing.
are shown in Fig. 5(c) and 5(d), respectively. It is found that
despite the much lowered Not after NH3 annealing, the peak IV. C ONCLUSION
trap density in the baseline sample is located at 1.2 ∼ 1.4 nm In summary, improved 1/ f noise characteristics have been
depth, while it is closer to the SiO2 interlayer in the MSA achieved in 14-nm FinFET devices through optimized thermal
treated samples. It is worth to mention that the dielectric bulk processing. By introducing MSA in NH3 environment after
trap density is higher in pMOS than that in nMOS devices PDA in HKMG module, N atoms are successfully doped in the
as shown in Fig. 4. This is because more electrons tend to HfO2 dielectric to compensate the oxygen vacancies lowing
transfer to the metal gate in pMOS due to the higher metal the bulk trap density. Meanwhile, such thermal processing also
work function, and more oxygen vacancies are left behind forms a SiON layer at the SiO2 interface which effectively
leading to flat-band voltage shift which is known as the Fermi- prohibits oxygen diffusion from HfO2 to the substrate. This
level pinning effect [13]. not only lowers the oxygen vacancies trap density in the bulk
In addition to the HfO2 bulk traps, the interface states HfO2 , but also minimize the growth of the SiO2 interlayer as
at the SiO2 interlayer can also contribute to the carriers well as the interface trap density. The 1/ f noise performance
trapping and de-trapping. By introducing thermal process- is thus enhanced based on the CNF model and the device
ing in NH3 , the interface trap density (Nit ) extracted by performance including the mobility and BTI reliability is also
charge-pumping method is lowered for both nMOS and improved. This provides a promising pathway for the device
pMOS devices (Fig. 6(a)). Fig. 6(a) also indicates that the study and process optimization in advanced semiconductor
inversion layer thickness (Tinv ) decreases with NH3 MSA technology nodes.

Authorized licensed use limited to: UNIVERSITY SABAH MALAYSIA. Downloaded on August 28,2021 at 04:15:02 UTC from IEEE Xplore. Restrictions apply.
ZHU et al.: IMPROVING LOW-FREQUENCY NOISE IN 14-nm FinFET 1115

R EFERENCES [8] F. N. Hooge, “1/f noise sources,” IEEE Trans. Electron Devices, vol. 41,
no. 11, pp. 1926–1935, Nov. 1994, doi: 10.1109/16.333808.
[1] A. V. de Oliveira, D. Xie, H. Arimura, G. Boccardi, N. Collaert,
[9] A. L. McWhorter and R. H. Kingston, Semiconductor Surface Physics.
C. Claeys, N. Horiguchi, and E. Simoen, “Low-frequency noise
Philadelphia, PA, USA: Univ. of Pennsylvania Press, 1957.
characterization of germanium n-channel FinFETs,” IEEE Trans. [10] G. Ghibaudo, O. Roux, C. Nguyen-Duc, F. Balestra, and J. Brini,
Electron Devices, vol. 67, no. 7, pp. 2872–2877, Jul. 2020, doi: “Improved analysis of low frequency noise in field-effect MOS tran-
10.1109/TED.2020.2990714. sistors,” Phys. Status Solidi A, vol. 124, no. 2, pp. 571–581, Apr. 1991,
[2] P. Kushwaha, H. Agarwal, Y.-K. Lin, A. Dasgupta, M.-Y. Kao, Y. Lu, doi: 10.1002/pssa.2211240225.
Y. Yue, X. Chen, J. Wang, W. Sy, F. Yang, P. C. Chidambaram, [11] M. Dai, J. Liu, D. Guo, S. Krishnan, J. F. Shepard, P. Ronsheim,
S. Salahuddin, and C. Hu, “Characterization and modeling of flicker U. Kwon, S. Siddiqui, R. Krishnan, Z. Li, K. Zhao, J. Sudijono, and
noise in FinFETs at advanced technology node,” IEEE Elec- M. P. Chudzik, “A novel atomic layer oxidation technique for EOT
tron Device Lett., vol. 40, no. 6, pp. 985–988, Jun. 2019, doi: scaling in gate-last high-κ/metal gate CMOS technology,” in IEDM Tech.
10.1109/LED.2019.2911614. Dig., Dec. 2011, pp. 28.5.1–28.5.4, doi: 10.1109/IEDM.2011.6131632.
[3] C. G. Theodorou, N. Fasarakis, T. Hoffman, T. Chiarella, G. Ghibaudo, [12] J. W. Lee, W. S. Yun, and G. Ghibaudo, “Impact of trap localization
and C. A. Dimitriadis, “Origin of the low-frequency noise in n-channel on low-frequency noise in nanoscale device,” J. Appl. Phys., vol. 115,
FinFETs,” Solid-State Electron., vol. 82, pp. 21–24, Apr. 2013, doi: no. 19, May 2014, Art. no. 194501, doi: 10.1063/1.4878456.
10.1016/j.sse.2013.01.009. [13] Y. Akasaka, G. Nakamura, K. Shiraishi, N. Umezawa, K. Yamabe,
[4] R. Talmat, H. Achour, B. Cretu, J.-M. Routoure, A. Benfdila, O. Ogawa, M. Lee, T. Amiaka, T. Kasuya, H. Watanabe, T. Chikyow,
R. Carin, N. Collaert, A. Mercha, E. Simoen, and C. Claeys, “Low F. Ootsuka, Y. Nara, and K. Nakamura, “Modified oxygen vacancy
frequency noise characterization in n-channel FinFETs,” Solid-State induced Fermi level pinning model extendable to P-metal pinning,”
Electron., vol. 70, pp. 20–26, Apr. 2012, doi: 10.1016/j.sse.2011. Jpn. J. Appl. Phys., vol. 45, no. 49, pp. L1289–L1292, Dec. 2006, doi:
11.007. 10.1143/JJAP.45.L1289.
[5] J. Woo Lee, M. ju Cho, E. Simoen, R. Ritzenthaler, M. Togo, [14] H. Arimura, K. Wostyn, L.-A. Ragnarsson, E. Capogreco, A. Chasin,
G. Boccardi, J. Mitard, L.-Å. Ragnarsson, T. Chiarella, A. Veloso, T. Conard, S. Brus, P. Favia, J. Franco, J. Mitard, S. Demuynck,
N. Horiguchi, A. Thean, and G. Groeseneken, “1/f noise analysis and N. Horiguchi, “Ge oxide scavenging and gate stack nitrida-
of replacement metal gate bulk p-type fin field effect transistor,” tion for strained Si0.7 Ge0.3 pFinFETs enabling 35% higher mobility
Appl. Phys. Lett., vol. 102, no. 7, Feb. 2013, Art. no. 073503, doi: than Si,” in IEDM Tech. Dig., Dec. 2019, pp. 29.2.1–29.2.4, doi:
10.1063/1.4793306. 10.1109/IEDM19573.2019.8993467.
[6] M. S. Akbar, H.-J. Cho, R. Choi, C. S. Kang, C. Y. Kang, C. H. Choi, [15] Y. He, D. W. Zhang, H. Liu, Y. Chen, G. Yu, Y. He, L. Jin, J. Wu,
S. J. Rhee, Y. H. Kim, and J. C. Lee, “Optimized NH3 annealing J. Zhao, W. Song, S. Yu, and J. Wu, “Investigation of different post HK
process for high-quality HfSiON gate oxide,” IEEE Electron Device annealing impact on HK film property and device performance,” in Proc.
Lett., vol. 25, no. 7, pp. 465–467, Jul. 2004, doi: 10.1109/LED.2004. 20th Int. Conf. Ion Implantation Technol. (IIT), Jun. 2014, pp. 1–4, doi:
830270. 10.1109/IIT.2014.6939969.
[7] P. Srinivasan, E. Simoen, Z. M. Rittersma, W. Deweerd, L. Pantisano, [16] E. Cartier, A. Kerber, T. Ando, M. M. Frank, K. Choi, S. Krishnan,
C. Claeys, and D. Misra, “Effect of nitridation on low-frequency (1/f) B. Linder, K. Zhao, F. Monsieur, J. Stathis, and V. Narayanan, “Fun-
noise in n- and p-MOSFETs with HFO2 gate dielectrics,” J. Elec- damental aspects of HfO2 -based high-k metal gate stack reliability
trochem. Soc., vol. 153, no. 9, pp. G819–G825, Jul. 2006, doi: and implications on tinv-scaling,” in IEDM Tech. Dig., Dec. 2011,
10.1149/1.2216455. pp. 18.4.1–18.4.4, doi: 10.1109/IEDM.2011.6131579.

Authorized licensed use limited to: UNIVERSITY SABAH MALAYSIA. Downloaded on August 28,2021 at 04:15:02 UTC from IEEE Xplore. Restrictions apply.

You might also like