Investigation of Stress Memorization Process on Low-Frequency

Document Sample
Investigation of Stress Memorization Process on Low-Frequency Powered By Docstoc
					                                                                                                             SS10038 41 Total pages 4
Japanese Journal of Applied Physics 50 (2011) 04DC20                                                                       REGULAR PAPER
DOI: 10.1143/JJAP.50.04DC20

Investigation of Stress Memorization Process on Low-Frequency Noise Performance
for Strained Si n-Type Metal–Oxide–Semiconductor Field-Effect Transistors
Cheng-Wen Kuo, San-Lein Wu1 , Hau-Yu Lin, Yao-Tsung Huang, Shoou-Jinn Chang,
De-Gong Hong1 , Chung-Yi Wu1 , Yao-Chin Cheng2 , and Osbert Cheng2
Institute of Microelectronics and Department of Electrical Engineering, Center for Micro/Nano Science and Technology,
Advanced Optoelectronic Technology Center, National Cheng Kung University, Tainan 70101, Taiwan
  Department of Electronic Engineering, Cheng Shiu University, Niaosong, Kaohsiung 833, Taiwan
  United Microelectronics Corporation, Ltd., Tainan Science-Based Industrial Park, Tainan 74145, Taiwan
Received September 16, 2010; accepted December 13, 2010; published online April 20, 2011

  The use of low-frequency (1=f ) noise to evaluate low-cost stress-memorization technique (SMT) induced-stress in n-type metal–oxide–
  semiconductor field-effect transistors has been investigated. As compared to device without SMT process, the comparable 1=f noise level
  obtained for strained Si devices with the low-cost SMT process indicates that adding the low-cost SMT process will not affect the Si/SiO2 interface
  quality. Moreover, through observing experiment result and Hooge’s parameter H , the mechanism of 1=f noise in the both devices can be
  properly interpreted by the carrier number fluctuations correlated mobility fluctuations (unified model).
  # 2011 The Japan Society of Applied Physics

1. Introduction
The low frequency (1=f ) noise is being considered regarding
continuously scaling down complementary metal–oxide–
semiconductor (CMOS) devices due to the 1=f noise
increases as the reciprocal of the device area.1) Excessive
1=f noise and fluctuations in the nanoscale transistors could
lead to serious limitation of functionality of the analog,
digital, mixed-signal, and radio frequency circuits.2) More-
over, 1=f noise is also a viable characterization tool that used
in the area of Si metal–oxide–semiconductor field-effect
transistors (MOSFETs) to examine quality of interfacial
physics,1) contacts, and might be also used for studying
degradation processes. As strain engineering have been
commonly incorporated in the state-of-the-art CMOS
                                                                             Fig. 1. (Color online) Schematic of NMOSFET structure with process
technologies to enhance the device performance,3–6) such                     flow sequences for SMT process.
as locally tensile strained channel technique induced by the
contact etch-stop layer (CESL) on top of the salicide,
embedded SiGe in the source/drain (S/D) area or stress-                      precursors. The silane/ammonia ratio (flow rate ratio) was
memorization technique (SMT). The use of the 1=f noise                       10, the power was set at 100 W, and the temperature was
characteristics to monitor the strain-induced interface                      300  C. This was followed by the standard activation anneal
properties in strained devices is particularly important.                    performed at 1050  C. The 40-nm tensile stress SiN film was
   On the other hand, the fabrication process in Si device                   deposited on all transistors as CESL to add up the stress
have a strong impact on 1=f noise characteristic.7–10) In                    effect. The other procedures were based on standard backend
addition, the strain CESL, which is known to have a high                     process. For comparison, the control devices without SMT
concentration of hydrogen that can degrade 1=f noise                         process were also fabricated. All the devices have the same
characteristics.11) However, relatively little literature is                 equivalent oxide thicknesses (EOT) of 1.7 nm. Prior to noise
available on the 1=f noise behavior in MOSFETs with                          measurements, the DC characteristics were measured using
SMT process. In this paper, we proposed an efficient method                    an Agilent 4156C semiconductor parameter analyzer. 1=f
to observes the interface behavior which use device process                  noise measurements were carried out using the battery-
with SMT and without SMT. Moreover, experiment result                        powered SR570 low-noise current preamplifiers and the
and Hooge’s parameter are used here as a figure of merit for                  Agilent 35670A dynamic signal analyzer. Threshold voltage
distinguishing the noise model.                                              is defined when drain current of 3 Â 10À7 W=L (A)
                                                                             according to the constant-current method. The 1=f noise of
2. Experimental Procedure                                                    n-MOSFETs were measured in linear operation (VDS ¼
The strained Si nMOSFETs used in this study were                             0:05 V), while varying the gate overdrive voltage (VGS À
fabricated by 40-nm technology CMOS process on (100)                         VTH ) from the subthreshold regime to strong inversion
substrates with h100i channel orientation. The SMT process                   regime.
sequence is illustrated in Fig. 1. A tensile SiN of 22 nm was
deposited as SMT film by plasma-enhanced chemical vapor                       3. Results and Discussion
deposition (PECVD). The SMT file was deposited by                             On- and off-current curve of nMOSFETs for different gate
utilizing silane (SiH4 ) and ammonia (NH3 ) as the reaction                  length devices (LGate ¼ 40{200 nm) with and without SMT
                                                                     04DC20-1                          # 2011 The Japan Society of Applied Physics
Jpn. J. Appl. Phys. 50 (2011) 04DC20                                                                                                C.-W. Kuo et al.

Fig. 2. (Color online) On- and off-current curve of nMOSFETs for              Fig. 4. (Color online) The drain current noise (SID ) at f ¼ 10 Hz as a
different gate length devices (LGate ¼ 40{200 nm) with and without SMT        function of drain current ID for SMT and control devices.

Fig. 3. (Color online) The drain current noise (SID ) versus frequency for   Fig. 5. (Color online) Dependence of the normalized drain current noise
SMT and control devices at VGS À VTH ¼ 0:2 V.                                      2
                                                                             SID =ID at f ¼ 10 Hz on the gate overdrive voltage.

process were illustrated in Fig. 2. It can be seen that an 8%                noise source at high current region, the normalized drain
Ion improvement at Ioff ¼ 10À7 A/m was obtained for                                             2
                                                                             current noise SID =ID at f ¼ 10 Hz versus the gate overdrive
nMOSFETs with SMT, as compared to control device                             voltage VGS À VTH was shown in Fig. 5. It shows ðVGS À
(without SMT process). Such an enhancement indicates that                    VTH ÞÀm with m $ 1:25 for control devices and m $ 1:15 for
the SMT process indeed induce additional tensile stress in                   SMT devices. It was indicated that noise and resistance
the channel.                                                                 contribution from the channel, which highlights that noise
   Figure 3 presented the drain current noise spectral density               contribution of the series resistance to the overall noise is
(SID ) for SMT device and control device taken from the                      negligible.12) On the other hand, Hooge’s parameter H is
average of six devices biased at VGS À VTH of 0.2 V. Both                    used as figure of merit for evaluating the noise model, we
devices show typical 1=f 
 noise type with the frequency                     use Hooge model to analyze the 1=f noise, the normalized
 close to unity (usually between 0.9 and 1.0 for                   drain current noise in linear operation then takes the form1,13)
our devices) and no evidence of generation–recombination                                                   SID   H
components are observed.10) Figure 4 shows the SID at                                                       2
                                                                                                               ¼    ;                              ð1Þ
                                                                                                           ID    fN
f ¼ 10 Hz versus drain current for SMT and control devices.
It can be seen that the increasing SID with square of ID at                  where f is the frequency at which the noise is measured, N is
small drain current region (ID . 10À5 A) indicated a carrier                 the total number of free electrons, and Hooge parameter H
number fluctuations dominated 1=f noise. At high current                      is an empirical dimensionless constant. It should be noted
region (ID > 10À5 A), the increasing SID with ID implied                     that H value is lower than 2 Â 10À6 , the noise is attributed
that carrier number fluctuations may correlate mobility or                    to mobility fluctuations, whereas H value is higher than
S/D series resistance fluctuations. In order to further clarify               1 Â 10À3 , the noise is attributed to number fluctuation. In
correlated mobility or S/D series resistance influence of 1=f                 this paper, the H value was estimated to be 1:67 Â 10À4 and
                                                                      04DC20-2                         # 2011 The Japan Society of Applied Physics

Jpn. J. Appl. Phys. 50 (2011) 04DC20                                                                                                  C.-W. Kuo et al.

Fig. 6. (Color online) The normalized drain current noise spectral density
      2                                                                        Fig. 8. (Color online) Interface state density (Nit ) measured using base
SID =ID at a frequency f ¼ 10 Hz and the transconductance to drain current
                                                                               level sweep charge pumping for control and SMT devices.
squared ðgm =ID Þ2 as function of drain current for SMT and control devices.

                                                                                                q2 kT Nt
                                                                                       SVG ¼              ½1 þ 0 COX ðVGS À VTH ފ2 ;               ð2Þ
                                                                                                LWC2 f 

                                                                               where kT is the thermal energy,  is the tunnel attenuation
                                                                               distance (%0:1 nm), Nt is the equivalent oxide traps per unit
                                                                               area (eV/cm3 ), Cox is the gate oxide capacitance, f is the
 is the noise frequency dependence with the
                                                                               slopes,  is a Coulomb scattering coefficient, and 0 is the
                                                                               low field mobility. Based on the result in Fig. 7, Nt is
                                                                               estimated to be 2 Â 1017 eVÀ1 cmÀ3 and  is around
                                                                               3 Â 104 VÁs/C for the control device. Comparable Nt of
                                                                               1:68 Â 1017 eVÀ1 cmÀ3 and  of 3:2 Â 104 VÁs/C are
                                                                               obtained for SMT devices. In addition, the charge pumping
                                                                               measurement was also used to evaluate the interface
                                                                               property. Figure 8 can be found that the comparable interface
                                                                               quality between control and SMT device, which is consistent
Fig. 7. (Color online) The input-referred voltage noise spectral density       with the result obtained from 1=f noise measurement.
SVG ¼ SID =g2 at a frequency f ¼ 10 Hz versus gate overdrive VGS À VTH
            m                                                                  Moreover, it should also be noted the Nt level is lower than
for SMT and control devices.
                                                                               that of nanowire nMOSFET with 3.5 nm SiO2 gate dielectric
                                                                               (1:7 Â 1018 eVÀ1 cmÀ3 ) and tripe-gate FET with 1.4 nm
                                                                               EOT SiON gate dielectric ($4 Â 1018 eVÀ1 cmÀ3 ),14,15)
4:06 Â 10À4 for control and SMT devices, respectively,                         and comparable with that observed from the biaxial
which were between 2 Â 10À6 and 1 Â 10À3 , and also                            strained-Si/SiGe nMOSFET with 2.2 nm SiO2 gate dielec-
indicates that the 1=f noise can be reasonably attributed to a                 tric.16) These observations bring us to the conclusion that
unified model.13)                                                               there are no serious interface quality degradation in the
   Figure 6 shows the normalized drain current noise spectral                  case of SMT devices and no evidence for a different
density SID =ID and the transconductance to drain current                      trap density from SMT process, compared to the top
squared ðgm =ID Þ2 versus ID for SMT and control devices. The                  interfaces.
SID =ID exhibits a fairly good proportionality with ðgm =ID Þ2 at

low drain current. However, at high current region, the                        4. Conclusions
SID =ID curves cannot follow this trend, which imply that                      We have investigated the impact of SMT process on the low
either of the correlated mobility fluctuation. In addition,                     frequency noise characteristics in strained Si nMOSFETs. It
the associated input-referred voltage spectral noise (SVG ¼                    was found that the mechanism of 1=f noise for both devices
SID =g2 ) shows a parabolic dependence with gate voltage at
      m                                                                        can be properly interpreted by the unified model. Further-
strong inversion (Fig. 7). These results also point to that the                more, SMT devices showed comparable noise level to the
1=f noise can be reasonably interpreted by the unified                          control device, indicating that the adding SMT process will
model.1) Moreover, the values of the SID =ID and SVG for SMT                   not degrade interface quality. In addition, the charge
and control devices are similar. It can be reasonably                          pumping measurement results also agreed with the 1=f
attributed to the lower concentration of hydrogen for SMT                      noise mesurement results. The extracted effective oxide trap
process, which brought out the comparable 1=f noise level                      density is lower than the nanowire nMOSFET and tripe-gate
between SMT and control devices.11) Based on the above                         FET, and comparable level with that biaxial strained-Si/
analysis, the unified model takes the following form:1)                         SiGe nMOSFET.
                                                                        04DC20-3                         # 2011 The Japan Society of Applied Physics

Jpn. J. Appl. Phys. 50 (2011) 04DC20                                                                                                         C.-W. Kuo et al.

                                                                                  7) W.-C. Hua, M. H. Lee, P. S. Chen, M.-J. Tsai, and C. W. Liu: IEEE
Acknowledgement                                                                        Electron Device Lett. 26 (2005) 667.
                                                                                  8) Y. P. Wang, S. L. Wu, and S. J. Chang: IEEE Electron Device Lett. 28
This work was supported by the National Science Council                                (2007) 36.
(NSC) of Taiwan under Contract No. NSC99-2221-E-230-                              9) M. von Haartman, D. Wu, B. G. Malm, P.-E. Hellstrom, S.-L. Zhang, and
019.                                                                                   M. Ostling: Solid-State Electron. 48 (2004) 2271.
                                                                                 10) M.-P. Lu, W.-C. Lee, and M.-J. Chen: Appl. Phys. Lett. 88 (2006) 063511.
                                                                                 11) T. Ueno, H. S. Rhee, H. Lee, M. S. Kim, H. S. Cho, H. S. Baik, Y. H. Jung,
                                                                                       H. W. Lee, H. S. Park, C. K. Lee, G.-J. Bae, and N.-I. Lee: Dig. Tech. Pap.
                                                                                       Symp. VLSI Technology, 2006, p. 104.
 1) G. Ghibaudo and T. Boutchacha: Microelectron. Reliab. 42 (2002) 573.         12)   L. K. J. Vandamme: IEEE Trans. Electron Devices 41 (1994) 2176.
 2) Y. Nemirovsky, I. Brouk, and C. G. Jakobson: IEEE Trans. Electron            13)   L. K. J. Vandamme and F. N. Hooge: IEEE Trans. Electron Devices 55
    Devices 48 (2001) 921.                                                             (2008) 3070.
 3) E. R. Hsieh and S. S. Chung: Appl. Phys. Lett. 86 (2005) 093102.             14)   J. Zhuge, R. Wang, R. Huang, Y. Tian, L. Zhang, D.-W. Kim, D. Park, and
 4) S. L. Wu, Y. M. Lin, S. J. Chang, S. C. Lu, P. S. Chen, and C. W. Liu:             Y. Wang: IEEE Electron Device Lett. 30 (2009) 57.
    IEEE Electron Device Lett. 27 (2006) 46.                                     15)   F. Crupi, B. Kaczer, R. Degraeve, V. Subramanian, P. Srinivasan, E.
 5) S. Orain, V. Fiori, D. Villanueva, A. Dray, and C. Ortolland: IEEE Trans.          Simoen, A. Dixit, M. Jurczak, and G. Groeseneken: IEEE Trans. Electron
    Electron Devices 54 (2007) 814.                                                    Devices 53 (2006) 2351.
 6) K. Fobelets, S. L. Rumyantsev, M. S. Shur, and S. H. Olsen: J. Appl. Phys.   16)   H. Y. Lin, S. L. Wu, S. J. Chang, Y. P. Wang, and C. W. Kuo: Semicond.
    103 (2008) 044501.                                                                 Sci. Technol. 23 (2008) 105022.

                                                                          04DC20-4                           # 2011 The Japan Society of Applied Physics


Shared By: