† Corresponding author. E-mail:
Project supported by the National High Technology Research and Development Program of China (Grant No. SS2015AA010601), the National Natural Science Foundation of China (Grant Nos. 61176091 and 61306129), and the Opening Project of Key Laboratory of Microelectronics Devices & Integrated Technology, Institute of MicroElectronics of Chinese Academy of Sciences.
The thickness effect of the TiN capping layer on the time dependent dielectric breakdown (TDDB) characteristic of ultra-thin EOT high-k metal gate NMOSFET is investigated in this paper. Based on experimental results, it is found that the device with a thicker TiN layer has a more promising reliability characteristic than that with a thinner TiN layer. From the charge pumping measurement and secondary ion mass spectroscopy (SIMS) analysis, it is indicated that the sample with the thicker TiN layer introduces more Cl passivation at the IL/Si interface and exhibits a lower interface trap density. In addition, the influences of interface and bulk trap density ratio Nit/Not are studied by TDDB simulations through combining percolation theory and the kinetic Monte Carlo (kMC) method. The lifetime reduction and Weibull slope lowering are explained by interface trap effects for TiN capping layers with different thicknesses.
High-k metal gate stacks have been used for improving MOSFET performance since 45 nm technology node.[1–4] High dielectric constant material successfully suppresses the leakage of gate, however, the reliability issue of the high-k metal gate device becomes a challenge. The gate stack degradation and process optimization for trap reduction have been widely studied.[5–8]
In this paper, TiN is used as the capping layer after being deposited with high-k dielectric. Due to work function tuning and reliability consideration, the thickness of TiN capping layer is a trade-off for process optimization.[9–11] It is reported that the oxygen diffusion model and multi-phase soft breakdown could be used to explain the TiN thickness effect.[9] However, no detailed relation between bulk trap and time to breakdown is given and the Weibull slope change is not well explained.
As one of the most important reliability issues, time dependent dielectric breakdown (TDDB) is related to the bulk and interface trap generation in ultra-thin EOT high-k/IL/substrate stacks. Longer time to breakdown and larger Weibull slope are of benefit to lifetime assessment.[6] In recent studies, simulations combining percolation theory and the Monte Carlo (MC) method were proved to be a useful tool in TDDB analysis.[12,13]
In this paper, the thickness effects of the TiN capping layer on the TDDB characteristic of NMOSFET are studied. TDDB and SILC characteristics are compared for different TiN thickness of 1.4 nm and 2.4 nm. Both transconductance degradation and charge pumping results show that the interface trap plays an important role in dielectric degradation. The origin of interface trap difference is studied by SIMS measurement. Finally, the effects of bulk and interface trap density on TDDB are simulated by combining percolation theory and the kMC method. The Weibull distribution parameters are systemically studied by varying the ratio between bulk and interface trap density.
In this paper, NMOSFET devices under tests are fabricated by a standard gate last process. Interfacial layer and HfO2 are chemically formed by ozone oxidation and atomic layer deposition (ALD), respectively. Then, post-deposition annealing (PDA) in nitrogen at 450 °C is performed. Ultra-thin TiN films with thickness values of 1.4 nm and 2.4 nm are deposited as a capping layer. Then 5-nm TiAl (PVD), 5-nm TiN (PVD), and 75-nm W (ALD) are fabricated as following gate stacks and the whole gate stacks structure as shown in Fig.
In this work, devices with TiN capping layers of different thickness and Lg = 0.5 μm have a similar threshold voltage in a range of 0.23 V∼0.25 V. The EOT results are 0.85 nm and 0.87 nm for 1.4-nm TiN and 2.4-nm TiN, respectively. The nominal W/L ratios of NMOSFET are 8 μm×2 μm and 1 μm×0.5 μm for SILC and TDDB measurement, respectively. All reliability measurements are done by an Agilent 4156C at 125 °C. All tests are done by constant voltage stress, with considering the worst condition for devices reliability.
As shown in Fig.
TDDB is measured at a gate bias 1.8 V, and 16 devices are used for making a statistical analysis, and the results are shown in Fig.
To confirm the cause of gate stack degradation, the transconductance and threshold voltage shift are monitored during SILC stress. As shown in Fig.
The charge pumping method is used to directly measure interface trap density in fresh devices. In charge pumping measurement, the pulse amplitude is constant and base level voltage sweep is applied. The frequency for charge pumping is 500 kHz. The amplitude of pulse is 1.2 V and base voltage sweep is from −0.6 V to −0.3 V. As shown in Fig.
To study the origin of the TiN thickness effect on the interface state, SIMS is used to analyze the elemental compositions in the gate stacks. The SIMS results are shown in Fig.
Based on the above results, the interface trap density difference could be described as the Cl passivation effect. Firstly, Cl is introduced when depositing the TiN. Then, in the subsequent annealing process, Cl diffuses from TiN layer to dielectrics and substrate. It is reasonably assumed that Cl passivation happens at the dangling bonds of the IL/Si interface, which is in detail studied by Park et al.[17] Thus, it is concluded that a thicker TiN capping layer will introduce higher Cl concentration and the passivation effect of Cl reduces interface trap density.
Based on the above results, the effects of interface trap density on the TDDB characteristic are studied in simulations. A cell-based simulator is set up by combining percolation theory with the kMC method as described in Ref. [13]. The physical layers of dielectric stacks are treated as 2 high-k layers +1 interfacial layer with the trap sizes of aHK = 10 Å and aIL = 6 Å for high-k and IL, respectively. Trap generation rate time exponent for bulk trap and interface trap are α = 0.38 and αit = 0.16, respectively. All these parameters are in accordance with those in Ref. [13].
Different ratios of interface and bulk trap density are simulated, and the distributions of time to breakdown are studied. As shown in Fig.
It is demonstrated that when interface trap density is much larger than bulk trap density, the interface trap will strongly influence TDDB characteristic. The scaling TBD follows a power law with the interface trap density as follows (see Fig.
Based on the above simulation results, interface trap influences on TDDB could be explained. As shown in Fig.
Thus, the effect of TiN thickness could be explained for both oxygen and Cl diffusion. In this work, the latter effect on interface trap is crucial for the reduction of lifetime and overall lower Weibull slope. While in Chen and King’s work,[9] oxygen vacancies are more important. The different determinants for each work may be due to a thinner high-k layer in this work (EOT ∼ 0.85 nm in this work and EOT ∼ 1.33 nm in Chen and King’s work). Thus, from the view of EOT scaling, interface trap degradation is more crucial for reliability concerns.
In this paper, the effects of TiN capping layer thickness on trap generation and TDDB characteristic are systemically studied. Samples with two thickness values of ultra-thin TiN capping layer are tested and have nearly the same electrical characteristics. However, the device with 2.4-nm TiN capping layers shows less SILC increment and longer time to breakdown than those with 1.4-nm TiN. The degradation of Gm indicates that the interface trap is related to dielectric degradation, which is proved by charge pumping measurement. Based on SIMS analysis, it is demonstrated that a thicker TiN layer introduces a stronger Cl passivation effect in the ALD TiN deposition process and reduces the interface trap density.
In addition, simulations combined with percolation theory and the kMC method are used to study the detailed relation between TDDB and interface trap density. It is demonstrated that when Nit ≫ Not, breakdown time decreases fast with Nit increasing. Meanwhile, the Weibull slope degrades compared with that under the bulk trap dominant conditions. It is well explained by the simulation results for lifetime reduction and Weibull slope changes due to oxygen and Cl diffusion of the TiN thickness effect. It is also concluded that with EOT scaling, interface trap degradation is more crucial for TDDB assessment.
1 | |
2 | |
3 | |
4 | |
5 | |
6 | |
7 | |
8 | |
9 | |
10 | |
11 | |
12 | |
13 | |
14 | |
15 | |
16 | |
17 |