next up previous contents
Next: 2. Semiconductor Doping Technology Up: Dissertation Robert Wittmann Previous: List of Abbreviations and

Subsections



1. Introduction

Information technology (IT) is one of the most important technologies, which has allowed to change the industrial society into an information and knowledge based society. The electronic industry is the largest industry in the world with a global sales volume of over 1 trillion US$ [1]. Microelectronics is the branch of electronics which deals with the miniaturization of integrated circuits (ICs). The evolution in microelectronics has resulted in complex System-on-a-Chip (SoC) devices which combine logic and memory units consisting of hundreds of millions of transistors packed on a single silicon chip. The production of ultralarge scale integrated (ULSI) circuits was only possible due to the CMOS (complementary metal oxide semiconductor) technology platform because of low-power and scaling properties.

The workhorse of integrated circuits is the MOSFET (metal-oxide-semiconductor field-effect transistor). This device is basically a switch where the electric current in the silicon between the source and drain electrodes is controlled by the potential of the gate electrode. For four decades, the semiconductor industry has achieved continuous performance enhancements by downscaling of the MOSFET device dimensions, as described by Moore's Law. That is, the number of transistors per chip doubles every eighteen months to two years. In the last three years it has become clear that the conventional transistor materials silicon and silicon dioxide have been pushed to fundamental material limits. The International Technology Roadmap for Semiconductors (ITRS) defines the current situation as material-limited device scaling [2] which requires the introduction of new materials. The incorporation of nitrogen into ultra-thin gate oxides to reduce the gate leakage current and the use of strain to enhance the carrier transport in silicon are two successful examples for the improvement of scaled bulk CMOS devices. In the next several years, either extensions of bulk CMOS technology or new approaches such as fully depleted SOI (silicon-on-insulator) and multi-gate devices must further reduce the cost-per-function and increase the performance of integrated circuits.

Technology Computer-Aided Design (TCAD) refers to the computer simulation of semiconductor process and device technologies. TCAD tools play a key role in the development of a new CMOS technology and they can help to reduce the development time and costs. The combination of process and device simulation tools in a TCAD framework enables analysis and optimization of the influence of process parameters on the electrical characteristics of a single device. Predictive modeling of a single process step in the IC manufacturing process requires to include the underlying physics of that step. An example of a very successful atomistic simulation approach is the Monte Carlo modeling of the ion implantation process. Each ion trajectory is simulated separately in this approach. In this way, an accurate prediction of the doping profiles after ion implantation is achieved. For deep-submicron devices two- and three-dimensional process simulations can provide a better insight than measurement techniques and have become indispensible for the design of advanced CMOS devices. Rapid technology enhancements, introduction of new materials, and increasing reliability problems caused by more and more shrinked device dimensions have led to the situation that commercial TCAD tools cannot keep pace with numerous newer developments. This situation becomes more critical in the field of process simulation tools due to the existing diversity in the process steps.

1.1 Manufacturing of Integrated Circuits

The fabrication of a modern IC in the CMOS process involves hundreds of sequential steps and can last up to 30 days of processing time [3]. The following basic process steps are used:

Figure 1.1: Schematic cross-section of a final CMOS integrated circuit with an interconnect structure of three metal layers (M1 - M3).
\includegraphics[width=.86\linewidth]{figures/cmos-cross}


The CMOS process flow starts with a uniformly doped silicon wafer, and after processing, the wafer contains hundreds of identical rectangular chips. The IC fabrication can be subdivided into the following three major phases:

Figure 1.2: Schematic of a CMOS NAND-gate (left) and its layout (right) [4].
\resizebox{0.9\linewidth}{!}{\rotatebox{0}{\includegraphics{figures/nand-comb}}}
The final result of such a process flow is shown in Fig. 1.1. This modern bulk CMOS structure is fabricated using a triple-well and a dual-polysilicon gate process. Both MOS transistors are isolated by an oxide-filled shallow trench isolation (STI). CMOS technology integrates n- and p-channel MOS devices on the same chip. An important characteristic of any CMOS circuit is that in either logic state, at least one device in the series path from the power supply to the ground is non-conductive. Therefore, the static power consumption of the CMOS inverter is given by the leakage current of the off device. A significant current flows only during the short transient period when both devices change their state. Due to the low power dissipation requirement in integrated circuits only the CMOS technology is currently used in advanced IC manufacturing. The NAND gate in Fig. 1.2 illustrates how additional n-MOS and p-MOS devices can be added to the inverter circuit to realize more complex logic functions. In the NAND circuit, the output Q will be grounded through the two series n-MOS devices only, if both inputs I$ _1$ and I$ _2$ are high. The layout of the NAND-gate shows that the series connection of the two n-MOS devices can be performed easily by a shared n$ ^+$-doped region, and the two p-MOS devices are connected by a shared p$ ^+$ region. It can be seen that the p-MOS device has a significantly larger area than the n-MOS device. At least double the width of the n-MOS device is used for the p-MOS due to the lower hole mobility in silicon. Finally, it should be noted that a complex IC circuit can contain over $ 10^9$ transistors, and each of them must work correctly.


1.2 Critical Issues for Device Miniaturization


The shrinking of MOS transistor dimensions is typically performed by a scaling factor of 0.7 per each technology generation.
Figure 1.3: Small devices with a gate length of 60nm (left) and 10nm (right) [5,6]. The gate dielectric thicknesses are 1.5nm and 0.8nm, respectively.
\resizebox{0.88\linewidth}{!}{\rotatebox{0}{\includegraphics{figures/small-mosfets-comb}}}
It has been demonstrated by Intel that a gate length of 10nm is possible in experimental devices, as shown in Fig. 1.3. The ongoing downscaling trend leads to some limitations and issues in modern CMOS technology, mainly due to small-geometry effects. For instance, the punchthrough effect must be suppressed in short channel devices [7], or the application of ultra-thin gate dielectrics requires the incorporation of nitrogen into the oxide, which in turn worsens the interfacial properties [8]. The issue of ultra-shallow source/drain junctions and their increased parasitic resistance is discussed in Section 1.3.

1.2.1 Short-Channel Effects

An MOS transistor is considered short when the effective channel length $ L_\mathrm{eff}$ is comparable to the source/drain junction depletion width [9]. In this case, the potential distribution in the channel depends on both the normal and the lateral electric field in the device. Experimentally, the short-channel effect is observed to degrade the subthreshold characteristics and to reduce the threshold voltage $ V_T$ with decreasing $ L_\mathrm{eff}$ and increasing drain voltage $ V_D$. When $ L_\mathrm{eff}$ is further reduced, the drain current finally cannot be turned off and the gate has no control over the charge. The so-called punchthrough effect poses a severe problem for miniaturized devices. Measures which are taken to suppress this effect are retrograde wells and halo implants [7]. The purpose of these background doping profiles is to prevent the expansion of the drain depletion region into the lightly doped transistor channel when the device is switched on [10,11].

For devices with very short channels an additional effect occurs which leads to increased leakage current. Due to the short distance between source and drain, the potential at the drain contact reduces the peak value of the energy barrier in the channel, which is called drain-induced barrier lowering (DIBL). It leads to a decrease of the threshold voltage with reduced channel length.

The required high channel doping to control short channel effects degrades carrier mobility, lowers the drain current, and increases band-to-band tunneling across the junction and gate-induced drain leakage (GIDL) [2]. Moreover, statistical fluctuation of channel dopants in small area devices causes increasing variation of the $ V_T$ value, posing difficulty in circuit design while scaling the supply voltage.

1.2.2 Hot-Carrier Effects and Drain Engineering

As a consequence of the power-supply voltage being reduced much less proportionally to the channel length $ L_\mathrm{eff}$ in practical scaling (deviation from constant-field scaling), the lateral electric field is increased in the device [9]. Carriers which move from the source to the drain in such a turned-on MOS transistor can get enough energy to cause impact ionization that generates electron-hole pairs in silicon and surmount the interfacial energy barrier. The carriers injected into a gate dielectric induce device degradation such as $ V_T$ shift and reduced drain current. Therefore, hot carrier injection (HCI) degradation significantly reduces the transistor lifetime. The n-MOS transistor is more sensitive to HCI than the p-MOS transistor, since electrons become hotter than holes due to their higher mobility and the energy barrier is lower for electrons compared to holes at the interface. This degradation effect was considered as the major reliability problem in former technology generations, in particular, if high electric fields were generated by constant-voltage scaling. However, to solve this issue, drain engineering is used to alleviate the peak of the lateral electric field located close to the drain edge by modifying the drain doping profile through the introduction of source/drain extension implants by a lower dose [7].

1.2.3 Gate-Dielectric Reliability

According to the CMOS scaling suggested in the ITRS roadmap [2], the gate dielectric thickness should be reduced with every new device generation. If the energy barrier between gate and semiconductor becomes too small, the quantum-mechanical tunneling effect comes into play [12]. One solution to this effect is to use dielectric materials which have a higher dielectric permittivity than silicon dioxide (SiO$ _2$). These materials allow to achieve a high physical thickness together with a small effective oxide thickness (EOT). The EOT is defined as the thickness of a SiO$ _2$ layer with equal capacitance. Since none of the alternative dielectric materials forms a native oxide on silicon, a thin interfacial layer of SiO$ _2$ can hardly be avoided. For a layer of SiO$ _2$ and a high-k dielectric, the EOT of the stacked dielectric is

$\displaystyle \mathrm{EOT} = T_{sio2} + T_{high-k} \cdot \frac{k_{sio2}}{k_{high-k}}  ,$ (1.1)


where $ T_{sio2}$ and $ T_{high-k}$ denote the thickness of the SiO$ _2$ and high-k layer, and $ k_{sio2}$ and $ k_{high-k}$ are the respective permittivities, respectively. With high-k dielectrics it is possible to retain good control over the inversion charge even with physically thick dielectrics to block tunneling currents. However, only nitrided gate oxides (SiON) are currently used in state-of-the-art CMOS technologies because of critical reliability issues of high-k dielectrics. On the one hand side, the incorporation of nitrogen into the oxide reduces gate leakage, avoids boron penetration into the dielectric, and improves HCI, and on the other hand side, it increases the negative bias temperature instability (NBTI) effect [13]. The NBTI mechanism leads to a rapid shift of the transistor parameters ($ V_T$, $ I_{Dsat}$) due to the buildup of charged interface traps during operation. The NBTI reliability will be investigated for a 90nm technology node in Section 5. CMOS scaling beyond the 45nm node requires the reduction of the gate dielectric thickness down to an EOT of $ \sim\!1$nm which will either be realized with SiON or high-k dielectrics.


Table 1.1: Doping requirements from the 2005 ITRS roadmap [2].
Year of introduction 2005 2007 2010
Technology node (nm) 90 65 45
Channel doping concentration (cm$ ^{-3}$) $ 3.7 \cdot 10^{18}$ $ 5.4 \cdot 10^{18}$ $ 8.9 \cdot 10^{18}$
Sidewall spacer thickness $ t_{sp}$ (nm) 35.2 27.5 19.8
Extension junction depth $ X_{j}$ (nm) 11 7.5 6.5
Extension lateral abruptness (nm/decade) 3.5 2.8 2.0
Contact junction depth $ X_{jc}$ (nm) 35.2 27.5 19.8



1.3 Trends in Doping Profiles for CMOS Technology

The ongoing scaling of device dimensions to increase packing density, to increase operating speed, and to reduce power consumption has posed difficult challenges for the doping of MOS transistors. As the channel length is reduced in a scaled device the threshold voltage decreases with the channel length (threshold voltage roll-off). The short channel effect can be minimized if the source/drain junction depth is reduced too. Table 1.1 shows the trend to very shallow junctions in future CMOS technology nodes. Shallower junctions require the introduction and activation of higher dopant concentrations to keep the resistance of the source and drain regions small so that the drive current is maximized. Table 1.1 also states that steeper lateral doping gradients of the source/drain extension regions under the gate are necessary to increase the device performance.
Figure 1.4: Doping problems in extremely scaled MOS transistors.
\resizebox{0.69\linewidth}{!}{\rotatebox{0}{\includegraphics{figures/doping-trend-comb}}}
Fig. 1.4 shows some major doping issues required to form an advanced MOS structure. Scaling of the contact area, junction depth, and contact silicide thickness leads to an increase in parasitic resistance. A possible solution is to use elevated source/drain contacts by selectively deposited silicon or germanium in the contact region to make more silicon available for the silicide formation process. In the polysilicon gate electrode the doping concentration has to be pushed beyond known limits in order to limit the depletion layer thickness [2]. Halo implants (also known as punchthrough suppression or ``pocket'' implants) avoid the punch-through between the source and drain through the bulk substrate in short-channel devices. This implantation places the dopants just below the active channel, adjacent to the source and drain regions, to precisely tailor the well background doping there (see Fig. 1.4). Operating frequencies above 1GHz have become possible for MOS transistors only due to the successful introduction of halo implants [4]. Finally, channel engineering with the doping profile of a modulated retrograde well structure will become increasingly important for the optimization of various device parameters (e.g. channel mobility, threshold voltage, source/drain junction capacitance, hot carrier control, substrate current, soft error control) in future CMOS technologies [7].

Virtually all doping profiles required for advanced CMOS processing are accomplished by ion implantation. The reasons that ion implantation has become the dominant doping technology in modern IC manufacturing are the flexibility in selecting the dopant species, spatial location, and amount of introduced dopant atoms within the device. This process provides a very precise control and reproducibility for the desired dopant distributions. Furthermore, the ion implantation process can be effectively modeled on computers. The accurate simulation of implanted dopant distributions in one-, two-, or three-dimensional structures requires only a few input parameters and can reduce the development time and costs for a new CMOS technology.


1.4 Overview of Ion Implantation Simulation Tools

The three-dimensional ion implantation simulator MCIMPL-II (Monte Carlo Implantation) [14] was used throughout this work for the investigation of implanted doping profiles in silicon and non-silicon materials. The simulator MCIMPL-II is based on a binary collision approximation (BCA) [15] and will be described in detail in Section 3. Beside MCIMPL-II, there are several ion implantation simulators available on the market today, both commercial and academic ones. Due to the high complexity inherent in BCA modeling, commercial ion implantation simulators are often based on academic simulation codes. An analytical ion implantation module is mostly used additionally to a Monte Carlo module to allow a fast simulation of implantation profiles.


TRIM


The Monte Carlo program TRIM (Transport of Ions in Matter) was developed by Ziegler, Biersack and Littmark in 1985 [16], and then rewritten to run on PCs [17]. This BCA program simulates the slowing down and scattering of energetic ions in amorphous targets. The projectile trajectory is statistically followed by randomly selecting a target atom, an impact parameter, and a distance (mean free-flight-path). It was developed for determining ion range and damage distributions as well as angular and energy distributions of backscattered and transmitted ions.


MARLOWE and UT-MARLOWE


The development of the MARLOWE simulator has been started in 1974 by Robinson and co-workers [18,19]. The program uses a BCA technique and considers crystalline target materials. The nuclear scattering is treated in a precise manner by numerically evaluating the classical scattering integral for realistic interatomic potentials. This calculation could be run only on a mainframe computer because of the tremendous computational effort. UT-MARLOWE is a highly modified MARLOWE code which was developed at the University of Texas at Austin [20,21]. Recently, an interpolation scheme of scattering events was developed to allow the implantation simulation of an arbitrary species into crystalline silicon [22].


Crystal-TRIM


The program Crystal-TRIM was developed at the Forschungszentrum Rossendorf at , based on the MARLOWE and TRIM codes. The current version 04/1D simulates ion implantation into crystalline silicon, germanium and diamond with up to 10 amorphous overlayers of arbitrary composition [24]. Not only atomic ions but also molecular ions may be considered. Dynamic simulation of damage accumulation in crystalline substrates and the formation of amorphous layers are possible. The simulator can be used to calculate implanted range and damage distributions as function of depth. An efficient splitting algorithm is employed in order to enhance the statistical accuracy of the simulation results without considerable increase of computing time. It is particularly useful, if channeling tails are of interest. Other versions of Crystal-TRIM (for calculating two- and three-dimensional range and damage profiles) were part of the process simulators TESIM, DIOS and FLOOPS, which were distributed by ISE Integrated Systems Engineering AG, Zürich. Now some of these simulators are part of the TCAD software of Synopsys.

1.5 Outline of the Dissertation


New materials have become more and more important in the development and implementation of advanced CMOS technologies. Process-induced strained silicon is currently used in the 65nm technology to enhance the drive current. Biaxially strained silicon, silicon-germanium (SiGe) alloys with high Ge concentrations or pure germanium offer larger intrinsic carrier mobilities compared to process-induced strain, which can be exploited for channel engineering.

The purpose of Chapter 2 is to provide a brief overview on semiconductor doping technology. The first part of the chapter explains the theoretical background of semiconductor doping and its impact on the electrical material properties. The ion implantation technique for introducing dopant atoms into semiconductors is presented in the second part.

Ion implantation is an extremely physical process, since the incoming dopant ions make way for themselves by knocking the target atoms out of their lattice sites. Chapter 3 starts with the description of the fundamental physical models required for the Monte Carlo calculation of ion implantation distributions. The following section of the chapter is focused on the Monte Carlo ion implantation simulator MCIMPL-II [14]. The process simulation environment, the principle of operation, and the models used for the trajectory calculation are described. Atomistic simulation provides a better insight into the ion implantation process, for instance, single ion trajectories can be visualized or the implantation induced vacancy and interstitial concentration profiles are available. The improvement of the simulation results by an advanced smoothing procedure is analyzed, and the accomplishment of a three-dimensional application is demonstrated.

In Chapter 4 the implantation of boron and arsenic is studied in novel crystalline materials such as SiGe alloys, germanium, and silicon-based heterostructures as a replacement for bulk silicon. For this purpose, the ion implantation simulator has been extended from silicon to these target materials on the basis of experimental results. For the simulation of a new material, the crystalline partner selection model has to be modified, the empirical electronic stopping model has to be calibrated, and the displacement energy for the damage generation has to be adapted. The investigations are focused on the influence of strain, the germanium content in SiGe alloys, the damage accumulation, and the channeling effect on the obtained doping profiles.

Negative bias temperature instability (NBTI) has emerged as a major reliability concern for newer CMOS technologies [25]. In Chapter 5, the impact of NBTI-driven degradation of transistor parameters on the lifetime of a high-perfomance p-MOSFET is investigated. Experiments for different gate voltages, frequencies, and duty cycles were performed to analyze the degradation behavior for the key device parameters, $ V_T$ and $ I_{Dsat}$. The presently leading reaction-diffusion (R-D) model is used for the numerical simulation of interface trap generation based on the diffusion and accumulation of released hydrogen in the gate oxide.

In Chapter 6, several simulation applications are presented. MCIMPL-II is used for the three-dimensional simulation of ion implantation applications for processing advanced MOS transistors and a high-speed photodetector. The applicability of enhanced mobility materials for processing of improved MOS devices with existing ion implantation equipment is shown. A calibrated reaction-diffusion model is used to investigate the impact of NBTI-induced transistor parameter degradation on the lifetime of a state-of-the-art SRAM cell by numerical simulations.

Finally, Chapter 7 summarizes the thesis with some conclusions.


next up previous contents
Next: 2. Semiconductor Doping Technology Up: Dissertation Robert Wittmann Previous: List of Abbreviations and

R. Wittmann: Miniaturization Problems in CMOS Technology: Investigation of Doping Profiles and Reliability