Next: Kurzfassung Up: Dissertation Ivan Starkov Previous: Dissertation Ivan Starkov
Transistors of nearly all technology nodes suffer from hot-carrier degradation (HCD) which is associated with the buildup of defects at or near the silicon/silicon dioxide interface. This detrimental phenomenon has been known for more than four decades and numerous modeling attempts have been undertaken. However, hot-carrier degradation is rather complicated to model because it includes three different but strongly connected aspects. In fact, carriers interacting with the Si/SiO2 interface break Si-H bonds, thereby generating traps and thus the microscopic mechanisms for defect creation must be properly described. The information on how efficiently these carriers trigger the bond dissociation process is provided by a thorough carrier transport treatment. Furthermore, these generated traps can capture carriers and thus distort the electrostatics of the transistor and degrade the carrier mobility. Due to the complicated nature, a comprehensive physics-based model is still missing and most existing HCD models are empirical.
The main task of this work is to design a physics-based model for hot-carrier degradation, which is able to represent HCD observed in metal-oxide-semiconductor field-effect-transistors (MOSFETs) with different channel length using a single set of physical parameters. The developed approach considers not only the damage produced by channel electrons but also by secondary channel holes generated by impact ionization. Although the contribution of the holes to the total defect creation is smaller compared to that of electrons, their impact on the linear drain current is comparable with the electron one. The reason behind this trend is that hole-induced traps are shifted towards the source, thereby more severely affecting the device behavior.
The model includes three main modules: a carrier transport module, a module for modeling of microscopic mechanisms of defect creation and a module for the simulation of the characteristics of degraded devices. The carrier transport module calculates a set of carrier energy distribution functions (DFs) at any position in the MOSFET for a particular device architecture and stress/operating conditions. For DF calculation a full-band Monte Carlo (MC) device simulator MONJU is employed. Then the information regarding the carrier DF is used to generate interface state density profiles. These profiles are loaded into a circuit and device simulator MiniMOS-NT, which calculates the characteristics of the degraded device. Since the MC method is very time-consuming the precise carrier transport module is substituted by a simplified treatment, i.e. a comparison between different realizations of the model carrier transport module is carried out. Namely Monte Carlo, hydrodynamic (HD) and drift-diffusion (DD) schemes for the solution of the Boltzmann transport equation are employed. A discrepancy between experimental results and simulations, which occurred while employing this simplified approach, is shown and explained.
As previously stated, the developed physics-based model for HCD includes three main modules. Each module is based on some assumptions acting as potential sources of error and includes a certain number of fitting parameters. In other words, due to the complicated structure of the model, these errors should be screened by properly evaluating the interfaces between the modules. Therefore, although a good representation of the degraded device characteristics could be achieved, one attempts to verify the microscopic model for the defect build-up in the following by employing the charge pumping (CP) technique for a particular device architecture. It is worth emphasizing that the algorithms for the extraction of the interface state profiles versus position along the device interface from the CP data are based on some assumptions and thus could be potentially inaccurate as well. Therefore, an exhaustive analysis and comparison of extraction techniques of the hot-carrier induced interface and oxide trap spatial distributions is performed. The advantages and limitations of the characterization algorithms improved by a new simple compact model for local oxide capacitance are discussed. It is demonstrated that by ignoring the spatial variation of oxide capacitance, a spurious result is produced, leading to an ambiguous picture of HCD. Additionally, a careful extraction of the initial interface state density profile for a pre-stressed MOSFET is undertaken. The impact of pre-existing interface states on the subsequent interface trap profile evolution is demonstrated in the context of predictive HCD modeling. A thorough comparison between simulated interface state profiles and those extracted from CP data is carried out.
Based on a rigorous technology computer-aided design (TCAD) version of a physics-based model for HCD, an analytical model, which suitably approximates the carrier acceleration integrals (AIs), is developed. Using such an approach one is able to represent the linear drain current degradation. One of the main advantages of this analytical approach is that it is based on a physics-based TCAD model rather than on an empirical fit to experimental data. The model also represents the saturation of HCD observed at relatively long stress times. The flexibility of the resulting expression allows us to employ this approach while considering the impact of fluctuating parameters of device topology on HCD. In this case the time-consuming MC based transport module would otherwise lead to extremely high computational costs.