industry has constantly improved computer performance by scaling a more or less unchanged device geometry. The over- all improvement closely follows Moore’s Law, which predicts that performance doubles approximately every 18 months. De- spite the successful history of device miniaturization, scaling is reaching the physical limits of traditional device materials. With the reduction of gate lengths and the use of more exotic materials such as metal gates, the influence of stress on diffu- sion becomes a more prevalent component in determining the final dopant profile and subsequent device performance. We present the development of a complete predictive sim- ulation capability for the effects of general anisotropic nonuniform stress on dopant diffusion in silicon as an exam- ple for modern physical process modeling. We derived the macroscopic diffusion equation from microscopic transition state theory, calculated the microscopic parameters from first principles, and predicted the feature-scale stress based on stress measurements in the relevant materials as a function of temperature. We used the developed methodology, imple- mented in a continuum solver, to investigate a titanium ni- tride (TiN) metal gate system. We also discuss how to effec- tively integrate predictive modeling tools such as this into the development of state-of-the-art semiconductor devices. Mosfet technology In a Mosfet (see Figure 1), the gate, the insulating oxide (between the gate and the channel), and the channel form a capacitor. A voltage on the gate attracts a conductive layer of charge in the channel region, which connects the source and drain electrically and switches on a source–drain cur- rent. Thus, the Mosfet acts as a switch, which the gate volt- age can turn on and off. Individual switches can be con- nected to form the basic building blocks for circuit design, which in turn form microprocessors and memory chips. Despite the successful history of device miniaturization, scaling is reaching the physical limits of traditional device materials. The gate oxide, for example, has reached a thick- ness of a few nanometers, causing excessive leakage currents (among other problems). One possible solution for this is to resort to so-called “high-k” materials (which have a much higher dielectric constant than SiO 2 ) instead of the tradi- tional SiO 2 , thus permitting a thicker insulator layer. How- ever, this might require other changes such as a different gate material (which currently consists of highly doped poly Si) and therefore a complete overhaul of Mosfet technology. Semiconductor companies and universities perform exten- sive, time-consuming, and costly (experimental) engineer- ing research in these areas. As an example of a metal gate system, Figure 1 shows a TiN metal gate integration on a subquarter-micron Mosfet. For the p-Mosfet, Motorola en- gineers have detected anomalous stress-dependent boron diffusion: Electrically measured lateral diffusion results in- dicate an enhancement in boron diffusion with increasing TiN thickness and, hence, gate stress. 1 (The “p” means that the source and drain are doped with acceptor impurities; ac- ceptors are generally group-III elements—in this case, boron—which introduce holes in the electronic structure.) Today’s Mosfet technology development has met similar hurdles for other device parameters. Examples are the dopant profiles in the extension regions of source and drain, which must be tailored to form ultrashallow junctions, where extremely high dopant concentrations are packed in very shallow distributions. Therefore, simulation of front- end processing is becoming an increasingly critical cost and timesaving component of integrated-circuit technology de- velopment, provided it is accurate enough. In addition, to- day’s electronics are so small that characterization of their material parameters is often difficult and expensive. Thus, simulation is often the only effective tool for exploring the lateral and vertical doping profiles of a modern device at the 92 COMPUTING IN SCIENCE & ENGINEERING Editors: Ken Hass, khass1@ford.com Thomas L. Tysinger, tlt@fluent.com C SE IN I NDUSTRY PREDICTIVE PROCESS SIMULATION AND STRESS-MEDIATED DIFFUSION IN SILICON By Wolfgang Windl, Matthew Laudon, Neil N. Carlson, and Murray S. Daw T HE SILICON-BASED METAL OXIDE SEMI- CONDUCTOR FIELD EFFECT TRANSISTOR (MOSFET) IS AT THE HEART OF TODAY’S SEMICON- DUCTOR INDUSTRY. BECAUSE THE SWITCHING SPEED OF A MOSFET INCREASES LINEARLY WITH SHRINKING DIMENSIONS, THE SEMICONDUCTOR