Silicon dioxide is used as masking material or as diffusion barrier in
multilayer device structures. Compared to single crystal silicon the
diffusivities of impurities in oxide is extremely low for almost all common
dopants. Chemical elements like He, Na, 
and 
are fast
diffusers in oxides and hence are influencing the dopant diffusivity. Due to
the void silicon dioxide structure and the variety of affecting parameters
it is not possible to define exact diffusion coefficients for the different
species. The diffusion coefficients depend on local process conditions and
have to be investigated for specific problems. The diffusion problem itself
is captured by the simplest approach, assuming Fickian diffusion without
field enhancement and constant diffusion coefficients. Recently, List et al.
published data for the diffusion coefficient of boron for multilayer
experiments [Lis95]
. They investigated the diffusion of boron across
the gate oxide initiated by the excessively high doped 
-poly gate. A
transient diffusivity effect for boron was observed, which could not be
simulated with a constant diffusion coefficient. This suggests a step
function for the oxide diffusivity 
as given by (3.3-1), where
denotes the start value for the diffusivity and 
the end value.
Values for the diffusion coefficients of dopants within oxides can be found in [Won86] [Mat93a] . They postulated a diffusivity enhancement at early stages of diffusion coming from hydrogen. Hydrogen is able to influence the dopant diffusivity significantly, but it is not clear where the hydrogen is coming from. Unfortunately, short time annealing experiments are not useful to investigate the boron outdiffusion. Only a small amount of dopants is able to diffuse across the oxide because of the the low dopant diffusivity. The resulting dopant profiles would be to steep for SIMS measurements. But even this annoying outdiffusion affects the threshold voltage in conventional CMOS device technology.
For two decades metal silicides such as 
or 
are used to
reduce sheet and contact resistance of polysilicon and diffusion
regions. For thick polysilicon layers and deep junctions the silicidation
process does not impact the active area of the devices. Indeed silicidation
technology gained rapid acceptance because of its reliability. With
decreasing device dimensions effects like dopant redistribution, anomalous
silicon consumption, and enhanced diffusivity of dopants caused by the
silicidation became important [Hov87] [Mae89]
. Therefore, the
dimensions of the silicide layers have been also reduced. Unfortunately, the
thermal stability of thinner layers is rather poor compared to earlier
technologies.
One of the most significant changes in silicide technology was to use
silicides as dopant diffusion sources. Silicidation is performed before the
junction formation. After fabrication of the metal silicide dopants were
implanted into the silicide layer, followed by a thermal cycle, usually RTA,
to diffuse out the dopants into the substrate [Hov89]
. Processes like
silicide grain growth, silicon consumption, silicon agglomeration, and
diffusion of dopants are involved during junction formation. The transport
of dopants within the silicide material is assumed to be based on grain
boundary and bulk diffusion. The dopants can either diffuse towards the
surface, where they accumulate and evaporate or towards the underlying
substrate, where they form shallow junctions. The diffusivities of the
dopants within silicide materials are in the same order of polysilicon
diffusivities or even higher for arsenic and phosphorus. Boron however,
shows quite contrary diffusion properties. While it is fast diffusing in
, it is nearly immobile within , which is attributed to a
stable alloy between the dopant and the metal silicide ( 
). Due to the
lack of understanding the details of diffusion within silicides, the
diffusion of dopants is modeled by the Fickian approach neglecting more
sophisticated processes like activation within the silicide. The influence
of the silicidation on the diffusivity of the dopants on the underlying
substrate material is modeled by inclusion of an enhancement factor, as
given for arsenic in (3.3-2) and (3.3-3).
denotes the well known arsenic diffusion coefficient for silicon, and
is the enhanced diffusivity factor, which is applied for a limited
time period onto the dopants diffusion [Osb93]
.
As a drawback of the silicidation technology the roughness of the layer during thermal treatment is known. Due to agglomeration the interface between the substrate and the silicide as well as the surface of the silicide is very rough. The roughness depends on the silicide thickness and the applied thermal budget and has to be carefully optimized during processing to obtain low sheet resistances, without degrading junction performance. The high segregation coefficients or in the case of phosphorus the high surface evaporation can lead to degradation of dopants within the substrate caused by the silicide layer. Generally, silicide layers are fast diffusion pathes for silicon point defects and other chemical elements.
Due to the different linear thermal expansion coefficients of silicide materials compared to other semiconductor materials stress will be induced during thermal treatment. In the worst case the silicide layer looses its adhesion and bends off the underlying material layers, or the induced stress affects the electrical properties of the substrate.
