Analysis and Calibration of LOCOS Simulation using VWF


Several effects typical in local oxidation (LOCOS) isolation technologies cannot be simulated without taking into account stresses generated during the oxidation process. Among these several are critical to sub-half-micron technology effects. These are decreasing of bird's beak length and its changing shape with the increase of nitride mask thickness, thinning of isolation oxide with reduction of the nitride window width (active area spacing), and pad-oxide punch through for narrow patterned nitrides (active area width).

Stress-Dependent Oxidation Model

Most of parameters involved in the stress-dependent viscous oxidation model cannot not be measured directly. Therefore, calibration through numerical simulation for wide range of local oxidation conditions is the only way to build predictive models. Virtual Wafer Fab (VWF) allows users to perform such calibration and model analysis very efficiently.

The standard Athena stress-dependent oxidation model was used throughout this study. There were several publications [1,2] pointing out that the model could be improved by implementing of several addition terms (e.g. the term describing plastic behavior of oxide and nitride layers). These terms could be important in accurate description of stress built in the case of thin nitride. However, even the incomplete stress-dependent model used currently in Athena is sufficient to describe LOCOS effects for quite wide variety of process parameters.

Temperature dependencies of nitride and oxide viscosity extracted in [1]have been used in this work. These viscosity parameters are different from Athena defaults and could be used as a recommended alternative set:

material oxide visc.0=5.1 visc.E=3.48
material nitride visc.0=5.96e5 visc.E=2.5625


Response Surface Models for LOCOS Parameters

In the first VWF experiment the Response Surface Models (RSM) for normalized nitride deflection (deflect) and normalized bird's beak length (normbbl) were built and stored in the VWF database for future reference (see Figure 1). Oxidation temperature T (within the range 900 - 1100C), nitride thickness Tnit (0.09 -0.21 microns), as well as empirical parameters of the stress dependent oxidation model Vd (8 - 60), Vc (100 -800), and Vr (10 - 70) were selected as variable parameters for building the RSMs. A Box Behnken experimental design was used. This required 109 simulation fragments 78 of which were time-consuming oxidation steps. Six processors on a SparcCenter1000 were used so the whole experiment took less than 20 hours.


Figure 1. Geometrical parameters of birds beak
showing definitions of deflect and normbbl.



The VWF Production Tools were used for visualization and analysis of the final RSMs. Figure 2 shows the simulated variation of normbbl as a function of two calibration parameters in the stress-dependent oxidation model. Figures 3a and 3b show how the RSMs could be used for physical model calibration by fitting simulated results to measured data. It is important that both critical geometrical parameters are fitted simultaneously. Unfortunately, there are no systematic experiments which measure LOCOS geometrical characteristics for different nitride thicknesses, temperatures, and pad oxide thicknesses. Such experiments together with even more elaborate RSMs would allow to complete calibration of the oxidation model including probable temperature dependencies of Vc, Vr, and Vd.


Figure 2. Normalized bird's beak length as a function of
stress dependent oxidation model parameters.

Figure 3. Normalized bird's beak length (normbbl) and nitride deflection (deflect) as functions of nitride thickness obtained from RSMs for T=1100 C.
left: default values : Vc=300, Vr=30, and Vd=25;
right: Optimized values : Vc=150, Vr=14, and Vd=10.
Experiment [3]

Field Oxide Thinning Effect

Overall increasing of chip density in ULSI technology leads to CD shrinkage of not only active but also isolation areas. The field oxide thinning effect associated with decreasing of field area CD brings about increasing concern to technology designers. Figure 4 shows the LOCOS shape for three different widths of nitride opening. Such behavior was confirmed in several experiments using different electron microscopy methods [4].


Figure 4. Simulation results of LOCOS oxidation for different
nitride windows (0.4, 0.8, and 1.5 microns).



A VWF simulation experiment was performed to analyze the oxide thinning effect.The split parameters were nitride window width (7 values in the interval between 0.4 and 2.5 microns), nitride thickness (3 values 0.1, 0.15, and 0.2 microns), oxidation temperature (1000 C and 1100 C), and oxidation time ( 90 and 150 minutes). The stretch function of Athena was used to facilitate the CD variation.

Figure 5 shows that simulation results are in good agreement with experimental data [4]. Analysis of other dependencies in the VWF worksheet also shows that the effect is more pronounced for thicker nitrides and lower temperatures.


Figure 5. Field oxide thinning effect for different nitride thicknesses.
Experiment for nitride thickness 0.1 micron [4] .



Pad Oxide Punch through Effect

It was found experimentally [5] that bird's beak deflection depends highly on patterned nitride width. Specifically, the dependency has a minimum at a nitride width of ~0.6 microns and then suddenly increases for narrower features. Analysis of [5] shows that such behavior can be attributed to coupling of nitride stresses at the two edges of patterned nitride and punchthrough effect is the result of oxidant profile merging under nitride. The pad oxide punch through effect is illustrated in Figure 6.


Figure 6. Simulation results of LOCOS structures for different patterned
nitride width (0.2, 0.4, 0.6, and 1.0 microns).


The third VWF simulation experiment of this study uses designed to confirm the effect as well as to analyze how it depends on other process parameters(nitride and pad oxide thicknesses). The split parameters were patterned nitride width (9 values between 0.2 and 2 microns, nitride thickness (0.1, 0.15,and 0.2 microns), and pad oxide thickness (0.015 and 0.025 microns).

Figure 7 shows that Athena simulation results demonstrate the same behavior as experimental data [5]. The differences in absolute values of deflection are probably due to different extraction techniques. Another reason could be non optimized parameters of the stress-dependent oxidation model.


Figure 7. Normalized nitride deflection versus patterned nitride width for
different nitride thicknesses (1000 C, 90 minutes, pad oxide 0.015 micron).
Experiment [5]




It is shown that Athena's stress-dependent oxidation model can adequately describe various effects associated with LOCOS structures common in sub-half-micron technologies. Calibration of empirical parameters is required but is demonstrated here using VWF. Future model developments in ATHENA with coefficients calibrated using this techniques will lead to predictive simulators for a wider selection of technologies.

It is also shown that VWF could serve as a very effective tool for detailed analysis of such complete effects. The response surface model generated by VWF and the analysis capabilities of Production Tools allow multi parameter calibration of complex process simulation models.


[1] P.Ferreira , Proceedings of ESSDERC'94, p.259, 1994.

[2] P.Griffin, in CAD for IC's - Process, Device, and Circuits, Stanford University, 1994.

[3] P. Griffin, IEDM Tech. Digest, p.741, 1990.

[4] P.Coulman, Proc. of 2nd Int. Symp. on VLSI Sci. & Tech., p.759, 1089.

[5] P.U. Kendale, IEDM Tech. Digest, p.479, 1993.