See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261354434

3-D simulation of silicon oxidation: Challenges, progress and results

Conference Paper · September 2013

DOI: 10.1109/SISPAD.2013.6650608

CITATION READS

1 388

7 authors, including:

Alp Gencer Zhiqiang Tan

Synopsys Synopsys
19 PUBLICATIONS 129 CITATIONS 26 PUBLICATIONS 407 CITATIONS

SEE PROFILE SEE PROFILE

Mark Johnson
Synopsys
44 PUBLICATIONS 1,861 CITATIONS

SEE PROFILE

All content following this page was uploaded by Alp Gencer on 15 December 2015.

The user has requested enhancement of the downloaded file.

 3-D Simulation of Silicon Oxidation:
Challenges, Progress and Results
Damrong Guoy, Alp H. Gencer, Zhiqiang Tan, Satish Chalasani, Mark Johnson, Luis Villablanca, Simeon Simeonov
Synopsys, Inc. Mountain View, CA, USA
alpg@synopsys.com

Abstract— We report a new algorithm for solving the moving conform and move together. Our design was motivated by the
boundary problem of oxidation of silicon in 3D structures. The need for accurate segregation, accurate dopant profile, and dose
algorithm solves many of the boundary and mesh quality conservation. We wanted the MovingMesh framework to
problems that makes 3D oxidation simulations challenging. Using support high fidelity simulation. An example of lower-fidelity
this algorithm, we demonstrate that complicated 3D oxidation design is the Eulerian description in which the bulk meshes are
steps can be performed and results can be obtained in a stationary, and the interface meshes move through the bulk
reasonable amount of time. without conforming to the bulk meshes. Another non-
conservative design is to use level sets to represent moving
Keywords— Oxidation, moving boundary, LOCOS, trench
interfaces and regenerate the bulk meshes all over again every
oxidation
time step.
I. INTRODUCTION Our design choice increases the difficulty in maintaining
Many modern device geometries (FinFETs, state of the art the quality of tetrahedrons in the bulk meshes, the quality of
bulk and SOI CMOS devices) require simulation in three triangles in the interface meshes, and the quality of the
dimensions (3D), either simply because the device cannot be underlying geometry, especially during long oxidation
simulated in 2D due to its operation, or because the 3D effects processes dictated by power device design. After a number of
are too large to ignore. Even though 3D implantation and time steps, a reasonable-shaped tetrahedral or triangular
diffusion simulations have been well established for some time, element can become a very poor element that harms the solvers
3D oxidation simulations lagged behind. and limits the time step to practically zero. Furthermore, the
underlying geometry can become so rough that no reasonable
It can be argued that for cutting edge CMOS devices there quality mesh can be generated. After many years of research,
is not much need for performing 3D oxidation simulation, we developed a set of tools that solved these problems, so that
since they tend to use shallow trench isolation (STI) instead of the simulation can run to completion with accurate,
local oxidation of silicon (LOCOS) for isolation and in such conservative, and trustworthy result.
devices the gate oxide has been replaced by high-k dielectrics.
However, for power and memory devices, the device engineers III. EXAMPLES
still wish to accurately model the topology of their 3D
structure, as well as the stress-buildup due to the oxidation. The A. 3D LOCOS Corner
oxidation processes used in power devices are especially One of the classic 3D oxidation examples is the LOCOS
challenging for simulation, since the grown oxides tend to be oxidation of a mask corner. Fig. 1 shows results for oxidation
several hundred nm thick and the surfaces tend to be non- of 120nm nitride hard mask corner at 950C under 6.5l/min O2
planar. and 6l/min H2 flow. The oxidation time is 25min, 1h 45min,
3.5h and 10h and the grown oxide thickness is 100nm, 300nm,
Previous attempts [1,2,3] in solving the 3D oxidation
simulation problem remained somewhat limited in scope and 500nm and 1μm, respectively. As expected, at the mask corner
could not handle the thick oxides required to manufacture the oxide is thicker since the oxidant can enter under the nitride
power devices. Maintaining interface and mesh quality mask from two directions. Therefore the nitride mask lifts up in
becomes critical as the oxide thickness increases, so a new the corner substantially more than predicted by 2D bird’s beak
algorithm had to be devised. simulations.
Fig 2 shows the same process, but using an opposite mask
II. 3D OXIDATION WITH MOVINGMESH ALGORITHM polarity, such that the inside corner of the mask is oxidized.
To support such 3D oxidation processes, we developed a Notice that due to the stress at the mask corner, created by the
new algorithm in Sentaurus Process called MovingMesh. The nitride layer pushing down from both sides, the oxide is thinner
algorithm uses a body-fitted tetrahedral bulk mesh, and as compared to the sides of the mask.
material interfaces are represented explicitly as triangulated
surfaces, each of which conforms to the neighboring bulk
mesh. Mesh motion is done in the Lagrangian description
meaning that the mesh vertices move together with the
material. Interface surface meshes and bulk volume meshes
B. Snow-plow effect deposition, mask and other structure modification steps.
Phosphorus preferably segregates into the silicon side at the Moreover, the surface must be sufficiently smooth such that no
silicon/oxide interface. During the oxidation process, the artifacts in device simulation occur.
moving silicon/oxide front pushes the phosphorus profile in the We demonstrate the superiority of our approach using a
silicon that is consumed by oxidation deeper into the substrate, challenging 3D vertical pillar MOS transistor example. The
creating a pile-up on the silicon side of the silicon/oxide process involves two 3D oxidation steps (a 30nm LOCOS
interface. This is called the snow-plow effect. corner and a 5nm gate oxide), as well as several mask, etch and
In a convex 2D trench corner, this effect is stronger at the deposition steps. In Fig. 5, the structure is shown before and
corner, since phosphorus is pushed from both top and side. In a after LOCOS, before and after gate oxidation and at the end of
3D convex trench the pile-up at the corner expected to be even the process flow.
bigger than the 2D case. This affects the doping distribution in IV. CONCLUSION
a 3D device significantly and will result in device
characteristics that cannot be predicted by using 2D oxidation A new MovingMesh algorithm was developed to address
results. moving boundary problems such as 3D oxidation. Since the
algorithm is capable of maintaining mesh and boundary quality
Fig. 3 shows results for wet oxidation of a 3D trench corner even after large changes in the structure, fairly thick oxidation
at 1150C, for 5, 11, 25 and 50 minutes respectively. The initial processes, such as those for power devices, can be simulated in
wafer is uniformly phosphorus doped at 5e16cm-3. The grown 3D using Sentaurus Process.
oxide is shown only as an outline for clarity and its thickness is
200nm, 300nm, 500nm and 700nm, respectively. The 3D REFERENCES
snow-plow effect is clearly visible. [1] V. Senez, S. Bozek and B. Baccus, "3-dimensional simulation of thermal
diffusion and oxidation processes," in International Electron Devices
There is also an inverse effect for a dopant that prefers to Meeting, pp.705,708, 1996.
segregate into the oxide, like boron. In this case, boron piles up [2] N. Strecker and D. Richards, “A 3D moving grid algorithm for process
on the oxide side of a bottom corner of a trench. This effect is simulation”. Journal of Computational Electronics, Volume 5, Issue 4 ,
shown in Fig. 4, under the same oxidation conditions as before. pp 297-300, 2006.
The silicon region under the oxide is not shown for clarity. [3] V. Suvorov, A. Hössinger, Z. Djurić and N. Ljepojevic, “A novel
approach to three-dimensional semiconductor process simulation:
C. Vertical pillar MOS transistor Application to thermal oxidation”. Journal of Computational Electronics,
As discussed previously, one of the challenges of Volume 5, Issue 4 , pp 291-295, 2006.
performing 3D oxidation is to maintain surface and mesh
quality. This is not only important for the oxidation step itself,
but is also required to ensure success of the subsequent etch,

Figure 1: 3D LOCOS corner with 100nm, 300nm, 500nm and 1μm oxide thickness
Figure 2: 3D inverse LOCOS corner with 100nm, 300nm, 500nm and 1μm oxide thickness

Figure 3: Phosphorus snow-plow effect at an oxidized 3D trench corner

 Figure 4: Boron pile-up in the oxide at an oxidized 3D bottom trench corner

Figure 5: Vertical pillar MOS transistor before and after LOCOS, before and after gate oxidation and final structure

View publication stats