Figure 4 - uploaded by Valeria V Krzhizhanovskaya
Content may be subject to copyright.
Source publication
In this article, the authors introduce a Grid-based virtual reactor, a High Level Architecture (HLA)-supported problem-solving environment that allows for detailed numerical study of industrial thin-film production in plasma-enhanced chemical vapor deposition (PECVD) reactors. They briefly describe the physics and chemistry underpinning the deposit...
Context in source publication
Context 1
... in accordance with the global evolution of the system to get correct overall behavior. The execution of distributed modules must be co- ordinated: each federate should treat its time correctly, and the events should be interpreted by federates in a correct order. Those issues are the concern of time management. Causality conflicts occur when a federate advances its time at a different rate than the other federates expect. Thus, federates exchange events with time stamps. Federate A can stop federate B to produce the event before federate A does not advance simulation time. Using HLA in a Grid environment requires additional support to ensure a high level of fault tolerance and ro- bustness. A good example is the Load Management Sys- tem architecture based on GT2 technology. The “services” concept provides a good starting point for building and connecting independent blocks of different functionality in the problem-solving environment. In Zajac et al. [23] and Rycerz et al. [24], we describe an architecture of Grid services that cooperate to set up distributed interactive applications. The user initiates the Flow Executor service and supplies it with the application code (already com- piled and dynamically linked with the HLA RTI library) as well as with information about the RTI library version used. This Flow Executor queries the Discovery Service, which checks the Information Service to find out sites with “HLA-Speaking” Services with the appropriate version of the RTI library installed. The RTI library itself is treated as a resource. The “HLA-Speaking” Service sets up all the necessary environmental variables to dynamically link the user code. The Discovery Service also finds out the location of the RTIexec service. Once supplied with necessary information, the Flow Executor transfers application code to the resources and asks the Local Flow Setup to set up all local services that support performance monitoring, check- pointing, and migration scheduling of the local HLA federate. When all the necessary services are set up, the HLA application is running within a robust and fault-tolerant environment. We also devised a mechanism to support migration of HLA federates [23].A Monitoring Tool monitors the execution of the HLA federate and resource load and makes decisions about migration. If there is a need to migrate, the Monitoring Tool informs the Migration Scheduler, which then asks the Discovery Service for available resources. The Discovery Service inspects the Information Service and responds to the Migration Scheduler. Based on that information, the Migration Scheduler makes a de- cision on where to migrate the HLA federate and asks the Local Flow Setup Service to set up all the necessary local services on that resource. When this is done, the Migration Scheduler asks the “HLA-Speaking” Service to save the state and transfers data to the new site. Then, it restarts the new “HLA- Speaking” Service with the available check- point file. We have incorporated the virtual reactor into the Grid using the above- described CrossGrid and HLA system. We achieved secure Grid access, resource discovery and regis- tration, Grid data transfer, application initialization, chemical reactions editing, parameter specification, job submission, distributed PECVD simulation, and advanced three- dimensional visualization (see Fig. 3). The resulting virtual reactor is a problem-solving environment with an advanced graphical user interface (GUI), integrating the basic modules for reactor geometry design, computational mesh generation, plasma, flow and chemistry simulation, editors of chemical reactions and gas properties connected to the databases describing species and possible chemical reactions, pre- and postprocessors, visualization modules, Web interfaces, and a Grid portal. The GUI allows one to vary the problem definition and reactor parameters, control the simulation process, visualize and analyze the simulation results, and access the databases and result archives. An advanced visualization system provides graphical representation of the results in real-time (or postponed) mode on different computer/graphic systems (personal computer, virtual reality DRIVE, three-dimensional immersive environment CAVE, “personal digital assistant,” or the Web) without the actual need to change the code [35]. The GUI was designed with the use of C/C++ program- ming languages, widespread platform-independent GTK+ graphic library (glib, gdk, gtk) [37], and the Glade user interface builder [38]. The three-dimensional immersive visualization system was developed using the VTK and CaveLib libraries. All modules of the virtual reactor are platform independent. To provide access to the virtual reactor, an interface for remote use was developed based on Web technology, using a client-server model. For this interface, we created HTML pages, Java applets, and Java and CGI scripts. This system was integrated into the Migrating Desktop (see Fig. 4). We have tested the problem-solving environment for a number of tightly coupled clusters as well as for the dis- tributed computer systems (the CrossGrid testbed), where separate modules of the PSE (databases, archives, uncou- pled solvers, visualization kernel, and the user interface) were located and operating at different sites. The actual network used as a testbed for our virtual reactor includes a local step inside the university via Fast or Gigabit Ethernet, a jump via a national network provider at speeds that range from 34 to 622 Mbits/s or even Gigabit to the national node, and a link to the Geant network at 155 Mbits/s to 2.5 Gbits/s. To validate the reliability of the model and the usability of the simulation environment, we set up a number of experiments with a PECVD reactor from the group of Dr. J. K. Rath from Utrecht University [39]. This allowed us to per- form detailed studies in a controlled environment. The geometry of the ASTER system is given in Figure 5. It shows the relevant reactor components such as inlet, outlet, electrodes, and overall configuration. We used detailed measurements of the actual geometry of the chamber, substrate, and the electrodes as input parameters to our virtual reactor. The virtual chamber was constructed from a number of simple blocks, and in each block, a regular mesh was generated, as shown in Figure 6. The initial gas mixture and plasma discharge parameters are chosen corresponding to the real experimental data available [40]. Used in our comparison are the following experimental parameters: discharge frequency = 50 MHz, applied power = 10 W, pressure = 0.15 Torr, gas temperature = 300 ◦ K, substrate temperature = 520 ◦ K, flow rate = 20-60 sccm, and inflow mixture composition = 50% silane and 50% molecular hydrogen. We studied in detail the simulation results for the electron distribution as well as the deposition rate on both of the electrodes. In Figure 7, we can clearly see that the area with the maximum electron concentration is located near the edge of the powered (lower) electrode, which corresponds to the edge effects observed in experiments. We analyzed the spatial distribution of the growing film thickness. The experimentally observed films have a parabolic variation in thickness, with the minimum in the middle of the substrate. The simulation results are quali- tatively in good agreement with respect to the film distribution shape and the observed thickness variation (Fig. 8). However, this variation is more pronounced in the experiment than in the simulation results. We report on a pilot study integrating a complex of simulation components for PECVD into a wide-area Grid computing environment. We showed that the Grid infrastructure developed in the CrossGrid research project facilitates advanced simulation, interaction, and visualization. To support seamless integration of legacy-distributed simulation systems, we developed and used Grid-aware HLA services. Apart from the successful nontrivial distribution of simulation functionalities in our system, we demonstrate the use and reliability of PECVD simulation applied to ex- isting experimental reactors. The multilevel parallelization and distributed mapping to the Grid infrastructure give way to scalability and performance tuning. Further detailed study of the parallelization of the numerical solvers and efficient load balancing on job and task levels needs to be addressed in future research. New insights into the physics of the PECVD process, such as formation of nanoparticles and their incorporation into the growing films, require additional research, the consequence of which can be further adaptation of the simulation environment. We have shown that the virtual reactor can aid the design of industrial reactors, the optimization of film properties, and the minimization of the prime costs. Potentially, it also opens up an opportunity for real-time control over industrial reactors, running the software in parallel with the real deposition processes, and tuning various parameters in interactive, manual, or automatic mode. Multidisciplinary work such as the one presented here can only be done through intensive teamwork. We specially acknowledge the following researchers: W. J. Goedheer (FOM Institute on Plasma Physics, the Netherlands), M. A. Zatevakhin and A. A. Ignatiev (St. Petersburg Polythechni- cal State University, Russia), M. Bubak and K. Zajac (Institute of Computer Science AGH, Krakow, Poland), J. K. Rath (Utrecht University, the Netherlands), and A. Tirado- Ramos, H. Ragas, and M. Scarpa (University of Amsterdam, the Netherlands). The help of the CrossGrid consor- tium has proven to be crucial for the success of the Grid implementation. The research was conducted with financial support from the Dutch National Science Foundation NWO and the Russian Foundation for Basic Research under grant numbers 047.016.007 and 047.016.018, as well as with partial support from the ...
Similar publications
In this paper we introduce a Grid-based Virtual Reactor, a problem-solving environment that supports detailed numerical study of industrial thin film production in Plasma Enhanced Chemical Vapor Deposition (PECVD) reactors. We describe the physics and chemistry underpinning the deposition process, the numerical approach to simulate these processes...
Citations
... In the next stage of our research we will investigate the effect of these boundary conditions. Our mathematical model is based on mass, momentum and energy balance equations for a laminar transient flow of a chemically reacting gas mixture at low Reynolds numbers [8]. A model of chemical reactions is based on [9,10]. ...
We developed a multiphysics mathematical model for simulation of silicon dioxide Chemical Vapor Deposition (CVD) from tetraethyl orthosilicate (TEOS) and oxygen mixture in a microreactor at atmospheric pressure. Microfluidics is a promising technology with numerous applications in chemical synthesis due to its high heat and mass transfer efficiency and well-controlled flow parameters. Experimental studies of CVD microreactor technology are slow and expensive. Analytical solution of the governing equations is impossible due to the complexity of intertwined non-linear physical and chemical processes. Computer simulation is the most effective tool for design and optimization of microreactors. Our computational fluid dynamics model employs mass, momentum and energy balance equations for a laminar transient flow of a chemically reacting gas mixture at low Reynolds number. Simulation results show the influence of microreactor configuration and process parameters on SiO2 deposition rate and uniformity. We simulated three microreactors with the central channel diameter of 5, 10, 20 micrometers, varying gas flow rate in the range of 5-100 microliters per hour and temperature in the range of 300-800 °C. For each microchannel diameter we found an optimal set of process parameters providing the best quality of deposited material. The model will be used for optimization of the microreactor configuration and technological parameters to facilitate the experimental stage of this research.
... The goal of our research is the development of a reliable model and investigation of correlation between process parameters and chemical composition of SiN x H y . We modeled a real PECVD reactor, and used experimental results and our previous studies [3,4,5] to build the chemical kinetics model. For process optimization at the macroscale level we have chosen a continuum modeling approach. ...
We developed a mathematical model of Plasma Enhanced Chemical Vapor Deposition (PECVD) of silicon nitride thin films from SiH4-NH3-N2-Ar mixture, an important application in modern materials science. Our multiphysics model describes gas dynamics, chemical physics, plasma physics and electrodynamics. The PECVD technology is inherently multiscale, from macroscale processes in the chemical reactor to atomic-scale surface chemistry. Our macroscale model is based on Navier-Stokes equations for a transient laminar flow of a compressible chemically reacting gas mixture, together with the mass transfer and energy balance equations, Poisson equation for electric potential, electrons and ions balance equations. The chemical kinetics model includes 24 species and 58 reactions: 37 in the gas phase and 21 on the surface. A deposition model consists of three stages: adsorption to the surface, diffusion along the surface and embedding of products into the substrate. A new model has been validated on experimental results obtained with the "Plasmalab System 100" reactor. We present the mathematical model and simulation results investigating the influence of flow rate and source gas proportion on silicon nitride film growth rate and chemical composition.
... We performed a series of targeted experiments varying the resource heterogeneity in the processor power and the network links bandwidth. As a sample of these tests, Fig. 4 shows the dependency of the load balancing speedup on the processing power heterogeneity metrics, analogous to that of the networks links heterogeneity introduced by Eq. (2). As it can be seen, the speedup grows superlinearly with the heterogeneity level, thus indicating that our approach is especially beneficial for strongly heterogeneous resources, such as the Grid resources. ...
... The proposed algorithm dynamically selects the resources best suited for a parallel application and optimizes the load balance based on the dynamically measured resource parameters and estimated requirements of the application. The performance of this load balancing approach is studied on a real-life Virtual Reactor application [2,3] and on a synthetic application with flexible user-defined application parameters and logical network topologies, on heterogeneous resources with a controlled level of heterogeneity. Some of the conclusions from our methodological experiments are as follows: ...
This paper suggests a hybrid resource management approach for efficient parallel distributed computing on the Grid. It operates on both application and system levels, combining user-level job scheduling with dynamic workload balancing algorithm that automatically adapts a parallel application to the heterogeneous resources, based on the actual resource parameters and estimated requirements of the application. The hybrid environment and the algorithm for automated load balancing are described, the influence of resource heterogeneity level is measured, and the speedup achieved with this technique is demonstrated for different types of applications and resources.
... All rights reserved. doi:10.1016/j.jpdc.2007.08.010 simulation of plasma enhanced chemical vapour deposition (PECVD), a multiphysics process spanning a wide range of spatial and temporal scales [15,16] . Simulation of threedimensional flow with chemical reactions and plasma discharge in complex geometries is one of the most resource-demanding problems in computational science, requiring both highperformance and high-throughput computing. ...
... Grid computing technologies opened up new opportunities to access virtually unlimited computational resources, and inspired many researchers to develop new methodologies and algorithms for parallel distributed applications on the Grid. The PECVD Virtual Reactor discussed in this paper has also been on its way to the Grid [15]. It serves as a test-case driving and validating the development of the Russian–Dutch Grid (RDG) computational for distributed high performance simu- lation [14,22]. ...
... Current work on porting the Virtual Reactor to the Grid started within the framework of the CrossGrid EU project [26] and the Virtual Laboratory for e-Science [27] . Some results of these efforts were reported in [15]. The RDG Grid is the successor of the CrossGrid in a sense that it uses many of the CrossGrid infrastructure services and operates as a testbed for the Virtual Reactor application. ...
We address the problem of porting parallel distributed applications from static homogeneous cluster environments to dynamic heterogeneous Grid resources. We introduce a generic technique for adaptive load balancing of parallel applications on heterogeneous resources and evaluate it using a case study application: a Virtual Reactor for simulation of plasma chemical vapour deposition. This application has a modular architecture with a number of loosely coupled components suitable for distribution over the Grid. It requires large parameter space exploration that allows using Grid resources for high-throughput computing. The Virtual Reactor contains a number of parallel solvers originally designed for homogeneous computer clusters that needed adaptation to the heterogeneity of the Grid. In this paper we study the performance of one of the parallel solvers, apply the technique developed for adaptive load balancing, evaluate the efficiency of this approach and outline an automated procedure for optimal utilization of heterogeneous Grid resources for high-performance parallel computing.
... For instance, in a multi-scale model for crack propagation, the basic units are continuum mechanics at the macroscale, modeled with finite elements, and molecular dynamics at the microscale [32]. Another examplex is provided by Plasma Enhanced Vapor Deposition where mutually coupled chemical, plasma physical and mechanical models can be distinguished [33]. In principle all basic modeling units can be executed on a single (parallel) computer, but they can also be distributed to several machines in a computational grid. ...
In this chapter we will argue that studying such multi-scale multi-science systems gives rise to inherently hybrid models containing many different algorithms best serviced by different types of computing environments (ranging from massively parallel computers, via large-scale special purpose machines to clusters of PC's) whose total integrated computing capacity can easily reach the PFlop/s scale. Such hybrid models, in combination with the by now inherently distributed nature of the data on which the models `feed' suggest a distributed computing model, where parts of the multi-scale multi-science model are executed on the most suitable computing environment, and/or where the computations are carried out close to the required data (i.e. bring the computations to the data instead of the other way around). We presents an estimate for the compute requirements to simulate the Galaxy as a typical example of a multi-scale multi-physics application, requiring distributed Petaflop/s computational power.
... Besides these immediate benefits, the computational resources of the Grid provide the required performance for large-scale simulations and complex visualization in collaborative environments, which are expected to become of major importance to the modern hi-tech society. Scientific communities have been investing great efforts into development of Grid-aware problem solving environments for complex applications [5]- [21]. ...
... The Virtual Reactor presented here as a case study application, was developed for simulation, optimization and control over the plasma enhanced chemical vapour deposition (PECVD) reactors [5], [6]. Simulation of three-dimensional flow with chemical reactions and plasma discharge in complex geometries is one of the most challenging and resource-demanding problems in computational science and engineering, requiring both high performance computing (one parameter setting calculated as fast as possible) and high throughput computing (sweeping various parameter settings), as well as interactive steering and advanced 3D visualization in collaborative mode. ...
... A complex problem solving environment has a modular architecture and consists of a number of loosely or tightly coupled components [8]. Our test case application, the Virtual Reactor, includes the basic components for problem description, reactor geometry design, computational mesh generation, plasma, flow and chemistry simulation, editors of chemical processes and gas properties connected to the corresponding databases, pre-and postprocessors, visualization, interaction, optimization and archiving modules [5]. The aim of our research is to virtualize separate modules of the PSE to run them efficiently as services and access them on the Grid. ...
This paper reports on our research into supporting collaborative distributed applications on the Grid. Our case study application, a Virtual Reactor problem solving environment, was built for simulation of industrially important technology of plasma chemical deposition. It incorporates a number of components distributed among different organizations and requires both high-performance and high-throughput computing, as well as interactivity and advanced visualization. The application has been ported to the Grid environment and tested on a highly heterogeneous infrastructure of the RidGrid testbed. We describe the Virtual Reactor architecture and approaches used to integrate distributed components of the application on the Grid, emphasizing the role of the Migrating Desktop Grid portal and tools that bridge the gap between Grid resources and end-user applications. We outline the main features and functionalities of our RidGrid testbed built within a larger European CrossGrid infrastructure, and describe novel solutions and technical approaches that facilitate the usage of Grid resources and optimize load balancing for parallel applications on heterogeneous resources.
... CrossGrid was officially concluded in 2005, although a large section of the original testbed still exists today. The RiDGrid [18] was launched in 2004, and aims at realizing a Russian-Icelandic-Dutch infrastructure as a testing ground for a highly distributed chemical application known as the Parallel Virtual Reactor [19], but is also used for other applications such as numerical simulations. The RiDGrid testbed is constructed using the CrossGrid middleware and consists of 6 different sites with over 65 CPUs of its own. ...
... Distributed simulations are an interesting field of research in modern computer science, being developed by various communities [63] -scientific (high performance computing) [95], defense [5], business (e.g. manufacturing [110]; transportation [153]; simulations for real-time planning, scheduling and control [50,19]), and the Internet gaming industry [105]. ...
... Examples of HAL-based applications can be divided into the following groups: Scientific simulations allows scientists to build agent-based frameworks for supporting interactive systems like ISS-Conductor, used for surgery planning [192] or 20 1 Research Area and Roadmap the combination of HLA with Java Agent DEvelopment Framework (JADE) to build Distributed Virtual Environments [176]. Another example is simulation of thin-film production [95]. ...
Grid computing may offer the potential for compute intensive simulations of systems and processes such as those associated with Computational Fluid Dynamics and protein design. Such potential is arguably difficult to realize due to some challenging issues associated with distributed computing systems; including latency, asynchronous communication, and the ubiquity of faults. Various research efforts have been focused on developing grid architectures and mechanisms to minimize the effect of latency and improve the fault-tolerance of resource management systems. In this respect, this paper describes a grid simulation platform geared towards the minimization of the effect of latency through an integration of the domain decomposition approach to process simulation with a chosen neighborhood-based grid architecture. The operation of the proposed platform is illustrated through an experimental simulation of a two dimensional diffusion process.
Thin silicon films are important semiconductor material widely used in modern microelectronics and solar cells. One of the
technologies employed for industrial production of these films is plasma enhanced chemical vapor deposition (PECVD). Modeling
and simulation of this industrially important technology is essential for optimizing physical and chemical parameters, reactive
chamber geometry and operating regimes of the installation, in order to reduce the costs of film production and to provide
a better growth rate and film quality in terms of layer composition and homogeneity. For the latter, spatial processes (convection,
diffusion, heat transfer, chemical transformations and plasma non-uniformity), as well as the variation of system behavior
in time, play a decisive role. We have developed a 3D transient multiphysics multiscale model taking into account all relevant
chemical kinetics, plasma physics and transport processes that occur in the bulk of a PECVD reactor and on the surface of
the growing film [Krzhizhanovskaya, V.V., Zatevakhin, M.A., Ignatiev, A.A., Gorbachev, Yu.E., Goedheer, W.J. and Sloot, P.M.A., In Proceedings
of the 5th International Bi-Annual ASME/JSME Symposium on Computational Technologies for Fluid/Thermal/Structural/ Chemical
Systems with Industrial Applications, ASME PVP-vol. 491-2, 59–68, 2004]. In addition to that we have built an efficient problem solving environment for scientists studying PECVD processes and
end-users working in chemical industry [Krzhizhanovskaya, V.V., Sloot, P.M.A. and Gorbachev, Yu.E., Simulation: Transactions of the Society for Modeling and Simulation
International, vol. 81, No. 1, 77–85, 2005].
