Faculty Interaction

Jayathi Murthy

School of Mechanical Engineering
Purdue University

Brief Project Description:

Parallelization of a molecular dynamics method for computing the thermal conductivity of materials such as Silicon.

A portable parallel program developed for molecular dynamics simulations is developed. This program computes the thermal conductivity of bulk silicon. In particular, the effect of the computational domain size (number of atoms) on the predicted accuracy is investigated. The thermal conductivity of thin film silicon at room temperature is explored. This program shows excellent scalability on a variety of architectures including IBM's BlueGene, IBM's Power4 P655+, and Linux clusters. A task that would have taken 30 days on a single processor can now be completed in 2.5 and 5 hours on 1024 Power4+ processors and 1728 BlueGene's PowerPC processors respectively, using our parallel molecular dynamics algorithm. Results show that, as the thickness increases up to 300 nm, the thermal conductivity approaches the bulk thermal conductivity. This work permits us to confidently apply molecular dynamics to the simulation of thermal conductivity of both bulk materials and thin films.

The computed thermal conductivity follows: Predicted thermal conductivity of bulk silicon using Green-Kubo method as a function of number of atoms. The experimental value is 148 W/mK. run.

The performance of this program on various national centers and Purdue computers is shown below for these computers. The average total time in seconds for one time step in the MD computation is plotted as a function of the number of processors, on a log-log scale.


Legend:

      BlueGene: San Diego Supercomputing Center (SDSC), IBM Blue Gene
      Cobalt: National Center for Supercomputing Applications (NCSA), SGI Altix
      Mercury: National Center for Supercomputing Applications (NCSA), IBM IA-64
      Bigben: Pittsburg Supercomputing Center (PSC), Cray XT3
      Lear: Rosen Center for Advanced Computing, Purdue (RCAC), Xeon cluster
      DataStar: San Diego Supercomputing Center (SDSC), IBM P655+
      Tungsten: National Center for Supercomputing Applications (NCSA), Xeon cluster
      Lonestar: Texas Advanced Computing Center (TACC), Intel Woodcrest

The initial results, including the BlueGene data, were presented in:

Lin Sun, Chinh Le, Faisal Saied, and Jayathi Y. Murthy, A Highly Scalable Simulation Model for Atomistic Calculation of Thermal Properties of Silicon, PDPTA'06 / RTCOMP'06: June 26-29, 2006.

An expanded version of the paper with results for additional platforms was presented at the Teragrid 06 conference in Indianapolis, in June 2006:

"Massively parallel atomistic simulations of thermal properties of silicon on the TeraGrid". Lin Sun, Chinh Le, Faisal Saied, Dave McWilliams, Jayathi Murthy. Presentation available in Powerpoint format at:
Massively parallel atomistic simulations of thermal properties of Silicon on the Teragrid