Argonne National Laboratory

Developing a Petascale Neural Simulation

TitleDeveloping a Petascale Neural Simulation
Publication TypeReport
Year of Publication2003
AuthorsHereld, M, Stevens, RL, van Drongelen, W, Lee, HC
Series TitleConf Proc IEEE Eng Med Biol Soc. 2004
Date Published12/2003
Other NumbersANL/MCS-P1147-0404

Simulations of large neural networks have the potential to contribute uniquely to the study of epilepsy, from the effects of extremely local changes in neuron environment and behavior to the effects of large-scale wiring anomalies. Currently, however, simulations with sufficient detail in the neuron model are limited to cell counts that are far smaller than scales measured by typical probes. Furthermore, future simulations are likely to follow the path of large-scale simulations in other fields and include hierarchically interacting components covering different scales and different biophysics. Problem solving in this domain calls for petascale computing---computing with supercomputers capable of performing 1015 operations a second and holding datasets pf 1015 bytes in memory. We present the structure of our simulation of epileptiform electrical activity in the neocortex, describe experiments and models of its scaling behavior in large cluster supercomputers, identify tight spots in this behavior, and project the performance onto a candidate next-generation computing platform.