Figure 1.

Scope and key concepts of the methods presented. (I) A spatial discretization of a neuron into two compartmental trees representing axons and dendrites branching; (II) A sample of a dendritic tree and its representative branching matrix structure in thumbnail. Simulation dependency parameters across compartmental connectivity and respective matrix are extracted from these structures; (III) A method for load balancing recursively tests for the optimal tree decomposition, based on the computational workload of each possible subtree. Initial compartmental tree is decomposed into a tree of subtrees and distributed across a distributed memory space. Matrix decomposition follows accordingly. Resolution of independent subtrees is computed asynchronously, with three variables dependencies on parent-children connectivity across subtrees (in dashed), synchronized throughout the execution; (IV) Subtree memory layouts are reorganized to provide SIMD-acceleration of mechanism (ion channel) state updates and solver resolution.