Abstract
We have witnessed recently a huge explosion of human brain big projects. Despite these immense efforts and acquisitions of enormous amounts of brain data, a model of the human whole brain at the nanoscale has not yet been developed. The goal of this work is to estimate high performance computing (HPC) requirements needed for morphology modeling of the entire human brain at the nanoscale including the segmentation of neurons and the calculation of chemical and electrical synapses.These requirements are estimated for different computers (1 giga-, 1 tera-, 1 peta-, and 1 exa FLOPS) and various spatial resolutions of volumetric brain data (1 µm as the reference, 0.5 µm to handle axons and large dendrites, 10 nm to handle chemical synapses, and 1 nm to handle electrical synapses). The determined numbers of operations are the lower-bound estimates needed for segmenting neurons by region growing and calculating synapses by multi-directional sampling.At the 1 µm spatial resolution, a giga FLOPS computer needs more than 2 years to segment the neurons and only a tera FLOPS (taking over 18 hours) or higher supercomputers are suitable for this task. At the 10 nm resolution, an exascale supercomputer is able to segment the neurons in no less than 18 hours (and at present there is only one operating exascale supercomputer in the world, Frontier; two more exascale supercomputers Aurora and El Capitan are under installation). The HPC requirements for the calculation of chemical synapses are negligible in comparison to those for neuron segmentation. To segment the neurons at 1 nm level enabling the calculation of electrical synapses, an exascale supercomputer requires more than 2 years.To my best knowledge, this is the first attempt to estimate the HPC requirements to compute a morphology model of the human whole brain at the nanoscale.