Implementation and acceleration of neuron simulator with CUDA C

Νεοφύτου, Αλέξανδρος; Neofytou, Alexandros

dc.contributor.author	Νεοφύτου, Αλέξανδρος	el
dc.contributor.author	Neofytou, Alexandros	en
dc.date.accessioned	2020-05-18T14:34:46Z
dc.date.available	2020-05-18T14:34:46Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/50631
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.18329
dc.rights	Αναφορά Δημιουργού - Παρόμοια Διανομή 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-sa/3.0/gr/	*
dc.subject	NVIDIA	en
dc.subject	Adaptive exponential integrate-and-fire model	en
dc.subject	GPGPU	en
dc.subject	CUDA	en
dc.subject	STDP	en
dc.subject	Προσομοίωση νευρώνων	el
dc.subject	Παράλληλος προγραμματισμός	el
dc.subject	Νευρώνες	el
dc.subject	Υπολογιστική νευροεπιστήμη	el
dc.subject	Επιτάχυνση	el
dc.title	Implementation and acceleration of neuron simulator with CUDA C	en
dc.contributor.department	Microlab	el
heal.type	bachelorThesis
heal.classification	Πληροφορική	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2019-06-20
heal.abstract	Neuroscience is the scientific study of the nervous system and the relation of nerves to behaviour and learning. The biggest effort for neuroscientists is focused on the brain as its increased understanding results in more clear knowledge about human consciousness. While most experiments in the past were conducted in specialised laboratories studying portions of actual brain neurons, in modern neuroscience computers are widely utilised to simulate biological neural networks in great detail and complexity. These simulations enable the visualization of networks of greater size, aiding them further in their research. In order for this discipline to continue advancing, even more sizeable and complex neuron networks are being used in simulations, generating the need for simulation acceleration in different platforms and architectures used by the corresponding laboratories. While plenty of simulators are available for a wide selection of neuron model simulations, the majority is not optimized for modern computer systems and subsequently doesn't achieve optimal performance, delaying neuroscientists from important conclusions drawn from simulation results. This diploma thesis aims to appease the need for accelerated simulation by utilizing the CUDA API for acceleration on an NVIDIA GPU system. The model this thesis is focused on is the Adaptive Exponential Integrate-and-Fire neuron model with Spike-timing Dependent Plasticity on its synapses, widely used in modern research. The original simulation was firstly imported from the Brian Simulator into a new simulator written in the C programming language and then developed for efficient GPU acceleration. The CUDA platform is a software layer that gives direct access to the GPU's virtual instruction set and parallel computational elements for general purpose programming applications. This implementation was successful in accelerating the neuron simulation a factor of over 100x times in comparison to the Brian Simulator, occasionally reaching even a 1000x acceleration rate while keeping the same functionality with the Brian Simulator. There is no theoretical limit in the amount of neurons contained in the network, though performance was observed to drop significantly as GPU memory limits were surpassed. This fact constituted a point of interest and was investigated further along with other observations in this diploma thesis.	en
heal.advisorName	Σούντρης, Δημήτριος	el
heal.committeeMemberName	Σούντρης, Δημήτριος	el
heal.committeeMemberName	Πεκμεστζή, Κιαμάλ Ζ.	el
heal.committeeMemberName	Ματσόπουλος, Γεώργιος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών. Τομέας Τεχνολογίας Πληροφορικής και Υπολογιστών. Εργαστήριο Μικροϋπολογιστών και Ψηφιακών Συστημάτων VLSI	el
heal.academicPublisherID	ntua
heal.numberOfPages	106 σ.
heal.fullTextAvailability	false