A&QT-R 2004 (THETA 14) 2004 IEEE-TTTC - International Conference on Automation, Quality and Testing, Robotics May 13 – 15, 2004, Cluj-Napoca, Romania Page 1 of 6 PRINCIPLES OF DESIGN FOR LARGE-SCALE NEURAL SIMULATORS Raul C. Mureşan 1,2 , Iosif Ignat 1 1. Technical University of Cluj-Napoca, Faculty of Automation and Computer Science, Gh. Baritiu 26-28 2. Nivis Research, Gh. Bilascu 85 3400 Cluj-Napoca, ROMANIA raul.muresan@nivis.com, iosif.ignat@cs.utcluj.ro Abstract We present a review of design principles that are to be used for large-scale neural simu- lators. This paper emphasizes the most important problems encountered in the simulation of biologically plausible neural systems and provides some solutions derived from modern simulation techniques. We stress upon the idea that a modern simulator should be able to perform generic simulations of as many as possible neural architectures, with various neural models. At the same time the amount of processing effort should be tunable, with the possi- bility of using various simulation methods simultaneously (e.g. iterative and event-driven). The neuroscientist should be able to use different types of electrophysiological models with complete inter-operability. We provide an example of a neural simulator that complies with modern design guidelines: ”The Neocortex Simulation Environment”. Key words: Neuron, synapse, simulator, large-scale simulation, neural spike. 1. INTRODUCTION As the field of theoretical neuroscience develops and the electrophysiological evi- dence accumulate, researchers need more and more efficient computational tools for the study of neural systems. In cognitive neuroscience and neuro-informatics the modeling of neural systems is the essential approach for understanding the complex processing performed by the brain. The most recent trend in these fields is to use spike-based or detailed models of neurons and detailed (or quasi-detailed) models for synaptic trans- mission. During the past five decades, the most frequently used types of artificial neural networks have been the perceptron-based models. However, the biological relevance of such models is very limited, the perceptron being a phenomenological model of the neu- ron. It has long been accepted that neurons encode information in the rate of their firing, which has led to rate-based models such as the perceptron. However, more recently, sci- entists began to understand that individual spikes of neurons could play a very impor- tant role in the encoding of information in the brain [15]. Moreover, millisecond proc- esses triggered by individual spikes shape the synaptic transmission efficacy, in a spike time dependent (STDP) way [6]. The modeling of detailed electrophysiological phenomena seems to be essential for the understanding of brain processes. At the same time, models inspired from biol-