Project NeuComp

Modeling, verification, simulation and hardware implementation of artificial neural networks

About the Project

Human Brain Project (HBP), FET Flagship funded by the EC and other international projects such as Synapse and BrainScale have highlighted the importance of large-scale simulations for understanding the functioning and structure of the brain. These simulations will enable real numerical experiments (as reproducible and reliable as physical or biological experiments) and will validate hypotheses in theoretical biology. This computational objective of biomimetic neural networks is one of the main issues of the C@UCA (Computational Brain and Cognition) - umbrella project funded by IDEX.

NeuComp -  a sub-project of C@UCA - is focused on the computational properties of the Leaky-Integrate-and-Fire (LIF) model. From the electronic point of view, it will be necessary to design a digital architecture reproducing the integration of the related synaptic potentials in the form of multiplexed neural processors, the synaptic connectivity between these neurons through a communication network on chip (NoC) using an  Address Event Representation (AER) and network learning in the form of embedded memory of synaptic weight.

NeuComp centres on three main objectives jointly targeting the Leaky-Integrate-and-Fire neuron model [LIF]:
1. Design an electronic architecture of neural network using  LIF model on FPGA circuit,
2. Modeling performances of the different computers, implementing simulation codes on some of them, and checking the temporal properties of neural archetypes,
3. Modeling their behavior with Luster language, a synchronous language for modeling of reactive systems.

Expected outcome 

NeuComp will be a step forward in design of a software and hardware basis for the computational study of the brain. Thanks to this basis, new research works can be initiated. With regard to the verification part (via model-checking), more formal work based on temporal logic will have to be carried out. As far as the neuronal computers part is concerned, a significant implementation work will have to be provided. A first validation on the HBP platform will have to be extended and generalized by the whole consortium.

Project completed - Extended with project ARTEFACT