Alain Cappy (Director of IRCICA)

Pierre Boulet & Philippe Devienne (LIFL)

Dominique Vuillaume & Alexis Vlandas (IEMN)



NEURAL SYSTEMS: 9h45-11h00

Spike-based Learning in Biological and Artificial Neural Networks

 Simon Thorpe (CerCo (Brain and Cognition Research Center) & SpikeNet Technology SARL, Toulouse, France)

Abstract: I will argue that Spike-Time Dependent Plascity mechanisms (STDP) could provide a key to understanding how biological neural systems are able to learn to recognize complex repeating patterns. A combination of experimental and simulation studies have demonstrated that, thanks to STDP, neurons can become selective to a given stimulus after a few tens of presentations. Since STDP like learning can potentially be implemented in a range of different hardware systems, this opens the possiblity of developing memristor based artificial systems that could reproduce some of the most interesting features of biological neural systems.


SpiNNaker: a spiking neural network architecture

Steve Furber (ICL Professor of Computer Engineering, the University of Manchester, UK)    &

Abstract: The SpiNNaker project is developing a massively-parallel computer, ultimately to incorporate over a million ARM processor cores, optimized for modeling large-scale systems of spiking neurons in biological real time. At present we have prototype systems with just under a thousand processors and a software suite that allows automated mapping of networks from a high-level description in a language such as PyNN or NENGO onto SpiNNaker, and various vision and robotics demonstrations on the system’s capabilities.  


BREAK (11h00-11h15)




Neuromorphic accelerators for computing

Rodolphe Héliot (Research scientist at CEA-LETI, Saclay, France)

Abstract: Because of power and reliability issues, computer architects are forced to explore new types of architectures, such as heterogeneous systems embedding hardware accelerators. Neuromorphic systems are good candidate accelerators that can perform efficient and robust computing for certain classes of applications. We develop a spiking neurons based accelerator, with its hardware and software that can execute a wide range of signal processing applications. A library of operators is built, and automated place-and-route tools map the application onto the hardware. A mixed-signal, 65nm integrated circuit implementation is designed that focuses on fault tolerance and low-power. Altogether, this system aims at providing to the user a simple, power-efficient way to efficiently implement signal processing tasks on neuromorphic hardware.


Memristor based neuromorphic architectures: from concept to reality.

Jacques-Olivier Klein 

Abstract : Neuromorphic architectures are frequently considered as ideal candidates to exploit emerging nano-devices that differ radically in their functionality from MOS transistors. Especially, their memristive behaviour is often compared to synaptic connexion in neural network and authors rely frequently on the supposed intrinsic robustness of neural architectures to overcome the defect of nanodevices. Thereby learning procedures have been proposed for ideal memristors.  Nevertheless, the actual characteristics of memristors are far from ideal and the supposed robustness of neural networks is not systematically guarantied. Thus, new strategies should be explored to move from
concept of memristor based neuromorphic architecture to real and functional circuits. 


BUFFET LUNCH at the IRCICA-IRI Hall (12h30-14h00)





Neuromorphic Computing with imperfect devices

Damien Querlioz (IEF, Univ; Paris-SUD, CNRS UMR 8622, France)
Nanodevices provide fantastic opportunities for electronics, as they can be fast, low power and provide novel functions like memristivity. They however come with challenges, like variability and unpredictability, which can be intrinsic features due to their small sizes. The brain can be an inspiration to exploit such nanodevices, since it itself relies on variable and partly unpredictable nanodevices. This talk will present different joint works between groups in the Orsay/Saclay area and Grenoble, where we propose to use memristive nanodevices in a way analogous to the brain’s synapses. We address specifically how such approaches can be tolerant to devices’ imperfections, and in some cases, can even benefit from them.  All these works illustrate how diverse neuromorphic approaches may be adapted to different memristive technology, and suggest their high potential.


Memristors – Artificial Nano-Synapses

Julie Grollier (Unité Mixte de Physique CNRS/Thales, Palaiseau, France)

Abstract: Memristors are nanometer size non-volatile tunable resistances.  The resistance state of these devices can be changed by varying the voltage or the current across the structure. Depending on the memristor type, the resistive switching effects can be due to several different physical effects, ranging from red-ox to charge-induced, phase change or purely electronic phenomena. Practically, due to the possibility to easily manipulate their resistance state, these electronic nano-devices have a number of extremely promising applications, such as digital memories (Resistive Switching RAMs), switches and latches for advanced logic functions etc.

In addition, since their resistance can in general be tuned over several orders of magnitude, it should be possible to stack densely these components in large-scale cross-bar arrays. One of the most fascinating properties of memristors is that they intrinsically behave like synapses, which could be a key to the future development of hardware Artificial Neural Networks (ANNs), and revolutionize non-conventional neuromorphic computing.

This talk will be a review of the state of the art of memristor devices and their applications, with a special focus on their implementation as artificial synapses for on-chip neural networks.


BREAK (15h30-15h45)




Molecular programming of logic gates in bacteria

Alfonso Jaramillo (CNRS, ISSB, Synth-Bio, Génopole Evry, France)

Abstract: The computational design of RNA interactions allows for the programming of complex information processing devices in bacteria. For this, we use evolutionary computation algorithms that automatically optimize the RNA sequences by minimizing the energy of formation and the activation energy. An automated design approach has been already reported in vitro, but never before in living cells. We have experimentally validated in E. coli fully synthetic RNAs displaying logic gate behavior. We also show in E. coli that our riboregulatory devices can be combined with known functional RNA fragments (ribozymes and aptamers) to create complex logic circuits in bacteria. We also characterized their in vivo RNA dynamics by using microfluidics time-lapse microscopy to track single-cells. Our work provides a new paradigm to program functional RNAs gates working
in living cells by allowing the computer to use first principles combined with a molecular interaction mechanism.


Designing in vitro reaction circuits

Yannick Rondelez (CNRS-LIMMS, Tokyo, Japan),_Dr

Abstract: The concept of circuitry at the molecular level, including loops, feedback and gates, to control complex dynamic functions has emerged from the study of living systems. Recent advances in biology are demonstrating the central role and the seemingly unlimited wealth of dynamic behaviors that can be performed by well-designed molecular systems. The next step is logically to start the construction of artificial, ex vivo molecular systems targeting specific dynamic purposes. We are developing molecular programming

techniques to achieve this goal. I will introduce the building of oscillators and switches, as well as spatially extended systems, rationally encoded from DNA interactions.


Tweet about this on TwitterShare on LinkedInShare on Google+Share on FacebookEmail this to someone