My PhD, which I defended in 2004, was undertaken in the neuromorphic systems engineering group (known as the ISN) at the IXL Laboratory, University Bordeaux 1, and it laid the fundamental groundwork for the design of neuromimetic ICs, i.e. integrated circuits that mimic the behavior of a nerve cell or a set of cells. I designed a library of mathematical operators for use in the Hodgkin-Huxley model which concerns the electrical activity of biological neurons, and developed an integrated system for use as a demonstrator using the component parts of the library. The library was designed to facilitate the subsequent design of neural circuits dedicated to neuroscientific applications. This work led to the first implantation on silicon of a wide variety of neuronal dynamics by using the Hodgkin-Huxley model.
During the year when I was an ‘ATER’ (temporary teacher/researcher) (2004-05), I decided to further my knowledge in the cognitive sciences and undertook a Masters at University Victor Segalen Bordeaux 2.
In September 2005, I was recruited as a Associate-Professor in electronics at University Bordeaux 1.
From the outset of my career, I supervised the work of highly promising students. One such student was Adel Daouzli (from 2005 to 2009), whose subject was based on the European project FACETS (FP6-2004-IST FETPI 15879). The objective of FACETS was to design new computing paradigms that represented a significant shift from the concept in vogue at that time, the Turing machine, by incorporating biological data observed in brain activity. My colleagues and I sought to design and operate a hardware simulator of small neural networks. The project was divided into three parts within Workpackage 6. My job was to implement our neuromimetic system and it was in this respect that Adel Daouzli worked under my supervision, especially with regard to characterizing the plasticity of small neural networks.
My activity as research supervisor continued with the supervision of Laure Buhry during her research masters in cognitive science (2006-07) and thereafter her PhD (2007-10). Her subject concerned the estimation of the parameters of the Hodgkin-Huxley model. In the first phase of the project, the method she developed used the Differential Evolution algorithm, which is a high-level problem-independent algorithmic framework. This metaheuristic was then used for the first time in computational neuroscience by applying it to our neuromimetic circuits in order to optimize each ion channel independently of each other. In the second phase, she extended the method to the simultaneous estimation of all the parameters by focusing on the dynamics of the voltage of the neuronal membrane. The multidisciplinary nature of the subject and the quality of her work were given official recognition in the form of a L’Oréal France/UNESCO/ Academy of Sciences award that she received in 2009.
In 2008, we hosted Dr. Hsin Chen from the National Tsing Hua University, Taiwan, for three months. I worked with him on implementing stochastic neural patterns in our analogic neuromimetic circuits.
In 2008-09, I was the leader of the ERNAM project (PEPS CNRS STI) which investigated the impact of electromagnetic fields on biological neurons. This study was conducted in collaboration with the Bio-EM team led by Dr Bernard Veyret from the IMS laboratory. ERNAM created a link between the Bio-EM team and the team at ISN headed by Prof. Sylvie Renaud. This in turn led to the creation in September 2009 of the Bio-Electronics Research Group team within the IMS.
The ISN then expanded its research and, given the growing number of researchers it was hosting, we decided in January 2010 to reorganize the Bio-Electronics group led by Prof. Sylvie Renaud into three teams: Bio-EM, Elibio and AS2N (Architecture of Silicon Neural Networks), which is the one I direct. The AS2N team at that time comprised two other tenured Associate-Professor in addition to me. Our research focus was the design and use of biologically realistic neural networks.
At that time, I was also the lead partner in the ECRéN project (PIR NeuroInf CNRS, 2009-11), whose aim was to calculate the transfer function of neuronal circuits in a combinatorial way and by using neuromimetic circuits. Neurons and their synaptic inputs were modeled realistically and on the basis of experimental data obtained both in vivo and in vitro by Dr Alain Destexhe from the Integrative and Cognitive Neurosciences Unit. From then on, it was up to the AS2N team to implement these models in integrated circuits, to generate randomly a large number of possible networks and to calculate their transfer functions.
From September 2009 to August 2013, I was lead scientist for the IMS European project FACETS-ITN (FP7-PEOPLE-ITN-2008 237955). This research and training program involved partners from 11 universities and research centers and two industrial companies from 6 European countries. It was within the framework of this project that I supervised the PhD of Filippo Grassia (January 2010 – January 2013), who was also involved in ECRéN. The work we did together allowed us to reproduce complex neuronal dynamics by using our neural circuits and to compare them with those in living organisms, thanks in particular to a methodology we developed based on the analysis of the phase plane.
In 2010 I got a Fulbright scholarship. Thanks to this award and to a 6-month research sabbatical that the university awarded me, I went to Johns Hopkins University (JHU), Baltimore (MD) as Visiting Associate Professor from February to July 2011. There I had the honor of collaborating with Professor Andreas Andreou, the world-renowned expert in neuromorphic engineering. Just after that in 2011, I received my full accreditation to supervise research.
Supervising Filippo Grassia for his postdoc in my team (January 2013 – August 2014) resulted in the implementation of a quartic neuronal model on FPGAs. This model takes into account subthreshold stimuli that generate oscillations under the threshold at which action potentials are triggered.
I also led the MHANN project (ANR P2N, 2011-15), a collaborative study with the Unité Mixte de Physique CNRS/Thales, INRIA Saclay and Thales TRT. Here, the aim was to combine analogic integrated circuits and neuromorphic ferroelectric memristors in order to build a medium-sized neural network. Memristors are currently the focus of much scientific interest, especially for researchers designing neuromorphic architectures. Their resistance decreases as the intensity of the current flowing through them increases. Like biological synapses, memristors facilitate the transport of information. Moreover, the last physical state of the component is retained after a power shutdown. The performance of the demonstrator developed in the MHANN project is currently being evaluated in order to establish whether memristors have potential value for creating a new generation of computers based on memristive neural networks. Within this project I was the PhD supervisor of Gwendal Lecerf (October 2011 – September 2014). Our work in the project MHANN allowed us to propose the first integrated solution for reproducing excitatory and inhibitory synapses using memristors. Furthermore, our solution can be used to create neural networks based on a crossbar of memristive synapses. This innovation has been patented internationally thanks to the CNRS patent policy concerning memristor-based technology. Thanks to this project, we have gained leadership in the design of memristive neuromorphic networks, and I led the team that produced a state-of-the-art article on the implantation of synaptic plasticity in memristors.
More recently I supervised the PhD of Matthieu Ambroise (October 2012 – July 2015) which had two objectives: firstly to create a digital neural network on FPGAs in order to perform living-artificial hybrid experiments as part of the ANR Hyrène and EU Brainbow projects; and secondly, to lay the foundations for a universal communication bus for neuromorphic systems that can take the form of integrated circuits, FPGAs or platforms for the acquisition/stimulation of the activity of biological cells. This has allowed our team to develop a central pattern generator based on the Izhikevich neuron model that was connected with a rat spinal cord in a living-artificial hybrid experiment.
Together with Dr Julie Grollier, I co-founded the GDR BIOCOMP project in 2015. This GDR aims to bring together the scientific community concerned (7 sections from 5 CNRS institutes: INB, INC, INP, INSIS and INS2I) on the issue of the hardware implementation of natural computing. We are striving to understand the mechanisms at work in biological systems in order to create chips based on natural computing and to build hardware architectures that can be used as test systems to better understand the biology. Since 1 January 2015 I have been the co-director of this GDR.
Since 1 January 2016, I have been the leader of the MIRA project (ANR 2016-2020). This project involving the Unité Mixte de Physique CNRS / Thales, the Intitut de la Vision and ChronoCam is seeking to develop a new neuromorphic computing paradigm based on ferroelectric memristors for processing high-speed visual information. The project should lead to a breakthrough in the field of artificial neuromorphic retinas. The latter do not use the conventional framework of serial pictures but encode visual information by events with a resolution of the order of a microsecond. Events (i.e. action potentials) will be sent from a camera to a neural network composed of silicon neurons and memristive synapses. It will become possible for the very first time to process visual information up to 100 kHz while consuming less energy.
Finally, I was elected Vice-Chairman of the Scientific Council of the IMS laboratory in 2016 by my colleagues in recognition of my multidisciplinary involvement in what our laboratory does.