On-Atom Chip Inertial Sensing Keywords : Atom interferometry; inertial sensors; cold atoms; atom chips; Partners: Systèmes de Référence Temps-Espace (SYRTE), Thales Research&Technology France (TRT) Requested grant: 300 000€ Project starting and ending dates: 2014 - 2017 The main goal of this project is to tackle the limits for inertial sensing using compact atom interferometers based on cold atom manipulation in atom chips. Our work will be essentially concentrated on the development and characterization of atom chip matter-wave interferometers for acceleration and rotation sensing. As compared to state-of-the-art atomic sensors, which typically use free-falling atoms, atom chip sensors would be more compact, while potentially keeping the high level of performance associated with cold atoms. The longer term objective supported by this project is the achievement of a compact atom chip based high-performance inertial measurement unit (IUM) hybridized with microelectromechanical (MEMS) sensors for higher measurement bandwidth. We will take into account the possibility of such hybridization from the very beginning of the project (at both hardware and software levels) in cooperation with experts from Thales Avionics. Such a compact high-performance IMU could find applications in several military systems usually relying on Global Navigation Satellite Systems, when the latter are not available (technical failure, urban canyon, intentional jamming…). Other possible applications are high performance inertial navigation with a compact IMU, terrain correlation with gravity sensing and several kinds of scientific applications. At the heart of the scientific project is the study and implementation of coherent beam splitters, and the investigation of different sources of decoherence in an atom chip interferometer, whether they are of technical or physical nature. In particular, we will investigate two very promising ways of implementing on-atom chip beam splitters, namely radiofrequency manipulation of the external degree of freedom of magnetically guided atoms and microwave manipulation of both internal and external degrees of freedom of magnetically trapped atoms. From the technology point of view, two different materials employed in the fabrication of atom chips (SiC and AlN) will be compared in terms of adequacy for the fabrication of inertial sensors. We will extensively study the technical issues leading to decoherence in atom chip interferometry, such as the roughness of the magnetic guide in guided atom interferometry and current noise leading to the fluctuations of trapping frequencies in trapped atom interferometry. From a physics point of view, we will address the questions of using sub-Doppler cooled thermal and Bose-Einstein condensed atoms for inertial sensing with guided and trapped atoms, the interaction induced decoherence, and the role of interatomic correlations. At the end of this project, we plan to establish a technology roadmap for the industrial development of a inertial sensors based on the hybridization of atom chips and MEMS sensors, which would be a technological breakthrough in the field.

In the general context of the field of eXplainable Artificial Intelligence (XAI), the IFP-in-RL project aims to propose a method for the automatic construction of a control system of a system, such as a drone, which takes take into account the interpretability constraint in its very design. This project takes place within the framework of systems based on fuzzy rules which, since their introduction, aim to facilitate the expression of knowledge in a linguistic form, natural for the user, and easily understandable by a human. Such a knowledge representation is an excellent way to promote human interaction with the computer system and to improve their understanding of how it works, thus offering the possibility of making their behavior transparent and easily validated. In the literature, different approaches to build or to fine-tune a fuzzy rule base to design a system exist, but they generally suffer from the drawback of not incorporating specific interpretability optimization. In this project, an innovative methodology is introduced for the design of such systems. This methodology is based on the implementation of a reinforcement learning approach using interpretability metrics. The objective here is to integrate the consideration and optimization of the desired interpretability during the learning itself, and not a posteriori as many methods currently do in the field of XAI. The IFP-in-RL project aims to achieve this upstream, a complete study, both theoretical and experimental, of interpretability metrics, including existing numerical criteria as well as user needs. This will involve proposing a taxonomy of existing metrics and defining new measures if necessary, in order to complete the previous ones and allow their exploitation in original reinforcement learning algorithms. An original feature of this project is to integrate a qualitative assessment, carried out on a human panel, of the proposed metrics but also of the rule bases obtained at the end of reinforcement learning. In application terms, the objective of the IFP-in-RL project is to implement these proposals for piloting a drone, navigating in complete autonomy to ensure a mission consisting of flying over points of interest and taking pictures, from data provided by a simulator.

Classical computed tomography scanners incorporate a rotating ring or “gantry” which holds the X-ray source and the X-ray sensors. They exhibit a high resolution but are large, heavy and are characterised by a low throughput. Using multiple X-ray sources, low cost, stationary (gantry free) and efficient scanners could be fabricated. Xintek (American company) proposes to use, for each X-ray tube, a carbon nanotube (CNT) cathode driven by a control circuit. In this case, the cathode is grounded and the anode is biased at high voltage. The anode cooling is then complex. Each electron source can be advantageously a CNT photocathode driven by a laser diode. As the circuit driving the laser power is insulated from the cathode, this cathode can be biased at high voltage and the anode can be grounded. The anode cooling is then facilitated. Moreover, for low/medium power scanners, one can use a transmitting target X-ray window (e.g. a tungsten film deposited on a beryllium window) that delivers wide angle X-ray beams. This allows to increase the analysed volume. The long term objective of this project is to develop a new generation of optically switchable multiple X-ray source for compact, efficient and low cost CT scanners for medical applications. High in depth resolution imaging implies the use of a large number of X-ray sources (around 50). When one X-ray tube is emitting X-rays, the emission from other tubes should be low. To satisfy the requirements of high resolution CT scanners, an ON/OFF ratio of 1000 is required. Other requirements relates to the ON current (10 mA/mm2) and to the fabrication of photocathodes operating with backside illumination (to ease their integration). CNT photocathodes based on silicon p-i-n photodiodes exhibit an ON/OFF ratio of 30. In order to meet the ON/OFF ratio requirement (1000), an alternative to p-i-n photodiodes is to use a photoconductor based on a high resistivity low-temperature-grown GaAs (LTG-GaAs) developed by IEMN. This material exhibit a resistivity of around 107 W.cm and a high breakdown voltage of 200 - 300 kV/cm. The CNT photocathode studied in this project will be an array of individual and vertically oriented multiwall CNTs, each nanotube being associated with a LTG GaAs photoswitch. Such arrays are already fabricated by different laboratories (including LPICM) on doped silicon substrates. The growth method is plasma enhanced chemical vapour deposition (PECVD) on Ni catalyst dots. The crystalline quality exhibited by the nanotubes is not sufficient to attain the objective of current density. To improve the CNT crystalline quality, LPICM has initiated the study of the growth of aligned CNTs on Fe (instead of Ni) catalyst using a new water vapour based PECVD process and showed that these nanotubes exhibit a significantly higher crystalline quality. Thus arrays of individual CNTs grown on Fe nanodots will be studied. The workprogram includes one task for management, dissemination and exploitation and another one to study the safety procedures of the photocathode process. Task 2 will deliver an accurate model of the photocathode. Task 3 will define the CNT technology, the photoswitch technology and the method to fabricate/transfer a photoswitch on transparent substrates. Task 4 will demonstrate a photocathode on a GaAs substrate (front side illumination) with an ON/OFF ratio of 1000. Finally Task 5 will show a photocathode operating with back side illumination and delivering an ON current of 10 mA/mm2 and an ON/OFF ratio of 1000. Thales has already shown that a CNT X-ray tube exhibit a good life time. This project will allow Thales Electron Devices to develop a new generation of optically switchable multiple X-ray source. The first application is the development of low cost, compact and efficient stationary (gantry free) CT scanners. The total market addressed by this new generation of multiple X-ray sources is estimated to be above $250 million per year.

ROMEO is a project which aims to develop a robust system for anomaly detection in drone trajectories based on the use of physics informed neural networks, statistical methods for anomaly detection and quantification of uncertainty through conformal inference methods. In this consortium Thales will bring its expertise in the use of physics informed neural networks and anomaly detection. The "Institut de Mathématiques de Toulouse" (IMT) will bring its expertise in uncertainty evaluation through conformal inference. The massive usage of drones open the path to multiple applications both civil and for defense, including surveillance or smart logistic missions. Such applications may require to use large numbers of drones and in this context, it is crucial to ensure a safe and secure usage of drones through unmanned traffic management (UTM) systems solutions that are both efficient and reliable. In this project, we propose a system which raises alerts for UTM operators. This system raise an alert when an anomaly is detected in the drone trajectory when compared with expected trajectory. The prediction of the normal trajectory will be based on physics informed neural networks, allowing to introduce prior knowledge on the flight dynamics. The anomaly detection will be performed with innovative statistical metrics. The uncertainty on the normal trajectory prediction will be estimated with conformal inference methods. This uncertainty bounds will be integrated in the anomaly detection method in order to provide the operator with trustable alarms.

Critical embedded systems are subject to time and security guarantee requirements, while requiring significant computing power that can only be offered by multicore processors. Off-the-shelf processors meet the performance requirements but do not provide the necessary safety and security guarantees, on the one hand because they were not necessarily designed for this purpose, and on the other hand because the details of their internal architecture are not known, thus prohibiting any reliable analysis of their behavior. The aim of the project is to propose a multicore processor that meets these two requirements. The availability and control of a safe (temporally predictable) and secure processor will allow a state and its public and private actors to carry out their missions in optimal conditions of trust. We will follow the path of the open hardware movement which offers the possibility to develop specific processors. The design of a specific processor enabled by the technological, software and organizational infrastructure of the RISC-V ecosystem meets the challenges of sovereignty by (1) ensuring components availability and (2) guaranteeing complete control of the hardware architecture. This full knowledge will enable the relevant implementation of formal verification tools for the expected properties in terms of predictability and security (complete and precise knowledge of the hardware implementation is the guarantee of being able to produce a correct model). Moreover, it will also make it possible to verify the authenticity and fidelity of the components.