Search

Keywords: Airborne molecular contamination, condensation defects, threshold voltage, high k metal gate, corrosion. Introduction During dies manufacturing, silicon surfaces cleaning continues to be challenged to reduce materials consumption. There’s a trade-off between such reduction and keeping a good enough cleaning efficiency to avoid defects creation. This paper gives a few examples of chemical traces deposited on various surfaces, degrading the dies quality such as corrosion concerns, device electrical properties shifts. Experimental and characterizations In this work, a 28nm node is considered in a high k metal gate, gate first integration. The metal gate is a stack of a HfSiON and TiN films, on FD-SOI wafers [1]. Various chemicals are hereby discussed, all dispensed on spin dry single wafer platforms from SCREEN. Contamination has been measured thanks to LPE-IC (Liquid phase extraction – ionic chromatography). Results and discussion Airborne molecular contamination can lead to various concerns. Such contamination can be divided in 4 categories: acids, bases, condensables, and dopants. Halogen based gas and liquids are widely used in the semiconductor industry: plasma etching (CF4, HBr, ...) [2], deposition and wet processes. First and foremost, halogen gas can adsorb on substrates during plasma, or deposition processes. If the gas is not properly purged away from the process chamber, this can lead to a significant surface contamination. First example consists in volatile salts left on a silicon nitride film after a LPCVD furnace deposition. The film surface is contaminated by salts due to inefficient dichlorosilane gas chamber purge. Although the salts are quite volatile, they can generate further patterning defects, blocking plasma etching processes (figure 1). Similar kind of defects, with an order of 20µm size are also observed after selective silicon epitaxy, that uses HCl gas. Such salts are analyzed by LPE-IC (table 1), revealing their nature: ammonium chloride, from adsorbed HCl having reacted with ammonia and humidity from air cleanroom. If unremoved, severe silicon corrosion is found few hours after the epitaxy process. Fortunately, a very mild SC1 (Standard Clean 1) wet or plasma process can either dissolve or degas such salts. Eventually, wet processes can also meet airborne molecular contamination concerns [3], especially when using concentrated acids (sulfuric [4], or phosphoric acid [5]). They can leave a significant amount of anions on the substrate surface, if not properly cleaned, forming ammonium based salts. Photo lithography patterning quality can be degraded (figure 2) but also thermal oxide growth kinetics [6]. As a last straw, single wafer cleaning platforms can deliver different processes between wafer frontside and backside. This is convenient to clean only the wafer backside without touching its frontside. Although the wafer is protected by a nitrogen curtain at few millimeters from the wafer surface, a device degradation is found when a wafer backside clean is operated after the high k film deposition. Indeed, the NMOS threshold voltage is shifted up to 40mV. A perfect correlation is drawn with a fluorine contamination measured on the high k surface, thanks to XPS (figure 3). This correlation is relevant towards some observations in the literature where fluorine can affect the device [7] [8]. This fluorine airborne contamination is finally confirmed thanks to a Picarro analyzer that confirms the chamber ambiance can be contaminated to levels of 0.6 ppbv. Conclusions Airborne molecular contamination can be linked with poor gas line purges, leaving adsorbed molecules reacting with cleanroom atmosphere. This can lead to severe yield degradation. Moreover, although cleaning tend to be milder and milder, concentrated acids still needs a relevant cleaning to avoid salts growth. Finally, a deeper attention is required to control the ambiance of single wafer cleaning platforms. References [1] J. Hartmann et al., Planar FD-SOI technology at 28nm and below for extremely power efficient SoCs, Las Vegas show (2012) [2] A. Kabansky, Solid State Phenomena Vols. 103-104, p 59-62 (2005) [3] J.M. Oh et al., Solid State Phenomena, vol. 145–146, Trans Tech Publications, p. 151–154 (2009) [4] P.J Clews et al., ECS, Vol. 97-35, p. 23-30 [5] P. Garnier et al., ECS, 80 (2), p. 101-109 [6]. W. Scott et al., ECS, 151 (1), G8-G12 (2004) [7] S. Lee et al., Solid State Technology 56(4), p. 22-24 (2013) [8] W. Xiong, IEEE international SOI conference (2007) Figure 1

The carbon footprint of astronomical research is an increasingly topical issue with first estimates of research institute and national community footprints having recently been published. As these assessments have typically excluded the contribution of astronomical research infrastructures, we complement these studies by providing an estimate of the contribution of astronomical space missions and ground-based observatories using greenhouse gas emission factors that relates cost and payload mass to carbon footprint. We find that worldwide active astronomical research infrastructures currently have a carbon footprint of 20.3$\pm$3.3 MtCO$_2$ equivalent (CO$_2$e) and an annual emission of 1,169$\pm$249 ktCO$_2$e yr$^{-1}$ corresponding to a footprint of 36.6$\pm$14.0 tCO$_2$e per year per astronomer. Compared with contributions from other aspects of astronomy research activity, our results suggest that research infrastructures make the single largest contribution to the carbon footprint of an astronomer. We discuss the limitations and uncertainties of our method and explore measures that can bring greenhouse gas emissions from astronomical research infrastructures towards a sustainable level., Comment: 48 pages, 6 tables, 2 figures

Since the Declaration of Salamanca and the Convention on the Rights of Persons with Disabilities, we have moved to an inclusive paradigm with regard to school. Mainstream schooling has become the rule. Thus, teachers in regular classes must work with all students, including those with special educational needs. It is not easy; many teachers face dilemmas in accommodating all types of students in the same classroom. This conceptual paper proposes a framework based on Mezirow’s transformative learning for understanding the transformations in teacher pedagogy needed to work in an inclusive school. It can be used as a basis for developing training mechanisms to support teachers in their transformation to work in an inclusive school.

Global dust storms at Mars engulf the entire planet in a dusty haze, causing increases in temperature and ion escape as dust is lifted up to 80 km in altitude. The two most recent storms occurred in 2007 (Mars Year (MY) 28) and 2018 (MY34), and have been observed by spacecrafts such as Mars Express (MEx). MEx has been operating at Mars since 2004, and has produced a long time-base of plasma measurements from as low as 250 km. Using MEx, we investigate whether the 2007 dust storm has influenced the magnetosphere of Mars by looking at the position of the bow shock and induced magnetospheric boundary, compared to the expected position provided by 3D magnetohydrodynamical models. To identify boundary positions, we use data from the ASPERA-3 instrument (Analyser of Space Plasma and EneRgetic Atoms) onboard MEx, which contains an electron spectrometer (ELS), ion mass analyser (IMA), neutral particle imager (NPI) and neutral particle detector (NPD). For this study, we use data from ELS and IMA. We consider a number of influences on the boundary position, including the solar wind conditions and the crustal fields. Our study period includes time before, during, and after the MY28 global storm, and we expected the bow shock and induced magnetospheric boundary to increase in altitude due to the storm. Out results show that the system is more complex, and multiple influences need to be distinguished to leave any change due to the dust storm itself.

The Martian interaction with the solar wind leads to the formation of a bow shock upstream of the planet. The shock dynamics appears complex, due to the combined influence of external (solar photons, solar wind plasma and fields) and internal (crustal magnetic fields, ionized atmosphere) drivers. The extreme ultraviolet fluxes and magnetosonic mach number are known major drivers of the shock location, while the influence of other possible drivers is less constrained or unknown such as crustal magnetic fields or the solar wind dynamic pressure and the Interplanetary Magnetic Field (IMF) intensity and orientation.We analyze and rank the influence of the main drivers of the Martian shock location, based on published datasets from Mars Express and Mars Atmosphere Volatile EvolutioN missions and on several methods such as the Akaike Information Criterion, Least Absolute Shrinkage Selection Operator regression, and partial correlations. We include here the influence of the crustal fields, extreme ultraviolet fluxes, magnetosonic mach number, solar wind dynamic pressure and various Interplanetary Magnetic Field parameters (intensity and orientation angles).We conclude that the major drivers of the shock location are extreme ultraviolet fluxes and magnetosonic mach number, while crustal fields and solar wind dynamic pressure are secondary drivers at a similar level. The IMF orientation also plays a significant role, with larger distances for perpendicular shocks rather than parallel shocks.

Keywords: hydrophobic films, drying marks, SiCN, amorphous carbon, chemical oxide growth Introduction Integrated circuits manufacturing requires use of hydrophobic materials, as low-k films. Maintaining their integrity during wet cleaning is challenging. Indeed, the aqueous liquid film break-up on these surfaces during the cleaning process can result in undesired drying marks defects [1] which can cause open-circuits and be yield killers by blocking proper patterning. Use of dry-cleaning methods instead of wet chemistries can be a solution to avoid drying marks formation. However, while relevant in particle removal they cannot remove chemisorbed defects. Therefore, the use of aqueous solutions cannot be always avoided to clean hydrophobic films. In this case, inducing surface modification of the hydrophobic surface before the aqueous-cleaning step can be a solution. Since drying is carried out after the surface becomes hydrophilic, the occurrence of drying marks can be prevented. In this work we first study how chemical treatments can modify hydrophobic surfaces, so they get a hydrophilic nature. Based on that, we evaluate the interest that can have such step before aqueous wafer cleaning techniques. Experimental and characterizations Silicon carbon nitrogen (SiCN) and amorphous carbon (a-C) films are deposited by plasma enhanced chemical vapor deposition method on 300mm p-type (011) silicon substrate. The a-C composition is of 65% carbon,35% hydrogen while the SiCN composition is of: 42% Hydrogen, 20% carbon, 14% nitrogen, 24% silicon. They have been obtained with elastic recoil detection analysis, nuclear reaction analysis and Rutherford backscattering spectroscopy. All wet chemical surface preparations have been carried out on spin dry single wafer platforms from SCREEN. Two wet chemistries have been evaluated: Standard Clean 1 (mix of ammonia, hydrogen peroxide and water with a respective volume ratio of 1/2/80, fresh chemistry at 65°C) and ozonated water (20ppm). A treatment with an excimer ultraviolet lamp with a 172nm wavelength has been also done. The reactions between these films and treatments are characterized by AR-XPS and XRR. Artificial contamination of SiCN wafers for particle removal tests has been done by spin-drying using isopropanol containing monodisperse SiO2 particles of 80nm. Results and discussion The surface preparations of a-C and SiCN films impact the water contact angles, as reported in Figure 1. A-C surface presents a contact angle of 5° with water, after UV or ozonated water treatments, while stays unaltered with the SC1. This difference between the ozonated water and SC1 dispense can be related to the oxidizing power difference as ozonated water features higher redox potential than ammonia. Proposed mechanism of amorphous carbon oxidation by ozone is the breaking of C=C bonds and the formation of carboxyl —COOH groups [2]. On the other hand, SiCN contact angle is reduced by SC1, ozonated water and UV (respectively 45, 30 and 5°). AR-XPS characterizations have been used to identify the bonds modification during these treatments. The results lead to the conclusion of an oxidation mechanism for ozonated water, with an oxide layer thickness of less than 5nm (XRR) (Table 1). SiCN bonds oxidation are thermodynamically favored, even though kinetics of these reactions has not been studied (Table 2). The development of a hydrophilic behavior on these surfaces suggests that we avoid watermarks formations. A comparison of SC1 spray cleaning on modified and as deposited SiCN wafers has been carried, to determine the impact of surface modification on particle removal efficiency. Nevertheless, in the specific case of low-k films two potential impacts of these surface modifications also need to be investigated. Indeed, electrical properties of low-k films are key parameters, that is why the consequences of surface modification on them need to be evaluated. And delamination of low-k interfaces is already a failure risk. It would be enhanced by an adhesion strength reduction due to surface energy change. Contact angle measurements and the Fowkes model have been used to deduce the influence of these treatments on surface energy. Conclusion The cleaning of hydrophobic films is a key challenge for yield benefits. These films are prone to drying marks formation during aqueous cleaning steps. In this work, we study how surface preparation can break non-polar bonds, leading to the development of a hydrophilic surface. Several treatments (UV, SC1 and ozonated water) and materials (a-C, SiCN) have been studied. A comparison of particle removal efficiency with SC1 spray method between as deposited and modified films have been carried out, to quantify the impact of these modifications on wet surface cleaning. References [1] Han, J. and al, Solid State Phenomena, 134, p295–298, 2007 [2] Mawhinney, D. and al. Carbon, 39(8), p1167–1173, 2001 Figure 1

Magnetic reconnection is a physical process that can occur when oppositely directed magnetic field lines interact with each other and change of the magnetic field topology, transferring the magnetic energy to the energy of the surrounding particles. This phenomenon can be found in numerous astrophysical contexts, one of which is at the magnetopause of the planets where the solar wind and the magnetosphere interact with each other. Many studies have been made on this matter since the 1980's using simulations and observations from missions like THEMIS or CLUSTER. However, the heart of the reconnection process, the electron diffusion region (or EDR), remained largely unstudied due to the difficulty to observe it until the launch of the Magnetospheric Multiscale (MMS) mission in 2015. Since then, more and more studies on the EDR emerge thanks to the unprecedented resolution of the data provided by MMS but still lack significant numbers of EDR cases to generalize the results, the number of identified EDRs was only of about 32 in 2018 and had not changed much in the next years. In this thesis, I detail the work I have done to identify new EDR events in the data of the MMS mission using deep learning techniques and to study some of the physical aspects of the EDR that are not yet fully understood by the scientific community such as the structure of the EDR or the energy conversion process at the heart of this region. I first developed a Multilayer Perceptron to automatically identify new EDR events using 32 cases previously reported in the literature as a training database. This model uses several classical physical parameters such as the magnetic field, the electric field or the magnetic current as well as one derived scalar that I created called the "MeanRL". This parameter was designed to characterize one of the key signatures of EDRs which is the presence of a crescent shape in the electron velocity distribution functions taken in this region. The trained algorithm was applied to the data of phase 1a of MMS and allowed us to identify 18 new EDR candidates. Four times as many EDRs with positive energy conversion (inner EDRs) as EDRs with negative energy conversion (outer EDR) were found in phase 1a, and the inertial term of the generalized Ohm's law was found responsible to the deceleration of the electrons in the outer EDR, leading to the negative energy conversion observed there. Then, I used the previously and newly reported EDRs to perform several studies about : the link between the solar wind parameters and the energy conversion at the EDR, potential new small scale parameters to identify EDRs, the position of several points of interest within the EDR as well as the intensity and the parallel/perpendicular components of the energy conversion at the EDR. Finally, I designed a Convolutional Neural Network (CNN) as an alternative way to look for EDRs. This model analyzes the electron velocity distribution functions to directly search for electron crescents and tries to be an improvement of the "MeanRL" parameter. Events showing such crescents are then "manually" investigated to confirm or not the EDR detection. This CNN was applied on the data of phase 1b of MMS that had not been treated by the first algorithm and allowed for the detection of 17 additional EDR candidates, bringing the total number of identified EDRs to about 70 as of the writing of this thesis.; La reconnexion magnétique est un processus physique qui intervient quand des lignes de champ magnétique opposées interagissent ensemble, résultant en un changement de la topologie magnétique et un transfert de l'énergie magnétique vers l'énergie des particules environnantes. Ce phénomène peut être retrouvé dans de nombreux contextes astrophysiques dont l'un est à la magnétopause des planètes où le vent solaire et la magnétosphère interagissent ensemble. De nombreuses études ont été réalisées à ce sujet depuis les années 80 en utilisant des simulations et des données d'observations de missions comme THEMIS ou CLUSTER. Cependant, le cœur du processus de reconnexion, la région de diffusion électronique (ou Electron Diffusion Region), restait largement non étudié à cause de de la difficulté de son observation jusqu'au lancement de la mission Magnetospheric Multiscale (MMS) en 2015. Depuis lors, de plus en plus d'études sur l'EDR ont vu le jour grâce à la résolution sans précédent des données produites par MMS mais manquent toujours d'un nombre significatif de cas afin de généraliser les résultats obtenus, le nombre d'EDR identifiées n'était en effet que d'environ 32 en 2018 et n'a pas beaucoup changé dans les années suivantes. Dans cette thèse, je détaille le travail que j'ai mené dans le but d'identifier de nouvelles EDRs dans les données de la mission MMS en utilisant des techniques d'apprentissage profond et pour étudier certains des aspects de l'EDR qui ne sont pas encore complètement compris par la communauté scientifique comme la structure de l'EDR ou encore la conversion d'énergie au sein de cette région. J'ai tout d'abord développé un Perceptron Multicouche pour identifier automatiquement de nouveaux cas d'EDR en utilisant 32 cas précédemment rapportés dans la littérature comme base de données d'entraînement. Ce modèle utilise plusieurs paramètres physiques classiques comme le champ magnétique, le champ électrique ainsi qu'un scalaire dérivé que j'ai créé appelé le "MeanRL". Ce paramètre a été conçu pour caractériser l'une des signatures clés de l'EDR à savoir la présence de formes de croissants dans les fonctions de distribution des vitesses des électrons prises dans cette région. L'algorithme entraîné a été appliqué aux données de la phase 1a de MMS et a permis l'identification de 18 nouveaux candidats d'EDR. Quatre fois plus d'EDR présentant de la conversion d'énergie positive (inner EDR) que d'EDR présentant de la conversion d'énergie négative (outer EDR) ont été trouvés durant la phase 1a, et il a été trouvé que le terme inertiel de la loi d'Ohm généralisée est responsable de la décélération des électrons dans l'outer EDR, causant la conversion d'énergie négative observée dans cette région. J'ai ensuite utilisé les anciens et nouveaux cas d'EDR rapportés afin d'effectuer plusieurs études portant sur : le lien entre les paramètres du vent solaire et la conversion d'énergie à l'EDR, de potentiels nouveaux paramètres de petite échelle pour identifier des EDR, la position de plusieurs points d'intérêts au sein de l'EDR ainsi que sur l'intensité et les composantes parallèles/perpendiculaires de la conversion d'énergie à l'EDR. Pour finir, j'ai conçu un réseau de neurones convolutionel (CNN) comme moyen alternatif pour rechercher des EDRs. Ce modèle analyse les fonctions de distribution des vitesses des électrons pour directement chercher des croissants électroniques et essaie d'être une amélioration du paramètre "MeanRL". Les cas montrant de tels croissants sont alors analysés "à la main" pour confirmer ou non l'identification d'une EDR. Ce CNN a été appliqué sur les données de la phase 1b de MMS qui n'avait pas été traitée par le premier algorithme et a permis la détection de 17 cas additionnels d'EDR, amenant le nombre total d'EDR identifiés à environ 70 lors de l'écriture de cette thèse.