Adaptive materials: devices and systems towards unconventional computing, sensing, bio-electronics and robotics

Resume : Introductory remarks

Resume : The search for brain-inspired solutions is driven by the need for architectures able to reach new levels of computing efficiency in tasks like data clustering and image recognition. The Brain represents a massively parallel and event-based system able to reduce the energy consumption by many order of magnitude in the aforementioned tasks. For this reason, investments in Neuromorphic Computation are gaining momentum, in particular toward the research for the specialized hardware that will enable the true exploitation of a brain-inspired architecture. The common major challenge in building artificial neural networks is efficient vector-by-matrix multiplication, which in turn requires compact implementation of synaptic weights. Classic CMOS-based circuits are insufficient for meeting these challenges, mainly because CMOS-implemented synapses are excessively large. One promising solution stems from the use of hybrid circuits combining the existing CMOS integrated circuits with crossbar add-ons using memristive devices. Such devices are able to process information in analog form, mimicking the basic biological synapse dynamics and strongly reducing the required area. In this presentation, we will discuss our recent progresses in both experimental realization of basic Neural Networks and simulations, grounded on experimental data, of scaled-up networks based on memristive crossbars. Limits and future challenges of such approach will also be presented.

Resume : Among unconventional computing solutions developed today, neuromorphic computing appears has a promising direction. Engineering computing systems inspired by the brain could be an efficient solution for low power processing of unstructured data. In this work we present how emerging memory devices belonging to electrochemical metallization (ECM) cells can reproduce key features observed in biological synapses. By controlling the stability of the filament, we show how short term to long term transition reported previously in this class of memory devices [1] can be efficiently used for unsupervised learning function implementation. A variation of Spike Timing Dependent Plasticity (STDP) is demonstrated based on a detail analysis of the switching dynamics. In addition to STDP, we show how filament dynamics can reproduce conveniently Spike Rate Dependent Plasticty (SRDP) observed in biological synapses and how STDP and SRDP can both be conveniently controlled in ECM cells. We demonstrate by a modeling approach how these features can be used for spike-based computing tasks: we report on a dynamic motion task learning in small sized neuromorphic network composed of a fully connected 9*9 input and 10 output neurons. [1] S. La Barbera et al., ACS Nano, 9 (1), pp 941–949, 2015

Resume : Since their prediction, memristive devices revolutionized the world of computing and nowadays they have been widely considered as promising candidate for mimicking synapses. In particular, organic-based memristors allow the construction of circuits capable of learning. Physarum Polycephalum (PP), a unicellular organism made by a miriad of nuclei dispersed in a cytoplasm, is well suited for the implementation of the functional properties of smart living systems into electronic devices. Here, we presents the characteristics of an hybrid memristor developed by interfacing a channel of (PEDOT:PSS), with PP. The device has memristive features resulting by electrochemical changes occurring into the polymer upon application of anodic potentials across the semiconducting PEDOT:PSS channel. We built a three-terminal device where PP is used as electrolyte and the gate electrode, placed in it, is kept at zero-bias and used as a controlling, silent electrode. The realized device is stable and its characteristics are reversible in the voltage range less than the over-oxidation potential; moreover, it showed memory effects induced by electrochemical changes occurring into the polymer. The device can be accounted as an element with properties similar to those of synapses, useful to implement bio-inspired computation. . Further steps will provide an integral response to Physarum?s metabolism as a consequence of environmental variations.

Resume : Conventional computing systems are facing severe limitations due to the increasing demands on data and processing potentials. Hence, the current challenge is to realize a new computing paradigm that is up to the stringent requirements of today’s applications. Particularly in terms of interpreting the perceived data and learning from it to solve unfamiliar problems. Inspired by the operation of the human brain, from the dimensionality, energy and underlying functionalities, neuromorphic systems are building upon circuit elements to mimic the neuro-biological activities. In abstract terms, the brain is mapped into its corresponding building blocks of neurons and synapses along with their corresponding interactions. Several circuit designs are available either to act as a close resemblance to the philological elements and their behavior or as a mere abstraction of the operation principles in an event based manner. However, adhering to the size and energy constraints along with the extensive scaling witnessed in the CMOS technology, emerging non-volatile memory technologies are rising up as suitable alternatives to conventional designs. Resistive Random Access Memories (ReRAM), in particular is a prominent candidate and has been integrated into several neuromorphic platforms. In its original operation form, with a continuous change of resistance upon input excitation, it has found diverse venues as analog synapses and neuronal elements. On the other hand, with the threshold-based devices, more abstract implementations are available benefiting mainly from a binary operation with the high (R_OFF) and low (R_ON) resistance values. Nonetheless, an intriguing feature of variability is apparent in the operation of the brain, where biological noise has been proven to be quite beneficial for the learning, information processing and decision making. Incorporating this stochasticity characteristic into the neural network operation is mainly narrowed down into having either the neuron or the synapse behaving in a nondeterministic manner. Traditionally, stochasticity is added into the neural networks through injected background noise into the circuit. However, recent studies on the material characteristics of the memristive elements have shown a high predisposition of variable behavior. It is due to the underlying switching mechanism between the ON and OFF states. Electron hopping principles and sub-threshold switching allow for a stochastic formation of a dominant conducting filament and subsequently the variable operation of the memristor. The switching process was found to follow a Poisson process that is akin to the firing of the neuronal element. In that regards, we proposed the use of the stochastic memristor as an inherent source of variability in the neuron that allows it to produce spikes stochastically. From the circuit perspective, the memristor, with its variable threshold, acted as a stochastic comparator and allowed for a more area and power efficient implementation of the neuron. It replaced the random number generator for the injected noise along with the operational amplifier in earlier designs. The proposed design was put further to the test with a Winner-take-all neural network. The stochastic memristive neurons were set as the input and output layers and MNIST hand written digits were used as the input data set for image recognition and learning application. The network performance showed high accuracy ranges and robustness to several characteristic metrics variation. An alternative approach to having stochasticity induced into the neural network is through having stochastic synapses; a feature where the communication channels between the neurons are not behaving in a deterministic manner. An approach we also investigated through the incorporation of stochastic memristors in a crossbar structure and building on the random switching between its two binary values. The synaptic weight was variably set to either an ON or OFF state according to the corresponding value of the memristor.The proposed solution was put to test in a Winner-take-all network as well with digit images used as the input data. SPICE level simulations showed high accuracy in the recognition and learning phases as well. Where the synapses adapted to the input patterns and the neurons were selectively firing to specific patterns rather than random sets. In comparison between the two proposed schemes, the stochastic neurons approach builds on non-memristive deterministic synapses. Furthermore, the synaptic weights are analog, in a sense where they can vary in a multi-level fashion based on the learning rule applied and depending on the resolution of the corresponding circuit implementation. In contrast to our earlier approach, the memristor is the source of stochasticity in the synapses while having deterministic neurons at the input and output levels. The synaptic weights are also confined and abstracted into two levels only. Aside from the size and area perspective, the impact on the system performance, in terms of accuracy, recognition rates, and learning, among these two approaches and where the memristor would fall into place are the main comparison points to be considered.

Resume : In recent years, formidable effort has been devoted to exploring the potential of Resistive RAM (RRAM) devices to model key features of biological synapses. This is done to strengthen the link between neuro-computing architectures and neuroscience, bearing in mind the extremely low power consumption and immense parallelism of biological systems. Here we demonstrate the feasibility of using the RRAM cell to go further and to model aspects of the electrical activity of the neuron. We focus on the specific operational procedures required for the generation of controlled voltage transients, which resemble spike-like responses. Further, we demonstrate that RRAM devices are capable of integrating input current pulses over time to produce thresholded voltage transients. We show that the frequency of the output transients can be controlled by the input signal, and we relate recent models of the redox-based nanoionic resistive memory cell to two common neuronal models, the Hodgkin-Huxley (HH) conductance model and the leaky integrate-and-fire model.

Resume : Neuromorphic Response in Pt/NbO Wire Arrays Using Continuum Resistance for Greyscale Image Recognition. Curtis J. O’Kelly, Yasmin Halawani, Baker Mohammad Khalifa University of Science and Technology Abu Dhabi, United Arab Emirates Abstract: Bio-inspired devices with adaptive and changeable properties in response to input stimuli present many advantages over conventional static digital silicon based technologies. The ability to compute, process and retain information in parallel without referencing other circuit elements offers enhanced speed, size and functionality benefits. One promising emerging application for bio-inspired neuromorphic devices is in image and pattern recognition [1]. The building blocks of this technology benefit from the memristor’s non-volatility and gradual change in conductivity. Conductance tuning over a dynamic range of synaptic weights in response to voltage pulses emulates the biological synapse and can be developed into a true bio-inspired device. In this paper a neuromorphic Pt/NbO nanojunction is presented as the building block for an adaptive synaptic array. The junction is composed of an Nb wire annealed to create a thin oxide layer shell, a Pt wire completes the device and facilitates the dynamic resistance response. The properties of a single junction are similar to that of a multilevel resistance memristor [2]. An individual junction displays continuum resistance levels with analogue current response. Initially the conductivity is low in the nA range (Glow) but can be tuned through the application of voltage pulses to achieve a large range of conductivity levels between 10-100 times Glow. The continuum of conductivity states between Glow and Ghigh are accessible through either altering the voltage input amplitude or the pulse timing width. By exploiting this device property there are many potential neuromorphic applications the single junction could enable. Data is presented on the repeatability and robustness of accessing the continuum of resistance levels in a single junction. The memristive relationship between current response and voltage flux pulses is explored for different voltage pulse shapes, timing and amplitude. In theory, equal amount of voltage flux should produce the same current response from a voltage controlled memristor [3]. In the Pt/NbO junctions presented here, equivalent flux from different voltage pulse shapes (square, triangle and sine waveforms, 4 Volts ~ Vamp) are shown to induce similar levels of conductivity change in the device provided the time period is similar (T~0.01-1 s). Shorter timed pulses with higher voltage amplitude (Vamp > 4 V) but with equivalent flux as the other waveforms are observed to induce greater levels of conductivity changes in the device suggesting a nonlinear relationship between conductivity change and the strength of the pulse Vamp. Preliminary data on a potential application of many of these junctions in image processing is presented. In a simple technology application the memristive junctions are incorporated into a small array consisting of many individual Pt/NbO junctions. In the simplest iteration the continuum resistance of each junction is mapped to a greyscale value pinned at Glow. A custom Matlab code is used to interpret and translate images into voltage pulses with the required parameters necessary to reproduce the image within the junction array. Once executed, the Matlab code controls the voltage source and pulsing parameters in a semi-autonomous manor enabling complex neuromorphic behavior when multiple epochs or iterations are performed. Other applications such as edge detection, pixel movement and image recognition will also be discussed. [1] Training and operation of an integrated neuromorphic network based on metal-oxide memristors [2] A Single Nanoscale Junction with Programmable Multilevel Memory [3] Resistance switching memories are memristors

Resume : Introductory remarks

Resume : For over 40 years, eumelanin has been the focus of increasing attention as a promising functional material for organic devices like transistors, sensors, charge storage, memory devices and drug delivery [1]. Eumelanins are an heterogeneous biopolymer with a combination of unique physical-chemical properties such as broad absorbance throughout the UV-visible region, strong non-radiative relaxation of photoexcited electronic states, hybrid ionic-electronic conduction, intrinsic free radical character and biocompatibility [1]. These features inspire it’s use in bioelectronics applications. However, the low solubility of eumelanin is an issue that narrows its full technological potential. Thus, in this study we propose a different synthetic approach from L-DOPA and oxygen pressure to obtain eumelanin soluble derivative. We studied the effect of O2 pressure (4 to 8 atm) in aqueous and organic environment on the chemical structure of the derivative. The polymer obtained was compared with the conventional synthesized melanin and characterize using UV-Vis spectroscopy, FTIR and 13C-NMR. In aqueous medium, it was obtain a water-soluble derivative without any exotic functionalization. It is also possible to observe that at 8 atm of oxygen pressure leads to an increase in the reaction velocity, reducing the time required for the synthesis by a factor of 16 and a higher DHICA/DHI ratio of 50% compared to conventional synthesis [2]. On the other hand, in organic medium (DMSO) we have a DMSO, DMF and NMP soluble material functionalized with sulfone groups. The time of the organic synthesis is reduced by a factor of 14 and 30% higher DHICA/DHI ratio with 8 atm of oxygen pressure compared to the traditional synthesis in DMSO [3]. A reaction mechanism based on higher oxidative nature of the reaction medium including oxygen-induced decomposition of H2O2 is proposed to explain the structural changes observed. Keywords: Eumelanin, solubility, oxygen pressure, synthesis. Acknowledgment: We thank Prof. Marcos Donate (USP) for the reactor system and CAPES, FAPESP and CNPq for financial support. References: [1] M. d’ Ischia, et al.. Pigment Cell Melanoma Res., 2015, 28; 520–544. [2] R.C. Seally, et al.. Radicals in Biology Academic, New York, USA (1980). [3] S. N. Dezidério, et al.. Journal of Non-Crystalline Solids. 2004, 338-340, 634–638.

Resume : The global economic impact of the five leading non-communicable diseases (NCDs) – cardiovascular disease (CVD), chronic respiratory disease, cancer, diabetes and mental ill-health – could total US$ 47 trillion by 2030, according to a study released by the World Economic Forum in 2011. According to the World health organization, the four major NCDs (cardiovascular diseases, cancer, chronic respiratory diseases and diabetes) are responsible for 82% of NCDs deaths which corresponds to 31 million deaths in 2012. All these diseases involve a variation of H2S concentration in blood. Indeed, H2S is the 3rd endogenously generated gaseous signaling compound and it’s involved in the biological systems under physiological and pathological conditions. What about the development of accurate point-of-care testing for H2S blood concentration using fluorescent MOF as sensor? Early diagnostic will allow avoiding millions of deaths every year. We developed a functionalized MOF, via post-synthetic modification, able to specifically respond by a fluorescence turn-on mechanism to physiological (ie healthy person) and pathological (ie sick person) H2S concentration. The presentation will address the synthesis and characterization of MOF bearing functional moieties. Results of the specific reaction leading to an increase of MOF fluorescence when triggered by various H2S concentration will be also presented.

Resume : Conventional silicon-based computing is reaching its limit due to fundamental physical and reliability issues and new approaches have to be explored to go ?beyond Moore?. New paradigms for unconventional computing, e.g. Reservoir Computing (RC) approaches have attracted a large interest from computer science community but only few results have demonstrated and explored the material implementation of such computing networks. We previously studied the photoswitching performances of Azobenzene derivatives/Au NanoParticle Self-Assembled Network (AzBT-NPSAN) nanodevices1. Here, we demonstrate the fabrication of a reservoir composed of AzBT-NPSAN. This network is connected to sub 100 nm multi-electrode devices to stimulate / sense the reservoir dynamics. This non-linear conductance pathways network is also controlled by trans-cis photo-izomerisation of the molecules. We show reconfigurable Boolean logic operations at room temperature, associated to the light controlled switching of the molecules. Moreover, we investigate how complex non-linearity of the device leads to High Harmonics Generation (HHG), one of the prerequisites to RC approaches. We also show that HHG can be modified by UV illumination which paves the way for the processing of multi-input signals through a device that acts as a reservoir computer. 1 Y. VIERO et al. J. Phys. Chem. C, 2015, 119 (36), 21173?21183. Financially supported by EU-FET projects SYMONE and RECORD-IT and by the French ANR project FOST.

Resume : Emerging synapse-like nanoscale devices such as synaptic transistors are of great interest to inspire new circuits and neuromorphic systems [1]. These devices called NOMFET (Nanoparticle Organic Memory Field Effect Transistor), combines a memory and a transistor effect in a single device. In this paper, we present for the first time an N type NOMFET based on a polymeric semiconductor, ActivInk PolyeraTM N2200. Our device exhibit a facilitating and a depressing drain current behavior, with a relative amplitude of about 50% and a dynamic response of about 4s. Studying the charging/discharging dynamics, we modulate the amount of charges trapped in the NPs by applying a train of pulses separated by different time intervals ?t, in order to investigate the response of the synapse as a function of the pre- and post- synaptic neurons spiking activity. We demonstrate a typical anti-STDP learning function, one of the fundamental mechanisms of the unsupervised learning in biological neural networks [2]. Combined with previous p-type NOMFET, our device can be a starting point to realize complementary circuits which bring us the capability to tune the STDP learning window. [1] F. Alibart et al. Adv. Funct. Mater., 22, 3, 609?616, 2012. [2] P. D. Roberts et al., Front. Comput. Neurosci., 4, 1?11, 2010.

Resume : The optimization of the geometric and electronic structure of polyacetylene with the different substituents of donor and acceptor functional groups was conducted with the basis set of 3-21G and 6-31 G**(d,p). The rates of these isomerizations and Arrhenius geometric parameters were determined. In the case of unsubstituted polyacetylene as a reference, the values of the equilibrium constant for the isomerization reaction were also determined at various temperatures between 300 and 500 K as well the value of the change in energy. The results show that the energy of the isomerization reaction are positive and the values of the gap are very sensitive to the effect of the substituent.

Resume : Gallium oxide nanomaterials have been fabricated via carbothermal reduction using a chemical vapor deposition (CVD) method and employing gallium oxide nanopowder as source material. The growth has been performed under argon gas stream. With few variations of the experimental conditions but depending on the distance from source to the substrate, different nanostructures and shapes have been obtained. Among them, nanowires have been synthesised, which tend to grow either in a 2D web distribution or vertically, and nanoplateletes. Nanowires have been structurally and optically characterised using X-ray diffraction, scanning and transmission electron microscopy, as well as photoluminescence and Raman spectroscopy, confirming their monocrystalline nature. To study the gas sensing properties, the nanowires were removed from the substrates applying sonication, followed by the deposition on suspended microhotplates with prepatterned electrodes and individual nanomaterials were contacted by combined Focused Electron- and Focused Ion-Beam assisted deposition techniques. The devices have been tested towards different gases relevant in air quality monitoring, like NO2 and CO, at different concentrations and temperatures. The sensing properties will be reported and the relation to the structural and chemical properties of the nanomaterials will be established.

Resume : Scientific interest in designing and making low dimensional electronic devices with traditional or novel materials is increasing since last two decades. Among the new materials, segments of DNA chains are highlighted. In fact, electronic properties of DNA strands have been widely study. However, the experimental results have contradictions that pointing to DNA has behaviors as conductor, semiconductor or insulator. The purpose of this contribution is to spread some light on this surprising problem, in spite that we use a theoretical toy model to study this apparent contradiction. We perform numerical calculations of electrical conductance on DNA segments. DNA is modeled as 1D disordered finite chains and described into a Tight binding model with nearest neighbor interactions and a s orbital per site. The inherent disorder of DNA is modelled as a random binary alloy with short-range correlations. Also, hydration effects are included as random variations of on-site self-energies. The electronic current as a function of applied bias is calculated using Launder formalism, where the transmission probability is determined into the transfer matrix formalism. We find a conductor-to-semiconductor-to-insulator transition as a function of the three effects taken into account: chain size, intrinsic disorder, and hydration. Authors like to thank to Universidad Nacional de Colombia for their financial support.

Resume : Neural processing in the brain is an extremely complex and yet efficient process, which, the powerful modern computers still struggle to capture. The reason for this limitation of modern computers lies in the heart of their architecture: static units are dedicated either to information processing or storing. This blueprint is determined during the design of a system and only predefined computational tasks are managed by such a static system. This strict definition of tasks, equips these systems with powerful computational capabilities, which although lack of fault tolerance. Moreover, signal processing in general purpose computing systems is limited in utilizing electrical signals of microelectronic devices, and the lack of interaction with biological substances such as living organisms is evident in these systems. On the other hand, a biologic system such as the brain offers a more generic address of information processing: spatiotemporal signals (electrical, ionic, chemical, optical and mechanical) from various sensory systems are processed in a microcircuitry environment that evolves structurally and functionally over time, resulting in memory and learning. Furthermore, the computational units of the brain are not strictly dedicated on specific tasks, in a way that processing and storing is merged on the same micro-regions of the brain. The human brain consists of approximately 1011 neurons communicating to each other in an interwoven network of almost 1014 synapses. Neurons process and transmit information and are regarded as the processing units of the brain, while are interconnected to each other through synapses (i.e., the memory units) in a complex network and their connection efficiency (synaptic plasticity) can be modified in a range of timescales. Synaptic plasticity can be broadly divided into two categories: short- (msec - min) and long-term (> min, days or even longer). Long-term plasticity underlies memory and learning, while short-term plasticity assists various computational tasks of the brain such as transmission, encoding and filtering of neural signaling. This complex network attributes unique capabilities to the brain, such as fault tolerance, massive parallelism and energy efficiency. The realization of neuroinspired circuits, able to emulate the processing capabilities of the brain, demands the development of neuromorphic devices that will be able to mimic the various synaptic plasticity functions of the brain. Recently, several functions of the neural circuitry, such as short- and long-term plasticity, Spike-Time-Dependent-Plasticity, short to long term memory transition and dynamic filtering were implemented in a single device level. Over the past decades, organic electronic materials are attracting increasing interest within the semiconductor industry due to a unique properties, such as compatibility with low-cost fabrication processes and large-area, mechanically flexible substrates, tunability of their electronic properties via chemical synthesis, and biocompatibility in some cases. Recently, organic devices have also entered the realm of neuromorphic devices and various synaptic plasticity functions were also implemented with two and three terminal devices. An organic-based device of particular interest especially in the field of bioelectronics is the organic electrochemical transistor (OECT). Regarding the interaction with biological substances, OECTs offer several interesting features such as operation in liquid electrolyte environment, ionic-to-electronic transduction, amplification and potential biocompatibility. In this work, we will present our recent findings in neuromorphic devices based on the OECT platform. The device channel consist on derivatives of poly(3,4-ethylenedioxythiophene) (PEDOT) such as poly(3,4-ethylenedioxythiophene): poly(styrene sulfonate) (PEDOT:PSS) and poly(3,4-ethylenedioxythiophene): poly(tetrahydrofuran) (PEDOT:PTHF). The volatile phenomena of PEDOT:PSS based OECTs are used for the implementation of short-term synaptic plasticity functions (e.g., Pair-Pulsed-Depression, adaptation and dynamic filtering), while the non-volatile memory phenomena of PEDOT:PTHF based OECTs, are used for the implementation of long-term plasticity functions (e.g., synaptic integration, short- to long-term memory transition and merged information processing and storage behavior). The results presented here pave the way for future, organic-based, neuroinspired circuits.

Resume : Microfluidic techniques have increased potential and broadened their application in the field of micro-reactor engineering, particulate systems formation to very challenging and complex lab-on-a-chip apparatus. Moreover, the fabrication techniques used in construction of microfluidic device enable to encode hard mathematical problems into networks consisting of micro-channels in form of nodes and junctions. The use of biological agent such as bacteria, fungi or protein motors has a great potential in solving such mathematical problems in highly parallel manner, and thus substitute conventional computing. In this work, bacterial strains of motile Pseudomonas putida, Magnetococcus marinus and Vibrio fischeri were tested in terms of their motility behavior and prioritization inside microfluidic channel networks. A pre-requisite parametric study of optimal channel design for each strain of studied bacteria were obtained and selected application of bio-simulation and bio-computation microfluidics chips were introduced.

Resume : Neurodegenerative diseases are consider to be a significant challenge of 21st century, due to the difficulty of the nervous system to be restored after s severe damage. Recently, application of Nanomedine as a valuable tool for the detection and treatment of Neurodegenerative diseases used in order to improve functional neuroprotective agents and enhance restoration of affected neural tissues. In this study we proceed towards the fabrication of a Polyvinyl Alcohol (PVA) /Poly (ε-caprolactone) (PCL) scaffold by Dual Electrospinning System, further treated with Oxygen Plasma and explored its application as nerve guide substrate and regenerative agent in vitro. The surface morphology, hydrophilicity and chemical composition were evaluated by Scanning Electron Microscopy (SEM), Contact Angle measurement and XPS respectively. SEM revealed the rough surfaces of plasma-treated scaffolds and the dual physicochemical behavior, combining both hydrophobic and hydrophilic properties. Contact angle on plasma-treated scaffolds appeared smaller than on untreated and the O/C atomic ratio was also found to be increased. Neural cell line was cultivated onto the engineered matrices and in vitro studies indicated that nano-fibrous scaffolds promoted neural cell adhesion, elongation and proliferation. Oxygen plasma treatment of the scaffolds further promoted cell differentiation with the presence of neuritis endings and the formation of a neural network. Due to its advantage of high surface area for cell attachment, it is believed that this electrospun nerve scaffold could find further application in cell therapy for nerve regeneration in order to improve functional regeneration outcome, especially for longer nerve defect restoration.

Resume : All oxide heterostructures composed with La0.9Hf0.1MnO3(LHMO) and ZnO thin layers have been integrated on (0001) sapphire substrates using pulsed laser deposition. The LHMO is known as electron doped manganite since the Hf dopants have unique 4 valence. X-ray diffraction shows that these thin films are of single phase and highly epitaxial. Similar to those heterojunctions composed with La0.9Hf0.1MnO3 and other manganites, the La0.9Hf0.1MnO3 /ZnO n-n junctions also exhibit excellent rectifying behaviors over the temperature range 60K-320K. Moreover, photoconductivity of the junctions appears to be significant under 240K. The temperature dependence of photoconductivity is also investigated. The open circuit voltage Voc is ~ 0.12V at 240K. When the temperature is lowered, it increases continuously.These phenomena can be attributed to the change of band structure due to lattice deformation.

Resume : La0.9Hf0.1MnO3 (LHMO) films was grown on 0.05%Nb-doped SrTiO3 (STON) substrates through pulsed laser deposition (PLD). X-ray diffraction measurements demonstrated that our samples were of good epitaxy and single-crystal. The temperature dependent resistance has been investigated. It was observed that the as-grown film of LHMO exhibited a typical paramagnetic-ferromagnetic transition. Similar to the previously reported LHMO/0.7% STON heterojunction, LHMO/0.05% STON heterojunction exhibited asymmetric current-voltage characteristics and a remarkable temperature-dependent rectifying behavior in a wide temperature range (from 40 to 300 K).While LHMO/ 0.05%STON heterojunction exhibited an abnormal behavior that a lower leakage current has been found in high temperature region. This behavior is different from our previous report on LHMO/ 0.7%STON heterojunction. Another significant feature on this heterojunction is under the light illumination with wavelength of 650nm and 532nm, the I-V curve of LHMO/0.05%STON heterojunction moved downward, and a photocurrent was observed. Such features may be applicable in functional device. The rectifying characteristics and photoelectric properties of these heterojunctions may be attributed on band structure of LHMO/0.05%STON heterojunction.

Resume : Resistance random access memory (ReRAM) is an attractive non-volatile memory, which has simple structure consisting of two electrodes and insulating layers between them. Recently, a synapse device for neuromorphic computing using multilevel ReRAM was proposed, in which ReRAM was used as an analog-resistance memory. In this work we show methods to control the resistance of MoOx ReRAM. We fabricated 1T1R-type ReRAMs of which 1T is an n-channel MOSFET. The stacked structure of MoOx ReRAM consisting of an aluminum bottom electrode, an AlOx layer, a MoOx layer (50nm), a Cu layer as an active electrode (30nm) and a Pt top electrode (100nm) was formed on the drain terminal of the MOSFET. The ReRAM device showed bipolar switching characteristics. By the use of the SET compliance current controlled by the gate voltage of the nMOSFET, we succeeded in controlling the low-resistance state (LRS) almost corresponding to the gate voltage. As a result, the resistance in the LRS decreased monotonically as the compliance current or gate voltage increased. This result is attributed to formation of thicker filament relying on the increased compliance current. However, LRS of the ReRAM shows some variability. So, we tried to apply a verify scheme and demonstrated that the verification makes the resistance close to the target value.

Resume : In this study, we used quantum chemistry calculations in order to determine some kinetic parameters of the isomerization reaction of the substituted dodecahexane. The studied molecules are: (C12H14, C12H7F7, C12H7Cl7,C12H7Br7 et C12H7I7) Cis and Trans. One of the adopted ways to access these parameters (activation energy, rate constant, etc ...) is looking for the transition state that is based on the exploration of intermediaries during the passage of Cis-Trans isomerization process. The study of a ten molecules series gives the following results:  The Trans conformer is more stable than the Cis.  The activation energy changes very greatly depending on the size and nature of the substituent according to the reaction profile.  The constants of the isomerization reaction rates are in the following order: kC12H14 >> kC12H7F7>> kC12H7Cl7>>kC12H7Br7>> kC12H7I7.  The geometrical parameters vary considerably according to intermediate products The calculation methods are DFT and Ab-initio methods at STO-3G *.

Resume : The results obtained through the optimization of molecules gave us the different distances and angles according to the methods and bases applied with a C2v symmetry. We were able to determine the total energies, the energy gap ΔE (HOMO-LUMO) of the four conformers trans-transoid; trans-cisoid; cis-transoid; cis-cisoid. (the semi empirical AM1+PM6 at 6-31G and 3-21G** levels) and finally a comprehensive analysis on the topological charges. The analysis of the results show that for the eight molecules, the trans-transoid conformer is energetically very stable compared to the cis-cisoid one, this stability is confirmed by the obtained values for the total energy. The increase in the stability energy leads to a less important Homo-Lumo energy gap. The analysis of the optimized geometrical parameters of the ten molecules using the AM1 and PM6 methods, are in agreement with the experimental structures characterized by X-ray diffraction. Finally, we were able to determine the reaction profiles of the Cis-Trans isomerization reactions of the decapenta-ene in the gas phase, and to calculate the activation energy (Ea), as well as the diagrams of energies E (eV) based on the coordinates of the isomerization reaction of its molecules, and molar absorption according to energy calculated by the method HF at 6-31G and 3-21G** levels. Thus we have to determine the Spectrum IR of all the molecules by (AM1) and PM6. Keywords: substituted decapenta-ene , semi empirical, HF (AM1+PM6),

Resume : In recent years, the nanoscale multigate MOSFETs have attracted more attentions in nanoelectronics technology due to their high electrical performance for advanced integration circuits in nanoscale domain. However, the accurate modeling of these devices is still an important challenge due to the complex quantum behavior that describes the transport mechanism. In this context, several approaches have been proposed to investigate the nanoscale Double Gate (DG) MOSFET using numerical and analytical approaches. But from the circuit simulation point of view even numerical modeling is an overkill approach in terms of complexity and computational cost. In addition, it is difficult to obtain closed and compact analytical models for nanotransistors due to the approximations taken during the models development. Therefore, modeling tools which can be applied to design nanoscale devices require new modeling approaches taking into account the model accuracy and computation efficiency. The aim of this paper is to investigate the efficiency of a new approach, built upon Kriging metamodeling and non-dominated sorting genetic algorithm II, for the optimal design in terms of RF and analog performances including the hot carrier effects. Data generated according to computer experiments, based on Atlas 2-D simulator, are used to identify and fit Kriging surrogate models. It is emphasized that the obtained models can be used accurately in a multi-objective context to offer several Pareto optimal configurations. Therefore, a wide range of selection possibilities is available to the designer depending on situations under consideration.

Resume : We prepared TeO2 branched SnO2 nanowires, which were obtained from the 2-step vapor-liquid-solid method. The changes in morphologies, microstructures, and compositions of the resulting TeO2 branched SnO2 nanowire were characterized by using scanning electron microscopy, X-ray diffraction, and transmission electron microscopy. The SnO2 nanowires with TeO2 branches provide a significantly enhanced gas detection, in comparison to undecorated SnO2. The attachment of p-type TeO2 branches on conventionally-fabricated n-type SnO2 stem nanowires is expected to bring about useful and economical composite nanostructures. The present study on composite-nanostructures will be a significant contribution to both the academic field and industrial applications, which will be useful in exploring new areas of multiple-component nanosystems.

Resume : The rapidly growing field of brain computer interface (BCI) has recently benefited from a significant coordination effort through the "BNCI Horizon 2020" project funded by the European Commission, which identified roadmaps for the development of future BCIs clustered around some well defined use cases: a BCI can replace some central nervous system (CNS) outputs as a result of injury or disease through the control of artefacts such as wheelchairs or prosthetic limbs, restore lost CNS outputs stimulating body parts such as muscles in paralysed persons, enhance natural CNS output like attention monitoring systems, improve natural CNS output such as in rehabilitation. The number of applications that in the future will benefit from BCIs is significant and broad, covering rehabilitation, medical and therapeutic applications, but also entertainment and human machine interaction.  Aspects like real-time performance, automatic adaptation to multiple users or to the variability over time of the same user, ease of train and cognitive effort needed to be operated, will deeply impact BCIs adoption and diffusion. A real interdisciplinary approach touching hardware, software, computational neuroscience appears to be essential to fulfill all these requirements: the aim of this review is to give the reader an update on the most recent developments in the rapidly evolving area at the intersection of BCI, machine learning and hardware implementation of neuromorphic computational paradigms.

Resume : This study explores the limits of the capability of ordered, micron-sized patterns, which alternate hydrophobic and hydrophilic areas, to confine water down to attolitre volumes on practically flat surfaces. The contact angle, volume, and geometry of the confined droplets have been precisely determined by Atomic Force Microscopy and, separately in silico, by phenomenological simulations (Surface Evolver). The combination of AFM topographical experiments and equilibrium droplet shape simulation was used to describe non-trivial wetting phenomena at micro/nano scale for rectangular structures of various wettability. Both experimental and simulation results demonstrate that the rectangular footprint of micro/nano-structured surfaces causes significant pinning of the nano-droplets to the rim corners. These findings can be used for development of the purposeful design of surface-addressable hydrophobicity arrays employed in digital microfluidics and high throughput screening nano-arrays.

Resume : The bio-interaction between human beings and machines has been one of the most fascinating challenges in the last decade. Several complications in terms of biocompatibility, type of chemical signal and the scalability of the electrical devices are still present. Within this context, memristor devices represent possible candidates in this direction, thanks to their particular electrical behavior similar to the neuronal response, their expectable biocompatibility and their nanoscale dimensions. In this work, a vacuum-based deposition technique, namely Pulsed Microplasma Cluster Source (PMCS) is proposed and exploited to synthesize TiO2-based materials. PMCS is based on a supersonic beam seeded by clusters of metal oxide and it is used to grow nanocrystalline TiO2 thin films at room temperature, controlling the oxide stoichiometry from metallic titanium to a significant oxygen excess. Morphological and chemical analyses of films produced with this technique are shown, together with relevant examples of metal-insulator-metal structures showing a pinched hysteresis loop in their current-voltage characteristics. Moreover, an important study on the biocompatibility and neurocompatibility of PMCS grown TiO2 having different stoichiometries is presented, as well as the approaches to improve the quality of the adhesion/proliferation of several cell lines growing on the nanostructured metal oxides. This study is the first fundamental stage for the realization of novel memristor-neuron interfacial based devices. This work is developed in the framework of MaDEleNA project (Provincia Autonoma Trento, Grandi Progetti 2012).

Resume : The organic memristor is an important element of a new generation of computational systems and neuromorphic networks considered as a synthetic analogue of the synapses of the neurons of living beings due to ability of changing its resistance depending on charge that has passed through it. The main component of the device – active layer – consists of thin-film polymeric materials or polymer-based heterostructures (such as polyaniline or polyaniline-polystyrene sulfonate) in contact with the solid electrolyte (lithium doped polyethylene oxide). Working principle of the organic memristor is based on the ability of such active layer to change its conductivity between the conducting and insulating states as a result of applied voltage. Since reversible redox reactions involving lithium ions are responsible for such processes, the response speed is one of the most important properties, defining usability of the device. Lithium ions diffusion speed therefore is the main value, characterizing working properties of the organic memristor. In the present work we have investigated theoretically lithium ions diffusion in polymeric materials, which are components of the memristor, and have compared the results with the pulsed field gradient NMR experiments. Moreover, we have explored transformations of the active layer between conducting and insulating states by means of IR spectroscopy and have compared the IR spectra with the theoretical calculations.

Resume : Introductory remarks

Resume : Which features make the brain such a powerful and energy-efficient computing machine? Can we reproduce them in the solid state, and if so, what type of computing paradigm would we obtain? I will show that a machine that uses memory to both process and store information, like our brain, and is endowed with intrinsic parallelism and information overhead - namely takes advantage, via its collective state, of the network topology related to the problem - has a computational power far beyond our standard digital computers [1]. We have named this novel computing paradigm “memcomputing” [2]. As an example, I will show the polynomial-time solution of prime factorization and the NP-hard version of the subset-sum problem using polynomial resources and self-organizable logic gates, namely gates that self-organize to satisfy their logical proposition [3]. These are examples of scalable digital memcomputing machines that can be easily realized with available nanotechnology components. [1] F. L. Traversa and M. Di Ventra, Universal Memcomputing Machines, IEEE Transactions on Neural Networks and Learning Systems, 26, 2702 (2015). [2] M. Di Ventra and Y.V. Pershin, Computing: the Parallel Approach, Nature Physics, 9, 200 (2013). [3] F. L. Traversa and M. Di Ventra, arXiv:1512.05064

Resume : Complementary resistive switch [1] is formed by two bipolar memristors connected anti-serially. Adding a third (output) terminal to a complementary resistive switch results in a device termed complementary resistive gate [3]. Two distinct uses have been proposed for the gate. First, it may be used for analog fuzzy computations (MIN and MAX gates) [2], when applied signals typically significantly exceed memristor’s switching threshold. Secondly, the gate was proposed in a communication scenario [3], where applied signals are below the threshold. In our contribution we propose a third application that assumes characteristics of both approaches. Much like in the second application, most of the time the gate will serve as a communication device delivering signals/energy to the output terminal. However, depending on relative timing of signals, the states of constituent memristors will change, like they do in fuzzy computation. We call this phenomenon coincidence adaptation. This behavior has parallels in neural circuits and may serve as a basis of passive pattern recognition subsystems complementing spiking neural networks. Our results are based on simulations done with ECM based model and experiments with ECM and metal oxide nanostructures. Finally, we review desirable memristor characteristics supporting the proposed application of the gates. [1] E.Linn, R. Rosezin, C. Kugeler, R. Waser, Complementary resistive switches for passive nanocrossbar memories, Nature Materials, 2010 [2] M. Klimo, O. Such, Memristors can implement fuzzy logic, arXiv:1110.2074 [3] O. Such, E.Linn, M. Klimo, P. Jancovic, M. Fratrik, K. Frohlich, On passive permutation circuits, IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 2015

Resume : The structure, energetics and transport properties of TiO2 anatase with different densities of oxygen vacancies are studied by computer simulations using a variety of techniques ranging from first-principles to Monte Carlo methods. This investigation is motivated by the recent development of memristive electronic devices, usually made of metal oxides in which arrays of defects control the resistance switching mechanisms. Anatase, in particular, emerged as one of the most promising candidate for memristor design. However, the microscopic behavior of these multi-vacancy systems is not yet entirely understood. This work investigates electronic and transport properties of TiO2 anatase structures containing from one to several oxygen vacancies both in neutral and charged states by adding an Hubbard-like term (U) to the the Generalized Gradient approximation to the electron density (GGA+U). Calculated observables are the formation energy of oxygen defects, vacancy cohesion energies and Local Density of States for different vacancy distributions. Furthermore, to demonstrate the correspondence between energetically stable states and the conductive phase of memristors, electron transport is studied first via the integrated local density of states (ILDOS), and then within a tight-binding approximation to density functional theory. Finally, a Kinetic Monte Carlo model of the conductive channel formation in bulk anatase, based on the nudged-elastic band (NEB) calculations of the vacancy diffusion rates, is reported.

Resume : Despite the significant number of works devoted to creating practically effective methods for spiking neuron networks learning, still none has been developed based only on the current knowledge of biological neural systems operating rules, namely, on STDP, a biologically inspired long-term plasticity model. From this point of view we consider different spiking neuron models (Leaky Integrate-and-Fire, Izhikevich, Hodgkin-Huxley), STDP forms (all-to-all and several nearest-neighbour spike pairing rules), ways of input coding, unsupervised and supervised learning implementations. With the help of spiking neurons simulator the influence of the amount of correlation between input and output spike trains on the learnability by different STDP rules is examined. Combined learning schemes, involving artificial and spiking neuron models, are discussed. We analyze a few examples of these learning mechanisms implementations and demonstrate how the neural network can recognize patterns of neural activities in dependence of their mutual correlation.

Resume : For many tasks, such as visual recognition or time series prediction, the brain functions in a much faster way and with a drastically lower power consumption than any classical computers. In the context of big data, an alternative approach to classical computers is to build hardware neural networks to process the information. Reservoir computing [1] is a recently introduced type of neural network, which was already implemented in optics. These setups use the transient dynamics of non-linear oscillators to process an input. Reservoir computing has demonstrated its high efficiency dealing with tasks such as spoken digit recognition or financial forecasting. Our recent research interest is focused on using non-linear magnetic microwave oscillators [2] for Reservoir computing. In comparison with optical implementations, magnetic microwave oscillators present the advantage of a nanometric size, highly tuneable non-linearity and compatibility with CMOS technology. In this talk, we will show that we obtain very promising classification results recognising unambiguously different patterns in an input signal. Our work opens the path to building dense and compact bio-inspired networks of nano-oscillators leveraging non-linear dynamics for pattern recognition. [1] L. Appeltant et al. Information processing using a single dynamical node as complex system. Nat. Commun. 2, 468 (2011) [2] N. Locatelli et al. Spin-Torque building blocks. Nat mat 13, 11-20 (2014)

Resume : Outlook: perspectives and initiatives on "Adaptive materials: devices and systems towards unconventional computing, sensing, bio-electronics and robotics"