A5010173029- Advanced Memory and Neural Computing
- Ferroelectric and Negative Capacitance Devices
- Neural dynamics and brain function
- Neural Networks and Reservoir Computing
- Neural Networks and Applications
- Neuroscience and Neural Engineering
- EEG and Brain-Computer Interfaces
- Conducting polymers and applications
- Analytical Chemistry and Sensors
- Context-Aware Activity Recognition Systems
- graph theory and CDMA systems
- Brain Tumor Detection and Classification
- Electrochemical sensors and biosensors
- Network Time Synchronization Technologies
- Functional Brain Connectivity Studies
- Advanced Sensor and Energy Harvesting Materials
- Bluetooth and Wireless Communication Technologies
- Electronic and Structural Properties of Oxides
- CCD and CMOS Imaging Sensors
- Transition Metal Oxide Nanomaterials
Polytechnic University of Turin
2018-2025
École Polytechnique Fédérale de Lausanne
2020
Spiking neural networks and neuromorphic hardware platforms that simulate neuronal dynamics are getting wide attention being applied to many relevant problems using Machine Learning. Despite a well-established mathematical foundation for dynamics, there exists numerous software solutions stacks whose variability makes it difficult reproduce findings. Here, we establish common reference frame computations in digital systems, titled Neuromorphic Intermediate Representation (NIR). NIR defines...
Spatio-temporal pattern recognition is a fundamental ability of the brain which required for numerous real-world activities. Recent deep learning approaches have reached outstanding accuracies in such tasks, but their implementation on conventional embedded solutions still very computationally and energy expensive. Tactile sensing robotic applications representative example where real-time processing efficiency are required. Following brain-inspired computing approach, we propose new...
Spiking Neural Networks (SNNs), known for their potential to enable low energy consumption and computational cost, can bring significant advantages the realm of embedded machine learning edge applications. However, input coming from standard digital sensors must be encoded into spike trains before it elaborated with neuromorphic computing technologies. We present here a detailed comparison available encoding techniques translation time-varying signals event-based signal domain, tested on two...
Hyperparameter optimization (HPO) is of paramount importance in the development high-performance, specialized artificial intelligence (AI) models, ranging from well-established machine learning (ML) solutions to deep (DL) domain and field spiking neural networks (SNNs). The latter introduce further complexity due neuronal computational units their additional hyperparameters, whose inadequate setting can dramatically impact final model performance. At cost possible reduced generalization...
Abstract Human activity recognition (HAR) is a classification problem involving time-dependent signals produced by body monitoring, and its application domain covers all the aspects of human life, from healthcare to sport, safety smart environments. As such, it naturally well suited for on-edge deployment personalized point-of-care analyses or other tailored services user. However, typical wearable devices suffer relevant limitations regarding energy consumption, this significantly hinders...
Resistive switching (RS) devices based on self-assembled nanowires (NWs) and nanorods (NRs) represent a fascinating alternative to conventional with thin film structure. The high surface-to-volume ratio may indeed provide the possibility of modulating their functionalities through surface effects. However, NWs usually suffer from low resistive performances in terms operating voltages, endurance retention capabilities. In this work, we report behaviour ZnO NW arrays, grown by hydrothermal...
Resistive switching (RS) devices are considered as the most promising alternative to conventional random access memories. They interestingly offer effective properties in terms of device scalability, low power-consumption, fast read/write operations, high endurance and state retention. Moreover, neuromorphic circuits synapse-like envisaged with RS modeled memristors, opening route toward beyond-Von Neumann computing architectures intelligent systems. This work investigates how zinc oxide...
Neuromorphic computing relies on event-based, energy-efficient communication, inherently implying the need for conversion between real-valued (sensory) data and binary, sparse spiking representation. This is usually accomplished by using as current input to a neuron model tuning neuron's parameters match desired – often biologically inspired behavior. To support investigation of models parameter combinations identify suitable configurations neuron-based encoding sample-based into spike...
In-liquid biosensing is the new frontier of health and environment monitoring. A growing number analytes biomarkers interest correlated to different diseases have been found, miniaturized devices belonging class biosensors represent an accurate cost-effective solution obtaining their recognition. In this study, we investigate effect solvent substrate modification on thin films organic semiconductor Poly(3-hexylthiophene) (P3HT) in order improve stability electrical properties Electrolyte...
Spiking neural networks and neuromorphic hardware platforms that emulate dynamics are slowly gaining momentum entering main-stream usage. Despite a well-established mathematical foundation for dynamics, the implementation details vary greatly across different platforms. Correspondingly, there plethora of software implementations with their own unique technology stacks. Consequently, systems typically diverge from expected computational model, which challenges reproducibility reliability...
Neuromorphic models take inspiration from the human brain by adopting bio-plausible neuron to build alternatives traditional Machine Learning (ML) and Deep (DL) solutions. The scarce availability of dedicated hardware able actualize emulation brain-inspired computation, which is otherwise only simulated, yet still hinders wide adoption neuromorphic computing for edge devices embedded systems. With this premise, we adopt perspective conventional present L2MU, a natively Legendre Memory Unit...
Neuromorphic computing has proved to be capable of remarkable gains in energy efficiency over traditional architectures. With the continuous development new software and hardware tools, benchmarking plays a crucial role measurement technological advancement field. Over last few years, some efforts for neuromorphic have been proposed focusing on specific tasks like noise suppression or gesture recognition, while micro-benchmarking approaches not widely investigated. In this paper, we present...
In-liquid biosensing is the new frontier of cells real time monitoring and biomarkers detection. In order to improve stability electrical properties an Electrolyte Gated Organic Field Effect Transistor (EGOFET) biosensor, in this study we investigate effect solvent substrate modification on thin films organic semiconductor Poly(3-hexylthiophene) (P3HT). The studied surface relevant interface between P3HT electrolyte acting as gate dielectric for in-liquid detection analyte. AFM XPS...
Resistive switching (RS) devices, also referred to as Random Access Memories (ReRAMs), rely on a working principle based the change of electrical resistance following proper external stimuli. Since demonstration first resistive memory binary transition metal oxide (TMO) enclosed in Metal-Insulator-Metal (MIM) structure, this class devices has been considered key player for simple and low-cost memories. However, successful large-scale integration with standard CMOS technologies still needs...
Neuromorphic computing relies on spike-based, energy-efficient communication, inherently implying the need for conversion between real-valued (sensory) data and binary, sparse spiking representation. This is usually accomplished using real valued as current input to a neuron model, tuning neuron's parameters match desired, often biologically inspired behaviour. We developed tool, WaLiN-GUI, that supports investigation of models parameter combinations identify suitable configurations...
Although initially conceived as a tool to empower neuroscientific research by emulating and simulating the human brain, Spiking Neural Networks (SNNs), also known third generation neural networks, are gaining popularity for their low-power sparse data processing capabilities. These attributes valuable power-constrained edge Internet of Things (IoT) applications. Several open-source FPGA ASIC neuromorphic processors have been developed explore this field, although they often require...
Spatio-temporal pattern recognition is a fundamental ability of the brain which required for numerous real-world activities. Recent deep learning approaches have reached outstanding accuracies in such tasks, but their implementation on conventional embedded solutions still very computationally and energy expensive. Tactile sensing robotic applications representative example where real-time processing efficiency are required. Following brain-inspired computing approach, we propose new...