Periodic Reporting for period 1 - SMART (Synchronised neuro-Memristive Architecture for Reinforced learning Technology)
Período documentado: 2023-09-01 hasta 2025-11-30
Artificial intelligence (AI) is becoming an essential part of everyday life, from smart sensors and mobile devices to autonomous vehicles and robotics. However, the computer hardware used today struggles to meet the growing demand for fast learning and decision-making while keeping energy consumption low. Current computers process and store information in separate units, which makes them inefficient for tasks that require rapid adaptation and real-time responses.
Inspired by the human brain, neuromorphic computing aims to overcome these limitations by developing new types of hardware that can process information and learn in a more brain-like and energy-efficient way. Instead of relying on complex software running on conventional processors, neuromorphic systems seek to embed learning and adaptability directly into the physical hardware. Despite its promise, this field still faces major challenges, particularly in identifying materials and device designs that can naturally reproduce the dynamic behaviour of biological neurons.
The SMART project addressed this challenge by exploring a new approach to brain-inspired hardware based on an emerging functional material, TbMnO₃. This material can exhibit a special electrical behaviour that enables devices to generate self-sustained electrical oscillations, similar to the firing activity of neurons in the brain. The overall objective of the project was to understand and exploit these properties to develop compact, low-power artificial neuron elements capable of synchronising with each other and supporting learning processes.
Specifically, the project aimed to investigate how these material properties can be controlled, how oscillatory behaviour can be stabilised and synchronised, and how such devices could serve as building blocks for future neuromorphic systems. By linking fundamental material research to practical computing concepts, SMART contributes to the development of more efficient and adaptive hardware for next-generation AI
The results of the SMART project contribute to long-term efforts to develop computing technologies that are faster, more energy-efficient and better suited to intelligent applications operating close to where data are generated, such as sensors and edge devices. By demonstrating how learning-related functions can be embedded directly into hardware, the project opens new pathways toward low-power AI systems with reduced environmental impact.
At a broader level, the project supports European priorities in digital innovation and advanced computing by strengthening knowledge in emerging materials and brain-inspired technologies. The findings are relevant for future applications in autonomous systems, smart electronics and intelligent sensing, and they contribute to Europe’s capacity to develop sustainable and competitive AI hardware technologies.
Inspired by the human brain, neuromorphic computing aims to overcome these limitations by developing new types of hardware that can process information and learn in a more brain-like and energy-efficient way. Instead of relying on complex software running on conventional processors, neuromorphic systems seek to embed learning and adaptability directly into the physical hardware. Despite its promise, this field still faces major challenges, particularly in identifying materials and device designs that can naturally reproduce the dynamic behaviour of biological neurons.
The SMART project addressed this challenge by exploring a new approach to brain-inspired hardware based on an emerging functional material, TbMnO₃. This material can exhibit a special electrical behaviour that enables devices to generate self-sustained electrical oscillations, similar to the firing activity of neurons in the brain. The overall objective of the project was to understand and exploit these properties to develop compact, low-power artificial neuron elements capable of synchronising with each other and supporting learning processes.
Specifically, the project aimed to investigate how these material properties can be controlled, how oscillatory behaviour can be stabilised and synchronised, and how such devices could serve as building blocks for future neuromorphic systems. By linking fundamental material research to practical computing concepts, SMART contributes to the development of more efficient and adaptive hardware for next-generation AI
The results of the SMART project contribute to long-term efforts to develop computing technologies that are faster, more energy-efficient and better suited to intelligent applications operating close to where data are generated, such as sensors and edge devices. By demonstrating how learning-related functions can be embedded directly into hardware, the project opens new pathways toward low-power AI systems with reduced environmental impact.
At a broader level, the project supports European priorities in digital innovation and advanced computing by strengthening knowledge in emerging materials and brain-inspired technologies. The findings are relevant for future applications in autonomous systems, smart electronics and intelligent sensing, and they contribute to Europe’s capacity to develop sustainable and competitive AI hardware technologies.
This project investigated the electronic properties of oxide thin-film devices based on TbMnO₃ and Nb-doped SrTiO₃, with a focus on understanding negative differential resistance (NDR). The work combined thin-film fabrication, device processing, electrical measurements, and theoretical analysis to study how material properties and device structure influence nonlinear electronic behaviour.
An experimental workflow for oxide thin-film device fabrication and characterisation was established. TbMnO₃ thin films were deposited using pulsed laser deposition, followed by structural and morphological characterisation with standard materials analysis techniques. Micro-scale devices were fabricated using cleanroom lithography and metallisation, and their electrical behaviour was systematically measured using semiconductor parameter analysis. Measurements on reference devices provided by the host group were used to benchmark and interpret the observed electrical response.
In parallel, a mathematical model was developed to analyse the role of Joule heating and heat dissipation in NDR behaviour. The model is based on established physical principles and was adapted to the specific material system studied in the project. Sensitivity analyses were performed using material parameters from the literature, allowing qualitative insight into how thermal and geometrical factors influence device characteristics.
An experimental workflow for oxide thin-film device fabrication and characterisation was established. TbMnO₃ thin films were deposited using pulsed laser deposition, followed by structural and morphological characterisation with standard materials analysis techniques. Micro-scale devices were fabricated using cleanroom lithography and metallisation, and their electrical behaviour was systematically measured using semiconductor parameter analysis. Measurements on reference devices provided by the host group were used to benchmark and interpret the observed electrical response.
In parallel, a mathematical model was developed to analyse the role of Joule heating and heat dissipation in NDR behaviour. The model is based on established physical principles and was adapted to the specific material system studied in the project. Sensitivity analyses were performed using material parameters from the literature, allowing qualitative insight into how thermal and geometrical factors influence device characteristics.
The project clarified practical constraints relevant to experimental research on complex oxide electronic devices, including equipment stability, material ageing, and limitations associated with advanced characterisation techniques. Identifying these constraints represents a useful outcome for planning future experimental studies and integrating modelling with experimentally accessible data. Overall, the project delivered experimental benchmarks, a tailored electro-thermal modelling framework, and a clearer understanding of the technical challenges associated with studying NDR in oxide thin-film devices. These outcomes provide a sound basis for future work in this research area without overstating technological readiness or application impact.