Nano-dispositivos para sistemas de computación
Loading...
Official URL
Full text at PDC
Publication date
2026
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
License

Citation
Abstract
Este trabajo de fin de grado expone los fundamentos físicos y tecnológicos de la computación neuromórfica como solución a las crecientes limitaciones energéticas y estructurales de la tecnología CMOS y la arquitectura clásica de Turing-von Neumann. Inspirada en el funcionamiento del cerebro, esta alternativa tecnológica propone el procesamiento masivamente paralelo y la co-localización de memoria y unidades de procesamiento mediante nano-dispositivos electrónicos basados en óxidos de metales de transición: los neuristores y los sinaptores. Por un lado, los neuristores emulan la dinámica neuronal de integración y disparo con pérdidas (LIF) aprovechando la conmutación resistiva volátil (VRS) en aislantes de Mott, donde se inducen transiciones de fase metal-aislante reversibles. Por otro lado, los sinaptores replican la plasticidad y memoria de las sinapsis biológicas mediante la conmutación resistiva no volátil (NVRS). Estos últimos operan bajo los principios teóricos del memristor, alterando su conductancia mediante procesos de migración iónica y formación de filamentos para implementar reglas de aprendizaje fundamentales como la plasticidad dependiente del tiempo entre impulsos (STDP). El trabajo concluye que la física subyacente a estos materiales tiene potencial para permitir reproducir analógicamente las capacidades de las redes neuronales con una extraordinaria eficiencia energética, allanando el camino hacia una inteligencia artificial sostenible en hardware.
This final degree project presents the physical and technological foundations of neuromorphic computing as a solution to the growing energetic and structural limitations of CMOS technology and the classical Turing-von Neumann architecture. Inspired by the functioning of the brain, this technological alternative proposes massively parallel processing and the co-location of memory and computation through electronic nano-devices based on transition metal oxides: neuristors and synaptors. On the one hand, neuristors emulate the leaky integrate-and-fire (LIF) neuronal dynamics by exploiting volatile resistive switching (VRS) in Mott insulators, where reversible metal-insulator phase transitions are induced. On the other hand, synaptors replicate the plasticity and memory of biological synapses through non-volatile resistive switching (NVRS). The latter operate under the theoretical principles of the memristor, altering their conductance through ion migration and filament formation processes to implement fundamental learning rules such as spike-timing-dependent plasticity (STDP). The study concludes that the underlying physics of these materials show potential for the analog reproduction of neural network capabilities with extraordinary energy efficiency, paving the way toward sustainable artificial intelligence in hardware.
This final degree project presents the physical and technological foundations of neuromorphic computing as a solution to the growing energetic and structural limitations of CMOS technology and the classical Turing-von Neumann architecture. Inspired by the functioning of the brain, this technological alternative proposes massively parallel processing and the co-location of memory and computation through electronic nano-devices based on transition metal oxides: neuristors and synaptors. On the one hand, neuristors emulate the leaky integrate-and-fire (LIF) neuronal dynamics by exploiting volatile resistive switching (VRS) in Mott insulators, where reversible metal-insulator phase transitions are induced. On the other hand, synaptors replicate the plasticity and memory of biological synapses through non-volatile resistive switching (NVRS). The latter operate under the theoretical principles of the memristor, altering their conductance through ion migration and filament formation processes to implement fundamental learning rules such as spike-timing-dependent plasticity (STDP). The study concludes that the underlying physics of these materials show potential for the analog reproduction of neural network capabilities with extraordinary energy efficiency, paving the way toward sustainable artificial intelligence in hardware.













