Publicação
Fabrication and characterization of neuromorphic photodetector devices
| Resumo: | With the advent of Large Language Models (LLMs), developing new computational paradigms is essential to meet ever growing demands for processing power and energy efficiency. Neuromorphic computation, taking inspiration from biological nervous systems, emerges as a promising solution by enabling highly parallel and power-efficient hardware. At the same time, Resonant Tunnelling Diodes (RTDs) have become a key candidate for such hardware due to their unique attributes. RTD scan generate neuron-like spiking signals, posses strong photo detection capabilities suitable for high-frequency optoelectronic systems, and exhibit low power consumption. This dissertation focused on the design, fabrication, and characterization of RTD devices specifically for neuromorphic applications. It begins with analysis of the quantum phenomena that give rise to resonant tunneling in double barrier quantum wells, which produces a macroscopic Negative Differential Resistance (NDR) region. This NDR is responsible for the device’s ability to create high-frequency spikes. To explore how RTDs can interconnect to form hardware neural networks, a dedicated simulator was developed that can model RTD behaviours ranging from individual devices to series and parallel configurations. Experimentally, the work involved fabricating III-V semiconductor devices from a GaAs/AlAs material wafer in a clean room. Several micro fabrication steps were performed, such as photolithography, wet and dry etching, passivation, and metallisation. Process control was achieved using a profilometer and a Scanning Electron Microscope (SEM). Additionally, a thorough electro-optical characterization assessed the photo detection and high-frequency modulation performance of the two fabricated samples, demonstrating strong results such as responsivity values above 1 AW−1 and a large optical modulation bandwidth. In summary, this research advances the understanding of the RTD’s role in neuromorphic computing, successfully outlining their complete fabrication workflow and validating their significant optoelectronic capabilities. This confirms their potential for efficient, brain-inspired hardware. |
|---|---|
| Autores principais: | Coito,Rui Alexandre de Barros |
| Assunto: | Resonant Tunnelling Diode Neuromorphic computation Negative Differential Resistance Photodetectors Optoelectronics |
| Ano: | 2026 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso embargado |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | With the advent of Large Language Models (LLMs), developing new computational paradigms is essential to meet ever growing demands for processing power and energy efficiency. Neuromorphic computation, taking inspiration from biological nervous systems, emerges as a promising solution by enabling highly parallel and power-efficient hardware. At the same time, Resonant Tunnelling Diodes (RTDs) have become a key candidate for such hardware due to their unique attributes. RTD scan generate neuron-like spiking signals, posses strong photo detection capabilities suitable for high-frequency optoelectronic systems, and exhibit low power consumption. This dissertation focused on the design, fabrication, and characterization of RTD devices specifically for neuromorphic applications. It begins with analysis of the quantum phenomena that give rise to resonant tunneling in double barrier quantum wells, which produces a macroscopic Negative Differential Resistance (NDR) region. This NDR is responsible for the device’s ability to create high-frequency spikes. To explore how RTDs can interconnect to form hardware neural networks, a dedicated simulator was developed that can model RTD behaviours ranging from individual devices to series and parallel configurations. Experimentally, the work involved fabricating III-V semiconductor devices from a GaAs/AlAs material wafer in a clean room. Several micro fabrication steps were performed, such as photolithography, wet and dry etching, passivation, and metallisation. Process control was achieved using a profilometer and a Scanning Electron Microscope (SEM). Additionally, a thorough electro-optical characterization assessed the photo detection and high-frequency modulation performance of the two fabricated samples, demonstrating strong results such as responsivity values above 1 AW−1 and a large optical modulation bandwidth. In summary, this research advances the understanding of the RTD’s role in neuromorphic computing, successfully outlining their complete fabrication workflow and validating their significant optoelectronic capabilities. This confirms their potential for efficient, brain-inspired hardware. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.