From Memristors to Minds: Enabling End-to-End Neuromorphic Systems with Spiking Neural Networks

This thesis explores advanced computer systems that mimic the human brain, known as neuromorphic systems. We're trying to understand both how these systems are built from the basic components and how they process information at a higher level.

We focus particularly on integrating special components called memristive devices into networks that operate similarly to neural networks in the brain. These devices can change their behavior based on the electrical signals they receive, which is similar to how connections between neurons in the brain adjust when learning.

The thesis delves into using a specific type of memristor made from a material called SrTiO₃ (Nb:STO) that acts like the synapses in the brain. These memristors can adjust their electrical resistance in response to input signals, mimicking the adaptability of biological synapses.

From a broader perspective, the thesis presents a model that simulates how the brain’s memory works. This model incorporates features that allow it to maintain and manage information effectively, helping the brain-like computer system remain stable and flexible in handling tasks.

Overall, this work helps bridge the gap between the physical hardware of these brain-like computers and the software that runs them. By understanding and developing both the hardware and software from two perspectives—building from the bottom up and designing from the top down—we enhance our knowledge of these neuromorphic architectures. This not only pushes the boundaries of traditional computing but also shows the potential of these systems to be more energy-efficient and powerful.