Student UCSB Biomedical Engineering Society Chapter Santa Clara, California, United States
Introduction: Neuromorphic computing serves as an intriguing and extremely interdisciplinarily driven field. The area desires to model computational systems based on the complex underpinnings and inner workings of biological neural systems and brain circuitry. The human brain serves as a culmination of billions of years of evolution. The brain has a nearly limitless capacity to process and analyze information via constant sensory input while maintaining unbeatable energy efficiency standards. Currently, the mainstream traditional computing models are structured based on the principles of Von Neumann architecture which poses a conundrum with regards to effective scalability and technological advances. These limitations include but are not limited to higher energy, unscalable speed in data processing, as well as the main von Neumann architecture limitation (segregation of memory and processing systems lends to deficits in computational speed and efficiencies. Due to the exponential growth and amalgamations of data daily, the requirement for innovative and efficient exponentially scaling systems is necessary. Neuromorphic computing enables expansive learning capabilities with maintenance on powerful energy efficiency tactics. The below ascribed proposal seeks to conceptualize a pathway to overcome the traditional computational limitations native to this outdated architecture. It brings together spintronics alongside bio-inspired neural computational networks to facilitate in improving efficiencies, scalability factor, and overall computational performance. Spintronics, an emerging field within nanotech, uses the intrinsic electron spin effects to store and process data. Through integration of spintronics with bio-based computation, an allowance is provided for much more energy efficient computation.
Materials and
Methods: Technical approach for this project will involve integration of multiple scientific disciplines. Understanding of both chemical and biological systems will assist in catalyzing information processes and enable guidance for design and implementation of neuromorphic and spintronic based systems. After the initial design phase, the systems will go through rigorous, data driven testing in order to provide validation and quantitative metrics to evaluate functional efficacy. The projected research plan structuring bifurcates into three sequential phases. The first phase will involve a comprehensive and overall exhaustive literature review to enable better understanding of neuromorphic systems, spintronics application, and just current bio-inspired computational directives. The information garnered within this phase will inform and construct foundations for the formulation of the initial stage theoretical modeling. The literature review will also cover biological neural systems in great detail to ensure effective emulation of artificial systems. The second phase involves more practical lab experimentation in order to take the theoretical model into a more tangible system setting. The fabrication as well as testing of the proposed system are critical within this phase. Experimental results derived from this phase will provide necessary information to optimize the system’s performance. The final phase will seek to evaluate final experimental results and refine the model based off observations. This means that it will heavily rely on iterative improvement and comprehensive objective completion. This phase will also involve testing to ensure performance metrics have been met. The research findings will then be published and hopefully disseminated out.
Results, Conclusions, and Discussions: Because this project is still in progress, it has yet to have fully concluded. To provide a more specific description of the trajectory of the project, I will describe the current path. I am initially focusing on spintronic devices and their nanoscale properties, specifically STT (spin-transfer torque) and SOT (spin-orbit torque). I will work to exploit these phenomena and modulate magnetization dynamics on MTJs (magnetic tunnel junctions). After this, I will work with semiconductor physics analysis and quantum mechanical modeling to further elucidate the behavior of said MTJs, especially under influence of STT and SOT. I want to look particularly at the stochastic nature of switching magnetic polarities and how this relates to the reliability of device performance. The next phase will involve designing a neuromorphic computing framework with non-volatile spintronic devices. In the design phase of this, I am going to start with defining the system’s architecture as a network of spiking neurons, stimulated through MTJs and configured to model after Hebbian plasticity as well as homeostatic systems observed in a biological context. MTJ based neurons leverage bidirectional tunneling and the magnetoresistance effect to correspond with neuron firing and resistant states. VHDL and FPGAs enable concurrent signal processing. I want to also focus on optimizing STDP to realize dynamic biological capacity. The architecture will be designed such that it mimics synaptic plasticity especially with respect to spintronic elements and overall creates an ultra low power, high adaptivity response computational system.
