Advances in computing technology have brought us to thousands of calculations performed in milliseconds, however moving data from memory to processor takes substantial time and energy. This begs the need to rethink data storage in high-performance systems to eliminate this bottleneck, particularly in machine learning workloads where millions of operands and parameters are used. With traditional memory candidates like SRAM, eDRAM, and Flash reaching physical limits as technology evolves, Magnetic Random Access Memory (MRAM) is a promising solution. By storing information using magnetic states, MRAM is non-volatile and illustrates the potential of compact, low power data storage scalable with technology. Voltage Control Magnetic Tunnel Junctions (VC-MTJs) are one demonstration of MRAM, having less power consumption and higher resistance, allowing for efficient read and write of the device. Our IC chip combines numerous VC-MTJ devices, utilizing CMOS fabrication, into four separate memory arrays. The IC is designed to perform efficient multiply-and-accumulate (MAC) operations which comprise upwards of 80% of all neural network computations. By computing locally with information stored in the memory, rather than transporting that data to a separate processor, this IC creates compute-in-memory. My project is to design a PCB that will interface with this chip to test the limits of reading, writing, and computing. Starting from scratch, the PCB has independent power supplies to operate each compute-in-memory array separately, a simplistic way to read the analog output signals, and constant current and voltage references. Once manufactured, the PCB will be tested using JTAG serial interfacing and python-controlled scripts

Ongoing research is considering the cooling of hypersonic vehicles via thermionic emission. Thermionic emission occurs when a material contains enough thermal energy that electrons escape from the surface. As high energy electrons escape from the surface, they are replaced with electrons of lower energy causing a net cooling effect. The work function must be known to accurately predict the cooling magnitude. While the work function for many materials is documented, emerging aerospace applications use newer materials or processes lacking a documented work function. In this project, we are creating an in-house system that can accurately measure the work function of materials. Our approach utilizes a high vacuum chamber with an electron energy analyzer, infrared camera, and thermocouples to measure the electron energy distribution from samples at elevated temperatures. The work function can be determined using the shape of the spectral emission intensity curve. Preliminary data has been used to verify proper operation of the equipment, and test procedures will be ready for full operation upon calibration with (100) tungsten in the coming weeks. The significance of this research relates to improving heat management in hypersonic flight. Our work is a precursor to employing the High Energy Flux Test Facility (HEFTY) to measure cooling in a high heat flux environment. By determining the work function, we will be able to predict net cooling and validate modeling efforts with experimental data. This work lays the foundation for innovations in aerospace technology, with the potential to ultimately drive advancements in hypersonic flight capabilities.

Radio frequency fingerprinting (RFF) is a physical layer identification technique that can be used for various purposes such as security and device tracking. Due to imperfections in hardware, each device will have its own distinctive “fingerprint” when transmitting signals. Using machine learning models such as support vector machines or convolutional neural networks, individual devices can be identified via their own fingerprints. Bluetooth signals are of interest to fingerprint. Bluetooth is a wireless communication used for low range and low power connectivity between devices. Specifically, BLE employs a modification of a technique called frequency hopping spread spectrum (FHSS). Frequency hopping is when a transmitter changes its carrier frequency across multiple channels in a large spectral band, adding reliability to a connection. First, transient-based features such as the Shannon entropy and transient energy were extracted from highly oversampled Bluetooth signals from 27 different phone makes-and-models. Then, these features were used in machine learning models to correctly identify different phones. Next, frequency hopping, Bluetooth Low Energy (BLE) like signals were generated in MATLAB and transmitted over the air using 10 different ADALM-PLUTO software defined radios (SDR) to create a dataset of undersampled signals. Steady-state features such as carrier frequency offset and in-phase quadrature offset were extracted, and machine learning models were then used to identify each radio through its fingerprint. Further research includes reducing the bandwidth of the receiver radio to only analyze specific channels.

When a wire is strung through a set of beads, applying tension to the wire compresses the beads together, increasing the rigidity of the structure. These bead-based wire jamming structures enable applications where a controllable level of stiffness is a key functionality, like deformable grippers. We aim to accelerate the design and research of these structures by developing software that allows the user to computationally generate 3D models of a large set of beads, eliminating the usage of CAD software. We created a web-app that supports the ability to quickly preview, modify, and export 3D bead models based on user-inputted numerical parameters defining the desired geometry of each bead. The app accepts parameters of the size, length, and interface angle/shape of each bead, and supports the generation of cylindrical and spherical beads as well as multi-bead line segments and arcs. Using these software tools and a resin 3D printer, we generated, 3D printed, assembled, and iterated through multiple versions of various bead structures. By applying several bead geometries and performing static analysis, we designed self-deploying polygonal structures with multi-angled joints, as well as shape-shifting structures controlled by running multiple wires through each bead. Furthermore, we implemented the generation of both polygonal and shape-shifting structures in our web-app, increasing the design space it supports. The capabilities introduced in our research provide a higher-level method of designing beads, accelerating the design work necessary to investigate the properties of complex bead structures and their applications in soft robotics and self-deploying structures.

When a wire is strung through a set of beads, applying tension to the wire compresses the beads together, increasing the rigidity of the structure. These bead-based wire jamming structures enable applications where a controllable level of stiffness is a key functionality, like deformable grippers. We aim to accelerate the design and research of these structures by developing software that allows the user to computationally generate 3D models of a large set of beads, eliminating the usage of CAD software. We created a web-app that supports the ability to quickly preview, modify, and export 3D bead models based on user-inputted numerical parameters defining the desired geometry of each bead. The app accepts parameters of the size, length, and interface angle/shape of each bead, and supports the generation of cylindrical and spherical beads as well as multi-bead line segments and arcs. Using these software tools and a resin 3D printer, we generated, 3D printed, assembled, and iterated through multiple versions of various bead structures. By applying several bead geometries and performing static analysis, we designed self-deploying polygonal structures with multi-angled joints, as well as shape-shifting structures controlled by running multiple wires through each bead. Furthermore, we implemented the generation of both polygonal and shape-shifting structures in our web-app, increasing the design space it supports. The capabilities introduced in our research provide a higher-level method of designing beads, accelerating the design work necessary to investigate the properties of complex bead structures and their applications in soft robotics and self-deploying structures.