<p>Prior to the mid-20th century, the idea of electric propulsion had been all but a foreign one that manifested itself in the topic of science fiction. It was around this time when companies and agencies like NASA began to take interest in the topic of space propulsion, as most famously seen in the landing of the Apollo 11 mission on the moon. It was not until the early-1960s where the idea of a pulsed plasma thruster was first realized, with its first test being in 1964 aboard the Russian Zond-2 satellite which contained 6 ablative Polytetrafluoroethylene (PTFE, or “Teflon”) pulsed plasma thrusters.</p> <p>In this paper, a new low-voltage, liquid-fed pulsed plasma thruster was developed, tested, and characterized. This project took influence from the previous low-voltage, liquid-fed pulsed plasma thruster in Purdue’s EPPL and desired to transition it from a current gas-fed system to its intended liquid-fed system. The two main objectives for this project included conducting direct studies of the cathode’s erosion rate using a simple weighing method after simulating a lifetime of discharging the thruster, and completing the initial design of the liquid-fed pulsed plasma thruster using AF-M315E as its propellant while gathering data on its required breakdown voltage, exhaust velocity, and specific impulse.</p> <p>Both objectives were successfully completed, with the following parameters being measured or calculated. The required breakdown voltage was seen to be less than 26kV to keep the ignition spark inside the chamber. For the subsequent results measured however, the breakdown voltage was kept between 10-16kV for all successive tests. The peak current measured for all discharges was an average of 11kA, far exceeding similar geometries such as MPD thrusters. The operational voltage was less than 200V, although an operational voltage closer to 100V is expected after further optimization of the system is completed. The erosion rate of the tungsten cathode at this operational setting was found to be 15.4046 +/- 0.592 microgram/Coulomb which is much less than the cathode spot erosion rate reported for tungsten in literature of about 60 microgram/Coulomb and is beneficial for extending system lifetime. The exhaust velocity was calculated to be 30.6 +/- 4.8km/s which is typical of state-of-the-art PPT electric propulsion devices. The specific impulse was also extrapolated from the ion’s exhaust velocity, calculating to be 3,119 +/- 489 seconds. Future work would require optimization of the propellant injection mechanism to minimize propellant loss.</p>

<p>Convolutional neural networks (CNNs) have become important workloads due to their<br> impressive accuracy in tasks like image classification and recognition. Convolution operations<br> are compute intensive, and this cost profoundly increases with newer and better CNN models.<br> However, convolutions come with characteristics such as sparsity which can be exploited. In<br> this dissertation, we propose three different works to capture sparsity for faster performance<br> and reduced energy. </p> <p><br></p> <p>The first work is an accelerator design called <em>SparTen</em> for improving two-<br> sided sparsity (i.e, sparsity in both filters and feature maps) convolutions with fine-grained<br> sparsity. <em>SparTen</em> identifies efficient inner join as the key primitive for hardware acceleration<br> of sparse convolution. In addition, <em>SparTen</em> proposes load balancing schemes for higher<br> compute unit utilization. <em>SparTen</em> performs 4.7x, 1.8x and 3x better than dense architecture,<br> one-sided architecture and SCNN, the previous state of the art accelerator. The second work<br> <em>BARISTA</em> scales up SparTen (and SparTen like proposals) to large-scale implementation<br> with as many compute units as recent dense accelerators (e.g., Googles Tensor processing<br> unit) to achieve full speedups afforded by sparsity. However at such large scales, buffering,<br> on-chip bandwidth, and compute utilization are highly intertwined where optimizing for<br> one factor strains another and may invalidate some optimizations proposed in small-scale<br> implementations. <em>BARISTA</em> proposes novel techniques to balance the three factors in large-<br> scale accelerators. <em>BARISTA</em> performs 5.4x, 2.2x, 1.7x and 2.5x better than dense, one-<br> sided, naively scaled two-sided and an iso-area two-sided architecture, respectively. The last<br> work, <em>EUREKA</em> builds an efficient tensor core to execute dense, structured and unstructured<br> sparsity with losing efficiency. <em>EUREKA</em> achieves this by proposing novel techniques to<br> improve compute utilization by slightly tweaking operand stationarity. <em>EUREKA</em> achieves a<br> speedup of 5x, 2.5x, along with 3.2x and 1.7x energy reductions over Dense and structured<br> sparse execution respectively. <em>EUREKA</em> only incurs area and power overheads of 6% and<br> 11.5%, respectively, over Ampere</p>