Browsing by Author "Nikdast, Mahdi, committee member"

Now showing 1 - 3 of 3

Open Access
3D localization of cytoskeleton in mouse spermatids using stochastic optical reconstruction microscopy
(Colorado State University. Libraries, 2022) Sunny, Reshma, author; Krapf, Diego, advisor; Nikdast, Mahdi, committee member; Prasad, Ashok, committee member
It is estimated by the World Health Organization that globally 186 million individuals live with infertility. Studies have shown that cause of male infertility is unknown in 30 to 50% of the cases. Over the last several years teratozoospermias have been investigated and have been backtracked to events in spermatogenesis. The development of the acrosome and the manchette, protein and vesicle transport in spermatids, and sperm head shaping are crucial steps in the formation of healthy sperms. The cytoskeleton in spermatids plays a crucial role in shaping the sperm head. The acroplaxome exerts forces on the nucleus and gives the mammalian sperm head its species-specific shape, and also facilitates the proper attachment of the nuclear cap called the acrosome, containing the enzymes required for sperm penetration of the oocyte. The manchette should be intact and formed properly to have shortened diameter as spermatids differentiate so that it can constrict the base of the nucleus to shape the head, and also facilitate the transport of cargo to the base of the cell. Thus as studies have confirmed, the disruption in the organization of the cytoskeleton is a concern for infertility. Hence it is crucial to learn more about the cytoskeletal structures in spermatids. The goal of this thesis is to 3D localize these structures. The major structures we are interested in are the acroplaxome and the manchette. For this, we use a super-resolution microscopy method called Stochastic Optical Reconstruction Microscopy to image spermatid cytoskeleton. Our experiments confirmed the presence of α-tubulin in the manchette and that of F-actin in the manchette and the acroplaxome, as previously observed by researchers with 2D confocal images. We observed that the manchette reduces in diameter and progresses to the caudal portion of the cell at the later steps of differentiation and that the structure forms completely at step 10 and disassembles after step 14.
Embargo
Analysis of LEAC biosensor for scalable manufacturing using BPM and FDTD simulation methods
(Colorado State University. Libraries, 2024) Holmes, Cameron Dane, author; Lear, Kevin L., advisor; Nikdast, Mahdi, committee member; Kipper, Matt, committee member
The increasing demand for rapid, scalable, and accurate diagnostic tools has driven the development of optical biosensing technologies. LEAC (Local Evanescent Array-Coupled) biosensors, which leverage the evanescent field generated by optical waveguides, are particularly well-suited for applications in biomedical diagnostics, environmental monitoring, and point-of-care testing. LEAC biosensors have previously been fabricated in incomplete and unoptimized near-commercial CMOS processes and fully custom processes in a university cleanroom but have not been implemented in suitable high-volume processes such as commercial silicon photonics. A primary motivation for the research presented in this thesis is to evaluate the ability to fabricate LEAC biosensors operating at 1550 nm wavelengths in the commercial AIM Photonics' active silicon photonics process. This thesis presents a comprehensive tolerance analysis of LEAC sensors for both bulk sample layers (400 nm thick) and protein monolayers (10 nm thick) in AIM's process, focusing on the impact of variations in key design parameters—specifically waveguide core thickness, cladding layers, and photodetector placement—on sensor sensitivity. Beam Propagation Method (BPM) and Finite-Difference Time-Domain (FDTD) simulation techniques are employed to assess how these tolerances affect optical field propagation, power dissipation, and flux into the photodetector, serving as proxies for sensor performance. Additionally, the study examines crosstalk between multiple sensing regions, evaluating how refractive index variations in one region influence adjacent regions—an important consideration for multi-region sensors. Results show that sensor sensitivity increases with cladding thickness and decreases with waveguide core thickness. A 25 nm manufacturing error in core thickness resulted in less than a 10% sensitivity shift, and a 300 nm cladding thickness error had a similarly small effect. Resonant absorption between the core and photodetector was observed across both bulk and monolayer samples. Sensitivity depends heavily on proximity to resonance; a 10% error in photodetector thickness at resonance caused a 600% change in sensitivity, while off-resonance, the same error had minimal impact. Coupled Mode Theory (CMT) explained these energy transfers and power fluctuations. ANOVA analysis of full-device FDTD simulations quantified forward crosstalk due to modulated absorption from sample regions closer to the optical source (upstream). Forward crosstalk was found to be negligible for protein monolayer samples but could be significant in bulk samples. However, even in bulk samples, forward crosstalk was largely mitigated using photocurrent ratios with a reference region. A crosstalk ration was used as a metric to determine the influence of each refractive index (n1, n3) on the photocurrent ratio. In the forward crosstalk direction, the use of photocurrent ratios decreased the magnitude of the forward crosstalk ratio; however, the use of photocurrents inherently introduce dependance on downstream indices (reverse crosstalk). Reverse crosstalk, caused by reflections at the dielectric boundary between sensing regions, was found to be negligible using photocurrent ratios with bulk analytes; however, with monolayers, the use of photocurrent ratios introduced a slight dependence on the downstream region, indicating minor backward crosstalk. This can be mitigated by using raw current values rather than current ratios. Raw currents eliminate backward crosstalk in region 1, while photocurrent ratios effectively eliminate forward crosstalk in region 3.
Open Access
Silicon photonic hardware accelerators for transformers and graph neural networks
(Colorado State University. Libraries, 2023) Afifi, Salma, author; Pasricha, Sudeep, advisor; Nikdast, Mahdi, committee member; Malaiya, Yashwant, committee member
The rapid growth of artificial intelligence (AI) applications has revolutionized the way we process data, make decisions, and interact with machines. Specifically, artificial neural networks (ANNs) have significantly evolved and now encompass various advanced neural networks such as transformers and graph neural networks (GNNs). This has enabled the development of innovative AI applications that can transform several industries, including healthcare, recommendation systems, and robotics. Transformer and transformer-based neural networks have outperformed multiple ANNs, such as convolution neural networks (CNNs) and recurrent neural networks (RNNs), across many natural language processing (NLP) tasks. Moreover, transformers are currently being integrated into vision tasks through using the vision transformer model (ViT). Similarly, GNNs have witnessed a surge of advancements over the past few years and have established their proficiency in dealing with graph-structured data. Nevertheless, each of these neural networks imposes unique challenges, hindering their inference and usage in resource-constrained systems. For instance, the transformer model's size, number of parameters, and complexity of operations lead to long inference times, large memory footprint, and low computation-to-memory ratio. On the other hand, GNNs inference challenges are due to their dense and very sparse computations. Additionally, the wide variety of possible input graphs structure and algorithms dictate the need for a system capable of efficiently adapting their execution and operations to the specific graph structure and effectively scaling to extremely large graphs. Accordingly, conventional computing processors and ANN accelerators are not tailored to cater for such challenges, and using them to accelerate transformers and GNN execution can be highly inefficient. ii Furthermore, the utilization of traditional electronic accelerators entails a number of limitations, including escalating fabrication costs due to low yields and diminishing performance improvements, associated with semiconductor-technology scaling. This has led researchers to start investigating other technologies for ANN acceleration such as silicon photonics which enables performing complex operations in the optical domain with low energy consumption and at very high throughput. While several hardware accelerators leveraging silicon photonics have been presented for networks such as CNNs, none have been customized for emerging complex neural networks such as transformers and GNNs. Due to the various challenges associated with each of these networks, designing reliable and efficient inference hardware accelerators for transformers and GNNs is a non-trivial problem. This thesis introduces two novel silicon-photonic-based hardware architectures for energy efficient and high throughput inference acceleration. As our first contribution, we propose a non-coherent silicon photonic hardware accelerator for transformer neural networks, called TRON. We demonstrate how TRON is able to accommodate a wide range of transformer and transformer-based neural networks while surpassing GPU, CPU, TPU, and several state-of-the-art transformer hardware accelerators. For GNN inference acceleration, we propose GHOST, a hardware accelerator that integrates various device-, circuit- and architecture-level optimizations which enable it to efficiently process a broad family of GNNs and real-world graph structures and sizes. When compared to multiple state-of-the-art GNN hardware accelerators, GPUs, CPUs, and TPUs, our experiments showcase how GHOST exhibits significantly better performance and energy efficiency.