Loading...
Search
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
Digital Object List
Bookmark
Review List
1023 viewed
بهینه سازی نگاشت توالی ژنوم با استفاده از پردازش ابربعدی برای پردازش درون حافظه
غلامی، اعظم Gholami, Azam
Genome Sequence Mapping Based on Hyper-Dimensional Computing Customized for In-Memory Processing
Gholami, Azam | 2025
44
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 58745 (19)
- University: Sharif University of Technology
- Department: Computer Engineering
- Advisor(s): Koohi, Somayyeh
- Abstract:
- The exponential growth of data across various domains of human society necessitates rapid and efficient data processing. In a number of machine learning models, an input query is searched across the trained class vectors to find the closest feature class vector based on the cosine similarity metric. However, performing cosine similarity computations between vectors on von Neumann machines involves a large number of multiplications, Euclidean normalizations, and divisions, thus incurring heavy hardware energy and latency overheads. Moreover, due to the memory-wall problem present in conventional architectures, frequent cosine similarity-based searches (CSSs) over class vectors require substantial data movement, limiting the throughput and efficiency of the system.To overcome the aforementioned challenges, this work proposes a ferroelectric FET (FeFET)-based time-domain cosine similarity search (TD-CSS) array for energy-efficient similarity computation. This platform is developed for genome sequence search based on hyperdimensional computing (HDC) for hardware-friendly computation and transforms inherently sequential processes of genome matching into highly parallelizable computational tasks. This work also introduces a time-domain architecture called TD-bundling, which implements the bundling operation of hypervectors using majority logic in the time domain. Such TD designs can convert their output (i.e., a time interval) into digital values with relatively simple sensing circuitry, thus saving a large amount of area and energy compared with conventional in-memory computing designs that process analog voltage/current signals. The variable-capacitance (VC) delay-chain structure in our design supports quantitative similarity computation and enhances robustness against variations. Benchmarking results on HDC applications show that the proposed FeFET-based TD-CSS achieves, on average, a 77× speedup compared to GPU-based implementations. The proposed TD-CSS promises energy-efficient, quantitative, and accurate cosine similarity search for diverse and intensive data-processing applications, especially in energy-constrained scenarios
- Keywords:
- Hyperdimensional Computing (HD) ; Ferroelectric Field Effect Transistors (FET) ; Genome Sequence Mapping ; In-Memory Computing ; Cosine Similarity Scoring ; Time Domain Computing
محتواي کتاب