{"id":{"repo_id":"clemson","oai_identifier":"oai:open.clemson.edu:all_dissertations-5259"},"canonical_url":"https://search.ndltd.org/etd/clemson/oai:open.clemson.edu:all_dissertations-5259","repository":{"repo_id":"clemson","name":"Clemson University","base_url":"https://open.clemson.edu/do/oai/"},"display":{"title":"Domain-Specific Design on FPGA and NPU for scientific computing and On-Device LLM Inference","abstract":"High-performance computing (HPC) is changing rapidly as scientific simulations and large language model (LLM) workloads push the need for higher performance under tight power and memory constraints. Conventional platforms such as CPUs and GPUs accelerate computation through instruction-driven parallelism, relying on multithreading, SIMD, and SIMT execution, but increasingly encounter scalability limits imposed by the power and memory walls. In contrast, Field-Programmable Gate Arrays (FPGAs) and Neural Processing Units (NPUs) offer a high-efficiency alternative through dataflow-oriented architectures that exploit deep pipelining and customized memory hierarchies to reduce data movement. However, the performance potential of these spatial accelerators remains largely unrealized when traditional, control-flow-centric algorithms are directly mapped onto them. This dissertation addresses this gap by developing domain-specific algorithm and dataflow designs tailored for FPGAs and NPUs, demonstrating that hardware–software co-design is essential for achieving high performance and energy efficiency on modern accelerators. Two representative and challenging applications are studied: electromagnetic simulation using the Finite-Difference Time-Domain (FDTD) method and on-device inference for large language models. FDTD simulations are critical for the design of photonic integrated circuits but are computationally intensive, while on-device LLM inference requires low latency and low power consumption. For FDTD, this work introduces a time-pipelined computation that significantly reduces data movement and enables scalable execution across FPGA networks and reconfigurable accelerators, substantially shortening photonic design cycles. For large language models, it demonstrates how reorganizing computation and dataflow allows NPUs to process long sequences efficiently, achieving lower latency and energy consumption than existing approaches.","abstract_html":"High-performance computing (HPC) is changing rapidly as scientific simulations and large language model (LLM) workloads push the need for higher performance under tight power and memory constraints. Conventional platforms such as CPUs and GPUs accelerate computation through instruction-driven parallelism, relying on multithreading, SIMD, and SIMT execution, but increasingly encounter scalability limits imposed by the power and memory walls. In contrast, Field-Programmable Gate Arrays (FPGAs) and Neural Processing Units (NPUs) offer a high-efficiency alternative through dataflow-oriented architectures that exploit deep pipelining and customized memory hierarchies to reduce data movement. However, the performance potential of these spatial accelerators remains largely unrealized when traditional, control-flow-centric algorithms are directly mapped onto them. This dissertation addresses this gap by developing domain-specific algorithm and dataflow designs tailored for FPGAs and NPUs, demonstrating that hardware–software co-design is essential for achieving high performance and energy efficiency on modern accelerators. Two representative and challenging applications are studied: electromagnetic simulation using the Finite-Difference Time-Domain (FDTD) method and on-device inference for large language models. FDTD simulations are critical for the design of photonic integrated circuits but are computationally intensive, while on-device LLM inference requires low latency and low power consumption. For FDTD, this work introduces a time-pipelined computation that significantly reduces data movement and enables scalable execution across FPGA networks and reconfigurable accelerators, substantially shortening photonic design cycles. For large language models, it demonstrates how reorganizing computation and dataflow allows NPUs to process long sequences efficiently, achieving lower latency and energy consumption than existing approaches.","abstract_has_math":false,"creators":["Yu, Miaoxiang"],"institution":null,"degree_name":"Doctor of Philosophy (PhD)","degree_level":"Dissertation","degree_discipline":"Electrical and Computer Engineering","degree_department":null,"school":null,"contributors":["Dr.Tao Wei","Dr. Fatemeh Afghah","Dr. Rong Ge","Dr. Judson Ryckman"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T07:34:07Z","subjects":["FDTD","FPGA","LLM Inference","NPU","Architectural Engineering","Electromagnetics and Photonics"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://open.clemson.edu/all_dissertations/4205","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://open.clemson.edu/all_dissertations/4205","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://open.clemson.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Aopen.clemson.edu%3Aall_dissertations-5259","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Dr.Tao Wei","Dr. Fatemeh Afghah","Dr. Rong Ge","Dr. Judson Ryckman"]},{"key":"dc:creator","label":"Author","values":["Yu, Miaoxiang"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2026-03-16T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2026-05-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical and Computer Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy (PhD)"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["FDTD","FPGA","LLM Inference","NPU","Architectural Engineering","Electromagnetics and Photonics"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://open.clemson.edu/all_dissertations/4205"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["High-performance computing (HPC) is changing rapidly as scientific simulations and large language model (LLM) workloads push the need for higher performance under tight power and memory constraints. Conventional platforms such as CPUs and GPUs accelerate computation through instruction-driven parallelism, relying on multithreading, SIMD, and SIMT execution, but increasingly encounter scalability limits imposed by the power and memory walls. In contrast, Field-Programmable Gate Arrays (FPGAs) and Neural Processing Units (NPUs) offer a high-efficiency alternative through dataflow-oriented architectures that exploit deep pipelining and customized memory hierarchies to reduce data movement. However, the performance potential of these spatial accelerators remains largely unrealized when traditional, control-flow-centric algorithms are directly mapped onto them. This dissertation addresses this gap by developing domain-specific algorithm and dataflow designs tailored for FPGAs and NPUs, demonstrating that hardware–software co-design is essential for achieving high performance and energy efficiency on modern accelerators. Two representative and challenging applications are studied: electromagnetic simulation using the Finite-Difference Time-Domain (FDTD) method and on-device inference for large language models. FDTD simulations are critical for the design of photonic integrated circuits but are computationally intensive, while on-device LLM inference requires low latency and low power consumption. For FDTD, this work introduces a time-pipelined computation that significantly reduces data movement and enables scalable execution across FPGA networks and reconfigurable accelerators, substantially shortening photonic design cycles. For large language models, it demonstrates how reorganizing computation and dataflow allows NPUs to process long sequences efficiently, achieving lower latency and energy consumption than existing approaches."]},{"key":"dc:title","label":"Title","values":["Domain-Specific Design on FPGA and NPU for scientific computing and On-Device LLM Inference"]}]}],"canonical_facts":{"dc:contributor":["Dr.Tao Wei","Dr. Fatemeh Afghah","Dr. Rong Ge","Dr. Judson Ryckman"],"dc:creator":["Yu, Miaoxiang"],"dc:date.available":["2026-03-16T07:00:00Z"],"dc:date.created":["2026-05-01T07:00:00Z"],"dc:description.abstract":["High-performance computing (HPC) is changing rapidly as scientific simulations and large language model (LLM) workloads push the need for higher performance under tight power and memory constraints. Conventional platforms such as CPUs and GPUs accelerate computation through instruction-driven parallelism, relying on multithreading, SIMD, and SIMT execution, but increasingly encounter scalability limits imposed by the power and memory walls. In contrast, Field-Programmable Gate Arrays (FPGAs) and Neural Processing Units (NPUs) offer a high-efficiency alternative through dataflow-oriented architectures that exploit deep pipelining and customized memory hierarchies to reduce data movement. However, the performance potential of these spatial accelerators remains largely unrealized when traditional, control-flow-centric algorithms are directly mapped onto them. This dissertation addresses this gap by developing domain-specific algorithm and dataflow designs tailored for FPGAs and NPUs, demonstrating that hardware–software co-design is essential for achieving high performance and energy efficiency on modern accelerators. Two representative and challenging applications are studied: electromagnetic simulation using the Finite-Difference Time-Domain (FDTD) method and on-device inference for large language models. FDTD simulations are critical for the design of photonic integrated circuits but are computationally intensive, while on-device LLM inference requires low latency and low power consumption. For FDTD, this work introduces a time-pipelined computation that significantly reduces data movement and enables scalable execution across FPGA networks and reconfigurable accelerators, substantially shortening photonic design cycles. For large language models, it demonstrates how reorganizing computation and dataflow allows NPUs to process long sequences efficiently, achieving lower latency and energy consumption than existing approaches."],"dc:identifier":["https://open.clemson.edu/all_dissertations/4205"],"dc:subject":["FDTD","FPGA","LLM Inference","NPU","Architectural Engineering","Electromagnetics and Photonics"],"dc:title":["Domain-Specific Design on FPGA and NPU for scientific computing and On-Device LLM Inference"],"thesis:degree_discipline":["Electrical and Computer Engineering"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Doctor of Philosophy (PhD)"]},"updated_at":"2026-11-19T07:34:07Z"}