{"id":{"repo_id":"usf","oai_identifier":"oai:digitalcommons.usf.edu:etd-12133"},"canonical_url":"https://search.ndltd.org/etd/usf/oai:digitalcommons.usf.edu:etd-12133","repository":{"repo_id":"usf","name":"University of South Florida","base_url":"https://digitalcommons.usf.edu/do/oai/"},"display":{"title":"Programmable and Scalable Bit-Sliced VLSI Architecture for Decision Tree Based Machine Learning Edge Inference","abstract":"As the volume and diversity of Internet-of-Things (IoT) data continues to grow, traditional cloud-based processing methods face significant challenges, including latency, bandwidth constraints, and privacy concerns. Our research focuses on employing decision trees (DTs) as an intelligent filtering mechanism on the edge. Preliminary comparisons across four datasets revealed DTs are significantly more efficient than multilayer perceptrons (MLP), saving 97-98\\% in area and power, leading to the selection of DT for our proposed architecture for lightweight IoT devices. We propose a novel programmable and scalable custom application specific integrated circuit (ASIC) architecture designed for Decision Tree based ML inference. Each bit-slice incorporates two 8-bit SISO input registers connected to an 8-bit comparator for data processing, the output of the comparator drives the select line of Mux, which selects the respective true and false paths. Each bit-slice can be programmed into either a leaf node or a regular node. A leaf node stores classification labels. A regular node compares a feature value with a weight value to decide between true and false paths. Given a DT model, the decision tree can be pre-programmed to store the model weights in respective tree nodes. In the inference phase, feature values are sequentially fed into the DT nodes. After the feature values are loaded the DT tree performs an inference with the classification value generated in the root node. We have implemented and validated the architecture at the layout-level using Cadence Virtuoso in 0.5$\\mu$m CMOS technology node. A 3-level DT occupies roughly 90 mm² area with 10.3 mW of power consumption at a maximum clock speed of 12.8 MHz","abstract_html":"As the volume and diversity of Internet-of-Things (IoT) data continues to grow, traditional cloud-based processing methods face significant challenges, including latency, bandwidth constraints, and privacy concerns. Our research focuses on employing decision trees (DTs) as an intelligent filtering mechanism on the edge. Preliminary comparisons across four datasets revealed DTs are significantly more efficient than multilayer perceptrons (MLP), saving 97-98\\% in area and power, leading to the selection of DT for our proposed architecture for lightweight IoT devices. We propose a novel programmable and scalable custom application specific integrated circuit (ASIC) architecture designed for Decision Tree based ML inference. Each bit-slice incorporates two 8-bit SISO input registers connected to an 8-bit comparator for data processing, the output of the comparator drives the select line of Mux, which selects the respective true and false paths. Each bit-slice can be programmed into either a leaf node or a regular node. A leaf node stores classification labels. A regular node compares a feature value with a weight value to decide between true and false paths. Given a DT model, the decision tree can be pre-programmed to store the model weights in respective tree nodes. In the inference phase, feature values are sequentially fed into the DT nodes. After the feature values are loaded the DT tree performs an inference with the classification value generated in the root node. We have implemented and validated the architecture at the layout-level using Cadence Virtuoso in 0.5<span class=\"etd-inline-math\">&mu;</span>m CMOS technology node. A 3-level DT occupies roughly 90 mm² area with 10.3 mW of power consumption at a maximum clock speed of 12.8 MHz","abstract_has_math":true,"creators":["Somesula, Raaga Sai"],"institution":"University of South Florida","degree_name":"MS in Computer Engineering (M.S.C.P.)","degree_level":"Thesis","degree_discipline":"Computer Science and Engineering","degree_department":null,"school":null,"contributors":["Srinivas Katkoori, Ph.D.","Hao Zheng, Ph.D.","Robert Karam, Ph.D."],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-03-15T07:00:00Z","date_published":"2024-03-15T07:00:00Z","updated_at":"2026-11-19T08:49:32Z","subjects":["DTs","Edge-AI","Internet-of-Things","H-Tree","Scalability","Computer Engineering"],"languages":["EN-US"],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.usf.edu/etd/10830","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.usf.edu/etd/10830","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://digitalcommons.usf.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.usf.edu%3Aetd-12133","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Srinivas Katkoori, Ph.D.","Hao Zheng, Ph.D.","Robert Karam, Ph.D."]},{"key":"dc:creator","label":"Author","values":["Somesula, Raaga Sai"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2024-03-15T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science and Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis","M.S.Cp."]},{"key":"thesis:degree_name","label":"Degree Name","values":["MS in Computer Engineering (M.S.C.P.)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of South Florida"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["DTs","Edge-AI","Internet-of-Things","H-Tree","Scalability","Computer Engineering"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["EN-US"]},{"key":"dc:rights.license","label":"Dc Rights License","values":["default"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.usf.edu/etd/10830"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["As the volume and diversity of Internet-of-Things (IoT) data continues to grow, traditional cloud-based processing methods face significant challenges, including latency, bandwidth constraints, and privacy concerns. Our research focuses on employing decision trees (DTs) as an intelligent filtering mechanism on the edge. Preliminary comparisons across four datasets revealed DTs are significantly more efficient than multilayer perceptrons (MLP), saving 97-98\\% in area and power, leading to the selection of DT for our proposed architecture for lightweight IoT devices. We propose a novel programmable and scalable custom application specific integrated circuit (ASIC) architecture designed for Decision Tree based ML inference. Each bit-slice incorporates two 8-bit SISO input registers connected to an 8-bit comparator for data processing, the output of the comparator drives the select line of Mux, which selects the respective true and false paths. Each bit-slice can be programmed into either a leaf node or a regular node. A leaf node stores classification labels. A regular node compares a feature value with a weight value to decide between true and false paths. Given a DT model, the decision tree can be pre-programmed to store the model weights in respective tree nodes. In the inference phase, feature values are sequentially fed into the DT nodes. After the feature values are loaded the DT tree performs an inference with the classification value generated in the root node. We have implemented and validated the architecture at the layout-level using Cadence Virtuoso in 0.5$\\mu$m CMOS technology node. A 3-level DT occupies roughly 90 mm² area with 10.3 mW of power consumption at a maximum clock speed of 12.8 MHz"]},{"key":"dc:title","label":"Title","values":["Programmable and Scalable Bit-Sliced VLSI Architecture for Decision Tree Based Machine Learning Edge Inference"]}]}],"canonical_facts":{"dc:contributor":["Srinivas Katkoori, Ph.D.","Hao Zheng, Ph.D.","Robert Karam, Ph.D."],"dc:creator":["Somesula, Raaga Sai"],"dc:date":["2024-03-15T07:00:00Z"],"dc:description":["As the volume and diversity of Internet-of-Things (IoT) data continues to grow, traditional cloud-based processing methods face significant challenges, including latency, bandwidth constraints, and privacy concerns. Our research focuses on employing decision trees (DTs) as an intelligent filtering mechanism on the edge. Preliminary comparisons across four datasets revealed DTs are significantly more efficient than multilayer perceptrons (MLP), saving 97-98\\% in area and power, leading to the selection of DT for our proposed architecture for lightweight IoT devices. We propose a novel programmable and scalable custom application specific integrated circuit (ASIC) architecture designed for Decision Tree based ML inference. Each bit-slice incorporates two 8-bit SISO input registers connected to an 8-bit comparator for data processing, the output of the comparator drives the select line of Mux, which selects the respective true and false paths. Each bit-slice can be programmed into either a leaf node or a regular node. A leaf node stores classification labels. A regular node compares a feature value with a weight value to decide between true and false paths. Given a DT model, the decision tree can be pre-programmed to store the model weights in respective tree nodes. In the inference phase, feature values are sequentially fed into the DT nodes. After the feature values are loaded the DT tree performs an inference with the classification value generated in the root node. We have implemented and validated the architecture at the layout-level using Cadence Virtuoso in 0.5$\\mu$m CMOS technology node. A 3-level DT occupies roughly 90 mm² area with 10.3 mW of power consumption at a maximum clock speed of 12.8 MHz"],"dc:identifier":["https://digitalcommons.usf.edu/etd/10830"],"dc:language":["EN-US"],"dc:rights.license":["default"],"dc:subject":["DTs","Edge-AI","Internet-of-Things","H-Tree","Scalability","Computer Engineering"],"dc:title":["Programmable and Scalable Bit-Sliced VLSI Architecture for Decision Tree Based Machine Learning Edge Inference"],"thesis:degree_discipline":["Computer Science and Engineering"],"thesis:degree_level":["Thesis","M.S.Cp."],"thesis:degree_name":["MS in Computer Engineering (M.S.C.P.)"],"thesis:institution_name":["University of South Florida"]},"updated_at":"2026-11-19T08:49:32Z"}