University of Twente Student Theses
Implementation and Evaluation of an Embedded AI System on a Resource-Constrained Platform
Sanad, N. (2024) Implementation and Evaluation of an Embedded AI System on a Resource-Constrained Platform.
|
PDF
1MB |
| Abstract: | There is increasing interest in deploying miniature AI systems on the edge, lessening reliance on cloud computing and improving security and latency. These systems are also increasingly being used in applications where high reliability and radiation resistance are a necessity, hence the need for studying the reliability of such systems. The current design flow for edge AI accelerators tends to involve the development of AI systems using the Xilinx design framework. SRAM-based FPGAs such as those from Xilinx are not ideal for such studies, as Flash-based FPGAs are inherently better suited than SRAM-based FPGAs to such environments and can provide radiation resistance at much more suitable prices. This necessitates the creation of a suitable system on a flash-based FPGA to assess its performance when subjected to reliability testing. Such a system is created on a Microchip SmartFusion 2 FPGA, incorporating a model built using the Conifer framework, which is designed to convert trained boosted decision trees into extremely low latency FPGA firmware. An adapter was designed to allow models produced by Conifer to interface with the ARM AMBA standard busses. This allows easy integration into microcontroller-based systems, allowing the ML model to interface with any memory on the bus matrix while being software-controllable. This results in a highly adaptable system to facilitate reliability testing of a flash FPGA-based AI system along with its internal and external memories. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 53 electrotechnology, 54 computer science |
| Programme: | Electrical Engineering BSc (56953) |
| Link to this item: | https://purl.utwente.nl/essays/101402 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page