12 Mars – Thesis defense - Jonathan Déchelotte

10 h30 Amphi Jean-Paul Dom - Laboratory IMS / Building A31 (Talence campus)

Study and implementation of an execution environment for heterogeneous reconfigurable architecture.

Today, embedded systems have taken a leading role in our world. Whether for communication, travel, work or entertainment, their use is preponderant. Together, research and industry efforts are constantly developing various parts that make up these systems: processor, FPGA, memory, operating system.
From an architectural point of view, the contribution of a generalist architecture coupled with a reconfigurable architecture positions SoC FPGA as popular targets for use in embedded systems. However, their implementation's complexity makes their adoption difficult. The abstraction of low-level layers seems to be an investigation's axis that would tend to reverse this trend. The use of an operating system seems suitable at first glance because they deliver an ecosystem of drivers and services for access to hardware resources, native scheduling capacities and libraries for security. However, this solution brings constraints and lead to evaluate other approaches.
This manuscript evaluates the ability of a high-level language, Lua, to provide an execution environment in such a case that the implementation does not provide operating system. It gives, through an ecosystem named Lynq, the necessary building blocks for the management and allocation of resources present on the SoC FPGA as well as a method for isolation between applications. Besides the adoption of this execution environment, our work explores the capacity of generalist architectures such as CPUs to become specialized when implemented on a FPGA. This is done through a contribution allowing the generation of a RISC-V CPU and its associated microcode.

Event localization