Approximate Computing Benchmarks for Embedded Systems
keywords APPROXIMATE COMPUTING, EMBEDDED SYSTEMS, HARDWARE DESIGN, SOFTWARE DEVELOPMENT
Reference persons ALESSANDRO SAVINO
External reference persons Prof. Michele Portolan (firstname.lastname@example.org)
Prof. Alberto Bosio (email@example.com)
Research Groups TESTGROUP - TESTGROUP
Thesis type IMPLEMENTATION, RESEARCH / EXPERIMENTAL
Description Goals: Identify and implement a meaningful set of multi-layer application for approximate computing techniques evaluation and comparison in Embedded System scenarios.
Approximate Computing paradigms is well suited for a particular subset of applications showing a so-called “intrinsic resiliency”. Approximation can be applied at all abstraction levels: from circuit design up to the final software application. However, the lack of a common multi-layer framework makes it difficult to estimate and compare the impact of different approximations choices.
The thesis concentrates on the identification of several approximate-aware algorithms to deliver their implementation both as hardware circuits (i.e., VHDL/Verilog and SystemC/SystemVerilog) and software applications (i.e., C and C++). The resulting benchmark will enable a clear comparison between the different approaches.
A particular emphasis will be placed on emerging computing paradigms-based algorithms such as artificial neural networks.
Learned Outcomes: High Level Synthesis (HLS), Alternative Computing Paradigms, HW/SW co-simulation (Verilog Programming Interfaces).
Required skills C/C++ programming and digital circuit design and simulation
Notes Possibility to spend half of the Thesis on the École Central de Lyon.
Deadline 23/01/2022 PROPONI LA TUA CANDIDATURA