| KEYWORD |
Area Engineering
An evolutionary-based environment to generate stress programs for a RISC-V processor
keywords BURN IN, RISC-V, STRESS TEST
Reference persons MATTEO SONZA REORDA
External reference persons Nikolaos Deligiannis
Research Groups DAUIN - GR-05 - ELECTRONIC CAD & RELIABILITY GROUP - CAD
Description The strict reliability requirements existing in some application domains (such as automotive, aerospace, or biomedicine) mandates for the adoption of a special test (named Burn In) aimed at stressing the device, so that possible weak points turn into faults, and the device can be labeled as faulty. In this way, the so called Infant Mortality (e.g., the phenomenon which causes some devices to stop working correctly in the first period of their operational life) is significantly reduced. In order to perform Burn In, a common technique is based on stressing the device either with external conditions (e.g., temperature or voltage) or with suitably created stimuli which induce a high internal stress. In this case, the stimuli are normally required to maximize the switching activity (SWA) inside the whole device, or in some specific module. The generation of such stimuli (corresponding to a specific program with suitable input data is far from trivial. One possible solution is based on the usage of Evolutionary techniques (e.g., Genetic Algorithms). This approach allows to start from random programs and data, which are gradually improved looking at the achieved SWA.
The student working on this project will benefit of the existing uGP environment developed by the CAD Group active within the Dept. of Control and Computer Engineering, and will set up an environment able to work on a RISC-V processor. The goal is to identify a sequence of assembly instructions (and input data) for each main module of the processor, able to maximize the SWA inside the module.
Required skills This proposal is suitable for students enrolled both in Computer and Electronic Engineering. Students from other other curricula will be evaluated case by case.
Deadline 16/01/2023
PROPONI LA TUA CANDIDATURA