Architectures for dependable modern microprocessors

Doctoral Dissertation uoadl:1309561 557 Read counter

Unit:
Τομέας Υπολογιστικών Συστημάτων και Εφαρμογών
Library of the School of Science
Deposit date:
2015-03-02
Year:
2015
Author:
Φουτρής Νικόλαος
Dissertation committee:
Δημήτριος Γκιζόπουλος Αναπληρωτής Καθηγητής (Επιβλέπων), Αντώνιος Πασχάλης Καθηγητής, Μιχάλης Ψαράκης Επίκουρος Καθηγητής
Original Title:
Architectures for dependable modern microprocessors
Languages:
English
Translated title:
Αρχιτεκτονικές Αξιόπιστης Λειτουργίας Σύγχρονων Μικροεπεξεργαστών
Summary:
Technology scaling, extreme chip integration and the compelling requirement to
diminish the time-to-market window, has rendered microprocessors more prone to
design bugs and hardware faults. Microprocessor validation is grouped into the
following categories, based on where they intervene in a microprocessor’s
lifecycle: (a) Silicon debug: the first hardware prototypes are exhaustively
validated, (b) Μanufacturing testing: the final quality control during massive
production, and (c) In-field verification: runtime error detection techniques
to guarantee correct operation. The contributions of this thesis are the
following: (1) Silicon debug: We propose the employment of deconfigurable
microprocessor architectures along with a technique to generate self-checking
random test programs to avoid the simulation step and triage the redundant
debug sessions, (2) Manufacturing testing: We propose a self-test optimization
strategy for multithreaded, multicore microprocessors to speedup test program
execution time and enhance the fault coverage of hard errors; and (3) In-field
verification: We measure the effect of permanent faults performance components.
Then, we propose a set of low-cost mechanisms for the detection, diagnosis and
performance recovery in the front-end speculative structures. This thesis
introduces various novel methodologies to address the validation challenges
posed throughout the life-cycle of a chip.
Keywords:
Dependability, Silicon debug, Testing, Errors, Bugs
Index:
Yes
Number of index pages:
31-37
Contains images:
Yes
Number of references:
111
Number of pages:
161
document.pdf (2 MB) Open in new window