Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Forthcoming manufacturing technologies hold the promise to increase multifuctional computing systems performance and functionality thanks to a remarkable growth of the device integration density. Despite the benefits introduced by this technology improvements, reliability is becoming a key challenge for the semiconductor industry. With transistor size reaching the atomic dimensions, vulnerability to unavoidable fluctuations in the manufacturing process and environmental stress rise dramatically. Failing to meet a reliability requirement may add excessive re-design cost to recover and may have severe consequences on the success of a product. %Worst-case design with large margins to guarantee reliable operation has been employed for long time. However, it is reaching a limit that makes it economically unsustainable due to its performance, area, and power cost. One of the open challenges for future technologies is building ``dependable'' systems on top of unreliable components, which will degrade and even fail during normal lifetime of the chip. Conventional design techniques are highly inefficient. They expend significant amount of energy to tolerate the device unpredictability by adding safety margins to a circuit's operating voltage, clock frequency or charge stored per bit. Unfortunately, the additional cost introduced to compensate unreliability are rapidly becoming unacceptable in today's environment where power consumption is often the limiting factor for integrated circuit performance, and energy efficiency is a top concern. Attention should be payed to tailor techniques to improve the reliability of a system on the basis of its requirements, ending up with cost-effective solutions favoring the success of the product on the market. Cross-layer reliability is one of the most promising approaches to achieve this goal. Cross-layer reliability techniques take into account the interactions between the layers composing a complex system (i.e., technology, hardware and software layers) to implement efficient cross-layer fault mitigation mechanisms. Fault tolerance mechanism are carefully implemented at different layers starting from the technology up to the software layer to carefully optimize the system by exploiting the inner capability of each layer to mask lower level faults. For this purpose, cross-layer reliability design techniques need to be complemented with cross-layer reliability evaluation tools, able to precisely assess the reliability level of a selected design early in the design cycle. Accurate and early reliability estimates would enable the exploration of the system design space and the optimization of multiple constraints such as performance, power consumption, cost and reliability. This Ph.D. thesis is devoted to the development of new methodologies and tools to evaluate and optimize the reliability of complex digital systems during the early design stages. More specifically, techniques addressing hardware accelerators (i.e., FPGAs and GPUs), microprocessors and full systems are discussed. All developed methodologies are presented in conjunction with their application to real-world use cases belonging to different computational domains.

Cross layer reliability estimation for digital systems / Vallero, Alessandro. - (2017). [10.6092/polito/porto/2673865]

Cross layer reliability estimation for digital systems

VALLERO, ALESSANDRO
2017

Abstract

Forthcoming manufacturing technologies hold the promise to increase multifuctional computing systems performance and functionality thanks to a remarkable growth of the device integration density. Despite the benefits introduced by this technology improvements, reliability is becoming a key challenge for the semiconductor industry. With transistor size reaching the atomic dimensions, vulnerability to unavoidable fluctuations in the manufacturing process and environmental stress rise dramatically. Failing to meet a reliability requirement may add excessive re-design cost to recover and may have severe consequences on the success of a product. %Worst-case design with large margins to guarantee reliable operation has been employed for long time. However, it is reaching a limit that makes it economically unsustainable due to its performance, area, and power cost. One of the open challenges for future technologies is building ``dependable'' systems on top of unreliable components, which will degrade and even fail during normal lifetime of the chip. Conventional design techniques are highly inefficient. They expend significant amount of energy to tolerate the device unpredictability by adding safety margins to a circuit's operating voltage, clock frequency or charge stored per bit. Unfortunately, the additional cost introduced to compensate unreliability are rapidly becoming unacceptable in today's environment where power consumption is often the limiting factor for integrated circuit performance, and energy efficiency is a top concern. Attention should be payed to tailor techniques to improve the reliability of a system on the basis of its requirements, ending up with cost-effective solutions favoring the success of the product on the market. Cross-layer reliability is one of the most promising approaches to achieve this goal. Cross-layer reliability techniques take into account the interactions between the layers composing a complex system (i.e., technology, hardware and software layers) to implement efficient cross-layer fault mitigation mechanisms. Fault tolerance mechanism are carefully implemented at different layers starting from the technology up to the software layer to carefully optimize the system by exploiting the inner capability of each layer to mask lower level faults. For this purpose, cross-layer reliability design techniques need to be complemented with cross-layer reliability evaluation tools, able to precisely assess the reliability level of a selected design early in the design cycle. Accurate and early reliability estimates would enable the exploration of the system design space and the optimization of multiple constraints such as performance, power consumption, cost and reliability. This Ph.D. thesis is devoted to the development of new methodologies and tools to evaluate and optimize the reliability of complex digital systems during the early design stages. More specifically, techniques addressing hardware accelerators (i.e., FPGAs and GPUs), microprocessors and full systems are discussed. All developed methodologies are presented in conjunction with their application to real-world use cases belonging to different computational domains.
2017
File in questo prodotto:
File Dimensione Formato  
Vallero PhD thesis.pdf

accesso aperto

Tipologia: Tesi di dottorato
Licenza: PUBBLICO - Tutti i diritti riservati
Dimensione 7.11 MB
Formato Adobe PDF
7.11 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/2673865
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo