| With the rise and development of Artificial Intelligence of Things(AIo T)and hardware acceleration technologies,reconfigurable systems are becoming a major alternative to general purpose embedded devices and computers.As a flexible programmable hardware configuration architecture that can meet dynamically changing application scenarios,and due to its high parallelism and low power consumption,reconfigurable systems are widely used in aerospace,industrial control,smart medical,smart driving,weaponry and intelligent automation.However,with the dramatic reduction in Complementary Metal Oxide Semi-conductor(CMOS)process size,the complex and harsh deployment environment,and the use of untrustworthy third-party foundries to save development costs,such systems are highly susceptible to reliability threats such as environmental impacts,ageing degradation and malicious hardware attacks.Therefore,how to ensure the reliable operation of reconfigurable systems against the various failure/attack threats they face is a current research hotspot in the field.In this dissertation,we first focus on the ”reliability issues” in reconfigurable systems and analyze the various reasons for their occurrence.Three major reliability threats to reconfigurable systems are identified: 1)the complex and harsh deployment environment makes the system prone to various types of soft/hard failures,2)the aging degradation of the system during long-term operation generates permanent failures and increases the probability of transient failures,and 3)hardware attacks caused by hardware trojans planted by adversaries cause device faults/failures.Then,adaptive redundancy fault tolerance methods,age-aware layout mapping strategies,and runtime fault tolerance management frameworks,as well as protection strategies against hardware trojans attack,are proposed to ensure the reliability of reconfigurable systems under different types of threats and effectively extend the system life cycle.To this end,the following innovative research results have been accomplished.1.Redundant fault-tolerant design methods can ensure the reliable operation of systems in complex applications,but there is a problem of high resource overhead.Given this,we propose a dynamic adaptive redundancy fault-tolerant method that dynamically matches redundant architectures at the granularity of a single task based on changes in the external radiation environment,thereby balancing system resource overhead and high reliability requirements.Firstly,using the sensitivity of Block Random Access Memory(BRAM)to high-energy particle radiation,a radiation monitoring module based on BRAM is designed to periodically monitor the radiant intensity of the external environment; Secondly,by combining the environmental radiant intensity and task attributes and other information,a task criticality level assessment strategy is proposed to provide theoretical support for a single task to dynamically and adaptively match the redundant architecture.Finally,simulation experiments were conducted to confirm the effectiveness of the proposed method.Compared with mainstream fault-tolerant methods based on redundancy design,the average on-chip task completion increased by 57.2% under the same external radiation level conditions.2.For the reliability threats caused by aging degradation and faults faced by reconfigurable systems,we first investigate existing aging mitigation and fault tolerance methods and explores the advantages and disadvantages of various technologies.On this basis,it is pointed out that the current offline aging aware layout mapping strategy does not consider the reliability requirements of ensuring tasks and lacks fault-tolerant operations.In this regard,we propose a reliability framework that combines aging mitigation and fault tolerance for reconfigurable systems.First,the Integer Linear Programming(ILP)method is used to model the offline aging awareness layout mapping problem,to achieve an aging mitigation mapping layout and ensure the reliability requirements of the task.Secondly,a runtime fault-tolerant management architecture is proposed to achieve fault-tolerant management,thereby ensuring the highly reliable operation of the system.Finally,a large number of simulation experiments were designed and conducted to evaluate in detail the effectiveness and applicability of the proposed framework in terms of Mean Time to Failure(MTTF),resource usage,Short-term Job Completion Time(SJCT),and redundant resources.3.Hardware trojan implantation is a common hardware attack,some of which can cause severe faults or even circuit failures,thus having a serious impact on system reliability.However,current research on hardware trojan protection focuses on detection methods but lacks runtime mitigation strategies.Given this,this paper tries to propose a runtime hardware trojan protection strategy based on evolving hardware.When the presence of a hardware trojan is identified in a running circuit,an evolutionary algorithm is used to automatically traverse and search for new circuit structures to circumvent the use of hardware trojan-infected parts.Furthermore,the evolutionary process has an unacceptable delay,which is undesirable for most electronic systems.For this reason,we propose an optimized evolvable region algorithm to accelerate evolutionary convergence.Finally,experiments demonstrate the effectiveness and practicality of this hardware trojan mitigation mechanism and verify that the proposed optimization algorithm can improve the evolution efficiency by more than 30% on average when the redundant resources reach 1.5 times. |
