FPGA故障容错方法综述：从架构冗余到实时自适应

需积分: 1 22 浏览量更新于2024-07-27 收藏 897KB PDF 举报

身份认证购VIP最低享 7 折!

领优惠券(最高得80元）

本文档是一篇关于FPGA（Field-Programmable Gate Array）故障容错方法的综述。FPGA作为可编程逻辑器件，因其灵活性和广泛应用在各种领域，如航空航天、通信和数据中心，对故障容忍度的需求日益增长。作者Jason A. Chatham和John M. Emmert来自 Wright State University，而Stan Baumgart则来自University of North Carolina, Charlotte，他们共同探讨了当前FPGA故障容忍技术的广泛研究。文章概述了多种不同的故障容忍策略，包括但不限于： 1. 简单的架构冗余：这种方法通过在关键逻辑路径上增加备份组件，如双重模块化或三取二（2x2N）设计，来提高系统的可靠性。这种策略主要针对制造过程中的缺陷，旨在提高芯片的总体良率。 2. 完全在线自适应实现：这些方法采用更高级的技术，如动态错误检测和纠正（DED/C）电路，以及自愈合或自我修复功能，能够在运行时检测并纠正错误，增强了系统在实际应用中的稳定性。 3. 故障隔离与恢复：通过设计冗余路径并利用硬件或软件逻辑来隔离故障区域，然后切换到备用路径，确保系统的连续性。 4. 冗余内存和数据一致性管理：在涉及大量数据处理的FPGA应用中，确保数据的一致性和持久性是关键，这可能涉及到分布式内存管理和一致性协议。 5. 软件级别的故障管理：除了硬件层面的措施，也研究了软件层面的故障管理策略，如实时操作系统（RTOS）和自适应编译器优化，以减少故障影响。作者们在比较这些方法时，使用了一些通用的评估指标，如故障覆盖率、修复时间、性能开销和成本效益。他们通过详细的描述和量化分析，让读者能够全面理解每种方法的优势和局限性。总结来说，本文的目的是为FPGA设计者、研究人员和工程师提供一个全面的参考，帮助他们选择最适合特定应用场景的故障容忍策略，从而提高系统的可靠性和整体性能。对于那些关注FPGA系统可靠性的人来说，这篇文章是深入了解当前技术趋势和挑战的重要资源。

资源详情

资源推荐

506

•

J. A. Cheatham et al.

Fig. 3. Conﬁguration of FPGA (a) before and (b) after a defect is tolerated [Hatori et al. 1993].

Fig. 4. Vertical routing resources (a) without and (b) with redundancy [Hatori et al. 1993].

There is no reconﬁguration time penalty for this method since reconﬁgura-

tion is performed during manufacture. Only logic faults in a single row may be

tolerated by this method, but multiple faults may occur in that row. No provision

is made for interconnect fault tolerance.

2.2 Reconﬁguration Network

Kelly and Ivey [1994] proposed a method for fault tolerance whereby faulty

resources could be bypassed by using a secondary interconnect network. The

method is described with relation to an idealized Xilinx XC3000 architecture,

but the authors state that it is applicable to all SRAM based FPGAs. Their ar-

chitecture has PLBs, a general interconnect matrix with switchboxes at cross-

points, and long lines. A PLB and the switchbox at its northwest corner are

treated as a single unit, and each unit has the set of long lines immediately to

its right associated with it. Four architectural changes are required to imple-

ment fault tolerance:

ACM Transactions on Design Automation of Electronic Systems, Vol. 11, No. 2, April 2006.

A Survey of Fault Tolerant Methodologies for FPGAs

•

507

Fig. 5. A PLB/switch pair is shown in black. Alternate routing resources for pair i are shown by

the dotted lines. They are (1) connect left; (2) connect right; (3) switch bypass; (4) long line spur; (5)

horizontal direct bypass; (6) vertical direct bypass. The small switch (7) controls the reconﬁguration

of switch S

—it can be used to bypass the switch directly above S

, to disconnect S

from the switch

directly above it, or to bypass S

horizontally.

(1) Additional routing resources (see Figure 5)

—direct connection bypasses bypass a faulty PLB both horizontally and

vertically

—general interconnect bypasses connect to the general interconnect of the

column to the right or left of a switch, bypass a a switch horizontally, or

disconnect a switch from the switch directly above it

—vertical long line “spurs” connect to a PLB to the right of a faulty PLB.

(2) Reconﬁguration memory to store the locations of faulty resources.

(3) On-chip circuitry to perform reconﬁguration routing (simple bypasses).

(4) Addition of one or more columns of PLBs.

There are three steps to this methodology. First, a custom testing process

is run on the FPGA to identify faulty resources; the results become a fault

map for the device. The fault map is loaded into the conﬁguration memory of

the FPGA and is used by an on-board router to set up a routing network that

avoids the faulty resources. The network is constructed by ﬁrst shifting all PLBs

in the fault-containing row towards the right (starting from the faulty PLB).

Connections are maintained between PLBs with the Reconﬁguration Network;

diagonal connections are made to the shifted PLBs, any necessary bypass con-

nections are made over the faulty row, and spur connections are made from

the shifted PLBs to maintain connections to their original vertical global rout-

ing. Once routing reconﬁguration is complete, the actual circuit conﬁguration

is loaded onto the chip. Because of the on-board reconﬁguration router, no

ACM Transactions on Design Automation of Electronic Systems, Vol. 11, No. 2, April 2006.

剩余32页未读，继续阅读

seanzhou12

粉丝: 0
资源: 4

FPGA故障容错方法综述：从架构冗余到实时自适应

A Survey of Fault Diagnosis and Fault-Tolerant Techniques 1

qt5.9为什么报错:fault tolerant heap shim applied to current process

FTH: (11616): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *** 程序异常结束。

Fault tolerant heap shim applied to current process. This is usually due to previous crashes.

FTH: (22440): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

FTH: (1220): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

FTH: (12924): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

FTH: (18176): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***.

qt编译运行时FTH: (340): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

以下报错：FTH: (14000): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

FTH: (5888): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

FTH: (10896): *** Fault tolerant heap shim applied to current process. This is usually due to previous crashes. ***

fault-tolerant network

Apache Mesos frameworks

redis cluster

c. fault-tolerant network

QEMU中Segmentation fault (core dumped)

A segmentation fault is a common run-time error for C programs. Use examples (several lines of C code for each cause) to show two causes of this problem.

最新资源

FTH: (11616): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. * 程序异常结束。

FTH: (22440): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

FTH: (1220): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

FTH: (12924): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

FTH: (18176): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *.

qt编译运行时FTH: (340): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

以下报错：FTH: (14000): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

FTH: (5888): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *

FTH: (10896): * Fault tolerant heap shim applied to current process. This is usually due to previous crashes. *