Fault tolerance and redundancy for reliability

Reliability is the probability that a system performs its required function for a specified time under stated conditions. Fault tolerance keeps the system working in spite of faults, using three approaches.

Careful design (fault avoidance)

Prevent faults from entering the design:

• Use high-quality components with derating (operate at 50-70 percent of rated stress), and design for worst-case tolerance and temperature.
• Keep the design simple with fewer components and connectors (fewer failure points).
• Add protection: fuses, TVS, reverse-polarity protection, watchdog timers and brown-out reset.
• Follow EMC and layout rules; thermal design and burn-in testing; design reviews and standards.

Testable architectures

A fault that cannot be detected cannot be handled, so the design must be testable:

• Provide test points, scan chains/boundary scan (JTAG), and built-in self test (BIST) at power-up.
• Use parity, CRC, checksums to detect data errors, and watchdog timers to detect software hangs.
• Provide diagnostics (LED/status codes), and partition the system into modules so that faults can be isolated.
• Monitor supply voltages and temperature.

Redundant architectures

Redundancy adds extra resources so that a failure does not cause system failure:

1. Hardware redundancy:
   • Passive (static): Triple Modular Redundancy (TMR): three identical modules and a majority voter mask one fault.
   • Active (dynamic): a standby spare is switched in on fault detection (hot or cold standby).
   • Hybrid: TMR plus spares.
2. Information redundancy: error-correcting codes (Hamming) and CRC.
3. Time redundancy: repeating the operation and comparing the results (catches transient faults).
4. Software redundancy: N-version programming and recovery blocks.

 In -> [M1] --\
 In -> [M2] ---> [Voter] -> Out   (TMR)
 In -> [M3] --/

Redundancy costs more in size, power and complexity, so it is used where the cost of failure is high (medical, aerospace, industrial control).