Hardware Metrics

ISO 26262-5 defines three quantitative metrics for random hardware failures. The Single-Point Fault Metric (SPFM) measures how well the architecture is protected against single-point and residual faults - faults that violate a safety goal directly. The Latent Fault Metric (LFM) measures coverage of latent multi-point faults that sit undetected until a second fault completes the failure. The Probabilistic Metric for random Hardware Failures (PMHF) expresses the average probability of a safety goal violation per hour of operation. Clauses 8 and 9 of Part 5 require these evaluations for ASIL B, C, and D designs, with targets that tighten as the ASIL rises.

ISO 26262-5 sets SPFM targets of 90% or more for ASIL B, 97% or more for ASIL C, and 99% or more for ASIL D. LFM targets are 60% or more for ASIL B, 80% or more for ASIL C, and 90% or more for ASIL D. For PMHF, the reference targets are below 10⁻⁷ per hour for ASIL B and C and below 10⁻⁸ per hour for ASIL D. ASIL A has no quantitative targets. These are evaluation criteria for the whole safety goal, so the numbers roll up across every element that can contribute to its violation.

A Failure Modes, Effects, and Diagnostic Analysis (FMEDA) is the worksheet-level analysis that produces the hardware metrics. For every component you list its failure modes, split the failure rate across them, classify each mode as safe, single-point, residual, or multi-point, and assign the diagnostic coverage of the safety mechanism that detects it. Summing the classified failure rates gives SPFM and LFM directly, and feeds the PMHF calculation. In practice the FMEDA is the single most audited hardware safety work product, which is why the course builds one interactively before showing the shortcuts.

A FIT (failure in time) is one failure per 10⁹ device hours. Base failure rates come from reliability handbooks such as SN 29500 or IEC TR 62380, from semiconductor vendor data, or from field return statistics, each adjusted for mission profile, temperature, and duty cycle. The choice matters: handbook values are conservative and defensible, vendor and field data are more accurate but need justification. The course covers the main sources, the mission profile adjustments, and how to document the derivation so the numbers survive an assessment.

The guide spans 16 chapters covering fault classification, the FMEDA methodology, SPFM, LFM, and PMHF calculation, architecture comparison (1oo2, 2oo2, 1oo2D), FIT data sources, the EEC method, and semiconductor FMEDA. It ships with 3 interactive calculators, a step-by-step FMEDA process simulator, and 7 real-world worked examples from brake controllers to ADAS systems. A free account starts you off, and the Pro and Expert plans unlock more of the library.