Two documents, one job split
NASA-STD-8719.13 addresses the who, what, when, and why of software safety — determining which software is safety-critical, scoping and tailoring the safety effort, documentation and traceability requirements, tool support, and waivers. It deliberately doesn't walk through specific techniques.
That job belongs to its companion, the NASA Software Safety Guidebook (NASA-GB-8719.13), which supplies the how-to — including detailed, step-by-step instructions for applying FMEA and Fault Tree Analysis to software specifically.
Software FMEA shows up twice
Paired with Software Fault Tree Analysis
The Guidebook treats SFMEA and SFTA as complementary halves of the same job, discussed together at both lifecycle points:
Bottom-up: start from a specific failure mode in a component or module, and ask what it does to the system. Good at finding failure modes nobody thought to ask about.
Top-down: start from an unwanted top-level event, and work backward to what combination of conditions could cause it. Good at proving a specific hazard is (or isn't) reachable.
The SFMEA process, per Appendix D
The Guidebook devotes an entire appendix to SFMEA specifically, laying out the process as a sequence:
Appendix D also addresses why to do an SFMEA and known issues with it — notably that, unlike hardware FMEA, software has no failure rate to anchor a quantitative severity/occurrence score to.
Scoped by risk, not run at full depth by default
How much SFMEA effort a project actually does is tied to the System Risk Index — a combination of hazard severity and likelihood established during the Preliminary Hazard Analysis:
Part of a bigger hazard-analysis chain
SFMEA doesn't stand alone — it's downstream of a Preliminary Hazard Analysis and a Software Subsystem Hazard Analysis that first establish which software is safety-critical at all. And like SAE1025, IEEE 1633, and the standards elsewhere in this series, the Guidebook is candid that software occurrence is "mostly qualitative, and as yet, not quantitatively expressed" — echoing the line it quotes from Dr. Nancy Leveson of MIT: software doesn't fail, it just doesn't perform as intended.