Skip to content

Adopting MEMO

Start small and keep it typed. Use the ontology and the methodology first; introduce automation and UI support once the project workflow has stabilised.

The first win is shared meaning, not tool replacement.

Start with one scenario-driven thread

Do not begin by modelling the device. Begin with one clinically meaningful use case, turn it into a concrete scenario, and walk that single scenario through every discipline. One thread that reaches evidence is worth more than four layers that reach nothing.

Step What you produce
1 · Use case The clinical goal — for example, patient-controlled bolus during active infusion
2 · Scenario One concrete path: lockout, alarm, timing, interface, or a user action that changes
3 · Operational analysis Clinical context, actors, therapy state, system response
4 · Functional flow Functions, mode transitions, alarms, timing, user and system actions
5 · Logical architecture Responsibilities, software items, interfaces, SOUP, data paths
6 · Implementation and realization Hardware/software boundary, sensors, actuators, UI, network, and deployment
7 · Change impact Affected requirements, hazards, interfaces, tests, and decisions
8 · Evidence thread Verification results, validation claims, DHF and RMF evidence state

One closed thread shows this walk on a real scenario from operational intent to evidence.

Cross-cutting work that runs alongside

These are not a ninth step. They accompany every step above:

  • Risk (ISO 14971 context) — identify hazard, hazardous situation, harm, risk control, verification, residual risk.
  • Software safety (IEC 62304 context) — classify, derive software requirements, update architecture, verify change impact.
  • Cybersecurity — identify assets, trust boundaries, threat scenarios, mitigations, and their safety impact.
  • V&V — define acceptance criteria, execute, and tie results back to the claims they support.

Four rules for the pilot

1. One product slice. Model one safety thread completely rather than every layer partially. The methodology scope determines which layers and modules the project includes.

2. Keep the core stable. Extend in your own package or profile. Device modes, organisation-specific kinds, and project vocabulary belong outside memo::core, which stays small so upgrades stay cheap.

3. Methodology before tooling. Agree the rules and the viewpoints your team will actually use before you automate anything. A rule nobody agreed to is a build failure nobody will fix.

4. Add automation gradually. Prove the value of the shared model first. Validation, CI, and visual review can be added without changing the ontology content.

Adopt by team context

MEMO should fit the team's current maturity, not the other way round.

Startup — keep it lightweight

  • Start with intended use and one safety thread.
  • Model only the key architecture, interfaces, and controls.
  • Add verification cases and lightweight review views.

Enterprise — bridge existing tools

  • Pick one change-impact problem worth solving.
  • Reference the requirements and risks you already have.
  • Connect architecture to evidence before integrating tools.

Review-driven — give each role a view

  • Architecture sees responsibilities and impact paths.
  • Safety sees controls and residual risk.
  • V&V sees targets, evidence state, and release impact.

The enterprise pattern is worth stressing: you do not have to migrate anything. Referencing existing requirements and risks from a MEMO model is a legitimate first step, and it is usually the one that demonstrates value fastest, because change impact is where document-based traceability hurts most.

What good looks like after one quarter

  • One scenario is modelled from use case to evidence.
  • The selected SysML v2 environment validates the model in CI and the team fixes what it reports.
  • At least one review question that used to take a day of searching is answered by querying the model.
  • The core ontology is unmodified; everything project-specific lives in your own package.

If those four hold, the thread can be widened. If they do not, widening the model will not help.

Next