Jump vs Normal Reset: When to Use Each

Decision matrix

[PHASE][UVM] reset strategy selection

power-on reset (POR) only:
  → single-pass reset_phase, NO jump
  → separate directed test per reset corner

mid-life soft reset / recovery:
  → phase.jump to reset_phase in one test
  → verifies continuity of simulation state

multiple independent reset corners:
  → separate tests often simpler than N jumps
  → jump when interaction BETWEEN resets matters

Prefer separate tests when

• Each reset corner is independent — no cross-reset interaction.

• Jump debugging cost exceeds value for simple directed cases.

• Regression farm needs parallelizable atomic tests.

Prefer phase jump when

• DUT must recover mid-scenario with state carried from earlier traffic.

• Soft-reset feature requires before/after traffic in one context.

• Power-fail recovery sequence is one continuous use case.

Complexity comparison

[PHASE][UVM] cost model

normal reset only:
  complexity: LOW
  debug:      standard phase trace

phase jump:
  complexity: HIGH
  requires:   drain, idempotent config, flush scoreboards
  debug:      phase trace + jump markers + txn counters

Key takeaways

• Default to single-pass reset; add jump only for genuine mid-life recovery.

• Independent reset corners → separate tests are often cheaper.

• Jump pays off when cross-reset interaction is the feature under test.

Common pitfalls

• Using jump for every reset directed test — unmaintainable.

• Restarting sim manually when jump would preserve context cheaper.

Test plan template

1. POR smoke — no jump, verify reset_phase on all agents.

2. Soft-reset directed — jump test with pre/post traffic checks.

3. Recovery error injection — jump after induced fault.

4. Regression: run jump test nightly; POR corners in parallel shards.