illegal_bins & ignore_bins

Two exclusions with opposite personalities

ignore_bins and illegal_bins both remove values from the coverage goal, but they react to a hit in opposite ways. A sampled value landing in an ignore_bin is silently discarded — no error, no count, no effect on percentage. The same value landing in an illegal_bin raises a runtime error the moment it is sampled. ignore says “out of scope”; illegal says “this must never happen, and if it does, fail loudly”. illegal_bins also take precedence: a value claimed by both an ordinary bin and an illegal_bin is illegal.

WHAT HAPPENS WHEN A SAMPLE LANDS IN...

                    ordinary bin    ignore_bins      illegal_bins
  hit count         increments      discarded        runtime ERROR
  in goal math      YES             removed from     removed from
                                    numerator AND    goal (and sim
                                    denominator      may abort)
  meaning           scenario        out of scope     must never occur
  on hit            progress ✓      nothing          test FAILS

The percentage math, worked

ignore_bins shrink the denominator — which means ignoring values makes the remaining coverage easier to close. That is exactly what you want for genuinely out-of-scope values, and exactly the abuse vector reviewers watch for.

covergroup cg with function sample(bit [2:0] mode);
  cp_mode : coverpoint mode {
    bins m[] = {[0:7]};            // start: 8 goal bins
    ignore_bins  rsv  = {6};        // reserved encoding, out of scope
    illegal_bins bad  = {7};        // spec says: must never occur
  }
endgroup

GOAL MATH, STEP BY STEP

  declared value bins:    m[0]..m[7]            8 bins
  ignore_bins {6}:        m[6] removed          7 remain
  illegal_bins {7}:       m[7] removed          6 goal bins

  simulation hits values: 0,1,2,4   → 4 goal bins hit

  coverage = 4 / 6 = 66.7%      ◄── NOT 4/8 (50.0%)

  later, a sample of mode==7 arrives:
    → runtime error from illegal_bins "bad"; test FAILS
    → coverage math unchanged (illegal never counts)

  a sample of mode==6:
    → silently discarded; still 66.7%

Is illegal_bins doing the checker's job?

There is a genuine methodology debate here, and interviewers love it. The case against illegal_bins as checkers: error detection belongs in assertions and scoreboards, which run on every cycle or every transaction regardless of sampling; an illegal_bin only fires if the covergroup happens to sample at the guilty moment, fires once where an assertion gives waveform-linked debug, and buries a check inside a measurement construct where nobody looks for it. The case for : the covergroup is already sampling exactly the fields in question, the illegal encoding is part of the same value-space story the bins tell, and a one-line illegal_bin documents the constraint precisely where a reviewer reads the bin plan. Pragmatic resolution: assert the property in the RTL or interface as the primary check, and keep the illegal_bin as documentation plus a redundant tripwire — defense in depth, with the assertion as the load-bearing layer.

Reserved-value exclusion patterns

covergroup cg with function sample(bit [3:0] burst, bit [2:0] size,
                                   bit [7:0] id);
  cp_burst : coverpoint burst {
    bins fixed = {0};
    bins incr  = {1};
    bins wrap  = {2};
    ignore_bins reserved = {[3:15]};    // spec: encodings 3-15 reserved
  }
  cp_size : coverpoint size {
    bins b[] = {[0:7]};
    // this DUT's bus is 64-bit: sizes above 3 unreachable by config
    ignore_bins over_width = {[4:7]};
  }
  cp_id : coverpoint id {
    bins used[4] = {[0:127]};
    ignore_bins out_of_pool = {[128:255]}; // ID pool limited by env cfg
  }
endgroup

• Reserved spec encodings → ignore_bins, with a comment citing the spec section.

• Configuration-unreachable values (narrower bus, smaller ID pool) → ignore_bins per build configuration.

• Must-never-occur protocol violations → assertion first, illegal_bins as documented backup.

• Every ignore_bins needs a justifying comment — uncommented ignores are presumed coverage-goal gaming until proven otherwise.

Key takeaways

• ignore_bins silently remove values from numerator and denominator; illegal_bins error at runtime on a hit.

• ignore_bins make coverage easier to close — legitimate for out-of-scope values, an abuse vector otherwise.

• illegal_bins only fire when the covergroup samples — assertions are the primary checker; illegal_bins are backup.

• Work the goal math: 8 bins, 1 ignored, 1 illegal → percentage is hits over 6, not 8.

Common pitfalls

• Using ignore_bins to bury bins the team failed to hit before a deadline — false closure, audit risk.

• Relying on illegal_bins as the only protocol check — violations between samples go completely undetected.

• Forgetting illegal_bins precedence — a value in both a named bin and an illegal_bin errors; it is not covered.

• Uncommented ignore_bins — reviewers cannot distinguish out-of-scope from coverage-gaming.