X-Pessimism vs X-Optimism

Chapter 6 · Section 6.2 · X-Propagation

Project thread — the FSM X from 5.5: is it a real bug (RTL optimism hid it) or gate pessimism (a false X)? This lesson gives the framework; 6.6 makes the call.

1. Why Should I Learn This?

Every gate-level X forces the same judgement: real bug, or pessimism? Getting it wrong is costly both ways.

• X-pessimism → false Xs; treat them all as bugs and you burn days.
• X-optimism → hidden Xs; trust RTL alone and a real bug escapes.
• RTL optimistic, gate pessimistic → the reason GLS finds what RTL misses (6.4).

This is the interpretive lens for the whole chapter (and for the FSM X in 6.6).

2. Real Silicon Story — the RTL "pass" that optimism produced

A design passed RTL simulation cleanly. Gate-level simulation then showed X propagating from a mux — filed as "gate-level pessimism, ignore."

It was not pessimism. The mux select was genuinely uninitialized. RTL's if (sel) ... else ... was X-optimistic — with sel = X it silently took the else branch, producing a definite value and hiding the uninitialized select. The gate-level mux was X-pessimistic — it propagated X, correctly flagging the real problem. Dismissing it as pessimism nearly let an uninitialized-control bug through.

Lesson: an X that appears in GLS but not RTL is often a real bug that RTL's optimism hid — not gate pessimism. Judge each on its source, not on which sim shows it.

3. Concept — the two failure modes, and the RTL/gate asymmetry

X-pessimism (safe but noisy):

• Model emits X when real HW would resolve to a definite value.
• Example: a UDP outputs X for an ambiguous input combination silicon would settle (2.3).
• Cost: false Xs → wasted debug, over-constrained analysis.

X-optimism (dangerous):

• Model hides an X that real uncertainty should produce.
• Example: if (sel) a; else b; with sel = X takes else → a definite value, masking the unknown.
• Cost: missed bugs → real uninitialized/uncertain logic looks fine.

The asymmetry (why it matters):

• RTL tends X-optimistic — behavioural if/case/?: resolve unknowns to a branch.
• Gate cell models tend X-pessimistic — UDPs propagate X (2.3).
• So a bug hidden by RTL optimism can be exposed by gate pessimism (6.4).

The anchor:

• Neither is silicon truth. Real HW resolves electrically to some level; X and its handling are modelling choices.
• Goal: catch real X (don't dismiss it) while managing pessimism (don't drown in false X, 6.5).

4. Mental Model — the over-cautious clerk vs the guesser

Two clerks handle an unclear order.

• The pessimist stamps everything uncertain as "UNKNOWN" — never wrong, but floods you with flags, some for orders that were actually fine (false X). That's the gate UDP.
• The optimist guesses an answer whenever unsure and moves on — tidy output, but a genuinely bad order slips through labelled "fine" (hidden X). That's RTL.
• Reality (silicon) always resolves to a definite outcome — but not necessarily the optimist's guess.

You want the pessimist's honesty (see the uncertainty) without drowning in its noise — that balance is X-propagation practice (6.5).

5. Working Example — the same unknown, optimistic vs pessimistic

The same mux, two behaviours on an unknown select:

// X-OPTIMISTIC (RTL behavioural): if/else picks a branch on X -> HIDES the unknown
assign y_rtl = sel ? a : b;      // sel = X -> often resolves to b -> DEFINITE (masks bad sel)
 
// X-PESSIMISTIC (gate UDP): propagates X when sel unknown and a != b -> EXPOSES it
// MUX2_udp row:  a=0 b=1 sel=x : x   (2.3) -> y_gate = X (flags the uninitialized select)

Practical context (representative, tool-neutral):

# Triage an X that appears in GLS but not RTL:
#   1) find the X SOURCE (6.1) — uninitialized? contention? model?
#   2) is the source a REAL unknown (bad reset, uninit control)?  -> RTL OPTIMISM hid a real bug
#   3) or would real HW resolve here (ambiguous UDP input)?       -> gate PESSIMISM (false X)
#   Default: investigate as REAL first; prove pessimism, don't assume it.

Same select, optimistic vs pessimistic, as a real waveform:

6. Debugging Session — is the GLS-only X real or pessimism?

7. Common Mistakes

• Dismissing every GLS-only X as pessimism. Often a real bug RTL optimism hid.
• Trusting an RTL "pass" on unknowns. RTL if/case can hide X (optimism).
• Treating every gate X as a real bug. UDP pessimism produces false X too (2.3).
• Calling either model "silicon truth." Both are modelling; silicon resolves electrically.
• Not tracing the X source. Source (6.1) is what settles real-vs-pessimism.

8. Industry Best Practices

• Triage GLS-only X by source — real unknown vs genuinely-resolvable.
• Distrust RTL optimism on unknown selects/conditions — use xprop (6.5) to shift-left.
• Manage, don't dismiss, pessimism — reduce it deliberately (6.5), not by ignoring X.
• Default to "real first" — prove pessimism before waiving an X.
• Remember X is a model value — a flag, not a silicon voltage.

Senior Engineer Thinking

• Beginner: "GLS shows X but RTL is clean — must be gate pessimism, ignore it."
• Senior: "RTL is optimistic — it hides X on unknown selects. Where does this gate X come from? If the select is really uninitialized, RTL lied and the gate is right. I prove pessimism; I don't assume it."

The senior treats a GLS-only X as probably real until the source proves pessimism.

Silicon Impact

The optimism/pessimism asymmetry is why gate-level simulation catches bugs RTL misses — and mishandling it cuts both ways at tape-out. Dismiss a real X as "pessimism," and an uninitialized-control or bad-reset bug reaches silicon, where it manifests as a power-up-dependent, intermittent failure (0.3) — the exact class RTL optimism is prone to hide. Conversely, treat unavoidable UDP pessimism as all-real and you drown real bugs in false X and may waive checks to cope. The discipline — judge each X by source, default to real, manage pessimism deliberately (6.5) — is what makes GLS's X findings trustworthy and protective.

Engineering Checklist

• For each GLS-only X, traced its source (6.1) before judging.
• Defaulted to "real bug" until pessimism is proven.
• Recognised RTL optimism on unknown selects/conditions.
• Managed pessimism deliberately (6.5), not by ignoring X.
• Treated X as a modelling flag, not silicon truth.

Try Yourself

1. Simulate a mux with an uninitialized select in RTL (sel ? a : b) — the output is a definite value (optimism).
2. Observe: RTL hides the bad select.
3. Change: run the gate-level mux (UDP) with the same uninitialized select.
4. Expect: the gate output is X (pessimism correctly flagging a real bug). Then initialize sel and confirm both agree — proving the X was real, not noise.

Any free Verilog simulator shows the RTL-vs-gate divergence with a UDP mux. No paid tool required.

Interview Perspective

• Weak: "Gate-level X is just pessimism you can ignore."
• Good: "X-pessimism emits X where hardware resolves (noisy); X-optimism hides a real X (dangerous). RTL tends optimistic, gate models pessimistic."
• Senior: "That asymmetry is why GLS catches bugs RTL hides — RTL if/case resolve unknowns to a branch, gate UDPs propagate X. So a GLS-only X is often a real bug RTL optimism masked. I triage by source, default to real, and manage pessimism deliberately. Neither model is silicon truth — X is a flag."

9. Interview / Review Questions

10. Key Takeaways

• X-pessimism: a model emits X where real hardware would resolve to a definite value — safe but noisy (false X, wasted debug); gate-level UDPs tend pessimistic (2.3).
• X-optimism: a model hides an X that real uncertainty should produce — dangerous (masks bugs); RTL tends optimistic (if/case/?: pick a branch on X).
• The asymmetry (RTL optimistic, gate pessimistic) is why GLS exposes bugs RTL hides (6.4).
• Neither is silicon truth — real hardware resolves electrically; X and its handling are modelling behaviours.
• Triage a GLS-only X by its source (6.1), defaulting to real — prove pessimism, don't assume it; then manage pessimism deliberately (6.5). Next: 6.3 — how X flows through muxes, flops & latches.

Quick Revision

X-pessimism = emits X where HW resolves (safe, noisy; gate UDPs). X-optimism = hides a real X (dangerous; RTL if/case). Asymmetry (RTL optimistic, gate pessimistic) = why GLS finds bugs RTL hides. Neither is silicon truth. A GLS-only X is often real — triage by source (6.1), default to real, prove pessimism. Next: 6.3 — X through muxes, flops & latches.