Skip to content
DenseChip
Learn
PracticeCode labInterview

DenseChip

Navigate

Menu

LearnPracticeCode labInterview

Tools

All toolsSV editorCode repoResume
DenseChip

Practice, courses, and chip-design tools in one focused workspace.

LearnPracticeCode labInterviewToolsLearning paths

© 2026 DenseChip

AboutContactPrivacyTerms
Skip to lesson

Course contents

  • How to Use This Course
  • Reading Path
  • STA Glossary
  • STA Metrics Reference
  • STA Mastery Cheatsheet
  • Timing Paths and Checks16
  • Constraints & SDC Fundamentals20
  • Delay Graph & Path Analysis16
  • Beyond Setup and Hold16
  • Multi-Corner Multi-Mode Signoff16
  • OCV, AOCV, POCV and Variation16
  • Crosstalk and Noise16
    • Delta Delay Mechanism
      • Coupling and Aggressors
      • Switching Windows
      • Delta Delay Arithmetic
      • Setup Versus Hold Impact
    • Noise and Glitch Analysis
    • SI ECO Fix Order
    • SI Signoff Reports
  • ECO and Closure16
  • STA Debug Playbook16
  • STA Interview Prep26
  • Capstone Timing Signoff Lab4
Loading lesson

‹Delta Delay Mechanism
  1. Topics
  2. Static Timing Analysis
  3. Crosstalk and Noise
  4. Delta Delay Mechanism
  5. Delta Delay Arithmetic

Part 7 · Crosstalk and Noise · All levels

Delta Delay Arithmetic

A path's SI slack is its nominal slack minus the data-path deltas plus the launch-clock and capture-clock path deltas, with the common clock segment credited rather than applied in opposite directions; a few picoseconds per net can flip a marginal path.

On this page

    Why it matters

    SI deltas are small per net and add along a path. A path with 50 ps of slack in a quiet analysis can lose 20 ps to crosstalk and a marginal path can turn negative. The arithmetic is simple and worth doing by hand once.

    Worked example (picoseconds, illustrative)

    example.py
    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    period, setup = 500.0, 50.0
    def slack(cell, net, delta=None):
        delta = delta or [0] * len(net)
        arrival = sum(cell) + sum(net) + sum(delta)
        return period - setup - arrival
    # Path A: four cells and four nets
    cellA, netA, dA = [60, 80, 70, 90], [20, 35, 15, 30], [4, 9, 0, 6]
    nominalA, siA = slack(cellA, netA), slack(cellA, netA, dA)
    assert (nominalA, siA) == (50.0, 31.0)
    # Path B: same total SI delta, but less nominal slack
    cellB, netB, dB = [60, 85, 75, 95], [30, 40, 20, 33], [5, 8, 3, 3]
    nominalB, siB = slack(cellB, netB), slack(cellB, netB, dB)
    assert (nominalB, siB) == (12.0, -7.0)
    assert sum(dA) == sum(dB) == 19
    print("PASS: path A", nominalA, "->", siA, "ps; path B", nominalB, "->", siB, "ps")

    Path

    Nominal arrival (ps)

    SI delta (ps)

    Nominal slack (ps)

    Slack with SI (ps)

    A

    400

    19

    50

    31

    B

    438

    19

    12

    -7

    Both paths pick up 19 ps of crosstalk delay (the sum of the four per-net deltas). Path A has plenty of margin and remains positive. Path B, with 12 ps of nominal slack, becomes a violation of 7 ps: it passed a timing run without SI and fails with it. This worked example models data-path deltas only; a full check adds the launch-clock and capture-clock path deltas, crediting the common clock segment rather than applying it in opposite directions. This is the common pattern: SI creates new violations on paths that looked fine, mostly those that were already close to zero.

    What the review looks for

    • Setup and hold reports are produced with SI on; a clean run without SI is not signoff evidence.

    • The per-path SI delta (the sum of deltas) is visible in the report, so a reviewer can tell how much of a violation is crosstalk.

    • Paths whose violation is mostly crosstalk are candidates for the SI fixes in this section, before sizing the data path.

    Common pitfall

    Fixing a crosstalk-dominated violation by speeding up the data path. When a path fails by 7 ps only because 19 ps of crosstalk was added, removing a few aggressors can be cheaper and safer than resizing cells.

    Interview

    Give the formula and one example where the same SI delta turns one path negative and not another.

    Check yourself What is path B's slack if only the two largest deltas (8 ps and 5 ps) remain after a fix?

    12 - 13 = -1 ps: still a small violation, so the fix must also remove another delta or the path needs a sizing change.

    Related topics

    • Switching Windows

    • Next: Setup Versus Hold Impact

    Previous

    108 / 183

    Next: Setup Versus Hold Impact

    Up next

    Up nextSetup Versus Hold Impact

    Sign in to save progress

    Practice this lessonReport an error

    See also

    • Coupling and Aggressors