Why it matters

Crossings close not when the RTL is written but when the review can answer five questions with evidence: is the inventory complete, is each mechanism right, do the constraints match, were the ratios swept, and is signoff clean. An unchecked crossing is a field failure with a long fuse, so the checklist is explicit and the evidence is attached, not promised. This lesson is that checklist. Its one job is the review; everything it checks was taught in the two previous lessons.

The core idea

First, inventory: the CDC tool's found crossings diffed against the ten rows, with any extra row explained and any missing row convicted as a tool-setup bug. Second, mechanism: each row's class-to-mechanism mapping restated, with depth arithmetic for the FIFOs and hold times for the levels. Third, constraints: the pair is declared asynchronous in the constraints (one clock-group declaration), the only extra constraints are the bounded-delay and skew limits on FIFO pointers and handshake data, and no other false path or multicycle crosses the pair; the exception audit from the switching lesson is rerun here. Fourth, ratio sweeps: each crossing simulated at the worst phase alignment, because NanoSoC ratios are fixed but phase is not. Fifth, signoff: the CDC methodology run clean with waivers that carry evidence and expiry, following CDC Signoff Methodology, and reset-domain crossing checked alongside, since reset releases cross these same boundaries.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39

# Audit model: the inventory every review re-runs. Each row has from, to and
# a mechanism, and the mechanism must match the signal class.
DOMAINS = {"clk_cpu", "clk_sys", "clk_per", "pad"}
INVENTORY = [
    ("X1", "clk_cpu", "clk_sys", "fifo", "multi"),
    ("X2", "clk_sys", "clk_cpu", "fifo", "multi"),
    ("X3", "clk_sys", "clk_per", "handshake", "multi"),
    ("X4", "clk_per", "clk_sys", "handshake", "multi"),
    ("X5", "clk_sys", "clk_per", "level-2flop", "single"),
    ("X6", "clk_per", "clk_cpu", "level-2flop", "single"),
    ("X7", "clk_per", "clk_sys", "level-2flop", "single"),
    ("X8", "clk_per", "clk_cpu", "level-2flop", "single"),
    ("X9", "pad", "clk_per", "level-2flop", "single"),
    ("X10", "clk_per", "clk_sys", "level-2flop", "single"),
]
OK = {"multi": {"fifo", "handshake"}, "single": {"level-2flop"}}

def audit(rows):
    errors = []
    if len({r[0] for r in rows}) != len(rows):
        errors.append("duplicate crossing id")
    for x, src, dst, mech, cls in rows:
        if src not in DOMAINS or dst not in DOMAINS:
            errors.append(x + ": clock not in the domain table")
        if src == dst:
            errors.append(x + ": not a crossing")
        if mech not in OK[cls]:
            errors.append(x + ": " + mech + " is wrong for a " + cls + "-bit signal")
    return errors

assert audit(INVENTORY) == []
kinds = [m for _, _, _, m, _ in INVENTORY]
assert (kinds.count("fifo"), kinds.count("handshake"), kinds.count("level-2flop")) == (2, 2, 6)
assert len(INVENTORY) == 10
# The audit can fail: a bus on two flops, an unknown clock, a same-clock row.
assert audit([("X1", "clk_cpu", "clk_sys", "level-2flop", "multi")]) != []
assert audit([("X1", "clk_cpu", "clk_dma", "fifo", "multi")]) != []
assert audit([("X1", "clk_sys", "clk_sys", "fifo", "multi")]) != []
print("PASS: 10 crossing groups, mechanisms match signal class; 3 bad rows rejected")

What the review looks for

• The tool-versus-table diff with zero unexplained rows, attached to the review record.

• The FIFO depth lines, the level hold lines and the exception audit, each current.

• A clean signoff report or waivers with evidence and expiry; a waiver without either is an open item, not a pass.

Verify and debug

Re-run the audit after every hierarchy or constraint change; crossings rot when constraints are edited without it. Treat a late-appearing eleventh crossing as a process failure, not a circuit problem: find which edit added the signal and why the inventory did not move with it, then fix the process that allowed the split.

Related topics

• Choosing a Mechanism per Crossing

• CDC and RDC: CDC Signoff Methodology

• Release Order and Timing