Why it matters

A floating test pin put a whole product line into scan on a noisy board. A tied-off address bit aliased high memory onto low. An unconnected error output hid the very failure the bring-up team was hunting. On NanoSoC the lines that matter are smaller still: the two reserved interrupt inputs, and the habit that connected the DMA interrupt to a spare input instead of input 1. Ties and unconnected ports are the smallest integration decisions and the most commonly skipped, because each one looks harmless. This lesson makes them explicit: the tie rules, the unconnected-port waivers, and an audit built from the netlist description itself, so the lists cannot drift from what is built.

The core idea

Three rules govern every top-level port. First, no input floats: every input has a driver, a reviewed slice, or a tie-list entry with a value and a reason, and the checker from the previous lesson enforces it. Second, ties default to the safe value for their meaning: interrupts and enables tie to inactive, which on NanoSoC means the reserved controller inputs 6 and 7 tie to zero. Third, unconnected outputs are waived individually, never by wildcard: each names the signal, the reason no logic needs it, and the reviewer who agreed. A tie is only a tie where a real input exists: a dropped address slice or an absent stream sideband is not a tie, because there is nothing to wire (the slice is recorded in the reviewed-slice list, the absence in the connectivity table). The audit below compares the ties and unconnected outputs found in the netlist description with the lists in the review, as a set difference in both directions, and requires both differences to be empty. It then mutates the netlist to show each kind of drift is reported.

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

# NanoSoC teaching values: audit built from the netlist, not from the list itself.
TIE_LIST = {"u_intc.irq6": (0, "reserved controller input, tied inactive"),
            "u_intc.irq7": (0, "reserved controller input, tied inactive")}
WAIVERS = {}        # unconnected output -> (reason, reviewer); none in the baseline
NETLIST = {"tied": {"u_intc.irq6": 0, "u_intc.irq7": 0}, "unconnected_out": []}

def audit(netlist, tie_list, waivers):
    bad = []
    found, listed = set(netlist["tied"]), set(tie_list)
    bad += ["tie not listed: " + s for s in sorted(found - listed)]
    bad += ["listed tie not in netlist: " + s for s in sorted(listed - found)]
    for sig in sorted(found & listed):
        if netlist["tied"][sig] != tie_list[sig][0]:
            bad.append("tie value differs: " + sig)
    unc, waived = set(netlist["unconnected_out"]), set(waivers)
    bad += ["unconnected, no waiver: " + s for s in sorted(unc - waived)]
    bad += ["waiver for connected port: " + s for s in sorted(waived - unc)]
    bad += ["wildcard waiver: " + s for s in sorted(waivers) if "*" in s]
    for sig, (val, reason) in tie_list.items():
        if sig.split(".")[1].startswith("irq") and val != 0:
            bad.append("spare interrupt not tied to 0: " + sig)
    return bad

assert audit(NETLIST, TIE_LIST, WAIVERS) == []                      # baseline clean

n = {"tied": {**NETLIST["tied"], "u_dma.test_mode": 0}, "unconnected_out": []}
assert audit(n, TIE_LIST, WAIVERS) == ["tie not listed: u_dma.test_mode"]
n = {"tied": {"u_intc.irq6": 1, "u_intc.irq7": 0}, "unconnected_out": []}
assert "tie value differs: u_intc.irq6" in audit(n, TIE_LIST, WAIVERS)
# a block revision adds an error output nobody connects
n = {"tied": NETLIST["tied"], "unconnected_out": ["u_dma.err_out"]}
assert audit(n, TIE_LIST, WAIVERS) == ["unconnected, no waiver: u_dma.err_out"]
w = {"u_dma.*": ("all outputs", "integrator")}                      # wildcard attempt
n2 = {"tied": NETLIST["tied"], "unconnected_out": []}
assert "wildcard waiver: u_dma.*" in audit(n2, TIE_LIST, w)
print("PASS: lists match the netlist both ways; drift, new unconnected output and wildcard are caught")

What the review looks for

• The tie list and the no-connect list read aloud at the review, each entry with its reason; silence means consent, recorded.

• Reserved and spare lines tied to their inactive value, verified in the audit, not by inspection of the schematic.

• No wildcard waivers anywhere in the lint setup; each waiver names one signal and one reviewer.

Verify and debug

In simulation, force each tied interrupt and confirm no handler runs, and add a monitor that fires if any listed tie changes value; a tie that toggles was never a tie, it was an undriven net with an opinion. For the physical tie-cell side see Tie Cells and Tie-Off Hygiene.

Related topics

• A Python Connectivity Checker

• Next: Top-Level SDC and Constraint Merge

• PD: Tie Cells and Tie-Off Hygiene