Ocean Netlist Regen

Ocean run() Reliability & Netlist Regeneration

Problem

Ocean's run() silently returns nil. All measurement results are nil. Spectre never runs. No error is reported.

Context / Trigger Conditions

• run() completes quickly and returns nil instead of the resultsDir path
• No spectre.out or stale one in psf directory
• createNetlist(?recreateAll t) returns nil
• Previously called sim setup or changed resultsDir to a different path
• Previously working simulation stops producing results

Root Cause 1: resultsDir Change Breaks run() ⚠️ MOST COMMON

The ADE session binds run() to the resultsDir that was active when the session was first established from the GUI. Changing it to ANY other path causes run() to silently return nil.

; ✅ Works — same path as ADE session
resultsDir("/tmp/opt_5t_ota")
run()   ; → "/tmp/opt_5t_ota"

; ❌ Breaks run() — different path
resultsDir("/tmp/new_dir")
run()   ; → nil (silent failure!)

; ✅ Restore original path, run() works again
resultsDir("/tmp/opt_5t_ota")
run()   ; → "/tmp/opt_5t_ota"

Find the canonical resultsDir from the runSimulation script:

cat <netlist_dir>/runSimulation | grep "\-raw"
# → -raw ../../../../../../../../tmp/opt_5t_ota/psf
# Canonical path = /tmp/opt_5t_ota

Fix in virtuoso-cli: sim run no longer auto-sets resultsDir to a temp dir (that was overriding the ADE session path). If resultsDir is nil, sim run now returns a clear error message.

Root Cause 2: sim setup Disrupts Active Session

Calling virtuoso sim setup sends design("lib" "cell" "view") which can reset the Ocean session state. After this, run() returns nil.

Symptom: run() worked → called sim setup → run() now returns nil.

Fix: Don't call sim setup again if a working session exists. Check:

virtuoso skill exec 'asiGetSession(hiGetCurrentWindow())'
# → stdobj@0x... means session is alive

Root Cause 2b: design() Not Called — evalstring() First-Expression Bug

Symptom: vcli sim netlist fails even on a registered library. si.foregnd.log has NO new entry (last entry is days old). design() from bridge returns nil.

Root cause: evalstring() in SKILL evaluates only the FIRST top-level expression in the string. The old setup_skill sent three newline-separated expressions:

unless(simulator()=='spectre ...)   ; ← evalstring stops here
design("lib" "cell" "view")         ; ← silently ignored!
resultsDir()                        ; ← silently ignored!

Result: design() was never called; createNetlist returned nil immediately (no ADE context).

Fix (v0.3.1+): setup_skill wraps everything in progn():

progn(unless(simulator()==... ...) design("lib" "cell" "view") resultsDir())

One top-level expression → evalstring evaluates all three.

Diagnosis: Run vcli skill exec 'design()' — if output is nil, design context is missing. Manually run vcli skill exec 'design("LIB" "CELL" "schematic")' to set it.

Root Cause 3: OSSHNL-109 — Extraction Timestamp Stale

Symptom: createNetlist returns nil. si.foregnd.log contains:

ERROR (OSSHNL-109): The cellview '…/schematic' has been modified since the last extraction.
Run Check and Save.

artSimEnvLog shows: generate netlist... ...unsuccessful.

Root cause: dbSave(cv) writes the OA file but does NOT update the extraction timestamp. schCheck(cv) is what updates it. Calling dbSave without schCheck leaves the timestamp stale; the next incremental createNetlist sees a mismatch.

Fix: Run schCheck(cv) before dbSave(cv). Or use vcli sim netlist — it auto-recovers by running schCheck + dbSave and retrying when createNetlist returns nil.

# vcli handles OSSHNL-109 automatically (v0.1.7+)
vcli sim netlist --lib FT0001A_SH --cell ota5t

Manual fix in SKILL:

; Get the cv (3-arg form opens "a" mode in IC23; fallback for Ocean-held cv)
cv = dbOpenCellViewByType("lib" "cell" "view")
unless(cv
  cv = car(setof(ocv dbGetOpenCellViews()
                 and(ocv~>libName=="lib" ocv~>cellName=="cell" ocv~>mode=="a"))))
chk = schCheck(cv)
when(car(chk)==0 dbSave(cv))   ; only save if no check errors

Also note: dbReplaceProp(inst "w" ...) sets the display parameter, NOT the netlisted simW. For SMIC PDK n12/p12 width changes, use:

cdfFindParamByName(cdfGetInstCDF(inst) "simW")~>value = "1.3u"

Root Cause 4: Stale or Missing Netlist

After editing a schematic (e.g., adding desVar variables), the netlist becomes stale. createNetlist() requires an ADE L window open for that cell — without it, returns nil.

Root Cause 5: Library Not Registered in Virtuoso Session

Symptom: vcli sim netlist returns an error like:

Library 'FT0001A_SH' is not registered in the current Virtuoso session.
Virtuoso was started from '/home/meow/git/virtuoso-cli'.

Or ddGetObj("FT0001A_SH") returns nil from the bridge.

Root cause: Virtuoso was started from a directory that has no cds.lib including the target library. The DD database is populated at startup from cds.lib files found via CDSHOME, site, and user paths — but NOT automatically from arbitrary directories.

Diagnosis: Check Virtuoso's working directory in CDS.log:

grep "Working Directory" ~/CDS.log
# Expected: meowu:/home/meow/projects/ft0001
# Wrong:    meowu:/home/meow/git/virtuoso-cli

Fix: Start Virtuoso from the project directory:

cd /home/meow/projects/ft0001 && virtuoso &

Distinguish from OSSHNL-109 (both produce err_count == -1 in vcli):

; Returns "found" if library is registered, nil if not
when(car(setof(l ddGetLibList() l~>name=="FT0001A_SH")) "found")

vcli v0.1.7+ runs this probe automatically and returns an actionable error message.

Solution A: vcli sim netlist (handles OSSHNL-109 automatically, v0.1.7+)

# Regenerate netlist — auto-recovers from OSSHNL-109 via schCheck+dbSave retry
vcli sim netlist --lib <lib> --cell <cell>
vcli sim netlist --lib <lib> --cell <cell> --recreate   # force full recreation

Solution B: Restore resultsDir (for Root Cause 1)

virtuoso skill exec 'resultsDir("/tmp/opt_5t_ota")'
virtuoso sim run --analysis dc --timeout 60

Solution B: Direct Spectre Invocation (most reliable, bypasses Ocean)

⚠️ Run from the schematic/ directory, NOT from netlist/ — relative paths inside the netlist resolve from schematic/. Running from netlist/ causes CMI-2011 (cannot open input file).

SCHEMATIC_DIR="/path/to/simulation/cell/spectre/schematic"
PSF_DIR="/tmp/my_results/psf"
mkdir -p "$PSF_DIR"

cd "$SCHEMATIC_DIR" && spectre netlist/input.scs \
  +escchars \
  +log "$PSF_DIR/spectre.out" \
  -format psfxl -raw "$PSF_DIR" \
  +mt -maxw 5 -maxn 5

Then read results:

virtuoso skill exec 'openResults("/tmp/my_results/psf")'
virtuoso sim measure --analysis dcOp \
  --expr 'getData("I0.NM0:gm" ?result "dcOpInfo")'

Advantages: ~200–300ms, no ADE session dependency, works even when Ocean is broken.

Solution C: Add Analyses to Netlist Directly

The ADE-generated input.scs may only have DC. Add AC without opening ADE:

sed -i 's/^dcOpInfo info what=oppoint where=rawfile$/dcOpInfo info what=oppoint where=rawfile\nacSweep ac start=1 stop=10G dec=20 annotate=status/' input.scs

Solution D: Regenerate via ADE GUI (for missing netlist)

1. Open testbench in Virtuoso → Launch ADE L
2. Simulation → Netlist and Run (or Netlist → Create)
3. Confirm input.scs is updated, then virtuoso sim run works again

PSF Signal Naming

Instances inside a subcircuit instantiated as I0 in the testbench:

Find actual signal names:

strings /tmp/results/psf/acSweep.ac | grep "^I0\."

Or in SKILL: openResults("/psf/path") selectResult('acSweep) VF("net1") — if it returns srrWave:0x..., the signal name is correct.

Diagnostic Checklist

1. grep "Working Directory" ~/CDS.log → is Virtuoso's CWD the project directory?
2. ddGetLibList() from bridge → does it include the target library?
3. cat runSimulation | grep -raw → what is the canonical resultsDir?
4. resultsDir() → does it match the canonical path?
5. run() → returns path (OK) or nil (broken)?
6. PSF dir has .dc/.ac files → simulation produced data
7. PSF dir only has simRunData, artistLogFile, variables_file → simulation failed
8. cat <netlist_dir>/si.foregnd.log → contains OSSHNL-109? → use vcli sim netlist to auto-fix

Key Files

<project>/simulation/<cell>/spectre/schematic/netlist/
├── input.scs       # Top-level: parameters, model includes, analyses
├── netlist          # Subcircuit from schematic
├── runSimulation    # Actual spectre command ADE uses (shows original PSF path)
└── netlistFooter    # Testbench instances and stimuli

Notes

• design("lib" "cell" "view") returns nil when no ADE window is open for that cell
• sim setup should only be called once per ADE session
• system("find / ...") hangs the SKILL daemon — blocked by virtuoso-cli bridge
• After bridge hang, restart: Ctrl+C in CIW, reload bridge.il + RBStart

See also: SKILL bridge scoping rules in ~/.claude/projects/.../memory/feedback_skill_bridge_scoping.md