VirtuosoClient — SKILL execution

Layout editing

from virtuoso_bridge.virtuoso.layout import (
    layout_create_rect as rect,
    layout_create_path as path,
    layout_create_param_inst as inst,
    layout_create_via_by_name as via,
)

with client.layout.edit("myLib", "myCell") as lay:
    lay.add(rect("M1", "drawing", 0, 0, 1, 0.5))
    lay.add(path("M2", "drawing", [(0,0),(1,0)], 0.1))
    lay.add(inst("tsmcN28", "nch_ulvt_mac", "layout", "M0", 0, 0, "R0"))
    lay.add(via("M1_M2", 0.5, 0.25))

Schematic editing

from virtuoso_bridge.virtuoso.schematic import (
    schematic_create_inst_by_master_name as inst,
    schematic_create_pin as pin,
    schematic_label_instance_term as label,
)

with client.schematic.edit("myLib", "myCell") as sch:
    sch.add(inst("analogLib", "vdc", "symbol", "V0", 0, 0, "R0"))
    sch.add(label("V0", "PLUS", "VDD"))
    sch.add(pin("VDD", 0, 1.0, "R0", direction="inputOutput"))
    sch.add_net_label_to_transistor("M0",
        drain_net="OUT", gate_net="IN",
        source_net="VSS", body_net="VSS")

Spectre simulation

from virtuoso_bridge.spectre.runner import SpectreSimulator

sim = SpectreSimulator.from_env(work_dir="./output")
result = sim.run_simulation("tb_inv.scs", {})

print(result.status)          # ExecutionStatus.SUCCESS
print(result.data.keys())     # ['time', 'VOUT', 'VIN', ...]
print(result.data["VOUT"][:5]) # first 5 voltage samples

ADE Maestro (Assembler)

from virtuoso_bridge.virtuoso.maestro import (
    snapshot, open_gui_session, close_gui_session,
    run_and_wait, read_results, export_waveform,
)

# Snapshot the focused maestro window — no session arg needed.
# Returns dict with raw_sections (list of (skill_probe, raw_output)
# tuples — no Python parsing of SKILL alists).
d = snapshot(client)
# Or also write the full disk dump (raw + filtered XMLs +
# state_from_skill.txt + per-point run files via tar). Files
# pulled are governed by snapshot_filter.yaml — edit to add/drop.
d = snapshot(client, output_root="output/")
# Pin a specific history instead of the auto-pick (mtime-first).
d = snapshot(client, output_root="output/", history="Interactive.160")

# Open / run / collect — GUI mode is required for run_and_wait
session = open_gui_session(client, "myLib", "myTB")
run_and_wait(client, session)  # non-blocking, SKILL callback + ssh polling
results = read_results(client, session, lib="myLib", cell="myTB")
export_waveform(client, session, 'dB20(mag(v("/OUT")))', "gain.txt")
close_gui_session(client)

SSHClient — independent SSH layer

from virtuoso_bridge import SSHClient, VirtuosoClient

# Remote mode
tunnel = SSHClient.from_env()
tunnel.warm()
bridge = VirtuosoClient.from_tunnel(tunnel)

# Local mode
bridge = VirtuosoClient.local(port=65432)

File transfer and shell

CLI commands

What snapshot -o pulls

One command dumps the focused maestro session's setup and the latest run's per-point files to a timestamped folder:

# auto-pick newest history
virtuoso-bridge snapshot -o output

# or pin a specific history (e.g. side-by-side comparison of two runs)
virtuoso-bridge snapshot -o output --history Interactive.160

Layout of the output directory:

output/<YYYYMMDD_HHMMSS>__<lib>__<cell>/
├── maestro.sdb, active.state                    # raw Cadence files
├── state_from_sdb.xml                           # sdb filtered (~10× smaller)
├── state_from_active_state.xml                  # active.state filtered
├── state_from_skill.txt                         # SKILL-probe setup summary
└── <history>/
    ├── <history>.{log,rdb,msg.db}               # run-level (rdb/msg.db are SQLite)
    └── <pt>/<tb>/
        ├── netlist/  → netlist, input.scs, qpInformation.ils, paramInfo.ils
        └── psf/      → spectre.out, logFile, dcOp.dc, *.ac, *.tran, *.noise, ...

Per-point netlist/ keeps only the 4 files that describe the design (main SPICE netlist, testbench top-level, FOM/output-expression definitions, corner label). psf/ keeps stdout + logs + non-binary analysis results.

File selection is governed by virtuoso/maestro/snapshot_filter.yaml — every entry carries a per-file tag ([KEEP], [REDUNDANT], [SCAFFOLDING], [METADATA], [DEBUG-ONLY], [GUI-CONFIG], [RARELY], [NEVER]). Uncomment a line to add a file, comment it to skip. No package reinstall needed; the YAML is reread on every snapshot call.

Deliberately not pulled, regardless of YAML: *.raw binary waveforms, wavedb/, and PDK info dumps (modelParameter.info, primitives.info.*). Read waveforms through reader.runs.read_results instead of copying the binaries.