Others

Find the netlist corresponding to param

1. Find a set of params that succeeded during the validation process

Params = [10 12 79 21 10 57 34]

2. The params output here is index, and the actual value is calculated according to the specification range in two_stage_opamp.yaml

The first 6 are: $1 + 1 \times x$, the last one is: $(0.1 + 0.1 \times x)\times 10^{-12}$

So the final params = [11 13 80 22 11 58 3.5e-12]

3. According to ngspice_wrapper.py, the changed netlist exists in the BASE_TMP_DIR path:

4. Enter /tmp/ckt_da/designs_two_stage_opamp/, enter cd two_stage_opamp_ + the params just calculated, and you can enter the corresponding path:

The Cir file is opened to get the netlist parameters. Its file name and path name are generated using the actual value of params.

Code error location (ac/dc file doesn’t exist)

In the file AutoCkt/eval_engines/ngspice/TwoStageClass.py around line 47, the following code checks for the existence of .ac and .dc simulation results:

if not os.path.isfile(ac_fname) or not os.path.isfile(dc_fname):
    log.warning("ac/dc file doesn't exist: {}".format(output_path))

This warning is often triggered when the simulation process fails to generate the required files. Here are the common causes and how to resolve them:

Reason 1: $SPECTRE_MODEL_PATH Not Set

The spectre log will show error as: ./design_wrapper.lib.scs not exist or cannot find ./design_wrapper.lib.scs.

If your netlist includes a model like:

include "$SPECTRE_MODEL_PATH/design_wrapper.lib.scs" section=mainlib

You need to ensure the environment variable SPECTRE_MODEL_PATH is correctly set before running the code. If not, Spectre will fail silently, and the .ac/.dc files will not be generated.

Fix: Add the following line in your C shell environment (csh):

setenv SPECTRE_MODEL_PATH /net/vlsiserver/usr11/library/CSM/gf55BCDlite_v1090_stack7A/55BCDlite/V1.0_9.0/Models/Spectre/55bcdlite/models/

Reason 2: Wrong Shell – Must Use csh

This codebase requires activating the Spectre simulation environment via conda in a C shell (csh). Running the script in bash will not activate the correct environment and lead to simulation failure.

Fix: Always launch your terminal with csh before activating conda and running the scripts.

Reason 3: Simulation Command Fails Silently

The simulation is performed in eval_engines/ngspice/ngspice_wrapper.py (lines 117–123) using this command block:

command = """
spectre "{0}" -o "{1}" -log >& /dev/null
ocean -nograph -restore "{2}" >& /dev/null
find "{1}" -name ".*.dep" -exec rm -rf  +
find "{1}" -name "*.raw" -exec rm -rf  +
""".format(fpath, output_dir, os.path.join(output_dir, "export.ocn"))

Note that both spectre and ocean logs are hidden by redirecting output to /dev/null. This means if any error occurs, you won’t see it by default.

Fix:

• Run spectre and ocean commands manually using the netlist and .ocn script to check:

  • If the netlist has syntax or model errors
  • If the ocean script is written correctly
  • If all required files exist in the specified output directory

✅ Summary

If you see ac/dc file doesn't exist:

1. Check that SPECTRE_MODEL_PATH is set in your environment.
2. Confirm you’re using csh shell with the correct conda environment.
3. Debug Spectre and Ocean commands manually to check for hidden errors.

This step is essential to ensure correct simulation output before proceeding with reinforcement learning training.

❗ Error: “- WARNING - results file doesn’t exist”

This warning typically appears when the result.txt file was not successfully generated after running simulation.

🔍 Root Cause

The most common reason is that Cadence log files (e.g., CDS.log) are already occupied by other processes. As a result, the ocean script fails to write logs properly and skips generating the result.txt file.

To confirm this, check the contents of ocean.log — it will usually show signs of resource conflict or failure to write logs.

🛠️ Solution

• Use the following command to forcefully kill leftover Python processes that might be occupying the logs:

  pkill -9 -f python
  

• After that, restart the conda environment and run the code again.