PPAAS: PVT and Pareto Aware Analog Sizing Analog Sizing via Goal-conditioned Reinforcement Learning (ICCAD 2025)

By Seunggeun Kim, Ziyi Wang, Sungyoung Lee, Youngmin Oh, Hanqing Zhu, Doyun Kim, and David Z. Pan

We introduce a goal-conditioned RL framework for analog device sizing that (i) generalizes across PVT corners and (ii) achieves ~1.6x higher sample efficiency and ~4.1× higher simulation efficiency than prior methods. Key ingredients:

• Pareto-Front Dominance Goal Sampling – builds an automatic curriculum by sampling from the current Pareto frontier of achieved goals.
• Conservative Hindsight Experience Replay – stabilizes updates and speeds convergence.
• Skip-on-Fail simulation – avoids unnecessary runs when it obviously violates spec in nominal corner.

---

Repository Structure

PPAAS/
├── train.py                   # main training entry point
├── gen_specs.py               # generate evaluation specs
├── envs/
│   ├── ngspice_env_base.py    # base Gym‑style environment
│   ├── ngspice_env.py         # discrete / continuous wrapper
│   └── ngspice_env_goal.py    # goal‑conditioned wrapper
├── eval_engines/              # SPICE wrappers, benchmarks & configs
├── scripts/                   # one‑line launchers
│   ├── TSA.sh
│   ├── CMA.sh
│   ├── LDO.sh
│   └── COMP.sh
└── environment.yml            # Conda spec

---

Getting started in this repository

To get started, create a conda environment containing the required dependencies.

conda env create -f environment.yml
conda activate PPAAS

Install ngspice (open-source circuit simulator) if not installed yet.

sudo apt-get update
sudo apt-get install ngspice

Or install any version from ngspice official website.

Fix SPICE input paths

cd eval_engines/ngspice/ngspice_inputs
python correct_inputs.py

---

Training & Logging

From the repository root, run the appropriate script:

./scripts/[TESTBENCH].sh

Replace [TESTBENCH] with one of:

• TSA (Two‑Stage Amp, GF180)
• CMA (Cascode Miller-compensated Amplifier, GF180)
• LDO (Low‑Dropout Regulator, SKY130)
• COMP (Comparator, GF180)

Logging is handled via Wandb; edit --entity in the scripts to use a different namespace.

---

Generating Custom Evaluation Specs

You can generate custom SPICE evaluation specifications with:

python PPAAS/gen_specs.py --num_specs <N>

Replace <N> with the number of specs you want to create.

---

Acknowledgment

Parts of this codebase were adapted from AutoCkt.

Contact

Open an issue or email [email protected].