Spicing up Genetic Netlist Generation with LLMs

Computer Science > Neural and Evolutionary Computing
[Submitted on 24 Aug 2026]
Title:Spicing up Genetic Netlist Generation with LLMs
View PDF HTML (experimental)Abstract:Analog circuit topology synthesis remains challenging because useful designs occupy a tiny fraction of a combinatorial search space, and small structural changes can induce highly nonlinear changes in behavior. Evolutionary algorithms are attractive because they can optimize over discrete circuit topologies using only black-box evaluations, but they often require many SPICE simulations and may converge prematurely. We introduce LLM-SPICEMixer, a hybrid synthesis framework that augments genetic netlist generation with IGEL (Inspiration-Guided Evolution with LLMs), an LLM-based proposal operator. During search, IGEL prompts an LLM with high-performing circuits from the elite set and instructs it to generate a new SPICE netlist, which is then evaluated by SPICE and selected using the same reward mechanism as conventional genetic operators. Thus, the LLM contributes structured topology proposals while simulation remains the source of truth. We evaluate LLM-SPICEMixer on a challenging benchmark task: synthesizing transistor-level circuits that implement a discriminant function for Iris classification. Compared with the genetic framework without LLM guidance, LLM-SPICEMixer improves the median final training reward by 8.4% and the median validation-selected test reward by 8.8%. The best validation-selected circuit achieves 93.3% test accuracy at the nominal tt corner and 85.9% average test accuracy across 17 process, voltage, and temperature corners.
Current browse context:
cs.NE
References & Citations
Loading...
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

* The framework is named LLM-SPICEMixer.
* It augments genetic netlist generation with IGEL (Inspiration-Guided Evolution with LLMs).
* IGEL functions as an LLM-based proposal operator.
* IGEL prompts an LLM using high-performing circuits from the elite set to generate a new SPICE netlist.
* The generated netlist is evaluated by SPICE and selected using conventional genetic reward mechanisms.
* Evaluation relies on SPICE simulations remaining the source of truth.
* The framework was evaluated on synthesizing transistor-level circuits for Iris classification discriminant function implementation.
* LLM-SPICEMixer improved the median final training reward by 8.4% compared to the genetic framework without LLM guidance.
* It improved the median validation-selected test reward by 8.8%.
* The best validation-selected circuit achieved 93.3% test accuracy at nominal tt corner and 85.9% average test accuracy across 17 corners.

A hybrid synthesis framework named LLM-SPICEMixer is introduced to augment genetic netlist generation in analog circuit design. The challenge lies in the limited search space and nonlinear behavior caused by small structural changes in circuit topologies, making traditional evolutionary algorithms slow due to the need for extensive SPICE simulations. LLM-SPICEMixer incorporates IGEL (Inspiration-Guided Evolution with LLMs), an LLM-based proposal operator, into the genetic synthesis process. During the search, IGEL prompts a Large Language Model with high-performing circuits from the elite set to generate new SPICE netlists. These generated netlists are then evaluated using standard SPICE simulations and selected via the same reward mechanism used in conventional genetic operators. This design ensures that the LLM provides structured topology proposals while SPICE simulation maintains the role of the source of truth for evaluation.

The core implication lies in leveraging the generative reasoning capabilities of LLMs to navigate complex, high-dimensional design spaces where traditional search methods suffer from computational cost and the discrete nature of topology constraints. The study establishes a critical tension between the creativity of LLM-generated proposals and the necessity of physical reality enforced by SPICE simulation as the ground truth. This introduces a pattern related to knowledge integration: combining abstract symbolic reasoning (LLM) with concrete, physics-based validation (SPICE). The success metrics—an 8.4% gain in reward compared to pure genetic methods—suggest that the LLM acts as an intelligent heuristic generator rather than a replacement for simulation fidelity. The process demonstrates that structured topological guidance can accelerate convergence in areas where manual or stochastic exploration is prohibitively slow. The missing piece is understanding the robustness of the LLM's generated *structurally sound* proposals versus novel, potentially infeasible ones, and whether this method scales effectively to more complex functional requirements beyond a simple classification discriminant. What mechanisms govern the selection criteria when an LLM proposes a solution that performs well in reward but has poor physical realizability outside the immediate SPICE evaluation context?

Spicing up Genetic Netlist Generation with LLMs

Facts Only

Executive Summary

Full Take

Spicing up Genetic Netlist Generation with LLMs

Facts Only

Executive Summary

Full Take