Search papers, labs, and topics across Lattice.
This paper introduces LLM-SPICEMixer, a hybrid framework that enhances genetic netlist generation for analog circuits by integrating an LLM-based proposal operator called IGEL. By leveraging high-performing circuit designs to prompt the LLM for new SPICE netlists, the framework optimizes circuit topology while maintaining simulation as the evaluative standard. The results show that LLM-SPICEMixer outperforms traditional genetic algorithms, achieving an 8.4% improvement in median final training reward and a 93.3% test accuracy on a challenging benchmark task for transistor-level circuits.
LLM-SPICEMixer boosts circuit design efficiency, achieving up to 93.3% accuracy by combining LLM insights with traditional genetic algorithms.
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.
