LaMi-0.4B: A Transformer-Based Large Model for Accurate Microwave Circuit Simulations

Inspired by the large language model (LLM), this paper proposes a large model dedicated to simulating microwave circuits. The model is based on a decoder-only Transformer architecture and can realize high-accuracy microwave circuit simulations. In addition, the paper proposes a quantized embedding method to map the geometrical parameters of microwave circuits to a vector and adapt it to the input of the Transformer. Finally, this paper compares the mean squared error (MSE) of the proposed large model and that of the multilayer perceptron (MLP) model through the simulation results of unequal-split branchline couplers.