This paper proposes a layout-level synthesis based frequency partial-space-mapped neuromodeling technique (FPSMN) for RF circuit.We exploit space-mapping technique to simplify the relation between the neural network (NN) training data and teaching data and therefore to reduce the cost for training in neuromdoeling.FPSMN establishes a partial mapping for the physical parameters

making even more efficient use of the implicit knowledge in the coarse model.This allows us a significant reduction in NN complexity and in training time.The coarse model is obtained from layout-level synthesis.Lumped-circuit modeling based on electromagnetic (EM) simulation describes the important parameters and parasitic parameters.It can serve as high quality coarse model in FPSMN.A Stoer-Bulirsch adaptive frequency sampling (S-B AFS) technique is used to decrease the cost for frequency sweeps. Numerical results of a low temperature cofired ceramic (LTCC) bandpass filter illustrate the advantage of the proposed method over existing techniques.