Quantum computers are moving from lab devices to larger, cloud-accessible systems, but scaling requires more than improving qubits. The control and readout electronics, often placed on the same wafer as the qubits, must work at temperatures just a few degrees above absolute zero. At those temperatures standard semiconductor models break down, making it hard to simulate circuits accurately.
A workflow described by Semiconductor Engineering tackles this by combining conventional compact models with machine learning. First, a baseline BSIM-BULK model is fitted to room-temperature measurements using Keysight's ML Optimizer, which can tune more than 80 parameters in one step. Then a neural network trained in IC-CAP learns the difference between the model's prediction and measured 4 K data — capturing effects such as dopant freeze-out and threshold shifts without requiring new physics equations.
This hybrid approach treats the neural network as a correction layer rather than a full replacement. That reduces the amount of training data needed compared to a standalone ANN and keeps the model interpretable. The trained network is converted to Verilog-A, so engineers can use it in existing circuit simulation flows for cryogenic low-noise amplifiers, readout electronics, and other critical quantum components.