| Abstract: | A novel modeling methodology for indium phosphide (InP) double heterojunction bipolar transistors (DHBTs) based on the theory of Bayesian inference, a well‐known method from the field of machine learning, is presented in this article. An extremely broadband small‐signal behavioral model, from 200 MHz to 325 GHz, is built, tested, and validated in this work, with excellent agreement obtained between the extracted model and the experimental data in the form of S‐parameters. A single finger InP DHBT device, with emitter size of 0.5 × 5 μm2 exhibiting an ft of over 550 GHz, is used in the verification example. Taking advantage of regression techniques based on machine learning concepts, the proposed black‐box behavioral model can more accurately predict the behavior of the device compared with the traditional equivalent circuit modeling method. Several sets of measured vs modeled data are shown, indicating the efficacy of the method. |
