Recently, generative models have become immensely popular in the machine learning community
as well as in other engineering disciplines.
Generative models are significantly more general than traditional discriminative models (e.g.,
polynomial chaos, neural networks, etc.) as they learn an approximation of the complete
distribution underlying a collection of designs or topologies, hence they can produce completely
new designs that satisfy the design specifications considered.
The main idea of this research project is to integrate generative modeling into the design phase by
studying, developing, implementing and testing new generative-based approaches for
electromagnetic and electronic applications. Considering the huge differences in input/output
behavior, frequency bandwidth and design constraints of different modern ICs, this research
proposal focuses only on linear and passive systems.
In particular, the key points of this project are:
(i) Use generative modeling to learn and propose valid designs (indirect approach).
(ii) Use generative modeling to directly generate valid designs that also satisfy the design
requirements (direct approach).
(iii) Characterize the effect of the manufacturing tolerances during the design phase in the
generative modeling strategies.
The proposed design framework aims at achieving short time-to-market and first-time-right design,
which are fundamental in the highly competitive market of modern electronic devices.