Efficient and accurate Technology Computer-Aided Design simulations with machine learning and their application to develop monolithic CMOS sensors for physics experiments

Student: Archie Hanlon
Supervisors: Eva Vilella-Figueras (UoL), Nicola Massari (FBK), Gianluigi Casse
Institution: University of Liverpool

Complementary Metal-Oxide-Semiconductor (CMOS) is one of the most popular commercial process technologies available to fabricate integrated circuits such as microprocessors, memories, and image sensors used in the vast majority of everyday life devices such as tablets and smartphones. CMOS is also an extremely attractive technology option for producing silicon sensors to measure charged particles in physics experiments that create extremelychallenging environments. An essential aspect in the R&D phase towards a new CMOS sensor is the ability to anticipate the performance of the device before its expensive and time-consuming fabrication, which is typically achieved with Technology Computer-Aided Design (TCAD) simulations.

Unfortunately, TCAD simulations are computationally very expensive. This limits the accuracy and depth of the simulations that can be done to guide the design of a CMOS sensor, and often results in sub-optimised devices and longer and more expensive R&D programmes.

In this project we propose to develop machine learning methods to support fast and accurate TCAD simulations in a commercial CMOS technology, and to apply the developed methods to guide the design of a real CMOS sensor for physics experiments. The student undertaking this project will collect training data from relevant TCAD simulations, explore a few neural-network models, and choose the most promising one to train and develop a machine learning model. The student will use the developed machine learning tool to run fast and accurate TCAD simulations to input the design of a novel monolithic CMOS sensor with fast-timing. Liverpool is a world leader in the development of advanced silicon sensors for physics experiments.

It has built substantial parts of the detectors for the ATLAS and LHCb experiments at the Large Hadron Collider (LHC) at CERN. It will do so again for the ATLAS, LHCb and ALICE upgrades at the High Luminosity LHC and for the mu3e experiment. Liverpool enjoys a central role in the CERN-RD50 collaboration. Since 2014, we also have an R&D Group dedicated to monolithic CMOS sensors for fundamental physics experiments (https://www.liverpool.ac.uk/particle-physics/research-strategy/rd/hv-cmos/). The R&D Group includes postdocs, engineers and students.

The student will join the Liverpool CMOS R&D Group and spend a minimum of six months at the partner organisation the Fondazione Bruno Kessler in Trento, Italy, where they will use their data science skills to contribute to a project beyond their PhD topic. This PhD will often involve teamwork and joint problem solving between colleagues with complementary skills. The student will be trained in state-of-the-art machine learning methodology, and the specifics of TCAD simulations and semiconductor devices. The student will acquire important transferable skills in data science and CMOS technologies design.