Valentin COPPOLA

Abstract: Nowadays, the dramatic increase in integration capabilities and performance of advanced nanometric technologies has enabled the development of mm-wave integrated circuits for novel applications in the context of imaging, communications, radars, … , where millimeter wavelengths (E or D bands, or even beyond) are particularly well suited. The key challenge for working at mm-waves is to design and implement suitable low-loss passive circuits [1]. The quality of the passives has a direct impact on system performance, often leading to additional power expenditures to compensate electrical losses. The problem, from a design perspective in a given technology, can be seen as a “simple” design optimization problem. However, if we look at the problem from the technology point of view, optimizing a technology to implement high-quality passives or qualifying a technology as a function of the quality of its passive devices are key open challenges for new advanced technologies. In fact, high-quality passive design and fabrication are key focus points of DTCO (Design-Technology Co-Optimization) for mm-wave technologies. To guide microelectronics technologists in their choices, an automatic tool for synthesizing passive circuits would enable better use of technological resources, would facilitate advanced exploration of the active part of circuits by reducing the time to access the highly-performing passives required for their optimal operation, and would make the qualification of new technologies more efficient and faster.
#Ambition#
This PhD project has the ambition to propose innovative approaches to address the design of distributed passive functions such as matching, power division, coupling, that are part of the most challenging and delicate aspects in a transceiver implementation. The challenge will be twofold: i) providing smart design tools and smart algorithms, based on deep-learning techniques and/or access to large language models, in order to ii) explore novel architectures of passive circuits in the mm-wave and sub-THz bands that would lead to more power- and area - efficient transceivers for current and future mm-wave and sub-THz applications. Last but not least, the care taken with the design steps must be exemplary. As highlighted by the European Commission in the White Paper Industry 4.E of June 2021 [2], one of the greatest challenges in our time is the execution of the twin transition strategy, i.e. green and digital. Electronics is not green but it can be virtuous. The digital transition, that concerns all economic and social sectors, will lead to a considerable growth in data exchange and storage, accompanied by an explosion in the number of connected devices. This could have a major impact on the green transition through the energy consumption. It is expected at the end of the PhD thesis to provide a life-cycle analysis embracing the impact, in terms of footprint reduction of the device, in terms of usage, in regards to the choice of the optimization algorithm through “IA”, and bring some judgement on energy efficiency.
#Objectives#
The final goal of the project is to define innovative architectures and design methods for performant and efficient passive functions in the context of mm-wave/sub-THz front-ends. To this end, this PhD will target a novel hybrid design flow based on the combination of expert design knowledge with state-of-the-art reinforcement learning (RL) algorithms [3]. Resorting to an RL approach will allow us to efficiently automate a dynamic exploration of the design space, including both classical architectures based on metal lines and patches, and innovative ones such as the multilayer metal pixel style in [4]. Incorporating expert design knowledge to guide the RL algorithm will allow us to optimize the search in the design space, reducing the required resources, while complying to strict physical requirements and design rules in advanced nanometric technologies.