OnThursday, March 8, 2018, Kevin Morot, a Ph.D. student in Optics and Radio Frequencies atthe Institute of Microelectronics, Electromagnetism, and Photonics and the Microwave and Characterization Laboratory (IMEP-LAHC), will defend his thesis titled “RF Integration and Modeling of 3D Interconnections for Photonic Interposers.” The defense will take place at 10:15 a.m. in the lecture hall of the Pôle Montagne Building on the Bourget-du-Lac campus.
thesis abstract
To deliver faster, less expensive, and more energy-efficient systems, microelectronics engineers have long relied on Moore’s Law, which predicts that the density of integrated circuits will increase at a steady rate. Now that this law is losing steam, manufacturers of these systems are forced to turn to disruptive solutions to keep pace in the race for performance. 3D circuit integration was developed to address this challenge. In data centers—the nerve centers of communication networks—signal propagation is distributed across optical and electrical networks. A component that converts between these two transmission modes is essential: the electro-optical transmitter. As transmission speeds increase, the optical signal must be routed closer and closer to the electronic components to reduce cost and energy consumption. To achieve the transmission capacities of 10 terabits per second targeted for 2020, the maximum distance traveled by the electrical signal between the transmitter and the processor is estimated at 10 mm. A new approach to integrating the transmitter directly into the digital circuit package is essential.
The photonic interposer is proposed to achieve this co-integration by leveraging three-dimensional integration. The first stage of the work presented in this thesis aims to evaluate the
limitations of the photonic interposer for the targeted applications. To this end, we demonstrate that “middle” integration of the through-silicon via (i.e., TSV) is preferable to its “last” version. Next, design rules ensuring the suitability of the interconnections are established. A methodology dedicated to extracting the maximum usable length of chains satisfying bandwidth criteria is implemented. It relies on 3D electromagnetic simulation of each interconnection and the cascading of the extracted models. The preferred routing paths are identified, and the results reveal significant routing constraints resulting from the limitations of current technology.
Developing an interposer capable of meeting the specifications requires optimizing the performance of the 3D interconnect chains. A broadband, configurable model is developed for each 3D interconnect. These models are based on electromagnetic simulation plans, whose definition domains are aligned with current technology and include longer-term technological prospects deemed promising. The models are validated by comparison with characterization results up to 110 GHz. An initial conclusion of the work identifies TSVs and RDL lines as the most detrimental elements of the 3D network. The parameters that most significantly affect their attenuation and delay are identified: these are the substrate conductivity and, to a lesser extent, the thicknesses of the oxide and conductors. Finally, the parameterizable models are cascaded to optimize the maximum usable length and
power consumption of the 3D chains of the photonic interposer in two technological contexts: one constrained by current processes and the other incorporating breakthrough solutions.
This study yields recommendations regarding technology, design, and routing. The benefits of thinning the substrate, thickening the RDL lines, and using high-resistivity substrates are quantified. The footprint of the interposer’s front side is thus reduced by routing high-speed signals through TSVs and RDLs up to 3 mm for a standard substrate. In the case of a high-resistivity substrate, transmission across the entire surface of the interposer is made possible without signal regeneration.