Project description
A novel process model will help us get more from Moore
Moore's law, first formulated in 1965 by Gordon Moore and later revised to reflect even greater growth, says that the number of transistors on an integrated chip will double every two years. This exponential growth is behind almost every electronic technology we use today. The realisation that Moore's law is nearing its physical limit is spawning a new era of innovation. Three-dimensional (3D) integrated circuits are one of the best ways to get more from Moore's law. Hybrid bonding is quickly gaining ground as the preferred way to form high-density interconnects and 3D integration. The EU-funded H-3D-SOC project is developing the first-ever instruction set (design flow) for the processing of hybrid-bonded 3D systems-on-chip. It focusses on the embedded metal pads in the bond interface that enable face-to-face wafer connections and device stacking and could significantly advance the development of high-end processors.
Objective
As the device size is shrunk to its quantum limit, Moore’s law is becoming more difficult to follow. By exploiting the z-direction in the gate, block or system level, three-dimensional (3D) integrated circuit (IC) technology has the potential to extend Moore’s law. Face-to-face wafer by wafer Hybrid-bonding (H-b) based 3D System-On-Chip (H-3D-SOC) is one of the most promising 3D IC solutions. However, an exclusive IC design flow for H-3D-SOC for high-end processor with parameters of H-b pads from process is not available yet. This proposal aims at developing the first H-3D-SOC IC design flow for a two-tier multi-core processor considering the impact of process variation of H-b, including the geometrical and defects variation, leading to its resistance and capacitance change. Several novel methods will be developed for the H-3D-SOC IC design: machine-learning inspired techniques exploited to guide the top and bottom tiers netlist partitioning; two-tier mutual-aware optimization realized during placement and routing; capacitance coupling between the top metal layers of the two tiers considered during parasitic extraction. Moreover, impact of H-b pad parameter variation will be evaluated from the device level up to the system level: electrical variability modelling for the H-b pad, calibrated by experimental data; incorporating the device model of the pad into the H-3D-SOC based multi-core processor characterized by various industrial benchmarks. To mitigate the H-b variability, the critical H-b pads of higher utilization and larger potential to fail during working can be first identified efficiently by using a heuristic search algorithm and then hardened by attaching a spare pad to each of them. The proposed research will enable the applicant to become an expert of 3D IC design of high-end processors for applications such as AI and cloud computing, having great impact on both academia and industry of semiconductor.
Fields of science
Not validated
Not validated
Programme(s)
Funding Scheme
MSCA-IF-EF-ST - Standard EFCoordinator
3001 Leuven
Belgium