Thermal Bottleneck Breakthrough: Peking University’s Predictive Modeling for BEOL Scaling

As the semiconductor industry pushes toward 2nm and sub-2nm nodes, the Back-End-of-Line (BEOL) interconnect stack has emerged as the primary bottleneck for performance and reliability. Peking University’s recent development of a structure-aware thermal conductivity (κ) modeling framework represents a critical advancement in addressing the 'thermal wall' that currently limits chip density and clock speeds. In advanced integrated circuits, the proliferation of complex metal layers, vias, and increasingly thin dielectric materials creates a chaotic thermal landscape. Traditionally, engineers relied on bulk approximations, which fail to capture the interfacial thermal resistance and phonon scattering characteristic of modern nanometer-scale stacks. By integrating extensive, layer-resolved thermal measurements into a predictive framework, this research provides a methodology to simulate heat dissipation with unprecedented accuracy. The industry impact of this modeling framework is profound. Thermal management is no longer a peripheral challenge; it is a fundamental design constraint that dictates the power-performance-area (PPA) scaling of future logic and memory devices. By accurately modeling how heat moves through a multi-layered BEOL structure, architects can perform ‘thermal-aware’ floorplanning earlier in the design cycle, reducing the risk of costly post-silicon failures or performance throttling. Furthermore, this research holds significant implications for the supply chain, particularly for EDA (Electronic Design Automation) providers like Cadence, Synopsys, and Siemens EDA. Integrating such sophisticated thermal models into industry-standard design tools will be essential for foundries like TSMC, Samsung, and Intel to maintain aggressive scaling cadences. Looking toward the future, this predictive capability is a prerequisite for the industry’s transition toward advanced packaging and heterogeneous integration. As chiplets become the norm, the ability to predict thermal behavior across diverse material interfaces—beyond just the monolithic BEOL—will be a key competitive differentiator. We expect this research to accelerate the adoption of new, high-thermal-conductivity dielectric materials and advanced metallization processes. By turning thermal management from an empirical guessing game into a deterministic engineering process, Peking University has provided a blueprint for sustaining Moore's Law in an era defined by extreme heat density.