Advanced packaging solutions are a key pillar in overcoming the existing limitations of monolithic integration, which has become more difficult and expensive at each node. Heterogeneous integration technologies, such as 2.5D/3D stacking, tiling, intra-die routing, through-silicon vias, chiplets and fan-out, have and significantly increased system capabilities and performance but have introduced new challenges to an industry accustomed to a “Moore’s Law inter-die scaling” approach. Quality and reliability are increasingly harder to achieve due to an inherent high density and high frequency nature, with limited visibility after assembly.
In this presentation, a new concept based on Deep Data chip telemetry will be introduced, offering a unique combination of embedded IPs (Agents), both core and I/O, and a software platform that implements machine learning algorithms and data analytics. Higher visibility for data-driven decision making is achieved, in production and when deployed in the field. Findings from a 7nm HBM ASIC controller use case will be presented.