|
Ever since IBM pioneered microelectronics packaging, IBM Research has continued to innovate to ensure that packaging and heterogeneous integration technology is available to satisfy the needs for performance, complexity, and memory and logic density with regard to high performance computing systems. In this era of ever-expanding need for high-performance computing and ever-pervasive artificial intelligence, traditional scaling economics headwinds combined with the need for versatility and fast product development mandate a system-level approach to generate the efficiencies the industry has been able to provide in the past through more traditional scaling approaches. This system-level approach renders imperative the use of chiplet-based architectures and the use of heterogeneous integration and advanced packaging to achieve the disaggregation with best-performing and most efficient IP components to sustain a viable economic model while achieving performance targets. In this talk we will discuss several key process and integration considerations that drive the use of various horizontal and vertical interconnection technology elements that must be implemented to enable the successful realization of efficient and cost-effective high-performance systems in the AI era. Future progress on these technology fronts will depend on disruptive innovation in two critical areas: (a) wafer-level and die-level processes, such as lithographic patterning, wafer-wafer bonding and debonding, bond and assembly processes, etc. that can enable packaging of these heterogeneous structures without compromising performance and reliability, (b) commensurate improvements and enablement of adequate metrology and inspection solutions to address the challenges stemming from these new chiplet-interconnecting methods and the associated topographic implications.
|
