Abstract: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.

Reconfigurable heterogeneous integration using stackable chips with embedded artificial intelligence

Monolithic 3D integration of 2D materials-based electronics towards ultimate edge computing solutions

Reconfigurable perovskite nickelate electronics for artificial intelligence

Reconfigurable Computing: a Promising Microchip Architecture for Artificial Intelligence

Energy-Efficient, Stable and Temperature-Tolerant Neuromorphic Device Based on Single Crystals of Halide Perovskites

A Reconfigurable Photonic Computing Chip with Easily Scalable Topology

Reconfigurable Neuromorphic Computing with 2D Material Heterostructures for Versatile Neural Information Processing

Phase-change Heterostructure Enables Ultralow Noise and Drift for Memory Operation.

Reconfigurable Neuromorphic Computing: Materials, Devices and Integration

Neuromorphic chip integrated with a large-scale integration circuit and amorphous-metal-oxide semiconductor thin-film synapse devices

Reconfigurable neuromorphic computing block through integration of flash synapse arrays and super-steep neurons

Innovative Architecture of Single Chip Edge Device Based on Virtualization Technology.

Reconfigurable Computing: Toward Software Defined Chips

The evolution of heterogeneous integration and packaging for the age of chiplets

Large-scale neuromorphic optoelectronic computing with a reconfigurable diffractive processing unit

IntAct: A 96-Core Processor With Six Chiplets 3D-Stacked on an Active Interposer With Distributed Interconnects and Integrated Power Management

In-Sensor Computing Realization Using Fully CMOS-Compatible TiN/HfO x -Based Neuristor Array

Neuro-inspired computing chips

Hybrid architecture based on two-dimensional memristor crossbar array and CMOS integrated circuit for edge computing

A 0.96pJ/SOP, 30.23K-neuron/mm^2 Heterogeneous Neuromorphic Chip With Fullerene-like Interconnection Topology for Edge-AI Computing

Extremely Low-power-consumption Reconfigurable Two-dimensional Tellurene Artificial Synapse for Bio-inspired Wearable Edge Computing