A New Smart-MicroSystems Age Enabled by Heterogeneous Integration of Silicon-Centric and AI Technologies-My Personal View

Abstract

After 60 years of development efforts since the 1960s to the current Giga-/Tera-Scale-Integration or System-on-a-Chip era [1]–[3], it is expected that Monolithic Silicon IC products using 2-nm CMOS devices will appear soon. The subsequent challenge is whether more novel device structures using heterogeneous materials and 3D-structures will be invented to realize manufacturable 1-nm ICs. On the other hand, through 20 years of efforts since 1999, many Heterogeneous Integration (HI) [4]–[6] products, each of which is composed of silicon and non-silicon materials/ dice/chiplets, diversified devices/circuits, innovative architectures and multi-dimensional arrangements of dice and other components inside either Chip-package or Module, have been increasingly emerging, especially recently benefiting from a strong driving force stimulated by the IEEE HI Roadmap unveiled in 2018 [5]. This paper presents an exciting, powerful and new Trend of Semiconductors, Intelligent Grand Scale Integration (IGSI), which is optimally utilizing Mixed Integration of Monolithic and HI Technologies (Si-4.0 [6]) with embedded 3A's (Algorithm, Architecture and AI) Design-Intelligences. A key target of IGSI technologies is to drive much higher energy efficiency of managing electronic information for more-effective/ intelligent future systems with better performance, lower power, higher reliability and smaller form-factor than those of our current systems. One effective way as proposed is to network multiple Self-Smart MicroSystems (S-SmS) each of which is designed with 3A's to a complete system level which can handle huge data processing smartly in its own compact multi-dimensional form factor like in a versatile solid-state micro-universe which has abundant self-contained intelligent functions with maximized speed-power efficiency due to close proximity of electronic/photonic/ micro-mechanical operations. It is projected that in such an S-SmS each Joule (energy unit) be able to operate more than 10^20 devices per die per joule allowed by thermodynamics (on the other hand, its performance can reach over hundreds of thousands of TOPS - Tera Operations Per Second) inside and/or across these MicroSystems to complete the final system need. Then how powerful a future system can be by networking enough S-SmS units and furthermore how many unprecedented and unexpected applications will be unleashed! To use AI computing systems as an example, it is expected that S-SmS be quickly applied to AI's edge, device or wearable applications. Moreover, just like the experiences of migrating from a Mainframe computer to networked PC Servers, Data servers used in AI Clouds may use such a networked S-SmS architecture to build large systems in order to optimize the energy efficiency and heat dissipation. The trend equally adds values to system's transformation and optimization in Autonomous Car areas, Industrial 4.0 Factory areas, Telecommunication and Computing areas and so forth.