信息学院学术报告：Emerging Technologies in Computing
报告人：Giovanni De Micheli
摘要：The evolution of computing over the years has lead to very different realizations, ranging from portable electronics to supercomputing. At the same time, embedded computing permeates every day’s life, as demonstrated by the increasing presence of computing in vehicles (e.g., self-driving vehicles), in the workplace (from smart office to smart factory), to health care, and – last but not least – to the Internet of Things (IoT). There is an increasing demand of computational power as well as of reduced energy consumption, two competing factors that are not obvious to reconcile. The current semiconductor technologies, mainly FinFETs and FDSoI, provide us with a robust physical means of addressing computing requirements, but a strong interest is placed now in how to enhance the current devices (in terms of performance and functionality) with new materials as well as how to use these materials to realize completely novel devices. Medium term enhancement of silicon technology includes tunnel FETs and NanoWire (NW) transistors, which both can address low-power solutions. Current realizations of Si NW transistors include both horizontal and vertical arrangements, the latter being very effective for increasing device density. Recent research has also shown the usefulness of Carbon NanoTubes (CNTs) in realizing simple but scalable computational engines, as well as 2-Dimensional materials, such and MoS2 to realize transistors. A different approach to increasing the computational effectiveness of transistors, as well as avoiding some variability problems, is electrostatic doping, i.e., the realization of transistors that can have electrically-programmable polarity. Such devices have been realized on various substrates, such as SiNW, CNTs and WSe2 in experimental forms. Electrical programmability is achieved by a secondary gate, thus making the atomic computing element into a 4-terminal device. From a logic and architectural abstraction standpoint, these devices are switches activated by a comparison, thus semantically more powerful than regular transistors. The mapping of computing structures on controllable-polarity substrates provides thus new challenges, and it has shown to be effective for arithmetic blocks where binate logic functions abound. Design algorithms, methods and tools, such as those based on majority algebra, are both important investigation and design means for these new technologies. Eventually, the evolution of semiconductor technology will bring us the fusion of sensors with computation, thus enabling new computing modalities and interaction with the physical environment. Namely, heterogeneous integration can be exemplified by the biological functionalization of Si NWs, that can make them into miniaturized and effective biosensors for medical and environmental needs. Broadly speaking, geometry downscaling, material hybridization and multi-layer integration are the ingredients for more diverse and powerful computing paradigms in the years to come.