Key points are not available for this paper at this time.
In the post exascale computing era, the energy efficiency improvement speed of traditional complementary metal-oxide-semiconductor (CMOS) process has slowed down significantly. In order to realize the zettascale computing capacity in 2035, a great innovation is needed in the microprocessor architecture and design. This paper selects four aspects, which are low power consumption technology, near data processing (NDP) technology, interconnection centered design method and domain specific architecture (DSA), which has a broad development prospect, and focus on the energy efficiency benefits of each technology. Firstly, we analyze various traditional low power consumption technologies and near threshold computing (NTC) technology; secondly, we analyze the NDP technologies such as near memory computing, in memory computing and in network computing; thirdly, we analyze the low overhead network on chip (NOC), NOC supported by new process and cache coherent NOC technology; finally, we take the popular artificial intelligence (AI) processor as an example to analyze the DSA.
Wang et al. (Fri,) studied this question.