类脑计算软硬件生态的现状与展望OA

With the rapid advancement of the Internet of Things(IoT)and edge intelligence,neuro-morphic computing has become a core direction of architectural research in the post-Moore era owing to its event-driven operation,massive parallelism,and ultra-low power consumption.However,a systematic quantitative analysis of 500 high-impact publications over the past decade(2016-2026)reveals a pro-nounced structural imbalance:hardware-oriented research accounts for over 42%of the surveyed litera-ture,whereas software stack and compiler toolchain research—which bridges hardware and applications—constitutes less than 5%.This disparity is fundamentally attributable to the absence of a unified in-struction set architecture(ISA)in the neuromorphic computing domain,which results in inconsistent software-hardware interfaces,fragmented ecosystems,and constraints on the large-scale deployment of applications.To address this gap,this paper presents a comprehensive survey of the full deployment pipeline from algorithmic models to hardware integration,encompassing spiking neural network(SNN)computational models,4 mainstream processor architecture paradigms(dedicated ASIC,programmable processor,computing-in-memory architecture,and heterogeneous hybrid processor),software stack tech-nologies(compilers,programming framework,and runtime environment),as well as the latest advances in 4 representative application domains:edge vision,robotic navigation,scientific computing,and biosig-nal processing.The central thesis of this paper is that a neuromorphic ISA constitutes the critical abstraction layer connecting algorithmic diversity with hardware execution efficiency,and represents the logical starting point for resolving the current ecosystem deadlock.Building on this foundation,the paper offers a forward-looking perspective on 2 emerging trends—neuromorphic ISA standardization and LLM-aided hardware design—with the aim of providing a systematic reference roadmap for the field's transi-tion from research prototypes to large-scale practical deployment.