基于生物分子的神经拟态计算研究进展  

Progress on neuromorphic computing based on biomolecules

作　　者：滕越 杨姗 刘芮存 Yue Teng;Shan Yang;Ruicun Liu(State Key Laboratory of Pathogen and Biosecurity,Beijing Institute of Microbiology and Epidemiology,Academy of Military Medical Sciences,Beijing 100071,China)

机构地区：[1]军事科学院军事医学研究院微生物流行病研究所,病原微生物生物安全国家重点实验室,北京100071

出　　处：《科学通报》2021年第31期3944-3951,共8页Chinese Science Bulletin

摘　　要：随着大数据、云计算和人工智能等技术为代表的计算新时代的到来,以构建人工神经网络为基础的神经拟态计算因具有低能耗、自适应学习和高并行计算等优点成为研究热点.生物分子计算伴随着合成生物学的兴起而不断发展, DNA等纳米材料不但可用于逻辑运算,还可以构造神经网络,并从训练数据中进行学习,为在分子层面实现神经拟态计算提供了可能.本文概述了神经拟态计算的基本原理,总结了DNA计算的研究进展,介绍了基于DNA计算的神经拟态计算,讨论了其发展所面临的挑战.基于生物材料构造人工神经网络是神经拟态计算迈向分子计算层面的重要一步,而由此集成的人工智能芯片则有望广泛应用于航空航天、信息安全及国防建设等领域.The current von Neumann computing system cannot satisfy the demand for the development of new technologies, such as big data, cloud computing and artificial intelligence. Neuromorphic computing based on the construction of artificial neural networks is considered a potential option because it has the advantages of low energy consumption, adaptive learning, and high parallel computing. Biomolecular computing continues to develop with the rise of synthetic biology. Nanomaterials such as DNA can not only be used for logic operations but also for constructing neural networks and learning from training data, offering the possibility of realizing neuromorphic computing at the molecular level. In this review, we summarized the basic principles of neuromorphic computing in the silicon-and bio-computing, which includes artificial neural network models, such as the deep learning networks and the SNN-spiking neural networks. Then, we focused on the research progress of DNA computing. The DNA circuit, which is the basis for DNA computing, is an important technology for the regulation and processing of molecular information. DNA computing leverage DNA circuits to accomplish a variety of tasks based on the enzyme, the principle of chain replacement reaction mediated by the viscous terminal, and the schematic diagram of the "seesaw" logic gate circuit. Furthermore, we introduced the neuromorphic computing based on DNA computing. It is utilized that Hopfield associative memory experiment, winner-take-all neural networks, and multi-cellular systems to complete classification tasks. The genetic circuit based on Hopfield associative memory experiment can remember four single-stranded DNA patterns and recall the most similar one when faced with an incomplete one. The genetic circuit of winner-take-all neural networks was used to learn and memorize the handwritten digits "1" to "9" of defined categories, and then classify the test patterns. An adaptive artificial neural network is constructed by using the similarity between the struc

关 键 词：神经拟态计算 神经网络 合成生物学 DNA计算 人工智能 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]