Construction And Application Of DNA Neurons Based On Strand Displacement

As silicon-based computers increasingly approach the limits of Moore's Law,exploring new computing devices has become a hot research topic.DNA computing has attracted extensive attention from researchers due to its high parallelism,large-scale storage capacity and low energy consumption.DNA circuits can be divided into digital and analog types.A basic work of digital DNA circuits is to construct DNA logic gates.An emerging research of analog DNA circuits is to construct DNA neural networks with capabilities of classification and recognition.Whether it is a digital DNA logic gate or an analog DNA neural network,DNA neurons that use DNA molecules to encode and deliver information can play a good role in it.Therefore,this paper studies DNA neurons based on strand displacement technology,and designs digital DNA logic gates and analog DNA neural networks by cascading different DNA neurons.The main research contents and results are as follows:In order to construct highly sensitive DNA logic gates and DNA neural networks capable of processing analog data,three types of DNA neurons with step function,Re LU function and Sigmoid function as activation functions were designed.First,a DNA matrix multiplier was constructed on the basis of a variable-gain DNA amplifier to solve the problem of weighted-sum operation of neurons.Then,to solve the problems of low sensitivity and uncontrollable steady-state output of traditional seesaw gates,a highprecision step function calculator is constructed;Finally,the Re LU function operator and Sigmoid function operator necessary for the analog DNA recognition network are proposed.Because existing DNA logic gates cannot distinguish between logic 0 signals and no input signals,and the composition of some logic gates are complex,a variety of digital logic gates were constructed based on DNA neurons,including AND,OR,NOT,XOR,XNOR gates,and one-bit binary full adder.The Logic gates based on DNA neurons represents logic 0 when relative concentration of DNA strands is 0.1,and logic 1 when relative concentration of DNA strands is 0.9,which can effectively distinguish between logic 0 signal and no input signal.By cascading multiple neurons,XOR gates,XNOR gates,and full adder circuits can be constructed directly without composing AND,OR,and NOT gates,which effectively reduces the complexity of logic gates.In order to solve the problem that existing DNA neural networks can only handle digital data,a DNA neural network capable of processing analog data is constructed based on DNA neurons.First,for the two-moons dataset,a DNA recognition network capable of binary classification is constructed;Then for the iris dataset,a DNA neural network for multivariate classification of analog data is realized based on the Winner-Take-All network.The simulation results have verified the effectiveness of the DNA neural network.