Essay Example on Concern and Possible Solutions in Neuromorphic Synaptic Models

Increases or decreases in their activity is extremely important in biological neural systems as it is generally accepted that learning in the brain and formation of memories arise from synaptic modifications 9 The ability of biological synapses to prominent STP and LTP is one of their essential features Concern and Possible Solutions in Neuromorphic Synaptic Models The main issue about neuromorphic synaptic models with learning capability is the storage of the synaptic weights referred as power of communication between two nodes which are confined as well as limited accuracy As a result when introducing new memories it makes memory recovery impossible 10 To overcome this problem various synaptic storage implementations have been suggested below 1 A vital solution is to introduce digital memory cells Meanwhile if the processing of the synapse is to be analog additional analog to digital and vice versa is required to interaction between the analog and digital system in this scenario needed more space on the chip die and enhances complexity and power budget 2 For digital memory cells the alternative of this is a capacitive storage That provides a better solution However if the storage capacitance is connected to a transmission gate the unavoidable leakage would call for mechanisms to lessen the issue by placing a using a big capacitor but have to suffers the occupying the space and power consumption 3 Floating gate transistors provides comprehensive analog conditions long term without support of external clock approach and continuous preserve storage while the break in the power supplies 11 

However the major concern while using the FG device is difficult to maintain accurate programming process 4 Problem solving of long time storage is done by using only two stable synaptic modes has described in 11 A bistable circuit is used to conserve memory when the presynaptic activity is low Working of this circuit is to restore the synaptic mode to high or low states considering the weight either above or below the threshold level 5 Now a day s progress in nanotechnology has lead in new devices that have capability to weight storage of multidirectional data meanwhile permit for synaptic plasticity Three devices are referred to the memristor 12 Transmission of Data Human brain contains a number of connections between neurons that is lurch Approximately 1011 neurons in brain on the other hand a cortical neuron makes 10 4 connections with other neurons In this way brain contains a total of 1015 serial connections among neurons The human brain routs connection by using 3 D volume while integrated circuits limited to 2 D layer 

The basic idea of address event representation AER protocol which is used to send spikes or transmission address beyond chip having analog and digital segments for local and long digital communication is to minimize this effect that is excessively used in figure As seen in figure one digital bus shared by two neurons instead of individual connection for each

An action potential from a neuron is encoded as a digital address i e a number that identifies the neuron producing the action potential and is transmitted on this time-multiplexed digital bus The receiving address by the receiver converted into pulses and distributed to the receiving neurons that are connected to the sender 13 When unit s internal mode has crossed a threshold level prompt request for transmission of address then request is granted in the form of sending address onto a common bus Arbitration circuits on the periphery of the chip ensure that the addresses are sent off sequentially After this AER protocol makes sure that when both the transmitter and the receiver allowed writing and reading from the bus Neural networks 

The first step towards neural networks developed after the introduction of simplified neurons by McCulloch and Pitts in 1943 Artificial Neural Networks ANN comes with a significant role in problem-solving of complex engineering algorithms by improving their computing ability Basically it consists of the computational network which is mimicking of complicated and parallel computing capability of a biological neural network An artificial neuron is a computational model stimulated in the natural neurons Natural neurons receive signals through synapses located on the dendrites or membrane of the neuron When the signals received are strong enough surpass a certain threshold the neuron is activated and emits a signal through the axon 

This signal might be sent to another synapse and might activate other neurons The Artificial neurons basically consist of inputs like synapses which are multiplied by weights strength of the respective signals and then computed by a mathematical algorithm function which determines the activation of the neuron ANNs combine artificial neurons in order to process information 14 Depicted in figure 3 Figure 3 An artificial neuron In comparison Stuttgart Neural Network Simulate SNNs differs from ANNs on the basis of two main facts SNNs combine the concept of time in neural simulation Spike based neurons and synapses mimic their biological counterparts The characteristics of SNNs can be