NEUROMORPHIC COMPUTING
Need help with a related project topic or New topic? Send Us Your Topic
DOWNLOAD THE COMPLETE PROJECT MATERIAL
NEUROMORPHIC COMPUTING
Introduction
Neuromorphic computing is now the preferred design for applications such as cognitive processing, replacing the von Neumann computing architecture.
Based on highly connected synthetic neurons and synapses, biologically inspired methodologies are used to develop theoretical neuroscientific models and demanding machine learning algorithms.
The von Neumann architecture is the primary computing standard for machines. However, there are substantial disparities in organisational structure, power requirements, and processing capacities when compared to the human brain’s functioning model [1].
In recent years, neuromorphic calculations have evolved as an auxiliary architecture for the von Neumann system. Neuromorphic calculations are used to build a programming framework.
These computations allow the system to learn and generate applications that simulate neuromorphic functions. These include neuro-inspired models, algorithms, and learning methods, as well as hardware, equipment, support systems, and applications [2].
Neuromorphic architectures have various critical and unique requirements, including increased connectivity and parallelism, low power consumption, memory collocation, and processing [3].
Its superior capacity to do complicated computing tasks at higher speeds than traditional von Neumann systems, while also saving power and having a smaller footprint.
These qualities are the bottleneck of the von Neumann architecture, hence the neuromorphic architecture will be evaluated as a suitable solution for implementing machine learning algorithms [4.
There are 10 primary reasons for employing neuromorphic architecture: real-time performance, parallelism, von Neumann bottleneck, scalability, low power, footprint, fault tolerance, faster, online learning, and neuroscience [1].
Real-time performance is the primary driving factor behind the neuromotor system. These devices can often outperform von Neumann architectures in neural network computing applications due to parallelism and hardware acceleration [5].
In recent years, the focus of neuromorphic system development has shifted towards low power consumption [5] [6] [7]. Biological neural networks are essentially asynchronous [8], and event-based computing models can be used to drive the brain’s efficient data processing [9].
However, coordinating the transmission of asynchronous and event-based tasks in big systems presents a barrier in the von Neumann design [10]. The hardware implementation of neuromorphic computing is advantageous to large-scale parallel computing architectures because it incorporates both processing memory and computation in neuron nodes and delivers ultra-low power consumption during data processing.
Furthermore, scalability allows for the creation of a large-scale neural network. Because of the aforementioned benefits, neuromorphic architecture is preferred over von Neuman for hardware implementation [11].
The fundamental issue with neuromorphic calculations is how to organise the neural network model. Biological neurons typically consist of cell bodies, axons, and dendrites.
The neuron models implemented by each component of the given model are classified into five classes based on the type of model, which is physiologically and computationally driven.
Artificial Neural Networks
An Artificial Neural Network (ANN) is a set of nodes inspired by the biological human brain. The goal of ANN is to do cognitive tasks like problem solving and machine learning. The mathematical models of the ANN were developed in the 1940s, but it remained silent for a long period (Maass, 1997).
ANNs have gained popularity with the success of ImageNet2 in 2009 (Hongming et al., 2018). The reason for this is the advancement of ANN models and hardware systems that can handle and implement these models. (Sugiarto and Pasila, 2018). The ANNs can be classified into three generations based on their computing units and performance (Figure 1).
Neuromorphic computing
Figure 1. Generations of Artificial Neural Networks.
The first generation of ANNs began in 1943 with the work of Mc-Culloch and Pitts (Sugiarto & Pasila 2018). Their approach was based on a neural network computational model in which each neuron is referred to as a “perceptron”.
Widrow and his students enhanced their model in the 1960s by adding extra hidden layers (Multi-Layer Perceptron) for increased accuracy, which they dubbed MAGDALENE.
However, first-generation ANNs were distant from biological models, producing only digital outputs. Essentially, these were decision trees with if and else conditions.
The second generation of ANNs improved on the prior generation by incorporating functions into the decision trees of the first-generation models. The functions interact with each visible and buried layer of the perceptron to form the structure known as “deep neural networks”. (Patterson, 2012; Camuñas-Mesa et al., 2019).
Thus, second-generation models are more similar to biological brain networks. The functions of second-generation models remain an active topic of research, and existing models are in high demand from markets and science.
Most recent breakthroughs in artificial intelligence (AI) are based on these second-generation models, which have demonstrated their accuracy in cognitive tasks. (Zheng and Mazumder, 2020).
Need help with a related project topic or New topic? Send Us Your Topic