Neuromorphic computing btech project .pptx

ece btech project

1. STANLEY COLLEGE OF ENGINEERING AND TECHNOLOGY FOR WOMEN (AUTONOMOUS), Abids , Hyderabad – 500 001 (Affiliated to Osmania University & Approved by AICTE) ( All eligible UG Courses are accredited by NBA & Accredited by NAAC with ‘A’ Grade) Department of Electronics and Communication Engineering M.E. (Embedded Systems) Academic Year 2024-25 NEUROMORPHIC COMPUTING INTERNAL GUIDE PRESENTED BY

2. CONTENTS Abstract Aim and Objective Introduction Literature Review System Architecture explanation with block diagram Flowchart IBM TrueNorth Neuromorphic Chip by IBM TrueNorth for Forecasting Epileptical seizures Software/Hardware tools used Outcomes/Results of the technology(concept) Advantages Applications Conclusion Future Scope References Acknowledgement

3. ABSTRACT Neuromorphic computing is an emerging field that seeks to mimic the structure and function of the human brain to create more efficient, adaptive, and scalable computing systems. Inspired by biological neural networks, neuromorphic systems integrate specialized hardware and algorithms designed to replicate the brain's architecture and processing methods. This approach aims to overcome the limitations of traditional computing, particularly in tasks involving pattern recognition, learning, and sensory processing, where the human brain excels. Neuromorphic computing are systems that use spiking neural networks (SNNs) , which more closely resemble the way neurons communicate via discrete spikes of activity. Neuromorphic chips, like IBM’s TrueNorth and Intel’s Loihi , utilize specialized circuits to replicate neural functions, offering significant improvements in energy efficiency, parallel processing, and real-time learning capabilities. As neuromorphic computing continues to advance, it holds the potential to revolutionize areas such as artificial intelligence, robotics, and autonomous systems, offering a path towards creating more intelligent machines that can learn and adapt in real-world environments.

4. AIM To explore the principles, technologies, and advancements of neuromorphic computing, highlighting its potential to revolutionize computing systems by mimicking the human brain's structure and function . OBJECTIVE To understand the core concepts of neuromorphic computing and its inspiration from biological neural networks. To analyze the role of spiking neural networks (SNNs) and their advantages in computation. To examine the architecture and capabilities of neuromorphic chips such as IBM's TrueNorth and Intel's Loihi . To identify key applications in artificial intelligence, robotics, and autonomous systems. To discuss the challenges, opportunities, and future directions of neuromorphic computing.

5. INTRODUCTION Neuromorphic computing is an innovative approach to computing inspired by the structure and function of the human brain. By emulating biological neural networks, it aims to create systems that are highly efficient, adaptive, and scalable. Unlike traditional computing architectures, neuromorphic systems utilize spiking neural networks (SNNs) that mimic the way neurons communicate via discrete spikes of activity. This enables energy-efficient processing and real-time learning for complex tasks like pattern recognition, visual perception, speech recognition, and decision-making. Neuromorphic hardware, such as IBM's TrueNorth and Intel's Loihi , leverages specialized circuits to replicate neural behavior, offering significant advancements in parallel processing and energy efficiency. With applications ranging from artificial intelligence to robotics and autonomous systems, neuromorphic computing holds the potential to revolutionize how machines learn and adapt in real-world environments.

6. LITERATURE SURVEY S.NO Year Name of the author Title of the paper Important findings 1. 2023 A. Roy, M. Nagaraj , C. Mihiranga Liyanagedera and K. Roy Live Demonstration: Real-time Event-based Speed Detection using Spiking Neural Networks Implemented DOTIE to detect a disk moving in a circular motion and identify the speed of rotation. Efficiently handle events by implementing DOTIE on Intel Loihi , a neuromorphic hardware suitable for spiking neural networks, and reveal a 14× reduction in energy consumption compared to the CPU implementation of DOTIE. 2. 2022 I. Sharma and Vanshika Evolution of Neuromorphic Computing with Machine Learning and Artificial Intelligence Paper discusses this benchmarking technology i.e. neuromorphic computing and how it is optimizing machine learning techniques by developing Spiking Neural Networks (SNNs) 3. 2020 Z. Yu, A. M. Abdulghani, A. Zahid, H. Heidari, M. A. Imran and Q. H. Abbasi An Overview of Neuromorphic Computing for Artificial Intelligence Enabled Hardware-Based Hopfield Neural Network Presents a comprehensive review and focuses extensively on the Hopfield algorithm's model and its potential advancement in new research applications 4. 2020 G. Tang, N. Kumar and K. P. Michmizos Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware Proposes a neuromorphic approach that combines the energy-efficiency of spiking neural networks with the optimality of DRL and benchmark it in learning control policies for mapless navigation 5. 2019 S. Yang et al Scalable Digital Neuromorphic Architecture for Large-Scale Biophysically Meaningful Neural Network With Multi-Compartment Neurons Presents a hardware efficient, scalable, and real-time computing strategy for the implementation of large-scale biologically meaningful neural networks with one million multi-compartment neurons (CMNs)

7. SYSTEM ARCHITECTURE WITH BLOCK DIAGRAM Spike input Spike output Neurons and synapses for processing & memory Neural Network Sensory input

8. EXPLANATION OF BLOCK DIAGRAM The system starts with sensory inputs that collect data from the environment, such as images, sounds, or touch. The input data is then preprocessed and converted into spike-based or event-based signals, mimicking how the brain processes information. The spikes or events are fed into a network of neurons , which process the data in a manner similar to biological neurons. Synapses , or the connections between neurons, control how information is transmitted, with adjustable strengths (synaptic weights). The processing units carry out the computation by simulating the behavior of interconnected neurons, making decisions based on the received inputs. Learning and memory mechanisms are incorporated to adjust the synaptic weights based on neuron activity, allowing the system to learn from experience.

9. The system then generates an output based on the processed information, which could be a decision, classification, or prediction. The neuromorphic computing system operates with high efficiency by using event-based communication instead of continuous signals, optimizing energy usage. Energy management ensures that the system consumes minimal power while processing data. This structure mimics how the human brain processes and learns from data, offering an energy-efficient and adaptable computational approach.

10. FLOWCHART Start Input data(sensory) Covert to spike Preprocessing layer(e.g., spike encoding) Input to neuron network Neuron activity (spiking neural network) Update synaptic weights (learning mechanism) Neuron firing (spikes generated) Decision/output End Yes Is the input data valid? No

11. TrueNorth Neuromorphic Chip by IBM TrueNorth was one of the neuromorphic (try to mimic human brain) chip released in 2015. Unlike the Von Neuman architecture, it does not use sequential programs that map instructions into linear memory. It implements spiking neurons. The program specifies the behavior of the neurons and the connectivity between them. The communication happened by sending spikes to each other. The encoding of data during communication was done using frequency, time, and the spatial distribution of spikes. By tiling 4096 neurosynaptic cores on a TrueNorth chip, it scales up to a highly parallel architecture, where each core implements 256 neurons and 64k synapses. Each subsection is just a bipartite graph of neurons. Hardware representation show in below figure.

12. The computation of a neurosynaptic core proceeds according to the following steps. A neurosynaptic core receives spikes from the network and stores them in the input buffers. When a 1 kHz synchronization trigger signal called a tick arrives, the spikes for the current tick are read from the input buffers and distributed across the corresponding horizontal axons. Where there is a synaptic connection between a horizontal axon and a vertical dendrite, the spike from the axon is delivered to the neuron through the dendrite. Each neuron integrates its incoming spikes and updates its membrane potential. When all spikes are integrated in a neuron, the leak value is subtracted from the membrane potential. If the updated membrane potential exceeds the threshold, a spike is generated and sent into the network. Applications of TrueNorth: IBM developed the TrueNorth neuromorphic chip. The chip is used in visual object recognition and has lower power consumption than traditional von Neumann hardware. The chip is used in applications that require the processing of sensory inputs like vision, sound, or touch. For example, in the development of hearing aids or advanced prosthetics that can respond to sensory inputs in real-time.

13. TRUENORTH FOR FORECASTING EPILEPTICAL SEIZURES IBM's TrueNorth chip can be applied to forecast epileptic seizures by analyzing brain activity patterns, leveraging its neuromorphic design for real-time, low-power processing. Here's how the process works: Input Data : Electroencephalogram (EEG) signals from the patient's brain are used as input. These signals capture brainwave patterns, which contain indicators of potential seizures. Preprocessing : The EEG data is preprocessed to filter out noise and extract relevant features, such as spike-wave discharges or other abnormal brainwave patterns. Neuromorphic Analysis : TrueNorth processes the EEG signals using its spiking neuron architecture. It detects and classifies abnormal patterns ( preictal states) that typically occur before a seizure. Each neuron focuses on specific features, mimicking how the brain processes information.

14. Pattern Recognition : Pre-trained machine learning models deployed on TrueNorth identify patterns associated with the onset of seizures. These models are trained on historical EEG data of the patient. Prediction and Alert : Once TrueNorth identifies a preictal state, it forecasts the likelihood of an imminent seizure. Alerts can be sent to caregivers or medical devices, enabling preventive measures, such as administering medication or initiating protective protocols. Real-Time and Energy-Efficient Processing : TrueNorth’s low power consumption and event-driven nature make it ideal for portable, real-time seizure monitoring systems, such as wearable devices. This application enhances epilepsy management by providing early warnings, improving patient safety, and reducing the impact of seizures on daily life. It is further used in Performing multiple tasks without a timekeeping mechanism.  It can help computers recognize faces and pick out specific voices.  Detecting people, bicyclists, cars, trucks, and buses in videos, Controlling a TV with gestures. Supercomputing calculations.

15. SOFTWARE/HARDWARE USED IN NEUROMORPHIC COMPUTING SOFTWARE : NEST Simulator : A simulator for large-scale brain-like networks of spiking neurons. Brian2 : A Python-based framework for simulating spiking neural networks. Matlab /Simulink : Used for modeling neuromorphic systems. PyNN : A Python package for building spiking neural network models. HARDWARE : Intel Loihi : A neuromorphic chip designed to simulate brain-like processes in real-time. IBM TrueNorth : A brain-inspired chip for low-power, high-performance computing. SpiNNaker : A supercomputing platform designed for simulating large-scale spiking neural networks. BrainScaleS : A neuromorphic system focusing on energy-efficient computation

16. OUTCOMES OF NEUROMORPHIC COMPUTING Neuromorphic computing leads to the creation of systems that are significantly more energy-efficient than conventional computing. It allows for real-time data processing and pattern recognition, similar to the brain. It mimics the brain's structure and function, allowing systems to adapt and learn from experience using biologically inspired learning rules. The technology leads to significant power savings compared to conventional computing by using event-based communication rather than continuous signal processing. Neuromorphic systems can process complex data, such as images and sounds, more efficiently, enabling faster and more accurate decision-making. It offers scalable solutions, allowing for the development of large, complex systems that can handle massive amounts of real-time data.

17. ADVANTAGES It supports real-time processing of sensory data, making it ideal for dynamic, time-sensitive applications. The system is adaptive, capable of learning from experience and improving its performance over time. It can handle complex, unstructured data, such as visual and auditory inputs, with greater efficiency. Neuromorphic computing mimics the brain's neural structure, enabling more natural and flexible decision-making processes. It is scalable, allowing for the creation of large networks capable of processing vast amounts of data. The use of spiking neural networks enables more biologically plausible models for intelligent systems It offers potential for improving AI systems by providing better pattern recognition, classification, and decision-making capabilities.

18. APPLICATIONS Neuromorphic computing has many applications, including:  Autonomous systems : Self-driving cars, drones, and automated guided vehicles Embedded systems : Control circuits, signal processing, and power electronics Internet of Things : Smart automation Remote sensing : High energy physics Computational sciences : Modeling and simulation in neuroscience and epidemiology Machine learning : Neuromorphic processors are expected to be low power machine learning accelerators

19. CONCLUSION Neuromorphic computing represents a revolutionary approach to computation, inspired by the structure and functioning of the human brain. By mimicking neural processes through spiking neural networks and event-based communication, it offers remarkable advantages such as energy efficiency, real-time data processing, and adaptive learning. With applications in robotics, AI, healthcare, and smart sensors, neuromorphic computing holds the potential to transform industries by enabling intelligent systems that are more natural, efficient, and scalable. As technology advances, it will pave the way for more sophisticated and powerful solutions in diverse fields, making it a promising frontier for the future of computing .

20. FUTURE SCOPE Improved Hardware : Development of more advanced neuromorphic chips that can handle even larger and more complex tasks with greater efficiency. AI Integration : Combining neuromorphic computing with traditional AI models to create hybrid systems that offer enhanced learning and decision-making capabilities. Brain-Machine Interfaces : Advancing technologies that allow for more intuitive and seamless interaction between the brain and external devices. Autonomous Systems : Expanding the use of neuromorphic computing in self-driving cars, drones, and robots, making them smarter and more efficient. Healthcare Applications : Utilizing neuromorphic systems for personalized medicine, brain health monitoring, and advanced diagnostics. Real-time Data Processing : Enabling more efficient processing of large volumes of real-time data, especially in fields like IoT, smart cities, and environmental monitoring.

21. REFERENCES A. Roy, M. Nagaraj , C. Mihiranga Liyanagedera and K. Roy, "Live Demonstration: Real-time Event-based Speed Detection using Spiking Neural Networks,"  2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , Vancouver, BC, Canada, 2023, pp. 4081-4082, doi : 10.1109/CVPRW59228.2023.00428. I. Sharma and Vanshika , "Evolution of Neuromorphic Computing with Machine Learning and Artificial Intelligence,"  2022 IEEE 3rd Global Conference for Advancement in Technology (GCAT) , Bangalore, India, 2022, pp. 1-6, doi : 10.1109/GCAT55367.2022.9971889 Z. Yu, A. M. Abdulghani , A. Zahid , H. Heidari , M. A. Imran and Q. H. Abbasi , "An Overview of Neuromorphic Computing for Artificial Intelligence Enabled Hardware-Based Hopfield Neural Network," in  IEEE Access , vol. 8, pp. 67085-67099, 2020, doi : 10.1109/ACCESS.2020.2985839. G. Tang, N. Kumar and K. P. Michmizos , "Reinforcement co-Learning of Deep and Spiking Neural Networks for Energy-Efficient Mapless Navigation with Neuromorphic Hardware,"  2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , Las Vegas, NV, USA, 2020, pp. 6090-6097, doi : 10.1109/IROS45743.2020.9340948. S. Yang  et al ., "Scalable Digital Neuromorphic Architecture for Large-Scale Biophysically Meaningful Neural Network With Multi-Compartment Neurons," in  IEEE Transactions on Neural Networks and Learning Systems , vol. 31, no. 1, pp. 148-162, Jan. 2020, doi : 10.1109/TNNLS.2019.2899936.