Software prototyping of neuromorphic ICs enabling brain inspired, ultra efficient parallel computation for low power Edge AI and predictive maintenance.
Supervisor: Dr Mohammed Waqas Mughal
School: Engineering
Industry Partner: MindSilicaAI
Description:
Artificial intelligence is increasingly embedded in everyday technologies, from smart sensors and wearable devices to autonomous robots and industrial automation. However, today’s AI hardware is reaching fundamental limits as traditional computing systems are based on the Von Neumann architecture, where memory and processing are separated. This causes excessive data movement, high energy consumption, and poor scalability, especially for applications that require real‑time decision‑making, low latency, and operation under strict power constraints. As AI workloads continue to grow, these inefficiencies present a major barrier to deploying intelligent systems sustainably and at scale.
The main need addressed by this project is for ultra‑low‑power, energy‑efficient AI hardware that can process data locally (“on‑device” or “at the edge”) rather than relying on energy‑hungry cloud infrastructure. This need is shared by AI hardware companies, edge‑AI device manufacturers, robotics and autonomous‑system developers, and sectors such as healthcare, industrial sensing, defence, and consumer electronics. The market challenge lies in delivering hardware that combines high performance, fast response, and adaptability while dramatically reducing power consumption and operating cost.
This project focuses on the software prototyping and validation of neuromorphic integrated circuit (IC) architectures designed to enable brain‑inspired, ultra‑efficient parallel computation for next‑generation low‑power Edge‑AI and predictive maintenance applications. Neuromorphic systems draw inspiration from the structure and operation of biological neural networks, offering a fundamentally different computing paradigm that emphasises event‑driven processing, massive parallelism, and energy efficiency.
The primary objective of the project is to develop a comprehensive software prototype and simulation into Python/MATLAB framework that models key neuromorphic hardware building blocks, such as spiking neurons, synapses, and memory, and evaluates their system‑level performance prior to full hardware implementation. This approach significantly reduces development risk while enabling rapid exploration of architectural trade‑offs, learning rules, and workload characteristics relevant to real‑world applications.
By targeting Edge‑AI scenarios, the proposed neuromorphic architectures aim to overcome the limitations of conventional von Neumann computing, particularly high-power consumption and memory bottlenecks. The software prototype will demonstrate how neuromorphic ICs can perform real‑time inference, pattern recognition, and anomaly detection locally at the edge, eliminating the need for continuous cloud communication and enabling ultra‑low‑latency operation under tight power constraints.
Other key application domain for this project is predictive maintenance, where continuous monitoring of sensor data is required to detect faults or degradation in industrial systems. Neuromorphic computing is particularly well suited to such tasks due to its ability to process sparse, asynchronous data streams efficiently and to learn temporal patterns directly within the hardware. The software prototype will therefore be used to evaluate representative predictive maintenance workloads, demonstrating reductions in power consumption and computational complexity compared to conventional digital processing approaches.
Overall, this project will establish a proof‑of‑concept software platform that bridges the gap between algorithm design and neuromorphic hardware realisation, providing a robust foundation for future IC development, hardware prototyping, and commercial spin‑out activities in the fields of Edge‑AI and intelligent sensing.