To main content

Neuromorphic computing

Spike based machine learning or what is called Neuromorphic computing might be the next step in evolution of AI.

Contact persons

The field of Artificial Intelligence (AI) has recently experienced significant advancements, which have fueled our desire for intelligent computing devices in all shapes and sizes. These devices range from recommender systems, predictive maintenance systems, and digital twins, to autonomous vehicles, robots, and more intuitive and predictive interfaces with our personal devices such as tablets, mobile phones and smartwatches.

As AI technology continues to develop, its limitations are becoming more apparent. In the AI method deep learning, deep neural networks (DNNs) are trained to solve complex problems. The downside is that deep neural networks require enormous amounts of computational power and pre-collected and labeled data. This type of data is not always available for industrial or real-world situations.

Also, many emerging AI applications are expected to operate in unpredictable and challenging environments with constraints on power, latency, and data access restricting the ability to perform edge computing.

Neuromorphic computing represents a new approach to computer architecture that is inspired by the structure and function of biological neural networks in the brain. Unlike traditional computing, which relies on algorithms and programming abstractions, neuromorphic computing is guided by the principles of biological neural computation.

This departure from conventional computing allows for a significant increase in efficiency and performance, as well as the potential to support the full range of intelligent information processing that living brains can perform, at microwatt power levels and millisecond response times. The ultimate goal is to develop a computer architecture that is inherently capable of supporting intelligent processing at levels that rival the processing power of biological neural circuits.

The advancements in neuromorphic hardware and training techniques for deep spiking neural networks (SNNs) have enabled the technology to move beyond the experimental phase and become viable for real-world applications. SINTEF, with its extensive expertise in various industries and applied domains, brings together domain experts and machine learning specialists to collaborate on projects that combine knowledge from neuromorphic sensors and computing.

Explore research areas