Scientists from the University of Münster in Germany have developed a new type of computing architecture for artificial intelligence (AI) applications. This architecture, called event-based architecture, uses photonic processors that transport and process data using light. Similar to the human brain, this architecture allows for the continuous adaptation of connections within the neural network, which is the basis for learning processes.
Benefits of Event-Based Architecture
This event-based architecture offers several advantages over traditional digital computer processes. By emulating the working principles of biological neural networks, it promises faster and more energy-efficient data processing. Compared to electronic processors, light-based processors have a higher bandwidth, enabling them to handle complex computing tasks with lower energy consumption.
The Study and its Findings
In the study conducted by a team of researchers from the University of Münster, nearly 8,400 optical neurons made of waveguide-coupled phase-change material were used to create a network. The researchers demonstrated that the connections between these neurons can strengthen or weaken (synaptic plasticity), and new connections can be formed or eliminated (structural plasticity). Unlike previous studies, the synapses in this architecture were not hardware elements but were coded based on the properties of optical pulses, such as wavelength and intensity.
By integrating several thousand neurons on a single chip and connecting them optically, this architecture shows promise for high-performance and energy-efficient AI computing.
Future Implications and Methodology
The researchers aim to develop an optical computing architecture that can compute AI applications quickly and efficiently. The use of non-volatile phase-change material allows for permanent data storage without the need for an energy supply.
In their study, the researchers tested the performance of the neural network by training it to distinguish between German and English texts using an evolutionary algorithm. They used the number of vowels in the text as the recognition parameter.
This research represents an important step towards the development of advanced AI systems and the advancement of computing technology.