Photonic circuits are of interest for all applications that require data to be processed very quickly and without large amounts of energy. Because light particles work in them instead of electrons. This opens the door to completely new optical computer architectures that do not work like traditional electronic computers, but are modeled on how the human brain works.
Calculating according to the model of the brain
The focus of our working group lies in the development and improvement of these computing architectures. Inspired by the way the human brain is built and works, neuromorphic computers are important in providing artificial intelligence (AI) hardware. AI applications have already arrived in many areas of our everyday lives. For example, they help with speech recognition, encrypt data on the smartphone, support search functions, are used for pattern recognition on the Internet and are essential for safety in autonomous driving.
All of these applications pose enormous challenges for conventional electronic computers because data storage and data processing are separate in them. This type of electronic architecture makes it necessary to exchange data continuously via special systems, so-called bus systems, which leads to sequential clocking and limits data throughput.
Many degrees of freedom – one calculation step
The human brain handles data completely differently: it is processed locally with a very high level of parallelism, and processing is often analog, i.e. continuous, and not digital, i.e. in stages. This is important for all arithmetic operations that are required in AI. The “photonic neuromorphic calculators” we are researching promise to meet these high standards.
Photonic neuromorphic calculators use light to perform addition and multiplication as elementary mathematical operations. The actual arithmetic task is translated into the coding of the numbers to be calculated and the measurement of the result. Photonic neuromorphic computers can do this in a single step because, unlike electrons, light particles have many degrees of freedom that allow inherent, simultaneous data processing – that is the special charm of optical computing methods.