In-memory computing with adaptive phase-change materials

The recently established collaborative research center (CRC) Intelligent Matter studies how responsive matter and its properties can be exploited to create intelligent systems. Among others, this broad topic includes the study of materials and devices for novel non von Neumann processors.

For this purpose, we collaborate with the AG Pernice to develop mixed electro-optical devices for in-memory computing. Such devices employ the pronounced optical contrast between the crystalline and amorphous states of phase-change materials to realize configurable weights of neural networks used, for instance, to classify objects in images.

Mixing electrical and optical pathways comes with a range of advantages: on one hand, switching these devices electrically allows efficient and precise adjustment of the weight. On the other hand, optical readout of the transmissive state enables a large bandwidth and the ability to feed multiple input data into the system at the same time. This can be achieved by using different wavelengths to encode information.

For developing the mixed electro-optical devices, our work group focusses on studying microscopically small heater structures for switching thin films of phase-change materials on waveguides. Furthermore, we are investigating new materials with suitable properties for this application.