Neuromorphic computing with halide perovskites
Modern neural networks achieve impressive performance in tasks such as image recognition and language processing. Nevertheless, this progress comes at a cost of high energy consumption. The human brain performs complex computations using very little energy in comparison. Neuromorphic computing aims to develop brain-inspired networks in hardware to reduce energy consumption.
In his thesis, Jeroen de Boer explores the use of halide perovskites as building blocks for neuromorphic hardware. These semiconductor materials are particularly attractive because they support high mobility of ions within the material, enabling devices whose electrical resistance changes in a controllable way. This behavior resembles how biological synapses and neurons operate. In addition, halide perovskites are excellent light absorbers, allowing devices to respond to both electrical and optical signals.
A key challenge has been that halide perovskites are difficult to process with standard microfabrication techniques, which has prevented their use in dense networks on a chip. De Boer introduces a fabrication procedure that is suitable for the integration of halide perovskites into large neuromorphic networks. Using this approach, artificial synapses and neurons are demonstrated that operate with very low energy consumption and exhibit brain-like behavior, including stochastic spiking.
De Boer further shows how combining electrical and optical signals can enable brain-inspired learning with an attention mechanism and reservoir computing. Simulations based on experimental measurements demonstrate that halide perovskite networks could efficiently process images and video data. Together, these results highlight the potential of halide perovskites for future low-power, brain-inspired computing systems.