On-chip analogue signal processing using molybdenum disulfide reservoir computing

In the era where edge intelligence powers robotics, autonomous systems, and various IoT sensors, real-time processing of the signals is vital yet challenging to achieve with the current digital systems, particularly when latency, reliability, and privacy are concerned. Here, we present a groundbreaking analogue reservoir computing prototype for on-chip parallel signal processing using solution-processed molybdenum disulfide (MoS2) fading memory. By harnessing the intrinsic nonlinearity and fading memory relaxation processes of the MoS2 memory, the prototype performs reservoir activation with rich dynamics while, notably, requiring no periphery circuits for virtual-node-coupling or recurrent connections. Benchmarked on chaotic system prediction, e.g. Lorenz-63 with NRMSE of 0.01, the prototype manifests the potential in enabling robust on-chip analogue signal processing in edge intelligence, e.g. assistive driving. The prototype proves successful human electrophysiological and vehicular signal processing on chip, with an accuracy of 80-95% achieved in the tasks, promising to enhance the driving safety. This analogue reservoir computing, with the compact hardware design along with the scalability of solution-processed electronics, is positioned to become a transformative signal processing technology for embedded edge intelligence systems.