Recent advances in physical reservoir computing (RC) are reviewed, focusing on physical implementations of reservoirs using various physical systems. RC is a computational framework suitable for temporal/sequential data processing, derived from recurrent neural network models like echo state networks (ESNs) and liquid state machines (LSMs). The reservoir is fixed, and only the readout is trained using simple methods like linear regression. This makes RC faster and more efficient than traditional RNNs, and suitable for hardware implementation using physical systems. Physical RC has gained attention in diverse research areas due to its potential for low-cost, energy-efficient hardware. The review classifies physical RC systems based on the type of physical phenomenon used for the reservoir. It discusses various physical reservoirs, including delayed dynamical systems, cellular automata, and coupled oscillators. Electronic RC systems are implemented using analog circuits, FPGAs, VLSIs, and memristive units. Photonic RC systems use optical node arrays and time-delay systems. Spintronic, mechanical, and biological RC systems are also explored, with biological RC involving brain regions and in-vitro cultured cells. Recent trends in RC include applications in machine learning, such as pattern classification, time series forecasting, and image recognition. New RC models have been developed to improve performance, incorporating diverse reservoir elements and learning algorithms. Physical implementations of RC have attracted significant attention due to their potential for energy-efficient hardware and real-time processing. Challenges remain in optimizing physical reservoirs for computational performance and practical applications. The review highlights the potential of physical RC for next-generation machine learning systems and emphasizes the importance of interdisciplinary research in advancing this field.Recent advances in physical reservoir computing (RC) are reviewed, focusing on physical implementations of reservoirs using various physical systems. RC is a computational framework suitable for temporal/sequential data processing, derived from recurrent neural network models like echo state networks (ESNs) and liquid state machines (LSMs). The reservoir is fixed, and only the readout is trained using simple methods like linear regression. This makes RC faster and more efficient than traditional RNNs, and suitable for hardware implementation using physical systems. Physical RC has gained attention in diverse research areas due to its potential for low-cost, energy-efficient hardware. The review classifies physical RC systems based on the type of physical phenomenon used for the reservoir. It discusses various physical reservoirs, including delayed dynamical systems, cellular automata, and coupled oscillators. Electronic RC systems are implemented using analog circuits, FPGAs, VLSIs, and memristive units. Photonic RC systems use optical node arrays and time-delay systems. Spintronic, mechanical, and biological RC systems are also explored, with biological RC involving brain regions and in-vitro cultured cells. Recent trends in RC include applications in machine learning, such as pattern classification, time series forecasting, and image recognition. New RC models have been developed to improve performance, incorporating diverse reservoir elements and learning algorithms. Physical implementations of RC have attracted significant attention due to their potential for energy-efficient hardware and real-time processing. Challenges remain in optimizing physical reservoirs for computational performance and practical applications. The review highlights the potential of physical RC for next-generation machine learning systems and emphasizes the importance of interdisciplinary research in advancing this field.