Speaker
Description
Physical reservoir computing offers an energy-efficient alternative to conventional neural networks, where the intrinsic memory capacity within the physical system plays a central role. In this work, we demonstrate that memory capacity can be engineered extrinsically in memoryless systems by exploiting the computational space-time tradeoff, substituting temporal memory with spatial degrees of freedom. Our approach utilizes multidimensional input nodes to function as a spatial memory axis, thereby removing the dependency on intrinsic history-dependent dynamics in the reservoir. We validate this framework through numerical simulations of a generalized single quantum dot, whose discrete energy states provide strong nonlinearity crucial for reservoir computing as well. To extract the processed information, we read out local observables equivalent to transport spectroscopy, accounting for the interaction between the system and measurement apparatus. By coupling this inherent nonlinearity with our extrinsic memory, we show that memoryless quantum reservoir can achieve high performance on both chaotic Mackey-Glass future prediction and nonlinear transformation tasks. Furthermore, by analyzing the geometry of the quantum state trajectories, we identify the physical mechanism underlying this memory emergence: extrinsic memory constructs a hysteresis loop within the quantum Hilbert space, and this loop becomes topologically stable when the evolution of the system state synchronizes with the input signal's frequency. Our work decouples reservoir computing from material-specific memory properties, significantly expanding the range of candidate systems for quantum neuromorphic computing.