マイクロキャビティレーザーベースのフォトニックリザーバーコンピューティングの確率的モデリング：情報処理能力の観点

The growing demand for efficient temporal processing has highlighted the limitations of von Neumann architectures, making photonic time-delay reservoir computing (TDRC) an attractive alternative. We develop a stochastic modeling framework for semiconductor microcavity laser-based TDRC, using Information Processing Capacity (IPC) to quantify the trade-off between linear memory and nonlinear computation. Systematic parameter sweeps reveal optimal virtual node spacing, resonant degradation at rational τ/T ratios, and laser-size-dependent performance via spontaneous emission coupling (β). While reduced cavity dimensions initially enhance nonlinearity, noise ultimately degrades prediction accuracy in Mackey-Glass and Santa Fe tasks. This work bridges theoretical metrics with hardware design, enabling optimized photonic neuromorphic systems for real-time forecasting.