<p>Reservoir computing (RC) has gained significant attention for the ability to process temporal data with high-dimensional dynamical systems. However, the existing photonic RC architectures have to face limitations in scalability, nonlinearity, and integration. In this work, we have proposed and numerically investigated a delay-based photonic reservoir computing architecture compatible with thin-film lithium niobate (TFLN) integration, which combines high-speed modulators, cascaded microring resonators, periodically poled lithium niobate (PPLN) waveguides, and photodetectors. This architecture exploits both the fundamental and second-harmonic frequency components simultaneously, thereby enhancing the effective dimensionality and dramatically reducing the prediction error. With numerical simulations, the performance of the proposed architecture has been evaluated through time-series prediction tasks, including the Santa Fe chaotic dataset and the 10-order Nonlinear Auto-Regression Moving Average (NARMA-10) benchmarks. According to the results, substantial improvements in prediction accuracy have been achieved with NMSEs of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">$2.6\times 1{0}^{-3}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>2.6</mn> <mo>×</mo> <mn>1</mn> <msup> <mrow> <mn>0</mn> </mrow> <mrow> <mo>−</mo> <mn>3</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">$3.9\times 1{0}^{-3}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mn>3.9</mn> <mo>×</mo> <mn>1</mn> <msup> <mrow> <mn>0</mn> </mrow> <mrow> <mo>−</mo> <mn>3</mn> </mrow> </msup> </mrow> </math></EquationSource> </InlineEquation>, respectively. The architecture reduces the prediction error by up to approximately one order of magnitude compared with single-component configurations, highlighting the benefits of combining fundamental and second-harmonic components. We believe that this work would set the stage for future all-optical reservoir computing systems, with potential applications in real-time data processing, machine learning, and complex system modeling.</p>

Delay-based photonic reservoir computing on thin-film lithium niobate with time–wavelength-coupled virtual nodes

期刊论文
出版模式：

开放获取
发表日期：

2026年7月21日

查看全文

Deyang Kong,
Yuxuan Liao,
Zhe Li,
Yongzhuo Li,
Xue Feng,
Yidong Huang

摘要

Reservoir computing (RC) has gained significant attention for the ability to process temporal data with high-dimensional dynamical systems. However, the existing photonic RC architectures have to face limitations in scalability, nonlinearity, and integration. In this work, we have proposed and numerically investigated a delay-based photonic reservoir computing architecture compatible with thin-film lithium niobate (TFLN) integration, which combines high-speed modulators, cascaded microring resonators, periodically poled lithium niobate (PPLN) waveguides, and photodetectors. This architecture exploits both the fundamental and second-harmonic frequency components simultaneously, thereby enhancing the effective dimensionality and dramatically reducing the prediction error. With numerical simulations, the performance of the proposed architecture has been evaluated through time-series prediction tasks, including the Santa Fe chaotic dataset and the 10-order Nonlinear Auto-Regression Moving Average (NARMA-10) benchmarks. According to the results, substantial improvements in prediction accuracy have been achieved with NMSEs of $2.6\times 1{0}^{-3}$ $2.6 \times 1 0^{- 3}$ and $3.9\times 1{0}^{-3}$ $3.9 \times 1 0^{- 3}$ , respectively. The architecture reduces the prediction error by up to approximately one order of magnitude compared with single-component configurations, highlighting the benefits of combining fundamental and second-harmonic components. We believe that this work would set the stage for future all-optical reservoir computing systems, with potential applications in real-time data processing, machine learning, and complex system modeling.