LAUSR

The accelerated advancement of artificial intelligence, particularly in large‐model‐based applications, intensifies the demands on processing systems. Integrated photonic computing, when executing the densest operations of matrix multiplication in AI tasks, shows promise in revolutionarily reducing the processing latency and energy consumption. Herein, a combined pulse optical weighting scheme is proposed for photonic synapses, which are indispensable in photonic computing. Employing a non‐volatile phase‐change material with 4 μm$\umu{\rm m}$ in length and 30 nm in thickness covered on a straight waveguide, a transmission contrast of 31.2 dB is achieved. By further exploiting the non‐Arrhenius behavior of the crystal growth rate in materials, the crystallization efficiency is substantially enhanced. Consequently, the energy required for complete crystallization of the device decreases to 107 pJ, with improved endurance (10 7$^7$ ) and weighting precision (7 bits). Moreover, by designing the energy configuration of the combined pulse, the formation of multiple crystalline phases is suppressed, thereby significantly increasing the weighting linearity. In convolutional neural networks under experimental conditions, the prediction accuracy for handwritten digits reaches 98.76 %$\%$ and exhibits strong robustness. This study provides novel insights into further exploring the potential and function of phase‐change materials, offering an efficient and powerful approach for high‐performance weighting in integrated photonic computing systems.