核心概念
此研究結合原位反射式高能電子繞射(RHEED)和機器學習(ML),開發出一種名為 ResNet-GLAM 的輕量級模型,用於即時分析 RHEED 數據,並動態調整磊晶生長參數,成功地實現了量子點雷射發射的原位自優化,顯著提高了量子點雷射的效能。
統計資料
在 InAs 生長速率約為 0.014 ML/s 的情况下,以不同的 V/III 比生長了一系列對照樣品。
數據集由總共 24 個樣本組成。
在模型的第一次部署過程中,生長溫度變化了 21 °C。
生長時間的波動很小,每個 InAs 量子點生長的變化僅為 1.87%。
圖 4j 中的 1 µm × 1 µm AFM 圖像顯示量子點分佈均勻,密度為 4.8 × 10^10 cm^-2。
相應的 PL 強度達到 14189.0,FWHM 為 34.30 meV。
我們已經實現了連續波雷射,其閾值低至 150 A cm^-2,室溫下的輸出功率超過 16.5 mW。
引述
"These results mark a significant step toward intelligent, low-cost, and reproductive light emitters production."
"Our approach enables in-situ characterization and optimization of parameters during material growth, marking a significant advancement in achieving precise control over material growth."
"This method has the potential for large-scale production, reducing optimization cycles and improving final yield."
"These results demonstrate that automated and in-situ QD laser self-optimization successfully achieves electrically pumped lasing."
"Our demonstration of the capability to grow high-quality III–V materials with tailored characteristics, along with the fabrication of electrically pumped lasers operating in continuous wave mode, opens new avenues for precise control over material growth."
"With further improvements in hardware, customized modelling, and other areas, this technology holds significant potential for large-scale production, which could enhance productivity and yield in the semiconductor industry."