Determining the polarization state of light is important in a variety of applications from optical communication to biomedical diagnostics. Polarimeters are, however, typically based on discrete bulky optical components, which can restrict miniaturization and limit wider application. Here we report the concept of an optoelectronic polarization eigenvector, which represents the linear relationship between the incident Stokes vector and the photocurrent of a detector. By configuring four of these eigenvectors to create an optimized optoelectronic conversion matrix, we establish a high-accuracy full-Stokes polarization detection method and use the approach to create a compact on-chip full-Stokes polarimeter. The polarimeter comprises four subpixels that share the same piece of few-layer molybdenum disulfide as the detection material. Each subpixel contains an integrated plasmonic metasurface and corresponds to a distinct optoelectronic polarization eigenvector. By tailoring the plasmonic metasurfaces and their geometric arrangement, the condition number of the optoelectronic conversion matrix can be minimized to achieve high-accuracy Stokes reconstruction. Using an optimized matrix, combined with a machine learning algorithm, the root mean square error of the full-Stokes reconstruction over the entire range of polarization states at arbitrary light intensities is less than 1%.

确定光的偏振状态在从光通信到生物医学诊断的各种应用中都很重要。然而，旋光仪通常基于离散的笨重光学元件，这可能会限制小型化并限制更广泛的应用。在这里，我们报告了光电极化特征向量的概念，它表示入射斯托克斯向量和探测器光电流之间的线性关系。通过配置其中四个特征向量来创建优化的光电转换矩阵，我们建立了一种高精度的全斯托克斯偏振检测方法，并使用该方法创建了一个紧凑的片上全斯托克斯偏振仪。旋光仪由四个子像素组成，它们与检测材料共享同一块少层二硫化钼。每个子像素都包含一个集成的等离子体超表面，并对应于一个不同的光电极化特征向量。通过定制等离子体超表面及其几何排列，可以最小化光电转换矩阵的条件数，以实现高精度的斯托克斯重建。使用优化的矩阵，结合机器学习算法，在任意光强度下，整个偏振态范围内的全斯托克斯重建的均方根误差小于 1%。