Light harvesting coating design accelerated by deep learning for semi-transparent polymer solar cells

Abstract

The reduction in optical loss in polymer solar cells (PSCs) plays a crucial role in the development of high-performance PSCs devices. Especially for the semi-transparent PSCs, high reflective transparent electrodes lead to low energy utilization. Optical multi-layer coating is proven to be an effective approach to reduce the reflection and transmission loss. In this work, a double-sided PSCs device coating strategy was used to reduce the device optical loss. Optical coating design on a multi-layer PSCs device is far more complex. The dispersion and thickness of each layer both have an impact on the optical property. Meanwhile, the illuminance spectrum is based on the solar AM1.5 spectrum rather than a common-used standard illuminance CIE-E spectrum. It brings many difficulties to the optical design, and the global optimization is generally time-consuming. To fast solve the optimization problem in optical design of the multi-layer coating for PSCs, we combine deep learning (DL) method with hybrid optimization algorithms. By designing a multi-layer device structure to achieve the highest light harvesting with tandem simplex simulated annealing and assisted simplex simulated annealing, we show unambiguously that DL is a powerful tool to minimize the computation cost and maximize the design efficiency for optical multi-layer design. The optical loss of the semi-transparent device is reduced from 52.71% to 27.95%, and the simulation time is reduced by a factor of 276 compared with standard simplex simulated annealing. This provides an efficient optical design strategy in multi-layer coating design for PSCs to achieve desired optical performance.