Edit/Keich
To make solar cells capable of drawing every bit of energy from sunlight, the researchers relied on computer modeling tools. These simulators allow them to evaluate the effect of minor adjustments in parameters such as equipment structure, material usage, and thickness of different material layers on the final power output.
Currently, there are some freely available solar cell simulator packages, but these tools are still slow and don't allow researchers to optimize different design parameters at the same time.
A team of researchers from MIT and Google Brain has developed new software that simplifies optimization and discovery of solar cells. An end-to-end differentiatable photovoltaic (PV) cell simulator based on a drift diffusion model and Beer-Lambert's law for light absorption is introduced. PV is able to calculate not only the efficiency of a solar cell, but also its derivative relative to any material properties set by the user.
This new computational tool enables extensive, efficient material optimization of photovoltaic cells and can be combined with standard optimization methods or machine learning algorithms.
The study, titled "PV: An end-to-end differentiable solar-cell simulator," was published in the journal Computer Physics Communications.
Differentiatable solar cell simulator
Traditional calculation tools take variables for a particular solar cell design as input and then output the final power rating.
In this work, the researchers proposed PV, a one-dimensional simulation tool for PV batteries that uses JAX automatic differentiation (AD) packets to solve the drift diffusion equation. Using the AD and hidden function theorems, calculate the power conversion efficiency (PCE) of the input PV design and the derivative of the PCE relative to any input parameters, all in comparable time to solve the forward problem.
PVs complement existing AD-based solvers, and relying on the composability of these tools makes it possible to achieve end-to-end reducability for more advanced multiphysics simulations.
With the new software, "we provide output, but also show how efficiency will change if we change any of the input parameters," says Giuseppe Romano, a research scientist at the Massachusetts Institute of Technology. "You can change the input parameters continuously and see how the gradient changes in the output."
This reduces the number of times developers run these time-consuming and computationally intensive simulations. "You only have to do one simulation and you automatically get all the information you need," he said. "That's the beauty of this approach."
Two ways to help solar cell development
The new tool can help with solar cell development in two ways. The first is optimization, Romano says, "Suppose an industry player wants to make high-performance solar cells, but doesn't know the impact of light-absorbing materials on overall efficiency." This material layer usually has an optimal thickness and can produce the most charge carriers from the light it absorbs. The software will help define optimal parameters to maximize efficiency.
The software can also be used to evaluate the optimal values for other variables, such as the amount of doping in the material layer, the dielectric constant of the band gap or the insulation layer.
Another way the tool works is to reverse engineer existing solar cells. In this case, the researchers can measure the I-V curve of the solar cell (a function of providing the current for each voltage) and pair these experimental measurements using a simulator. Based on the data, the software can help calculate the values of unknown specific material parameters.
Research cases
Next, the researchers present examples of perovskite solar cell optimization and multiparameter discovery, and compare the results to random searches and finite differences.
Traditionally, solar cell optimization has been done through various gradient-free black box optimization techniques, such as particle swarms and genetic algorithms. Without any additional information, the solar cell simulator is treated as a black box, and optimization becomes a data-intensive task. In particular, optimizing the combination of multiple parameters at the same time is often tricky. With the introduction of analytical gradients, a well-established set of nonlinear optimization algorithms became available.
To illustrate this, take the example of optimizing p-i-n perovskite batteries. For nonlinear constraint optimization problems, choose the Sequential Least Squares Programming (SLSQP) method. SLSQP has been implemented in several open source tools, including Optim, NLOpt, PyOpt, and Scipy. For this work, the last one was chosen.
Starting with the initial design of a random sampling with a PCE of 6.49%, the algorithm terminates only after 306 PDE solutions, reaching the optimal point of 21.62% PCE. This is higher than the result of all 200 random sampling designs, which equate to approximately 4000 PDE solutions.
Use SLSQP to optimize the results of the ETL and HTL material properties of the PSC.
The results show that the optimization target increases compared to random searches. The SLSQP best point (above: red) is better than all samples and is more than 10 times more efficient at PDE solving. Overall PDE solving saves about 25%.
Romano says others may have developed similar solar cell simulators, but "this is the first open source simulator with such nuances." The package is located on GitHub, he said, and anyone can easily use it and make improvements.
Package GitHub address: https://github.com/romanodev/deltapv
Researchers can combine this with their own optimization algorithms or machine learning systems. By allowing for the rapid evaluation of a variety of possible materials and equipment structures, this will accelerate the development of more efficient solar cells.
Thesis link: https://doi.org/10.1016/j.cpc.2021.108232
Reference: https://spectrum.ieee.org/new-simulator-to-speed-up-solar-cell-development
Artificial Intelligence × [ Biological Neuroscience Mathematics Physics Materials ]
"ScienceAI" focuses on the intersection and integration of artificial intelligence with other cutting-edge technologies and basic sciences.
Welcome to follow the stars and click Likes and Likes and Are Watching in the bottom right corner.