Back to search

University of Washington

Modeling, Machine Learning, and Additive Printing for the Solar Cell Grid

Abstract

dc:description.abstract

The front electrode grid of a solar cell solves a simple problem: it must move current from the solar cell area to a sink. In doing so, it must minimize the resistive power losses incurred in transit while minimizing the size of the shadow it casts on the underlying photovoltaic material. The optimal geometry to achieve this may become complex depending on the material and scaling properties of the system. It may share morphological aspects with the natural systems that solve similar problems, such as leaves, veins, and waterways. The underpinnings of solar cell grid optimization are a set of well-known equations describing the principal sources of shadowing and resistive power loss. We challenged two assumptions that have been long entrenched in the field: first, that colinear line arrays are strictly optimal when designing solar cell grids. In fact, there is a class of isotropic grids that can outperform linear arrays in certain conditions. Second, that constant grid line height describes most grid applications. Instead, in the era of additive and printed electronics, many grids may be subject to virtuous scaling - the property that larger wires become more efficient conductors. Under virtuous scaling, complex ramified electrode structures become optimal. A dynamic graph approach was developed to model these electrode patterns, and a locally greedy approach to optimize them realized novel electrode patterns with ~1\% performance over standard linear grids. The continuous width variation inherent in these patterns presented a control challenge for an additive process attempting to print them. An electrohydrodynamic (EHD) inkjet printing approach was proposed; the powerful but difficult-to-control additive technique would be combined with machine learning (ML) model-based control. Rapid serial experiments on the EHD produced a large dataset of training samples. ML models were trained on this data to predict the jetting and deposited feature characteristics of the EHD as a function of the input waveform. Transfer learning was demonstrated between EHD task datasets using a follow-the-leader gated mixture-of-experts ensemble, and zero-shot models were trained for use in a model-based control algorithm. Model-based control of the EHD, paired with a single proportional tuning adjustment, was able to achieve ~5 micron error when printing lines and ~1 micron error when printing dots, both over at least one decade of feature sizes. This capability enabled EHD printing of micron-scale ramified solar cell electrodes with ML-produced recipes. These steps demonstrate a complete journey from concept to prototype, in which computational tools model, optimize, and ultimately manufacture a novel engineered electrode structure to improve solar cell performance.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Nakano-Baker, Oliver
Advisor dc:contributor.advisor
  • MacKenzie, J. Devin

Subjects

dc:subject × 10

Rights

dc:rights
Statement dc:rights
  • CC BY-SA
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1773/52203

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Nakano-Baker, Oliver. Modeling, Machine Learning, and Additive Printing for the Solar Cell Grid. 2024. https://hdl.handle.net/1773/52203