Next-generation light-emitting diode (LED) lamps, similar to the lasers used in Blu-Ray players, can emit brighter, whiter light than conventional LEDs, but they’ve yet to replace them on a large scale because the gallium nitride (GaN) substrate they are built on is expensive and difficult to make.
Through a project funded by the Department of Energy’s High Performance Computing for Manufacturing (HPC4Mfg) program, which aims to provide supercomputing resources to industry partners, Lawrence Livermore National Laboratory (LLNL) is working with LED manufacturer SORAA to create a new computer model of the company’s research-scale process for growing GaN crystals, in hopes that improving the process could lead to widespread adoption of gallium nitride for use as a substrate in solid state lighting and power electronics, among other applications.
“The higher-quality crystals you have, the better your product will be,” said Nick Killingsworth, the LLNL scientist heading the project. “Gallium nitride can handle more energy per unit volume than existing technology leading to smaller and more efficient devices, so high quality low-cost GaN would reduce energy usage across many sectors, like electronics, LED lights and sensors. It’s really promising.”
SORAA, a Fremont, California-based company, was co-founded by Nobel Prize-winning physicist and UC Santa Barbara professor Shuji Nakamura, who invented the first high brightness LED. The company builds lamps using GaN layers deposited on a GaN substrate, and says the resulting high-powered violet LEDs are not only brighter and whiter than conventional LEDs built on substrates such as sapphire or silicon carbide, but also are safer, because blue light LEDs can cause health problems with long-term exposure. However, their research process for creating the single crystal GaN needed for a substrate is complicated, requiring high-pressures and high-temperatures. It also occurs within a sealed reactor, making it difficult for researchers to analyze the process.
“The goal through the simulations is to better understand what is happening inside the reactor,” Killingsworth said. “We have big computers and the expertise in using them in a way that can handle the physics involved. SORAA has been running simulations on a workstation and so is limited on what physics they can incorporate in their models and how many cases they can look at. We built on their models incorporating more detailed physics models.”
After a year of work, Killingsworth and team have used a commercial code to develop a computational fluid dynamics model to simulate the high pressure and intense heat needed for the growth process to occur. Killingsworth said the models will allow engineers to explore how changes to the reactor’s boundary conditions affect the growth process, gaining a better understanding and hastening the time it takes to improve growth.
“The LLNL simulations allowed an examination of the complex gas flows — mapping the velocity and temperature fluctuations of the fluid inside the reactor — experienced by the crystals,” said Mark D’Evelyn, vice president for bulk crystal growth at SORAA. “Our team is beginning to tie together the simulation results with experimental observations. SORAA is committed to further improving upon the models produced by LLNL and testing various configurations of the reactor in the computer and selecting the best ones for testing experimentally.”
LLNL researchers have been able to use finer meshes to better capture the physical conditions occurring within the reactor and have run time-dependent simulations to understand how the environment changes with time. They found that the process is not steady with time, thus time-dependent simulations are needed.
However, before time-dependent simulations can be developed, Killingsworth said, scientists will need better experimental data because the conditions inside the reactor are at the extreme edge of what is known.