Simulations on the OLCF Titan Supercomputer combined with complimentary Advanced Visualization will enable Rational Design of Organic Photovoltaics. Visualization by Jamison Daniel (ORNL).

Simulations of unrivaled accuracy reveal most efficient configurations of solar cell molecules

If only Thomas Edison had been able to discern the inner-workings of the light bulb before he hit the work bench, he may have saved himself the time and money that went into more than 1,000 failed bulb designs.

Now, scientists have computer simulations to help them engineer devices, medicines, and fuels, but as more technologies are designed on the atomic and molecular level where there are millions of operating parts, the computational demand toreplicate these systems skyrockets.

A research team studying organic photovoltaic (OPV) cells is using the world’s most powerful supercomputer for science, Titan, to simulate the molecule-by-molecule transformation of solar energy into electricity on the most realistic scale ever achieved for this system.

Their work is detailed in a paper recently accepted by the journal Physics Chemistry Chemical Physics.

Managed by the Oak Ridge Leadership Computing Facility (OLCF) located at Oak Ridge National Laboratory (ORNL), Titan is a Cray XK7 high-performance computing system with a CPU/GPU hybrid processing architecture that is unprecedented at the petascale.

“Titan has clear benefits for complicated models,” said Mike Brown, OLCF computational scientist who developed codes to run efficiently on Titan’s new architecture. “The CPUs and GPUs perform differently. CPUs have complicated cores that can perform computations with small amounts of data very rapidly, whereas GPUs achieve their speed through parallel computation with a large amount of data on thousands of energy-efficient cores.”

Titan is 10 times faster than its 2.3-petaflop predecessor Jaguar, topping out at a theoretical peak performance of 27.1-petaflops and a Linpack benchmark of 17.59 petaflops. Yet, it is five times more energy efficient.

When Titan went online in November 2012, the OLCF admitted early research projects on the machine. Although projects range from climate change modeling to nuclear engineering simulations, they have one thing in common: heavy loads of computation that benefit from the speed and endurance of GPUs. The team of OPV researchers needed this computational power to study a system that depends on careful organization at the molecular level to succeed.

The efficiency of inorganic (typically silicon) solar cells has increased slowly over the years to about 20 percent for widely used technologies and up to as much as 40 percent for developing technologies still under laboratory testing. But scientists with ORNL’s Center for Nanophase Materials Sciences (CNMS) are pursuing the potential in OPVs, which have a theoretical peak efficiency of 20 percent. (Currently, they push 10 percent at the best of times).

Despite the disparity in their efficiencies, organic solar cells could be more impactful than inorganic ones, according to Rajeev Kumar, CNMS research scientist.

“Inorganic photovoltaics have a higher efficiency, but they’re expensive and their function and design is limited because they’re not very flexible,” Kumar said. “But organic photovoltaics are cheaper to make, and the processing uses the same technology as printing so you could roll them out then use these flexible panels on all different kinds of surfaces.”

Still, Kumar and the rest of the team simulating OPVs on Titan—including Jan-Michael Carrillo, OLCF post-doctoral researcher, Monojoy Goswami, CNMS research staffer, and Bobby Sumpter, Nanomaterials Theory Institute group leader in CNMS—want to discover the most efficient OPV design possible. Although some OPV designs have already been researched and developed, the OLCF and CNMS team is starting from scratch and focusing on the essential layer of organic molecules that drives energy production for the entire OPV cell.

Using Titan, the team is simulating interactions between two molecules blended together to strike up an interaction that creates energy. The molecular blend, also known as a bulk heterojunction, includes donor molecules (long, semi-flexible chains of the P3HT polymer) and acceptor domains (tiny clusters of a molecule called PCBM). These donor and acceptor molecules absorb and convert solar energy into electricity, determining the baseline efficiency of the panel.

“The donor molecules absorb sunlight and get into an excited state called an exciton,” Brown said. “In order for the energy from the exciton to be usable, it needs to travel to an acceptor.”

Once the exciton reaches the acceptor, it connects at an interface where it splits into an electron and a hole (which has the opposite charge of an electron). From there, the negative and positive charges follow paths formed due to the phase segregation of donor and acceptor molecules, much like the current flow in batteries and transistors.

After only a few months, the team believes these early simulations on Titan have helped settle a long-standing debate over the morphology of the P3HT and PCBM blend and whether or not these donors and acceptors can be configured to improve their efficiency. The team believes they can.

To simulate this system, Brown and OLCF staff modified a widely-used molecular dynamics model known as LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) to perform better on GPUs. Over the course of the OPV simulations, they have observed a 2 to 3 speed up compared to what comparable high-performance computing CPUs could accomplish alone.

The problem confronted in each simulation is ensuring the exciton is delivered to an acceptor interface as fast as possible. The exciton has a limited life, and if it travels too long without reaching an acceptor, the absorbed energy is wasted. The simulations show the molecular domains as they morph into different configurations of varying efficiencies based on variables introduced by the team.

Seeking to improve efficiency, the team simulated conditions— including different P3HT-to-PCBM volume ratios, chain lengths, and other variables—that resulted in more interfaces between donor and acceptor molecules reached by short, direct pathways.

They didn’t know exactly what to expect based on prior OPV simulations. P3HT and PCBM don’t mix well under all conditions, and simulations carried out by different research teams on less powerful machines showed inconclusive and varying efficiencies.

“The debate over whether you could even mix P3HT and PCBM is an old physics problem,” Kumar said. “But with Titan, within three months, we could simulate a bigger system, and we could see when they do mix and when they don’t.”

The team realized that previous computer simulations had, to some degree, compromised their understanding of the phase segregation between P3HT and PCBM. More conventional computer architectures cannot tackle millions of molecules so they simulate smaller numbers and draw conclusions about how the entire system would develop over time. This new result in OPV research is enabled because these simulations on Titan are 27 times larger and run 10 times longer than previous state-of-the-art calculations.

“Previous results varied greatly depending on when in the process simulation stopped,” Kumar said.

OPV simulations on Titan, however, reached equilibrium—the reliable plateau where the system has been completely rendered so its behavior is no longer dependent on size and simulation run-time but only the effects over time. Now the team has more accurate projections for what combination of donor and acceptor molecules pushes maximum efficiency, as well as better predictions for how external forces like temperature change will affect the system.

“We are getting something very realistic because using Titan we can simulate the entire film of polymers, and we are free of any artifacts, or assumptions, coming from the simulation,” Kumar said.

The OLCF and CNMS will continue OPV simulations on Titan throughout the year and explore other barriers and opportunities to high-efficiency OPVs, including analyses of the substrates that reinforce the P3HT-PCBM film and potential co-polymers that could work directly with P3HT to create more organized pathways between donors and acceptor interfaces.

Katie Elyce Jones