Aspuru-Guzik took as a starting point an organic semiconductor called DTT, which has already demonstrated promise in the lab. First, the computer program generated several possible variations on this large carbon-based molecule by adding and subtracting components. The program then predicted how these variations would behave and screened for the most promising—those that seemed likeliest to conduct charges very fast. These predictions were informed by a huge amount of data on how similar molecules and molecular building blocks have performed in past experiments, and by theoretical physics and chemistry.
When Bao’s group synthesized the molecule predicted to have the best properties, it behaved as expected. Transistors made from the material operate 10 times faster than transistors made from amorphous silicon, the material used in today’s display electronics; the new material is the second-fastest organic semiconductor yet made. The work is described in the journal Nature Communications. While Bao and Aspuru-Guzik say the material could be used in industrial applications, the experiment has more significance as a proof of principle for their methods for using computers to develop new organic materials.
Aspuru-Guzik is using a similar computational approach for the Harvard Clean Energy Project, which aims to discover better solar material. For this endeavor, he has a lot of computational power at his fingertips: his calculations are being run on the almost two million computers of users signed on to the IBM World Community Grid. Aspuru-Guzik is taking advantage of this brute force to screen about 2.6 million molecules that haven’t been made for their solar potential, using experimental data on approximately 200 previously made molecules. The program predicts what color of light a material will absorb, and how strongly, as well as other factors that make a good solar material.
“There’s no way you can try all of the possible materials experimentally,” says Geoffrey Hutchison, a professor of chemistry at the University of Pittsburgh who’s also working on computationally predicting characteristics of possible solar materials. “As time goes on, the experimentalists are starting to rely more on prediction.” Results like Aspuru-Guzik’s should make them more confident in doing so, he says.