Editor's note: this article is by David Turek, IBM's vice president of High Performance Computing Scalable Systems.
Today, I am participating in a White House event highlighting the first results and next steps of the Materials Genome Initiative (MGI), which President Obama announced almost one year ago.
The name of this initiative is a riff on the Human Genome Project
because it intends to marshal and organize significant scientific
resources to gain a deep understanding of the structure and behavior of a
vast array of materials. The goal is to help U.S. companies become more
economically competitive by the application of discoveries in
materials science to the development of new and improved products in a
host of industries at a far greater speed and much lower cost than is
currently possible.
IBM is well aware of the challenges in advancing materials science. IBM Research started the Battery 500 Project
in 2009 to develop a new type of lithium-air battery technology that is
expected to improve energy density tenfold -- dramatically increasing
the amount of energy these batteries can generate and store. And we
invented semiconductor silicon germanium, laying the groundwork for
explosive advancement in wireless products.
There are a
host of other projects in materials science that could lead to new
desalinization membranes, development of biopolymers for medical
applications and new materials used to break the memory bottleneck in
advanced computers. The list goes on for considerable length, but the
pioneering insight from IBM has been to advance material research by
linking experimental techniques with large scale simulation and
modeling.
To realize the goals
of the MGI, it is essential that we build the right kind of supporting
infrastructure. It needs to have three key characteristics:
- Massive compute power: Deep understanding of materials
depends on an understanding of molecular structure and behavior under a
wide array of stresses and forces. Modeling and simulation at the atomic
level has been shown to generate keen insights into many materials, but
massive amounts of compute power is often required. Using powerful
supercomputers like the Blue Gene system has been very effective in
conducting these types of simulations because we are able to explore
millions of atoms in models of diverse materials. As the scale of the
problem increases, understanding of the macro behavior of the target
material deepens and the path to commercialization accelerates.
- Built for data and analytics: Modeling does not occur in a
vacuum; the data that describes the underlying processes must be
accommodated in the MGI infrastucture. In some cases, the sheer volume
of data will present challenges to store and analyze. In other cases,
the complexity of the data will warrant new models of organization and
analysis. In all cases, the MGI infrastructure must ensure that data,
analytics, modeling and simulation are inextricably linked in a way that
leads to near real time understanding of very complex scientific
problems. Compressing time to solution is what leads to competitive
advantage.
- Collaborative: Material scientists must have the ability to
collaborate widely. Sharing data and compute resources is a
foundational requirement of the MGI. The computing infrastructure
enabling this will also need to be secure and accommodate the presence
of proprietary processes and insights from many of the likely industrial
partners.
Argonne National Lab (ANL) is an example of how the right infrastructure can support materials science breakthroughs. Their IBM BlueGene/Q
supercomputer, named Mira, will be a 10-petaflop computer -- meaning it
will be capable of performing 10 quadrillion calculations a second,
making it one of the fastest in the world. ANL researcher Larry Curtiss
plans to use this added compute power to the aforementioned Battery 500 project.
A key
factor in these kinds of experiments is being able to use enough atoms
that scientists get a realistic response from the simulation. Working
with catalytic processes, for instance, the team at ANL has been able to
model reactions involving about 1,000 atoms. With Mira, they’ll be able
to model reactions involving tens of thousands of atoms.
Over
at the Lawrence Livermore National Lab (LLNL), they will soon flip the
switch on a 20-petaflop IBM Blue Gene/Q supercomputer named Sequoia. In
2005, researchers from LLNL and IBM were awarded the Gordon Bell Prize
for pioneering materials science simulations, and the level of
performance achieved, conducted on the Blue Gene/L supercomputer at
LLNL. In that project, simulation capability was increased from
thousands of atoms to millions of atoms, and the simulations still took
many hours. With the advent of Sequoia, LLNL scientists will be able to
run the same simulations in a few minutes -- or increase the fidelity of
the model by adding 100s of millions more atoms.
In
the late 1990's when the Human Genome Project was coming to fruition, a
whole new industry -- bio-informatics -- was born. Hundreds of new
companies burst into existence seemingly overnight. My belief is that we
are at the cusp of a similar phenomenon with the MGI and IBM plans to
be present at the dawn of a new age in materials science.
Other Materials Genome Initiative Projects
World Community Grid: Clean Energy Project
Labels: Argonne National Lab, battery500, bluegene, high performance computing, human genome project, Lawrence Livermore National Lab
