Three professors in the Luddy School of Informatics, Computing and Engineering Intelligent Systems Engineering department will engineer the next generation of computers as part of a three-year, $5.4 million government project that will help the U.S. intelligence community execute data analysis missions.
Martin Swany, professor and chair in the Intelligent Systems Engineering department; Ariful Azad, assistant professor; and Thomas Sterling, professor and the founder of Simultac LLC, an Indiana-based, high-performance computing company, have received one of six contracts from the U.S. Army Research Office and the Intelligence Advanced Research Projects Activity (IARPA).
The goal is to design a computer architecture that will include intelligent mechanisms for accessing, moving, and storing complex data streams and structures that enable efficient execution of data-analytic algorithms and Artificial Intelligence applications – capabilities that are crucial to U.S. national security.
“We are essentially designing an innovative, bespoke architecture, optimized to deal with a special class of problems” Swany said. “Custom microelectronic design is a key aspect of the recently announced new degree programs of microelectronics, semiconductors and nanofabrication at Luddy and Indiana University.”
The system they are creating for the project is called Pilot-CRIS.
“The Luddy School is inherently cross-disciplinary, so the idea of designing hardware and algorithms is natural for us,” Luddy Dean Joanna Millunchick said. “We are in a great position to take advantage of this emerging trend.”
The government contract was awarded through the Advanced Graphis Intelligence Logical Computing Environment (AGILE) program, which seeks innovative computer architecture designs that are efficient for large-scale data analytics, with extreme performance beyond current high-performance computing systems.
“The innovative Pilot-CRIS computer architecture offers a promising strategy to achieve the AGILE program objectives for reliable and energy efficient parallel computation of dynamic graph-based applications and other large-scale problems,” Sterling said. “Our system is designed to meet the needs of the AGILE program, and to seed a new generation of computer architectures that will generate significant performance advantages for the intelligence community.”
The researchers will design a system that is specialized for ingesting high-volume data, training graph neural networks, finding patterns in graphs and assembling DNA sequences.
“These applications are at the heart of national security and prosperity,” Azad said. “We are very excited to work alongside other leading universities and chip-making companies to define the trajectories of the next generation computer architectures.”
Researchers will design computers that can process exponentially increasing amounts and sources of data. They will use outside-the-box thinking to design computer architectures that address data-intensive problems while generating important analytical insights in a timely way.
Intelligence gathering operations range from the Internet to virology to social media information to disinformation and beyond. Much of that information is modeled using time-varying graphs.
“Interactions among entities drive almost all physical, social, and cyber-physical systems on earth,” Azad said. “Such interactive systems are mathematically modeled as graphs where a set of vertices represents entities, and a set of edges represents their interactions.”
Azad said current computing systems are predictably inefficient when used to learn from irregular graphs. The project aims to address this challenge with a completely new computer architecture designed for rapid graph processing.
The Active Memory Architecture (AMA) at the core of Pilot-CRIS enables a revolutionary execution model, with memory-centric operation as a fundamental advance. Memory-centric computing is an approach for operating on data where it exists, in memory. It distributes processing to the memory themselves. This is different than traditional computing that loads data from memory into a central processing unit, and then stores results from the CPU back into memory.
Swany’s research is in high-performance networking and his work with Pilot-CRIS lies in connecting the huge arrays of smart memory and avoiding traffic jams.
Sterling and Swany have been collaborating since coming to IU in 2011.
“We have distinct, but complimentary approaches and expertise,” Sterling said.
They were previously awarded IARPA Seedlings grants to develop technology upon which their current work builds. If this project is successful, they plan to enter a commercialization phase.