January 15, 2018
As the computing industry struggles to maintain its historically rapid pace of innovation, a new, $32 million center based at the University of Michigan aims to streamline and democratize the design and manufacturing of next-generation computing systems.
The Center for Applications Driving Architectures, or ADA, will develop a transformative, "plug-and-play" ecosystem to encourage a flood of fresh ideas in computing frontiers such as autonomous control, robotics and machine-learning.
Analysts worry that the industry is stagnating, caught between physical limits to the size of silicon transistors and the skyrocketing costs and complexity of system design.
"The electronic industry is facing many challenges going forward, and we stand a much better chance of solving these problems if we can make hardware design more accessible to a large pool of talent," said Valeria Bertacco, Arthur F. Thurnau, professor of electrical engineering and computer science, and director of the ADA Center.
"We want to make it possible for anyone with motivation and a good idea to build novel high-performance computing systems."
The center is a five-year project led by U-M and includes researchers from seven universities, pending final contracts: Harvard University, MIT, Stanford University, Princeton University, the University of Illinois and the University of Washington.
ADA is funded by a consortium led by the Semiconductor Research Corp. and includes the Defense Advanced Research Projects Agency. The center is one of six new centers recently announced as part of the Joint University Microelectronics Program, organized by Semiconductor Research.
ADA aims to democratize the development and deployment of advanced computing systems in several ways: It will develop a modular approach to system hardware and software design, where applications' internal algorithms are mapped to highly efficient and reusable accelerated hardware components. This faster and more effective approach will require that the entire design framework — from system software to architecture to design tools — be reimagined and rebuilt.
"You shouldn't need a Ph.D. to design new computing systems," Bertacco said. "Five years from now, I'd like to see freshly minted college grads doing hardware startups."
Computing has had a monumental impact on society, but the path forward is uncertain. Researchers are looking for creative approaches to extend the utility of traditional silicon beyond the era of Moore's Law, a long-standing but waning trend in which chips become cheaper to manufacture, and more powerful, each year.
ADA researchers see customized silicon for specific applications — like chips optimized for image search or data analytics — as a promising approach. But the biggest industrial customized silicon successes to date, such as smartphone systems-on-a-chip or graphics processing units, have required the immense resources of large, deeply integrated, vertical design teams. ADA's goal is to change that.
The center is organized into three themes:
• Agile system development: The team will identify patterns in the core algorithms of emerging applications — such as virtual reality, machine learning and augmented reality — in order to map those algorithms to new, tailored computational blocks. This approach would slash design costs by building ready-to-use components that usher designs all the way from high-level computational languages to fully packaged systems.
• Algorithms-driven architectures: The researchers will develop reusable, highly efficient algorithmic hardware accelerators for the computational blocks. Instead of targeting the application itself, designs will target the underlying algorithms. Special-purpose hardware designs can improve the efficiency-per-operation by several orders of magnitude over a general-purpose chip. Such special-purpose hardware design occurs today, but it can take a decade after a need is identified before mature and efficient solutions are available, and it requires extremely specialized expertise, the researchers say.
• Technology-driven systems: A key aspect of this theme involves developing an open-source chip scaffold for these new, accelerator-centric systems. The scaffolds would include all the necessary support subsystems — such as general-purpose cores, on-chip communication fabric, and memories — to facilitate a "plug-and-play" flow.
"One will no longer need to send a design to the fab and wait for a chip to come back. They may still need a clean room to assemble a system, but this will be much simpler and more economical," Bertacco said. Researchers will also explore technology innovations independent of silicon scaling.
"This is a daring and progressive approach to system design that stands to revolutionize the computing industry," said Alec Gallimore, the Robert J. Vlasic Dean of Engineering, Richard F. and Eleanor A. Towner Professor, Arthur F. Thurnau Professor, and a professor of aerospace engineering and of applied physics.
"The work of this new center will empower generations of engineers and computer scientists to design and build the systems that can bring their ideas to life."
DARPA and Semiconductor Research will contribute $27.5 million to this project, with the remaining funds provided by the participating institutions.
Semiconductor Research is a global, high technology-based consortium that serves as a crossroads of collaboration between technology companies, academia, government agencies, and SRC's engineers and scientists.