Is photonics the disruptive answer to turbocharging digital speed and efficiency?
As chip technology and compute capacity continue to advance – laid over the trajectory of Moore’s Law – digital interconnect speed and utility need to progress along with them. Continuing improvements in signals efficiency, bandwidth density, and integration are needed to accommodate future demands for connectivity.
To address this challenge, the Defense Advanced Research Projects Agency (DARPA) developed the Photonics in the Package for Extreme Scalability (PIPES) program, which intends to expand the use of optical signaling for data transfer. It also aims to insert high-bandwidth photonics inside the packaging of application-specific integrated circuits (ASICs) and field-programmable gate arrays (FPGAs). PIPES is part of the second phase of DARPA’s Electronics Resurgence Initiative (ERI) – a five-year, $1.5 billion-plus investment in the future of U.S. government and defense electronics systems. ERI Phase II is focusing on developing new types of manufacturing capabilities and crafting a road map for supplying high-performance electronics for the U.S. Department of Defense (DoD) and its commercial partners.
Bottlenecks and limited performance occur when data moves between optical transceivers and advanced ICs in the electrical domain. Integrating photonic solutions into the microelectronics package, DARPA researchers posited, would remove this limitation and enable new levels of parallel computing.
DARPA selected teams to take on three research areas under the PIPES program: Development and integration of optical signaling technology for next-generation digital microelectronics, with a particular focus on defense applications, creation of technologies and concepts that will lead to even better technical performance, and the exploration of new approaches to signaling that system architects can put into practice.
The PIPES project’s first research area – focused on developing high-performance optical input/output (I/O) technologies packaged with advanced ICs like FPGAs and ASICs – is being spearheaded by teams led by industry bigwigs Xilinx and Intel. The new technologies that come out of this piece will enable ICs with unprecedented bandwidth density, energy efficiency, and reach. The agency also says that researchers from Lockheed Martin, Northrop Grumman, Raytheon, and BAE Systems will be looped into the development of these optical I/O technologies to ensure that the results address the requirements of current and future defense needs. The researchers will also investigate which defense applications could benefit most from this technology.
“The benefits of optical signaling in digital systems have been recognized for a long time,” states PIPES program manager Dr. Gordon Keeler. “The integration of photonics within the package will have enormous benefits for commercial and defense applications, but it comes with considerable challenges. PIPES researchers are working to solve practical technical problems to meet the ambitious goals of the program, which include enabling I/O data rates as fast as 100 terabits/sec at signaling energies below one picojoule per bit [one trillionth of a joule]. At the same time, the teams are studying how to tailor their technologies to address national security applications where operating conditions may be very demanding.”
The second PIPES research area aims to push the optical I/O technologies an order of magnitude beyond even what Xilinx Corporation and Intel seek to accomplish, Keeler says:
“To help establish appropriate benchmarks for this research area, we first projected how much data will need to be transported from leading-edge ICs in the 2028 timeframe. Compared to the data capacity of a modern chip today, we may need up to 100 times more off-chip I/O. That’s a petabit [1,000 terabits] per second – roughly the equivalent of the entire world’s internet traffic today – but from a single chip. This is an aggressive benchmark, and we expect the technologies developed in this research area will be less mature at the program’s conclusion, but, if successful, we’ll position photonics to enable disruptive change in future microelectronic systems.”
The research teams selected to explore component technologies and facilitate on-package optical I/O include groups from Sandia National Laboratories, UCSD, UCSB, Columbia University, and University of Pennsylvania.
The final research area of the program wants to ask and answer the question of how system architects can both solve the problems and seize the opportunities created by high-performance optical I/O technologies. Researchers from the University of California, Berkeley, are handling this portion.
“If we can seamlessly integrate optical I/O with advanced ICs – and reduce the energy and latency of data movement enough – we eliminate the need to keep data local. It is a major paradigm shift, an opportunity to employ completely different system architectures,” Keeler asserts. “Take optical switching, for example. As data increasingly moves on optical fibers and can be routed long distances, how should we use distributed, disaggregated, and flexible system concepts? This research area will focus on creating novel optical packaging approaches and optical switching technologies to support potential opportunities that emerge through PIPES.”