By Wayne Gillam | UW ECE News
UW ECE Assistant Professor Sajjad Moazeni was recently named recipient of a National Science Foundation (NSF) CAREER award, one of the most prestigious awards in the nation for early-career faculty. The award will support Moazeni’s work developing ‘smart,’ energy-efficient links between thousands of networked computer processors and memory units connected by optical fiber — the ‘brain’ of an AI supercomputer. These smart, optical interconnects will not only function as a bridge between electrical and photonic (light) signals but will also increase computing speed and capacity needed to develop the next generation of artificial intelligence and machine learning applications. Moazeni’s research will help to empower AI supercomputers to tackle some of the most complex problems facing our world today, while at the same time, reduce the technology’s environmental impact through greater energy efficiency.
“I am thrilled and honored to have my proposal selected for funding since it allows my group to work on future AI supercomputers,” Moazeni said.
The NSF selects award recipients who are faculty members at the beginning of their careers to lead advances in the mission of their department or organization. The intent of the NSF CAREER program is to provide stable support, enabling awardees to develop not only as outstanding researchers, but also as educators demonstrating commitment to teaching, learning and dissemination of knowledge. The award spans five years, and it will enable Moazeni to better integrate his research and education efforts.
Smart, energy-efficient optical interconnects
Our world’s problems are increasingly complex, but AI and machine learning applications, including AI supercomputers, are being developed to help us address many of these tough challenges. For example, more accurate climate-change modeling, sustainable supply-chain logistics and faster drug development could all be made possible by faster and more powerful AI and machine learning applications. AI supercomputers could also facilitate significant advances in natural language processing and speech recognition, enabling our smartphones and other devices such as Alexa to better understand and enhance our everyday lives.
However, the size and complexity of AI and machine learning models and datasets needed for these sorts of applications are growing at an exponential rate. This means substantial increases in computing speed, capacity and energy efficiency will be needed to make the next generation of AI supercomputers a reality.
“Advanced learning models for AI and machine learning systems are so large that they have recently reached trillions of parameters, terabytes of data that need processing,” Moazeni said. “Pretty much the only way to handle this type of computation is to run it over thousands of processors that exist in data centers in the cloud — a supercomputer. Recent research has shown that we need to improve communication links between all these processors to better support this level of computing.”
Moazeni’s research seeks to address computing speed and capacity by redesigning the optical interconnects on a chip (known as ‘co-packaged optics’) and by using a novel system-level architecture to enable the interconnects to function as more than simply a converter between electrical and photonic signals. These new, co-packaged optics will have enhanced capabilities, enabling them to assist processors with computing tasks, as well as memory access and data retrieval across a large network of devices. Energy efficiency will be improved by moving power-hungry circuitry from the electrical domain into reconfigurable photonic devices, radically transforming signal equalization techniques and markedly reducing power consumption.
“I would consider co-packaged optics commercialized over the next five to 10 years the ‘current generation,’ even though it’s not yet in the marketplace,” Moazeni said. “For the next generation after that, we’ll need to improve the energy efficiency by at least a factor of 10, so that’s what I and my research team are setting out to achieve.”
Teaching about the intersection of integrated circuits and photonics
As part of the education component supported by the award, Moazeni will be developing a course for graduate and advanced undergraduate students that will help bridge the gap between integrated circuits and photonics.
“In this course, I’ll be giving an overview of the state of the art of electronic and photonic design, and we’ll go over applications from communications, computing, all the way to bioelectronics and neural photonics,” Moazeni said. “The course is meant to cover the entire space of electronics and photonics. We’ll talk about the challenges, how you can integrate the technology efficiently and the methodology of designing such systems.”
Moazeni has already been teaching a version of this course at UW ECE over the last two years, but he plans to update and expand the course according to the research that is supported by the NSF award. Curriculum and materials for teaching the course will also be made publicly available.
“We’re going to put all the course materials online, and I would be happy to help any other school or university that would like to offer this course,” Moazeni said. “As far as I know, there is no other school that offers a course like this, so it could be something that would potentially benefit a lot of other universities and schools.”
Moazeni is also in the midst of forming plans for other educational outreach efforts, which potentially will include school visits by members of his research team and lab tours for college-level and K–12 students. According to Moazeni, inclusive access for undergraduate and graduate students will continue to remain a priority for him and his research lab, which already includes a high percentage of students from underrepresented groups.
“It’s clear that we need more powerful computers for AI and machine learning development, and that the type of computers we’ll need in the future could benefit greatly from co-packaged optics. Marrying electronic and photonic technology together will be necessary for solving many upcoming challenges,” Moazeni said. “There is also a real need to train researchers and engineers from diverse backgrounds to develop skills in both areas, so they can build more powerful computers capable of transforming all our lives for the better.”
To learn more about the research, education and outreach programs supported by this NSF CAREER award, contact Sajjad Moazeni.