Improving Connectivity and Energy Consumption After the AI “Inflection Point”

The annual review for the Center for Ubiquitous Connectivity (CUbiC) was held on June 24-25. Researchers from 11 universities, industry, and government partners all gathered to discuss connectivity — from developing next-generation wireless connection to more efficient AI data centers. 

Shih-Fu Chang, dean of Columbia Engineering, delivered opening remarks and praised the center for translating research into industry. CUbiC is partnered with 15 leading electronics and communications companies, including Samsung, Intel, and Boeing.

“I was on a trip to a few regions last week in Turkey, Thailand, and Taiwan. Everybody was talking about the topic you are addressing here today: How to advance AI, computing, communication, connectivity, bandwidth computing, but then also at the same time cut the demand on power,” Chang said.

Keren Bergman, the director of CUbiC, delivered the opening remarks on Wednesday, June 24. She started by talking about the “inflection point,” a period of time between 2018 and 2020 where AI skyrocketed in popularity. Since that time, AI companies and researchers have been demanding more memory and bandwidth to train their models. CUbiC researchers are working to offer that memory and bandwidth by improving connectivity. 

“We want to increase the connectivity and communication bandwidth, but reduce the energy consumption,” said Bergman, a Charles Batchelor Professor of Electrical Engineering at Columbia Engineering.

CUbiC is uniquely positioned to address connectivity issues because of its remit to research the entire communication systems. Bergman highlighted this in her opening remarks as well.

“We are developing technologies, we are developing circuits and subsystems,” she said. “But the ability to look at the algorithms, the reconfigurability, and the adaptability of the system as a whole allows us to leapfrog these solutions.”

Looking ahead

At the annual review, representatives from CUbiC partner agencies presented their updates and priorities for the final year of the center. 

Ali Niknejad, the associate director of CUbiC from UC Berkley, delivered a presentation about new results coming out of CUbiC, including AI-assisted models that can accurately predict signal power. He also presented a promising development in getting signals through obstacles, like buildings. 

“You use your side lobes to discover alternate communication paths, and do that in the background, and if you get your line of sight blockage, you can automatically switch to a new path,” Niknejad said. 

In his presentation, Carl Naylor, executive director of JUMP 2.0, said he was seeing CUbiC students hired in record numbers by their industry partners. “These students have the skillset, are being trained to integrate in the industry quickly and make an impact rapidly,” said Naylor. 

Jonathan Hoffman, deputy director of the Multi X Office (MXO), a part of DARPA, spoke about the next steps for the MXO, the government office that runs the JUMP 2.0 program. They are widening their research goals from solely microelectronics, to other technologies like photonics and robotics. Hoffman said, “The question here is, what is that next generation of investment we need to make in materials and materials processing?”

CUbiC Annual Review 2026 Highlights

Alex Meng, CUbiC’s technical director and associate professor of professional practice in the Department of Electrical Engineering at Columbia Engineering

Go to the previous slide1/14Go to the next slide

Student achievements 

The annual review also commemorated the student research being done across the three themes of CUbiC. Theme 1 is research that takes a systems-level approach to improving bandwidth and connectivity. 

The first award went to Michael Cullen, a fifth-year PhD student at Columbia. He’s been working with CUbiC since the start of the center on the incorporation of photonics to improve bandwidth for AI training systems. His project tested photonic and electronic architectures to test how the two could work together in a simulated environment. 

“We can leverage the best of our photonic design and marry it with the best of other circuit designs and achieve really nice results in hybridized communications,” said Cullen. 

Anriban Banik, a third-year PhD student from UC Santa Barbara, took home the next award for Theme 1. Banik worked on a framework to improve millimeter-wave radar networks, a system that uses electromagnetic waves to map spaces.

“We wanted to utilize multiple views and introduce multiple radars so we can scale up to a huge network,” he said. Banik’s research leveraged techniques including real-time processing and one-shot fusion to improve calibration of the radar networks. 

CUbiC’s second theme is improving bandwidth by developing more efficient electric and photonic links. The first award went to Yunping Wang, a PhD student from UC Berkeley. His project tested a new photonic chiplet for high-bandwidth connectivity with a host system. 

Omar Bekdache, a third-year PhD student from the University of Illinois, also won the award for Theme 2. He tested a new framework for compute connectivity. “My biggest takeaway was the set of new insights that emerged from our COCO working group discussions and experiments, such as modeling the COCO fabric and recognizing that compute energy often dominates connectivity energy,” Bekdache said.

The third theme for CUbiC research is improving wireless connectivity, including developing the technology that could lead to next-generation wireless networks. 

Jurui Qi, a PhD student from UC San Diego, won an award for Theme 3. He presented a hybrid reconfigurable intelligent surface (RIS). Qi describes his hybrid RIS as a “hardware platform for future communications.” This technology could be used in future 6G networks, and Qi reported promising results indicating that his hybrid RIS could enable real-time device localization. 

The second award for Theme 3 went to Nagesh Patle, a fifth-year PhD student at UC Berkeley. His project was tackling limits in power delivery for eventual use in data centers.

 “We can shrink the inductors, make our transients really fast, and increase our efficiency,” he said. “The trade-off, of course, is higher cost and more complexity.” Patle proposed a hybrid switch capacitor that could deliver high voltage in a design only a few millimeters tall. 

The CUbiC annual review concluded after the awards ceremony. In the next year, the team of researchers, students, and industry partners will continue to develop the connectivity solutions. At the conclusion of her opening remarks, Bergman summed up the last year of CUbiC. “We’ve delivered,” she said. “We've delivered these immensely energy-efficient, bandwidth-dense connectivity solutions at the system level. We have this new technology now.”

Read original story here.