As the JUMP 2.0 center nears the end of its timeline, work on neurosymbolic AI—an approach that blends machine learning with human‑like reasoning—is showing results.
(text and background only visible when logged in)
When the proposal for the Center for the Co-Design of Cognitive Systems (CoCoSys) was written four years ago, its leaders made a “big bet” on neurosymbolic artificial intelligence (AI).
As the center wrapped up its final Annual Review on March 25 at the Georgia Tech Global Learning Center, they feel like that commitment has paid off.
“There were many gaps in neurosymbolic codesign research when we first started, and in the last four years, we’ve made significant progress,” CoCoSys Director and Georgia Tech School of Electrical and Computer Engineering Chair Arijit Raychowdhury said. “Some of the earliest papers and prototypes on this subject have come from this center.”
CoCoSys is one of several JUMP 2.0 research centers funded in 2022 through the Joint University Microelectronics Program, a national initiative co-sponsored by the Semiconductor Research Corporation (SRC) and the Defense Advanced Research Projects Agency (DARPA).
Together, the JUMP 2.0 centers address different long-term challenges in microelectronics and computing. CoCoSys tackles the neurosymbolic AI challenge through a strategy Raychowdhury calls “software to silicon,” a framework that connects interdisciplinary research across AI models and software systems down to the underlying hardware that enables them.
“The collaboration is fundamental because this research problem is a moonshot, in that it's very complex, so we need expertise in many different areas,” he said. “This cross-stack expertise is impossible to find in one school or company.”
Hundreds of researchers have worked on research in pursuit of the Center's goal of enabling the next generation of AI.
A New Type of AI
Nothing embodies the success of CoCoSys better than its first-of-its-kind neurosymbolic AI chip. It’s an accelerator that can natively support neurosymbolic AI algorithms and was taped out —meaning its design was finalized for fabrication— just before the 2025 Annual Review.
Getting there, however, required the center to navigate a rapidly shifting AI landscape. Just months before CoCoSys started its work, ChatGPT was released to the public, thrusting large-scale AI into the mainstream and reshaping expectations for what AI systems should look like.
“We thought this data-driven approach could be augmented with symbolic reasoning models,” Raychowdhury said. “And if you look at some of the interesting problems industry and academia are taking on today, many aren’t solved by ChatGPT-like models, they’re solved by neurosymbolic algorithms.”
Widely used Large Language Models (LLMs) like ChatGPT are neural models that excel at summarizing information and generating responses, but struggle with reasoning. Neurosymbolic AI aims to add symbolic reasoning that uses explicit symbols, rules, and logic to make more complex deduction and reasoning possible, predictable, human-understandable, and verifiable.
CoCoSys Director Arijit Raychowdhury addresses the researchers who came to Atlanta for the 2026 CoCoSys Annual Review.
Putting it to Test
Over the past year, CoCoSys researchers have continued to test the chip, achieving promising results.
“The neurosymbolic AI chip has demonstrated substantial improvements in efficiency over today’s widely used hardware,” CoCoSys co-director and Purdue University Professor of Electrical and Computer Engineering Anand Raghunathan said. “We hope that this first generation of hardware enables applications that use neurosymbolic algorithms to become a reality and sparks continued research from the broader community.”
Not only is the work novel, but specifically important and necessary as AI systems increasingly interact with people and the physical environment, according to Raghunathan.
“As AI gets deployed into the physical world, there are very real consequences to mistakes,” he said. “The new generation of algorithms has to be more reliable, trustworthy, transparent, and explainable.”
A major focus over the final months left in the Center’s timeline is to apply what they learned from the initial testing to making a second generation of the chip.
As AI gets deployed into the physical world, there are very real consequences to mistakes. The new generation of algorithms has to be more reliable, trustworthy, transparent, and explainable.
CoCoSys Director Arijit Raychowdhury
Developing Future Leaders
The neurosymbolic AI chip isn’t the only measure of success for the center.
Equally as important has been CoCoSys’s contribution to developing the next generation of AI researchers.
“We’ve trained over 150 students as part of this center,” Raychowdhury said. “The human capital that has been developed will have a huge impact on society and the industry over the coming decades.”
The benefits of this talent pipeline are already apparent.
“It’s very gratifying to see student researchers become colleagues,” Raghunathan said. “CoCoSys students have become professors conducting their own research, as well as gone into industry and come back to sponsor research.”
Yan Fang is one of those researchers. He started as a postdoctoral CoCoSys researcher at Georgia Tech when the center was first established and is now an assistant professor of electrical and computer engineering at Kennesaw State University.
“SRC resources have been great and helped foster me from student to postdoc to now faculty,” Fang said. “Without it, I would not be in my current position.”
The annual reviews are a good opportunity for researchers to get meaningful feedback from a variety of experts.
Mohamed Ibrahim also started as a Georgia Tech postdoc researcher at CoCoSys. He is now an assistant professor of electrical and computer engineering at the University of Texas at Dallas, where he leads a lab studying AI systems that involve human interaction.
“Once I became part of the CoCoSys community, I started interacting with high-level researchers from other areas than my own, and that’s when I started getting interested in co-design,” he said.
CoCoSys has also given students direct exposure to industry. Collaboration with companies like IBM, Intel, TSMC and RTX among others has provided students with valuable practical feedback on how AI research can move from theory to application.
Feedback from CoCoSys industry partners helps student researchers to move their research in a direction that can impact the field beyond academia.
“Industry is where new problems are discovered, so as an academic working with them, it has been a great opportunity to direct my research in a more meaningful and impactful direction,” said third-year Georgia Tech ECE Ph.D. student Akshat Ramachandran, whose research investigates making AI models smaller.
During a series of poster sessions and lightning talks, researchers present their work to experts from both academia and industry. At the end of the annual review, a Best Poster Award is given out for each of the Center's four themes.
The winners this year were:
- Theme 1 Winners: Justin Lukose (Kennesaw State University) and Yoni Friedman (Massachusetts Institute of Technology)
- Theme 2 Winner: Varun Raghuraman (Kennesaw State University)
- Theme 3 Winner: Kwangsoo Lee (Cornell Tech)
- Theme 4 Winner: Ginny Xiao (University of Southern California)
Continuing the Work
While the 2026 Annual Review may be one of the last times this group gets together in person, it’s not the end.
“I think we’ve changed the narrative in the broader research community to the point where they’re saying, ‘Hey, we may want to start thinking about neurosymbolic AI,’” Raghunathan said.
For many, it’s the start of a new community, and one they believe will continue well beyond CoCoSys.
“It’s been an honor to be part of CoCoSys,” Ibrahim said. “And I know this community will continue to push the field, in other forms of course, but CoCoSys will always be an important part of this research.”
(text and background only visible when logged in)
Related Content
CogniSense Holds Third Annual Review
Researchers at the Georgia Tech-led Joint University Microelectronics Program 2.0 center, focused on creating smarter, energy-efficient cognitive sensors, shared the progress they’ve made as it passed the halfway point of its five-year timeline.
How a Lens Is Pushing the Limits of Near-Zero‑Power Wireless Communication to Gigabits‑Per‑Second Speeds
Emmanouil Tentzeris and Marvin Joshi’s new work demonstrates how a lens‑enabled backscatter system can deliver modern wireless capability without traditional transmitters.