When Georgia Tech was awarded the Center for the Co-Design of Cognitive Systems (CoCoSys) in 2022, it came with the challenge and excitement of developing the next generation of collaborative human-AI systems.
In 2025, enthusiasm remains high as the center held its third annual review on March 24-26. This unofficial milestone, marking the halfway point of the Center’s lifespan, showcased significant progress in their multi-million-dollar initiative funded by the Semiconductor Research Corporation (SRC) and the Defense Advanced Research Projects Agency (DARPA).
“Our job is to work on the next big idea in artificial intelligence (AI) technology,” CoCoSys co-director and Georgia Tech School of Electrical and Computer Engineering (ECE) Chair Arijit Raychowdhury said.
For CoCoSys, the “next big idea” is developing a novel approach called neuro-symbolic AI, which combines the learning ability of neural networks with the logical reasoning of symbolic AI to create smarter, more human-like artificial intelligence.
It’s exactly the kind of boundary-pushing, high-risk, high-reward research the JUMP 2.0 program encourages, according to Raychowdhury.
“AI is getting agency,” added CoCoSys co-director and Purdue Professor of Electrical and Computer Engineering Anand Raghunathan. “To be able to trust it with that agency and give it the ability to be able to make decisions, we need to make sure the AI can explain its decision.”
Developing of the First Neuro-Symbolic AI Chip
Through the work of 21 principal investigators, five of which come from Georgia Tech, and over 150 students, that idea is becoming a reality.
CoCoSys recently got back the first rendition of an integrated circuit that they believe represents a tangible embodiment of the Center’s mission.
The team was able to successfully tapeout, meaning they finalized the design and readied for fabrication, the first accelerator that will natively support neurosymbolic AI algorithms.
“We essentially put the best of human ingenuity, in terms of reasoning algorithms, into silicon,” Raychowdhury said.
As Raghunathan explains, designing hardware from the ground up to run neurosymbolic workloads, like this chip is an important step in creating the next generation of AI. Currently, new AI algorithms are dictated by the hardware that can run them, which aren’t designed for cutting-edge work like that at CoCoSys, limiting their output.
The chip, which does not have a name yet, is capable of reasoning and solving International Mathematical Olympiad level problems, according to Roychowdhury.
“This kind of innovation across multiple layers of technology wouldn't be possible within a single research group or institution,” he said. “When we created the vision for the Center back in 2022, we made a big bet that this approach and this team would enable the next generation of AI.”
CoCoSys researchers are hopeful the chip will have a significant impact in the next generation of 3D scene perception, digital assistants, agentic and physical AI, among other things.
Preparing Future AI Engineering Leaders
Technology isn’t the only thing coming from CoCoSys. It’s also producing the next generation of AI engineers.
Hundreds of students are brought into the research process, getting a hands-on education and exposing them to the practical needs coming from industry.
“The students coming out of these five-year centers are the cream of the crop,” JUMP 2.0 Executive Director Roman Caudillo said. “They’ve been working on Ph.D. level topics for five years that industry really cares about. They understand what research looks like in academia but also what it looks like in industry within a company.”
Students got to present their research during the review's multiple poster sessions.
Around 20 percent of the semiconductor industry’s Ph.D. students have been part of an SRC project, according to Caudillo. Many researchers leading JUMP 2.0 programs also worked on SRC projects as students themselves, including Raychowdhury.
This part of JUMP 2.0’s mission, preparing students for the workforce, is just as important as the research, especially as the semiconductor industry continues to ramp up in the United States. The CHIPS and Science Act, along with other initiatives, is driving significant investments in domestic semiconductor production to reduce reliance on international suppliers and bolster national security.
“What that means is we need three times as many semiconductor-trained students over the next 10 years than we are currently producing now,” Caudillio said.
Eighty of these future semiconductor researchers made the trip to Atlanta for this year’s review, presenting their work to industry and academia.
Four students took home top prizes for their exceptional research:
- Theme 1 Winner: Timur Ibrayev (Purdue University) - Robust Representations Enabled by Active Vision
- Theme 2 Winner: Zishen Wan (Georgia Tech) - A Cross-Layer Software-Architecture-FPGA-SoC Approach for Neuro-Symbolic AI
- Theme 3 Winner: Sharad Nag (University of Minnesota) - A 22nm Reconfigurable Chiplet for FFT and AI Inference
- Theme 4 Winner: Arghadip Das (Purdue University) - ECO: Designing Energy-Efficient Multimodal Cognitive Systems Using Efficient Context Handling
Theme 1 winner Timur Ibrayev with CoCoSys Co-Director Arijit Raychowdhury and JUMP 2.0 Executive Director Roman Caudillo.
Theme 2 winner Zishen Wan with CoCoSys Co-Director Arijit Raychowdhury and JUMP 2.0 Executive Director Roman Caudillo.
Theme 3 winner Sharad Nag with CoCoSys Co-Director Arijit Raychowdhury and JUMP 2.0 Executive Director Roman Caudillo.
Theme 4 winner Arghadip Das with CoCoSys Co-Director Arijit Raychowdhury and JUMP 2.0 Executive Director Roman Caudillo.
Finishing Strong
The first three years of CoCoSys have shown the promise of the center’s mission. The final two will be dedicated to pushing it even further.
Among many other goals, researchers plan to continue testing the new chip and start exploring hyperdimensional computing. This type of computing uses hypervectors, which are high-dimensional vectors that can represent complex data in a way similar to how the brain encodes information. The approach represents a paradigm shift when it comes to the future of encoding information for AI reasoning.
“The plan is to double down on our neurosymbolic AI,” Raychowdhury said. “The most exciting thing is what we don’t know yet. But as long as we keep looking ahead, adapting to the changing landscape, and continuing to collaborate with all the talented engineers out there, we will stay ahead of the curve.”
For more information on CoCoSys and their research, click here.
Related Content
CogniSense Holds Second Annual Review
The event highlighted research and student projects conducted through the Center on Cognitive Multispectral Sensors, a JUMP 2.0 center, over the past year and previewed its upcoming goals.
Efficiency and Affordability Drive Groundbreaking ECE Projects at Spring 2025 Capstone Design Expo
The winning ECE project advances UAV technology using gallium nitride semiconductors, while the interdisciplinary winner creates affordable, precise reaction wheels for better satellite orientation control.