Six ECE faculty are involved in three new multi-institution research centers, all focused on next generation semiconductors. These centers are among the seven in total funded as part of the Joint University Microelectronics Program 2.0 (JUMP 2.0) program. Each center is funded at a level of approximately $30M over five years.
“This research is at a scale that you wouldn’t be able to do individually,” said Prof. Zhengya Zhang, who is involved in two of the centers. “It is only possible at the center level, which taps into all the resources that we already have.”
The seven new centers complement the goals of the Creating Helpful Incentives to Produce Semiconductors (CHIPS) and Science Act of 2022. But whereas most of the CHIPS Act money is going toward semiconductor foundries, the JUMP 2.0 program is focused on high-risk, high-payoff research “spurred by an increasingly connected world and a rapidly changing microelectronics landscape.”
“We’re on the edge of the next industrial revolution,” summarized the Semiconductor Research Corporation (SRC) in their overview of the Decadal Plan for Semiconductors, published in 2021. “Innovation in semiconductor technology is needed to advance information and communication technologies (ICT) critical to our economic growth and national security.”
After identifying the seismic shifts that will define the future of semiconductors and ICTs in their Decadal Plan, the SRC joined together with the Defense Advanced Research Projects Agency (DARPA) and a consortium of semiconductor companies to fund JUMP 2.0 to address the anticipated paradigm shifts. The 2.0 program is a continuation of the original SRC-organized JUMP program; one of those centers, the Applications Driving Architectures Center (ADA), was based at Michigan.
Each JUMP 2.0 center involves researchers from 12-13 institutions, while maintaining close collaborations with at least as many companies. These companies include ARM, Analog Devices, Boeing, EMD Electronics, Global Foundries, IBM, HRL Labs, Intel, MediaTek, Micron, Qorvo, Raytheon, Samsung, SK hynix, and TSMC.
“The industry partners will play a critical role in all the tasks that we and our students will be doing,” said Prof. Hessam Mahdavifar. “At the same time, working with our industrial partners will be a very valuable experience for the students, and help prepare them for their future professional career.”
Following is a description of the three centers involving ECE faculty:
- CogniSense: Center on Cognitive Multispectral Sensors
- ACE: Center for Evolvable Computing
- CUbiC: Center for Ubiquitous Connectivity
COGNISENSE: Center on Cognitive Multispectral Sensors
Participating Universities: Columbia University, Cornell University, Georgia Institute of Technology (Lead), Iowa State University, Massachusetts Institute of Technology, Purdue University, University of Delaware, University of California-Davis, University of California-Santa Barbara, University of Illinois-Chicago, University of Maryland, University of Michigan
Michigan PIs: David Blaauw, Michael Flynn, Hun-Seok Kim (Co-team Lead)
Funded by a $28.2M million grant from JUMP 2.0, Georgia Institute of Technology will lead the Center on Cognitive Multispectral Sensors (CogniSense) to develop sensors that can effectively “perceive” everything around them and, like humans, efficiently attend to only the information that really matters. Today’s electronic sensors sample everything they “see” and generate an abundance of digital data, sometimes way too much for a machine to store, process, and make sense. The CogniSense center’s goal is to change this paradigm by learning from biology.
The key novelty is the adaptive sensing. The entire system will change on the fly depending on the application and the scenario.Hun-Seok Kim
“Our research at Michigan involves the sensor interface and digital processing to do sensor fusion and information processing,” stated Prof. Hun-Seok Kim, who is co-leading the Analog-to-Insight theme with Justin Romberg at Georgia Tech.
Prof. Mike Flynn will figure out how to extract only the most relevant features from the sensors, which will include data acquired by radar, lidar, and cameras. This computation will be done in the analog domain so that the features are extracted before converting the signal into digital representations, which will greatly reduce the amount of data fed into the backend digital processing.
Kim and Prof. David Blaauw will work on the digital feature processing engine which will perform sensor fusion to extract a refined representation of the scene from multiple sensors.
“The key novelty is in the adaptive sensing,” explained Kim. “The back engine will dynamically find out where the interesting things are happening in the temporal and spatial domains, and control the analog sensor frontend based on that kind of analysis. The entire system will change on the fly depending on the application and the scenario.”
For example, explained Blaauw, a typical imager pixelates every square of the image equally. An enormous amount of power could be saved by capturing in high resolution only the relevant portion of the image, possibly acquiring no data at all from an area where nothing is happening.
“We are making sensors smarter, to do more AI processing on the data as you measure it,” added Blaauw. “An important impact on the entire system is to greatly lower the amount of data that needs to be transmitted, which makes it much lower power.”
ACE: Center for Evolvable Computing
Participating universities: Cornell University, Georgia Institute of Technology, Harvard University, Massachusetts Institute of Technology, Ohio State University, Purdue University, Stanford University, University of Texas-Austin, University of California San Diego, University of Kansas, University of Illinois-Urbana Champaign (Lead), University of Michigan, University of Washington
Michigan PI: Zhengya Zhang (Team Lead)
Funded by a $31.5 million grant from JUMP 2.0, the University of Illinois Urbana-Champaign (UIUC) will lead the ACE Center for Evolvable Computing to advance distributed computing technology, from cloud-based datacenters to edge nodes, so it operates with orders of magnitude more energy efficiency than today.
“Distributed computing refers to the fact that we have compute resources located everywhere,” said Prof. Zhengya Zhang, who is one of the team leads in the ACE Center. “These locations include where the data is being captured, which is the edge, to the network switches where the data is being routed, and ultimately the server rooms. How do you optimally provision resources so that you can meet the demands of potentially millions of users all at the same time?”
We want to dramatically limit the power required to process the data and the cost of the hardware, while still being able to process the data in real time.Zhengya Zhang
Zhang is focused on designing evolvable heterogeneous computing hardware that will accelerate the tasks, so that results can be obtained faster than could be accomplished on a conventional computer.
“Data for many applications is only useful when it can be processed in time,” added Zhang. “Real-time decision making is the key aspect.”
Also – the hardware needs to be evolvable in order to adapt to new applications as they arise, as opposed to constantly designing new architectures and fabricating new chips. That process is simply too slow and expensive to meet the demand, says Zhang.
Two applications that are the target of the ACE Center are edge computing and cloud computing.
“We want to dramatically limit the power required to process the data and the cost of the hardware, while still being able to process the data in real time,” said Zhang.
Zhang is the demonstrator team leader. “We plan to put all our new techniques into these demonstrators,” said Zhang, “from hardware to software to networking to security – to see if we can achieve something we won’t be able to do individually.”
CUbiC: Center for Ubiquitous Connectivity
Participating Universities: Columbia University (Lead), Cornell University, Duke University, Massachusetts Institute of Technology, Oregon State University, Princeton University, Stanford University, University of California-Berkeley, University of California-San Diego, University of California-Santa Barbara, University of Illinois at Urbana-Champaign, University of Michigan, and University of Southern California.
Michigan PIs: Elaheh Ahmadi, Hessam Mahdavifar, Zhengya Zhang
Funded by a $35M million grant from JUMP 2.0, Columbia University will lead the Center for Ubiquitous Connectivity (CUbiC) to advance energy-efficient communications technologies for addressing the vastly growing connectivity bottlenecks between data-hungry wireless devices and deluged data centers. CUbiC will strive to flatten the computation-communication gap, delivering seamless Edge-to-Cloud connectivity with transformational reductions in the global system energy consumption.
We are thinking 10 years into the futureHessam Mahdavifar
“The key word is seamless.” said Prof. Hessam Mahdavifar. “We want to provide seamless on-demand connectivity from edge to cloud – from densely populated areas to remote parts of the world.”
Mahdavifar will contribute to connectivity networks and systems by providing resiliency and security on these platforms. His group aims to develop new forward error correction modules to provide resiliency against the physical layer noise, as well as a wide range of other imperfections that are expected to arise in future technologies.
He will also focus on providing security modules that will be robust even with the speed of quantum computing, while also protecting against spoofing attacks.
“We are thinking 10 years into the future,” said Mahdavifar, “to provide solutions for the entire connectivity ecosystem – from hardware to optical links, from data centers to satellites.”
Mahdavifar will also team up with Prof. Zhengya Zhang to design super fast decoders, and then implement them in hardware.
Zhang and his group will focus primarily on developing the signal processing hardware to support the massive scale of future wireless communication systems. These wireless systems are expected to involve a large number of distributed antennas. He will use signal processing to leverage those many antennas to improve the data bandwidth.
Prof. Elaheh Ahmadi and her team will be working on the next generation of materials and devices for beyond 5G applications.
CogniSense (press release)
CUbiC (press release)
Joint University Microelectronics Program 2.0: with a summary of all 7 centers