The advent of artificial intelligence, or AI, has brought new challenges and new opportunities to the microelectronics industry.
“When it comes to making progress in AI, energy is the showstopper,” says Sarma Vrudhula. “Artificial intelligence computations take incredible amounts of electrical energy. For us to continue to advance, and for overall sustainability, we must find ways to reduce power consumption.”
Vrudhula is a professor of computer science and engineering in the School of Computing and Augmented Intelligence, part of the Ira A. Fulton Schools of Engineering at Arizona State University. To power the next phase of the AI revolution, he will design new microchips and new systems to meet these emerging needs.
The engineer and semiconductor expert is a notable figure in the development of computer-aided design software for circuit systems and microelectronics. Vrudhula has spent the better part of the past 20 years researching ways to automate the process of creating new types of microchips.
A microchip is a collection of semiconductors, which are usually tiny pieces of silicon that conduct electrical energy and serve as a computer’s processor or memory unit. For years, these chips were designed by hand with electrical engineers drawing patterns and circuits to be etched on the silicon in the way architects would draft blueprints.
Thanks to the work of experts in operations research and design automation, including Vrudhula, software was developed to take over this planning. Today, engineers can feed design and performance expectations into sophisticated software programs, such as those from Cadence and Synopsys, to receive optimized designs that meet their needs.
An image of one of Vrudhula’s earlier chip designs. His new work is funded by the CHIPS and Science Act of 2022, which was developed to revive domestic manufacturing of semiconductors. Photo credit: Erika Gronek/ASU
New ideas to power new technology
Vrudhula and his team have now received a $2 million grant from the National Science Foundation, or NSF, to develop an ultra-energy-efficient chip to power AI work.
For the project, Vrudhula will collaborate with colleagues Marwan Krunz from the University of Arizona and Sanmukh Kuppannagari from Case Western Reserve University. Their work is part of Future of Semiconductors, or FuSe2, an NSF initiative aligned with the CHIPS and Science Act of 2022.
The CHIPS and Science Act is a groundbreaking collaboration between U.S. governmental agencies and industry partners to address the global microchip shortage that was triggered in 2020 by the COVID-19 pandemic. The program seeks to restore the manufacturing capabilities of the American semiconductor sector, which began to diminish in the 1980s as production moved overseas.
It’s an important effort. Experts say the demand for semiconductors is rising and the global market for them will reach $1 trillion by 2030. The demand for AI-powered products and solutions is also increasing and it’s driving a corresponding surge in energy consumption.
Vrudhula’s new methodology uses complementary metal-oxide semiconductor, or CMOS plus X, technology. He expects the resulting design to be 100 times more energy-efficient than current generation technology, in part because AI calculations can be performed on the chip itself. Vrudhula’s design ideas have been awarded 18 patents, and the new chip is slated for use in devices on 5G and 6G cellular networks, such as mobile phones and tablets.
In most computer systems, different chips function in different ways, Vrudhula explains. In a typical scenario, a logic chip, known as a CPU, will perform the computer’s calculations while the results will be stored on a memory chip until they are transferred to the system’s hard drive. All of this consumes energy; but if some of the logic function can be combined onto the memory chip, power can be saved.
As the team winds down their initial design process, they are preparing to fabricate numerous prototypes and seeking manufacturing options with the Southwest Advanced Prototyping Hub, or SWAP Hub, as well as with industry partners.
Software designs chips, computer scientists design software
But as industry builds new semiconductor manufacturing facilities, it needs talent trained to enable production.
So, a key goal of the FuSe2 program is addressing critical workforce shortages in the semiconductor field. According to the Semiconductor Industry Association, the U.S. is at risk of having 1.4 million unfilled jobs in the microelectronics sector by 2030.
To help fill these gaps, Vrudhula knew just where to look.
“We’re going to work hard to get computer science students interested in hardware design,” he says.
ASU has one of the nation’s largest computer science programs. The university’s School of Computing and Augmented Intelligence welcomed more than 12,000 students this fall, and it broke graduation records at this spring’s convocation ceremonies where it awarded more than 1,800 degrees.
Vrudhula says students might be surprised to learn how much overlap exists between computer science, electrical engineering and applied mathematics. Computer scientists play vital microelectronics industry roles helping to develop hardware and software systems, operating on the forefront of innovation in areas such as computational lithography, which is essential for producing more powerful chips.
As part of his grant, Vrudhula is creating a new course that will provide a link between the computer science curriculum and an understanding of chip design. Students will have access to on-campus paid internships and meetings with important figures in the semiconductor industry.
Vrudhula, who also serves as the director of the Intelligent, Distributed, Embedded Applications and Systems Center, hopes to demonstrate that computer science students searching for opportunities to create AI technology, especially in the development of deep neural networks, can have exciting and rewarding careers in microelectronics.
Ross Maciejewski, director of the School of Computing and Augmented Intelligence, says that Vrudhula’s work reflects the school’s efforts to connect students with good job opportunities and to provide employers with a highly skilled workforce.
“Not only is Sarma Vrudhula designing an excellent piece of microelectronics, but he is also teaching students to do the same,” Maciejewski says. “This CHIPS-funded project honors the FuSe mission of producing new technology and training the next generation of technologists.”