CU Boulder to lead million-dollar DARPA computational microelectronics research
Sanghamitra Neogi has earned a key Department of Defense contract to tackle a big problem with tiny electronics: microchips crippled by heat.
An assistant professor in the Ann and H.J. Smead Department of Aerospace Engineering Sciences at the University of Colorado Boulder, Neogi is leading a multi-university research team to revolutionize how manufacturers model and deal with heat in computers.
“Thermal challenges are very much known, but right now management of it is very trial and error,” Neogi said. “Intel published a paper in 2021 on how they’re doing it and the thermal modeling approach they use is 40+ years old. It’s an area that’s just been left behind.”
It is well documented that microchips and transistors fail due to heating challenges. Mitigation to this point has primarily been through bigger fans and cooling channels, but as chips have gotten smaller to pack in more processing power, heat has become a larger issue.
“With microelectronics, we are moving away from planar chips to 3D stacked chips because it makes memory and processing quicker, but you can’t cool the inner channels using regular methods because you don’t have the real estate. The current ideas don’t work very well,” Neogi said.
To find new solutions, the Defense Advanced Research Projects Agency (DARPA) has awarded Neogi’s team a $1 million contract over 18 months to create an atomistic thermal model of microelectronic systems. In addition to CU Boulder, the team also includes Prof. Sayeef Salahuddin from the University of California, Berkeley and Prof. Kaushik Roy from Purdue University.
Neogi and her team will start by creating computational thermal model of individual transistors at the deeply scaled nanometer level, one millionth of a millimeter in size, and will then expand the model to a millimeter-scale circuit element with 300,000 transistors.
“We’re going to predict how the temperature map looks like; which zones are hot and which zones are cold. But most importantly, why certain zones are hot and cold,” she said.
Although the chips are extremely small, the modeling is a significant undertaking, requiring supercomputing resources, machine learning, and artificial intelligence Neogi said.
“Inclusion of AI at different length scales will be a major component of this research. Right now thermal modeling is very trial and error. We want to be able to instead predict how things will fail. If we are successful, we will have a new thermal approach not just for chips, but microelectronic circuits, sensors, devices. We are building a method that scales dramatically,” she said.
Although DARPA is interested in the research from a military application perspective, the work could also have broad applications across all electronic devices.
Neogi is especially excited about the project’s alignment with the federal CHIPS Act of 2022, which seeks to dramatically expand semiconductor research and development in the United States. Although her project is funded separately, the work is highly synced with CHIPS research.
“This is a fundamental thing that is at the heart of all electronics. Thermal challenges affect all of them at the very core,” she said.
The full title of the DARPA program is Thermal Modeling of Nanoscale Transistors (Thermonat). The contract officially begins August 14, 2023.