Researchers at Cornell University have developed a new type of computing device that stores information electrically and reads it through small mechanical movements. This is an unusual approach that could pave the way for more energy-efficient hardware for artificial intelligence and scientific computing.
The device is Published in the journal Nano Letters In April, By combining ferroelectric materials and microscopic vibrating beams, stored analog information can be accessed without relying on traditional electrical readouts. that is designed for neuromorphic computing, a brain-inspired approach to information processing, and pervasive analog in-memory computing, where memory and computation are tightly integrated.
“Today’s computers are very powerful, but they usually separate memory and computation,” the PhD student said. Shubham Jadhavled the research with Amit LalRobert M. Scharf 1977 Professor, Department of Electrical and Computer Engineering, Cornell Duffield Polytechnic Institute. “For AI and scientific computing, this means that systems spend a lot of time and energy just moving numbers around. We’re asking whether the materials themselves can store values and at the same time support computations with those values.”
Many ferroelectric devices use the same electrical path to write, store, and read information, which can lead to unnecessary current paths and read-related failures. The researchers wanted to move the readout to a mechanical channel while preserving electrical write functionality to reduce electrical disturbances and allow access to stored state with very low idle power at the device level.
To achieve that, the team built a ferroelectric microelectromechanical system (FeMEMS) using a 20-nanometer layer of hafnium zirconium oxide incorporated into suspended beams. Electrical pulses program the material by reorienting microscopic ferroelectric domains within the beam. A small readout signal causes the beam to vibrate, resulting in a stored value being displayed.
The researchers demonstrated roughly 200 distinguishable electromechanical states, rather than simple binary ones and zeros, allowing fine control over analog values. Accuracy is important because analog calculation errors can accumulate when many operations are performed simultaneously.
“If each value stored is just an approximation, these small errors can accumulate over many calculations,” Jadhav says. “By creating many distinguishable states, we can more accurately represent analog weights.”
Because the received signal and stored state interact directly within the device, beam motion provides a physical analog of multiplication, one of the most common operations in AI systems. In a simplified analogy, if the programmed beam state represents the number 6 and the received signal represents the number 8, the beam movement corresponds to their product, 48. In real AI hardware, this type of multiplication is repeated many times to process information.
Although the research is motivated by neuromorphic computing, Jadhav said the approach could also be extended beyond AI hardware. Electrically programmable beam motion could help researchers explore new ferroelectric materials and develop adaptive microsystems that combine ferroelectricity with capacitive, optical, or mechanical sensing.
The next step is to develop larger arrays of devices that can perform more complex matrix operations while integrating control circuitry and capacitive sensing systems.
“Before CMOS became a mainstream technology, computing was a much more open playground,” Jadhav says. “Researchers looked at different physical ways to store and process information. CMOS eventually became so powerful and scalable that many of those ideas faded into the background. Now, as traditional scaling becomes more difficult, we can revisit some of these concepts using modern materials, micro- and nanoscale manufacturing, and multiphysics design. That’s what makes this platform exciting.”
This research was supported by the Defense Advanced Research Projects Agency’s Nanowatt Platform for Sensing, Analysis, and Computing program. The device was manufactured at: Cornell University Nanoscale Facilityis supported by the National Science Foundation. Electron microscopy was performed at Cornell University. A platform that accelerates the realization, analysis, and discovery of interfacial materials and Cornell Materials Research Center.
Syl Kacapyr is an Associate Director of Marketing and Communications at Duffield Engineering.
