Key Takeaways
- Scientists are exploring a more energy efficient approach to optical sensors that is inspired by how the retina works.
- In the approach, some data processing is conducted in the sensor itself, before the data is sent to a computer or processed by edge computing devices.
- The research aligns with broader efforts at Berkeley Lab to increase the energy efficiency of microelectronics.
Today's optoelectronic devices, from smartphone cameras to scientific instruments that image the night sky or microscopic cells, are pushed to extract more and more information from photons - an energy-intensive process that limits device performance and some commercial and industrial applications.
To improve this process, a team of scientists led by the Department of Energy's Lawrence Berkeley National Laboratory (Berkeley Lab) is exploring a more efficient way to crunch photons into images that's inspired by the human eye. The project also includes scientists from Sandia National Laboratories, UC Berkeley, UC Davis, and the University of Texas at Arlington.
There's significant room for improvement. Most optical sensors record data from light and then transmit all of the raw data to a computer for processing. This typically consumes much more energy than necessary, because in most applications, only a small amount of information relative to the raw data is needed to render the final output.
Instead, the team is developing a less power-hungry approach in which some data processing is conducted in the sensor itself, before the data is sent to a computer or processed by edge computing devices.
"Our approach works like a retina, which both takes in light and stimulates neurons to generate encoded information from that light," said Maurice Garcia-Sciveres, the project's lead and a senior scientist in Berkeley Lab's Physics Division. "Likewise, we want to create an optical sensor that also incorporates processing, so that only sought-after and information-dense data is sent to a computer for further processing."
"This could save a lot of power in optoelectronic devices because you wouldn't have to transmit a huge amount of data for processing, and once the data is transmitted, you wouldn't have to do as much processing," he added.
The Berkeley Lab-led project is part of the Microelectronics Energy Efficiency Research Center for Advanced Technologies (MEERCAT), which is one of three Microelectronics Science Research Centers recently announced by DOE. The centers, which bring together multi-institutional, multidisciplinary projects in partnership with industry, are organized around making microelectronics more energy efficient and able to operate better in extreme environments.
The research also aligns with Berkeley Lab's broader efforts to increase the energy efficiency of microelectronics.
As outlined in the DOE announcement, MEERCAT is committed to revolutionizing energy-efficient microelectronics by advancing integrated innovations across materials, devices, information-carrying modalities, and systems' architectures. Focusing on intelligent sensing, data bandwidth, multiplexing, and advanced computing, the center will explore transformative solutions that seamlessly bridge sensing, edge processing, artificial intelligence, and high-performance computing.
"Our project is very synergistic with the other projects within MEERCAT," said Garcia-Sciveres. "In our approach, we want to improve the energy efficiency of computing by avoiding some of the computation in the first place."
Garcia-Sciveres and colleagues have previously investigated ways to create optical sensors by stitching together nanostructures such as nanotubes and nanowires. These "nanoscale hybrids" are highly sensitive in part because the sensor's nanoscale components are smaller than the wavelength of light. The components of nanoscale hybrid sensors can also be tweaked in myriad ways, making them great platforms to explore how to best incorporate processing directly into a sensor.
"Our research will explore how to tailor nanoscale hybrid sensors so they can be optimized for different kinds of light detection, such as color or wavelength, with the goal of developing light sensors that only capture properties of interest," said Garcia-Sciveres. "In this way, the sensor will get the optical data we want at the outset, and in doing so will process the data before sending it to a computer. In the future, energy-efficient light sensors based on this approach could be trained for specific scientific or commercial applications."
Other Berkeley Lab scientists contributing to the project include Tevye Kuykendal, Archana Raja, and Ricardo Ruiz in Berkeley Lab's Molecular Foundry, a DOE Office of Science user facility; Mi-Young Im in Berkeley Lab's Center for X-Ray Optics in the Materials Sciences Division; Ali Javey and Feng Wang in the Materials Sciences Division; Aikaterini Papadopoulou in the Engineering Division; and Andrew Nonaka and Zhi Jackie Yao in the Applied Mathematics and Computational Research Division.
The Microelectronics Science Research Centers are funded by the DOE Office of Science.
