Photonic chips have revolutionized data-intensive technology. Alone or in conjunction with conventional electronic circuits, these laser-driven devices transmit and process information at the speed of light, making them promising solutions for data-intensive artificial intelligence applications.
In addition to unmatched speed, photonic circuits use significantly less energy than electronic circuits. Electrons move relatively slowly through the hardware, colliding with other particles and generating heat, while photons flow without losing energy and generate no heat at all. Integrated photonics are poised to play a leading role in sustainable computing, freed from the energy loss burden inherent in electronics.
Photonics and electronics draw on different fields of science and use different architectural structures. However, both rely on lithography to define circuit elements and connect them in sequence. Photonic chips don’t take advantage of transistors placed in increasingly shrinking, increasingly layered trenches in electronic chips, but their complex lithographic patterning directs laser beams into coherent circuits to run computational algorithms. form a photonic network that can
But now, for the first time, researchers at the University of Pennsylvania School of Engineering and Applied Sciences have created photonic devices that offer programmable on-chip information processing without lithography, enhancing photonics with greater precision and flexibility for AI applications. speed. .
Offering unparalleled optical control, the device consists of spatially distributed optical gain and loss. Lasers project light directly onto semiconductor wafers without the need for a defined lithography path.
Liang Feng is a professor of Materials Science and Engineering (MSE) and Electrical Systems Engineering (ESE) and holds a Ph.D. His student Tianwei Wu (MSE) and postdoctoral fellows Zihe Gao and Marco Menarini (ESE) introduced microchips in a recent study published in Nature Photonics.
Silicon-based electronic systems have transformed the computing environment. However, they have distinct limitations. They process signals slowly, process data serially rather than in parallel, and can only be miniaturized to a certain extent. Photonics is one of the most promising alternatives as it can overcome all these shortcomings.
“However, photonic chips for machine learning applications face obstacles in complex manufacturing processes with fixed lithographic patterns, limited reprogrammability, error- and damage-prone, and costly.” says Feng. “By eliminating the need for lithography, we are creating a new paradigm. Our chips overcome these obstacles and remove all kinds of constraints from predefined functions, resulting in improved accuracy. and provide ultimate reconfigurability.”
Without lithography, these chips are powerhouses of adaptive data processing. Since the pattern is predefined and not etched, the device is essentially defect free. Perhaps more impressively, the lack of lithography has made microchips impressively reprogrammable, allowing laser cast patterns to be adjusted for optimal performance, making the task simpler (fewer inputs, larger datasets). small) or complex (many inputs, large datasets).
In other words, device complexity and minimalism are a living thing, adaptable in ways that etched microchips can’t.
“What you have here is incredibly simple,” says Wu. “It is very quick to build and use. It can be easily integrated with classical electronics. can also be changed and reprogrammed on the fly.”
An unassuming slab of semiconductor, this device couldn’t be easier. Manipulation of the material properties of this slab is key to the research team’s breakthrough in projecting lasers into dynamically programmable patterns to reconfigure the computational capabilities of photonic information processing processors.
This ultimate reconfigurability is important for real-time machine learning and AI.
“What’s interesting is how you control the light,” says Menarini. Conventional Photonic His chip is a technology based on passive materials, which scatter light and bounce it back and forth. Our material is active. A beam of pumping light modifies the material to release energy and increase the amplitude of the signal when the signal beam arrives. ”
“This active nature is the key to this science and the solution we need to realize technology that does not require lithography,” adds Gao. “We can use this to reroute optical signals and program optical information processing on-chip.”
Feng likens the technique to an artistic tool, a pen for drawing on a blank page.
“What we have achieved is exactly the same. Pumping light is a pen for drawing photonic computational networks (pictures) on unpatterned semiconductor wafers (blank pages).”
But unlike indelible lines of ink, these beams of light can be drawn and rewritten, their patterns taking countless paths into the future.
