Eighty years after the University of Pennsylvania gave the world ENIAC, the first general-purpose electronic computer, researchers at that same institution are working on something that could reshape computing at its most fundamental level. Not by making chips faster in the conventional sense, but by rethinking what carries information inside them in the first place.
The proposal: replace electrons with light.
Why the Hardware Powering Today’s AI Is Running Into a Wall
Every computer built since ENIAC, from the smartphone in a pocket to the data centers running large AI models, operates on the same basic principle. Electrons move through materials, and that movement encodes and processes information. It has worked extraordinarily well for decades. But electrons carry an electrical charge, and that charge creates problems that compound as systems grow more complex.
As electrons move through chip materials, they generate heat and encounter resistance. Energy is wasted. At the scale of modern AI systems, which process enormous volumes of data continuously, those losses are not trivial. The physical limits of electron-based hardware are becoming harder to engineer around, and the energy demands of AI are accelerating faster than conventional chip design can accommodate.
This is the context in which photonic computing, the idea of using light particles called photons instead of electrons, has attracted serious scientific attention.
The Particle That Bridges Two Worlds
Photons have properties that make them appealing for computing. As Li He, co-first author of a paper published in Physical Review Letters and a former postdoctoral researcher in the Zhen Lab at Penn, explains it: photons are charge-neutral and have zero rest mass, which allows them to carry information quickly over long distances with minimal energy loss. That is why light already dominates communications technology.
The problem is the flip side of those same properties. Because photons barely interact with their environment, they struggle with the switching operations that computing depends on. A system that can carry information efficiently but cannot make decisions with it is not yet a computer.
This is what makes the work of physicist Bo Zhen and his team at Penn’s School of Arts and Sciences significant. Their approach does not simply use light instead of electrons. It creates a new kind of particle that combines properties of both.
The particle is called an exciton-polariton. It forms when photons are strongly coupled with electrons inside an atomically thin semiconductor material. The result is a quasiparticle that retains light’s efficiency for carrying information while gaining enough interaction with its environment to perform the switching logic that computing requires.
The team demonstrated all-light switching using approximately 4 quadrillionths of a joule of energy. That figure is worth pausing on: it is far below the energy required to briefly power a tiny LED light. The switching happened without converting light signals back into electronic ones, which is precisely the conversion step that currently limits photonic AI chips.
Here is what most coverage of photonic computing misses: many experimental photonic AI chips already use light for certain calculations. The bottleneck is not the fast, linear parts of computation. It is the nonlinear activation steps, the decision-making operations, where existing photonic systems have to fall back on electronics. That conversion slows processing and increases energy use, partially negating the advantages of using light in the first place. Exciton-polaritons are a proposed solution to that specific problem.
What This Means Beyond the Laboratory
The implications, if the technology can be scaled, extend in several directions.
AI systems are among the most energy-intensive computational workloads ever built. The ability to perform more of that computation in the photonic domain, without repeated conversions between light and electricity, could substantially reduce the power demands of large AI infrastructure. That matters not just for cost but for the broader question of how societies sustain AI development as it grows.
There is also a more immediate application the Penn researchers point to: photonic chips capable of processing information directly from cameras without conversion steps. Vision-based AI systems, which are central to robotics, autonomous vehicles, and medical imaging, currently require that incoming light data be converted to electronic signals before processing begins. A chip that works natively in the photonic domain could handle that pipeline more efficiently.
The research also points toward quantum computing. Exciton-polaritons could potentially support basic quantum functions on future chips, though that application remains further from practical implementation.
The work was supported by the US Office of Naval Research and the Sloan Foundation. Additional authors include Zhi Wang and Bumho Kim from Penn’s School of Arts and Sciences. Li He, who co-led the research, is now an assistant professor at Montana State University.
In Short
Electrons have powered computing since ENIAC. They are reaching physical limits that matter most for AI. Photons carry information efficiently but cannot perform the switching logic computers need. Penn researchers have created a hybrid particle, the exciton-polariton, that combines properties of both, enabling all-light switching at extraordinarily low energy levels. If this approach scales, it could reduce the energy cost of AI systems and remove a key bottleneck in photonic chip design. The physics is new. The problem it addresses is urgent.
Based on reporting from ScienceDaily AI.