Portuguese 
English 
Spanish 
3 min of reading 
0 comments 
Researchers from the University of Pennsylvania and Montana State University presented a fully optical switch made with MoSe₂ and photonic nanocavity, capable of controlling light with energy close to 4 femtojoules and operating on a scale of a few picoseconds, paving the way for photonic chips aimed at AI, neuromorphic computing, and quantum technologies.
A light-controlled switch achieved fully optical switching using about 4 femtojoules of energy, an advancement described in 2026 in the Physical Review Letters by researchers from Pennsylvania and Montana for future AI photonic chips.
Switch uses light to control light
The proposal seeks to address a central obstacle of fully optical computing: making photons interact with each other. In electronic systems, logical operations depend on the movement of electrons. In photonic devices, the goal is to process information using light.
This path is of interest because light can travel faster and generate less heat than moving electrons. Still, photons typically do not interact, which makes it difficult to build components capable of switching light signals with low consumption.
-
Hidden hundreds of kilometers deep, two rivers of molten rock are devouring the ancient base of the North American continent towards the Black Hills, overturning the old certainty that this core was an untouchable and immovable anchor of the Earth.
-
A rare species of fish draws the attention of scientists for challenging evolutionary theories by thriving for over 100,000 years without sexual reproduction, thanks to a mechanism known as genetic conversion; meet the Amazon molly.
-
A solar panel created in the United States transforms real seawater into drinking water without generating the toxic brine that pollutes the oceans, and it also recovers salt and valuable minerals like lithium, an advancement that could simultaneously alleviate water scarcity and the search for batteries.
-
A cannibal coronal mass ejection, formed by the fusion of two solar plasma clouds, hit Earth in the early hours of June 5, 2026, but with a weaker impact than expected, causing a moderate to strong geomagnetic storm instead of the severe one anticipated.
The group worked with a system based on molybdenum diselenide, or MoSe₂, a 2D semiconductor formed by a single layer. The material was combined with a photonic crystal nanocavity, a structure designed to confine light on an extremely small scale.
How MoSe₂ creates interactions between photons
The operation depends on exciton-polaritons, hybrid quasiparticles formed when photons couple with excitons, bound pairs of electrons and holes within the semiconductor. This state combines properties of light and matter.
For their light part, polaritons can propagate at high speed. For the material part, inherited from excitons, they begin to exhibit interactions that allow altering the optical behavior of the system with little energy.
The nanocavity acts as a precise trap for light. By confining the polaritons in a sub-wavelength region, it increases the interaction strength between the particles and enhances nonlinear optical responses.
In the study, the charge-tunable MoSe₂ monolayer was coupled to a planar photonic crystal nanocavity. The excitonic resonance of the material favored robust hybridization between excitons and photons.
Result may aid photonic AI chips
The experiment demonstrated fully optical switching of the cavity spectrum with excitation energies as low as approximately 4 femtojoules. The paper states that this value is several orders of magnitude below previously reported limits in 2D exciton-polariton systems.
Pump-probe spectroscopy analysis indicated operation on an ultrafast scale, of a few picoseconds. This time reinforces the switch’s potential for integrated photonic platforms that require fast response and low energy.
Li He, assistant professor at Montana State University and senior author of the study, told Tech Xplore that the motivation was to advance fully optical computing, an area that seeks to process information with light, not electricity.
Platform designed for large-scale integration
Besides performance, researchers highlight the possibility of mass production. The platform uses materials and structures that can be standardized by standard manufacturing techniques, which favors integration into larger photonic circuits.
This approach could open the way for chips with thousands of interactive optical components. Among the cited applications are fully optical neuromorphic computing, quantum photonic information processing, and high-speed hardware for artificial intelligence models.
The 4 femtojoule limit does not yet represent a fundamental physical barrier. The team claims to have ways to reduce this threshold by orders of magnitude and explore the quantum regime, where a single photon could control another.
Be the first to react!