Researchers at Colorado State University and the University of California Berkeley have developed an innovative way to compute and store data using the orientation of light waves rather than their brightness. By pairing micro-optical tracks with an ultrathin, two-dimensional material, this technology allows electronic signals to dynamically switch light paths with almost zero wasted energy. The resulting system enables ultra-fast, high-capacity computing without the thermal overheating issues common in traditional processors. This breakthrough opens new avenues for compact hardware tailored for next-generation artificial intelligence and data networks.
Today’s fast light-based memory and computing chips try to store and process digital information by turning light brightness up and down or by melting and cooling special glass-like materials. Unfortunately, melting these materials with heat takes a lot of electricity, runs slowly, and causes tiny microchips to overheat. To solve this, this team of researchers used a special, microscopically thin material that can be controlled directly with tiny pulses of electricity instead of heat. By using electricity to instantly shift the material’s internal crystal pattern, the chip changes how light waves vibrate—rather than just making them brighter or dimmer—which avoids the massive energy waste and heat buildup of older methods.
This technology builds a brand-new kind of light-based memory and computer chip by placing an ultra-thin layer of advanced electronic material on top of microscopic tracks (waveguides) that carry light. When a tiny electric voltage is applied to the material, its internal crystal structure instantly flips between two natural patterns. As light travels along the track underneath, this crystal flip twists the light waves, allowing the system to store digital data or perform math based on the angle of the light wave rather than its brightness.
Because the system tracks how the light wave is twisted instead of trying to dim or block the light, it suffers far less signal loss and makes fewer reading errors. When these light tracks are linked together into a complete chip network, they can perform thousands of complex artificial intelligence calculations at once. Compared to traditional light-based computing networks, this system improves math accuracy by up to 83%. Best of all, once the crystal structure is flipped to store a piece of data, it stays in place forever without needing constant electrical power, cutting operational energy use from nearly a watt down to tiny milliwatts.
• Zero Static Power Retention: The non-volatile ferroelectric crystal states maintain stored data permanently without requiring continuous electrical bias or refresh cycles.
• Ultra-Low Operating Power: Replaces energy-intensive thermal melt-quench mechanisms with direct low-voltage electrical field switching, dramatically reducing total programming energy.
• High Computation Accuracy: Minimizes optical insertion losses and noise sensitivity associated with light intensity modulation, achieving higher precision in matrix-vector calculations.
• Ultrafast Switching Speed: Exploits acoustic phonon lattice dynamics to switch crystal orientations rapidly without thermal delays.
• Active Dynamic Control: Enables real-time, in-situ electrical tuning of light polarization within a single material layer rather than relying on passive, fixed-angle stacked structures.
• Foundry Compatibility: Seamlessly integrates thin 2D materials directly onto established silicon and silicon nitride photonic integrated circuit manufacturing platforms.
• AI Hardware Accelerators: High-performance, energy-efficient photonic tensor cores for artificial intelligence inference and deep learning workloads.
• Data Center Interconnects: High-speed, low-latency chip-to-chip and optical transceiver communication bridges.
• In-Memory Computing Modules: Non-volatile optical memory arrays for neuromorphic processing and edge computing systems.
• Smart Optical Routing: High-throughput photonic network switches, optic firewalls, and dynamic signal routing tables.
• On-Chip Optical Sensing: Compact, polarization-sensitive image processing arrays and compute-in-imaging hardware.
Available for Exclusive Licensing
TRL: 3
US Provisional Patent
Mahdi Nikdast
Amin Shafiee
Jie Yao
Linhong Chen
Jessy McGowan
Jessy.McGowan@colostate.edu
970-491-7100