Zinc Doping Enables Visible-Light Programming of Ferroelectric Memristors for Neuromorphic Computing

Key Takeaways

• Zinc doping cuts switching energy by 69%
• Visible light programs memristors at 10 mW cm⁻²
• Device retains states indefinitely after illumination
• Multilevel resistance offers 24 analog states
• Optical reservoir computing achieves 98.6% accuracy under noise

Pulse Analysis

Ferroelectric memristors have long promised brain‑inspired computing because their bistable dipoles store data without power. Yet lithium niobate, prized for its optical qualities, required intense ultraviolet light to flip polarization, limiting practical use. Introducing zinc ions reshapes the crystal lattice, narrows the bandgap, and eliminates defect‑mediated barriers, dropping the switching energy from 146 meV to 45 meV. This atomic‑scale engineering makes modest visible light sufficient to program the device, delivering non‑volatile states that persist without power—a critical step toward truly photonic synapses.

Beyond the material science, the researchers devised a pulsed laser‑magnetron sputtering co‑deposition method that produces uniform 85‑nm zinc‑doped films on strontium titanate substrates without the multi‑step ion‑implantation and wafer‑bonding pipelines that have hampered scale‑up. The resulting memristors exhibit less than 3 % voltage variation across 100 cycles, survive over 10⁸ switching events, and maintain a thousand‑fold resistance contrast. Such reliability, combined with 24 discrete conductance levels, positions the technology as a viable alternative to conventional CMOS‑based analog accelerators, especially for applications where space and power are at a premium.

The ability to encode visual information directly with light opens new horizons for edge AI and machine‑vision systems. In reservoir‑computing tests, the optically programmed array classified noisy MNIST digits with 98.6 % accuracy, demonstrating inherent noise tolerance and rapid inference. As data centers and autonomous platforms seek to cut energy consumption, integrating sensing, memory, and processing in a single optoelectronic chip could dramatically reduce data movement overhead. Industry players eyeing neuromorphic processors may soon adopt zinc‑doped lithium niobate memristors to build compact, low‑power vision modules that operate on ambient illumination, accelerating the shift toward photonic‑centric AI hardware.