In the world of electronics, devices are typically designed to sense, compute, and store information separately. However, the human brain accomplishes all these tasks seamlessly in a single glance. Inspired by this efficiency, we sought to create a device that integrates vision and memory.
Our journey began with a serendipitous observation in the KAUST nanofabrication facility. While testing devices, we noticed that chips stored in the humid lab air exhibited clear memory effects under light, whereas identical chips kept dry did not. This wasn't noise—it was our missing ingredient: water molecules.
Upon further investigation, we discovered that under voltage, ambient H₂O molecules on the hBN surface split into H⁺ and OH⁻ ions. Under illumination, silicon generates electrons that tunnel through a thin interfacial oxide into the hBN. These ions and electrons meet at the hBN grain boundaries, forming microscopic conductive filaments—the physical basis of memory. Suddenly, humidity wasn't a contaminant but a feature.
Armed with this insight, we engineered a robust, low-temperature photonic memristor:
-
Wafer-Scale hBN at 250 °C:
We replaced the 1000 °C furnace with plasma-enhanced chemical vapor deposition (PECVD) at a gentle 250 °C, fully compatible with back-end-of-line silicon processes. Overnight growth yielded uniform, nanocrystalline hBN on 4-inch wafers. -
“Goldilocks” Grain Boundaries:
-
Single-crystal hBN: Too few defects → no ion paths
-
Amorphous hBN: Too many defects → erratic switching
-
Nanocrystalline hBN: Just the right grain boundaries → stable, repeatable ion highways
-
-
Light-Power Dial for Memory Modes:
By tuning laser intensity, a single device toggles among three states:-
Dark (0%): No switching → Blank slate
-
Dim (≈5%): Volatile memory → Short-term memory
-
Bright (100%): Nonvolatile memory → Long-term memory
-
Our optimized memristor achieved:
-
Spectral Omnivore: Reliable switching from 375 nm (UV) to 1064 nm (IR)—the broadest bandwidth reported for any two-terminal photonic memristor.
-
Thermal Titan: Endured >10⁶ switching cycles at 300 °C—ideal for automotive, aerospace, and other extreme-environment electronics.
-
Blazing Contrast: On/off resistance ratio > 10⁹—over 10,000× better than previous hBN devices, ensuring ultrahigh precision.
By collapsing sensing, memory, and computing into one device, we can build smaller, faster, and dramatically more energy-efficient systems. Potential breakthroughs include:
-
Real-Time AI Vision: Smart glasses that analyze your environment on the lens itself, without off-chip processing.
-
Neuromorphic Edge Computing: Cameras that detect edges, recognize patterns, and make decisions—all in hardware, bypassing von Neumann bottlenecks.
-
Harsh-Environment Electronics: Electronics that “see and remember” in engines, spacecraft, and downhole oil sensors at temperatures up to 300 °C.
We demonstrated a playful proof-of-concept: programming a 5 × 5 array to spell “KAUST” with colored light—a vivid illustration of parallel, in-sensor computation.
Our next milestones:
-
CMOS Integration: Merging memristors with on-chip read-out circuits and micro-LED arrays for fully self-contained vision chips.
-
Multi-Modal Sensing: Functionalizing hBN surfaces to detect heat or chemical species, turning each pixel into a miniature, multi-sensor hub.
-
3D Stacking & In-Memory Computing: Building vertical stacks of memristor layers to boost density and perform complex AI tasks directly in hardware.
We invite collaborators in photonics, materials science, and AI hardware to join us as we push toward a future where machines see, think, and remember just like we do.