Designable Van Der Waals Crystals Mimic Neuronal Cells with Light

Researchers have developed a designable van der Waals crystal that mimics the behavior of biological neuronal cells using light, according to a study published in Science on June 19, 2026. This artificial neuron enables light-driven learning by simulating the way biological brains process and store information, providing a hardware foundation for neuromorphic computing that avoids the energy inefficiencies of traditional silicon-based AI.

The system uses van der Waals (vdW) crystals, which are materials composed of layers held together by weak intermolecular forces. By engineering these layers, the researchers created a device that triggers electrical spikes in response to light pulses, mirroring the “fire-and-forget” mechanism of human neurons. This process allows the hardware to perform computations and memory storage in the same physical location.

How does the vdW crystal mimic a biological neuron?

Biological neurons operate by integrating incoming signals until they reach a specific threshold, at which point they fire an action potential. The artificial vdW crystal mimics this through a process called optoelectronic spiking. According to reporting from Tech Xplore, the crystal responds to light stimuli by accumulating charge. Once the light intensity or duration hits a predefined limit, the crystal releases a sharp burst of electrical current.

The “designable” nature of the crystal is a key technical advancement. Researchers can tune the material’s properties—such as layer thickness and chemical composition—to change how the artificial neuron reacts. This means they can set different “firing thresholds” for different cells, allowing for the creation of complex networks that can distinguish between various light patterns.

This light-driven learning capability is achieved through synaptic plasticity. In biological brains, the connection between neurons strengthens or weakens based on activity. The vdW crystal simulates this by altering its conductance based on the history of light exposure. As reported by ChemEurope, this allows the device to “learn” and remember patterns without needing a separate software-based memory bank.

Why is this different from current AI hardware?

Most modern AI runs on the Von Neumann architecture, which separates the central processing unit (CPU) from the memory. Data must constantly travel back and forth between these two points, creating a bottleneck that consumes massive amounts of energy and slows down processing. This is known as the Von Neumann bottleneck.

The vdW crystal implements “in-memory computing.” Because the crystal itself stores the “weight” of the connection (the memory) and processes the light signal (the computation) simultaneously, the need for data transport is eliminated. This approach mimics the brain’s efficiency, where neurons and synapses are integrated.

Compared to standard silicon transistors, which require constant power to maintain states or move data, the optoelectronic vdW neuron uses photons to trigger state changes. This drastically reduces the thermal output and power requirements. While GPUs rely on billions of transistors switching on and off to simulate a neuron via mathematics, this crystal is a physical neuron that behaves like one by nature.

What are the technical implications for the industry?

The ability to control these crystals with light opens doors for integrated photonic circuits. Instead of using copper wires to move electrical signals, which generate heat, these systems could use light to communicate between artificial neurons. This would potentially increase the speed of AI hardware while further lowering energy costs.

The research suggests several immediate applications for this technology:

• Edge Computing: Low-power AI sensors that can process visual data locally without sending it to a cloud server.
• Optical Pattern Recognition: Hardware that can identify shapes or signals in light streams in real-time.
• Biocompatible Interfaces: Potential for future integration with biological systems due to the low-energy, light-based signaling.

The study in Science notes that the designable aspect of the vdW crystal allows for scalability. Researchers aren’t limited to a single type of neuron; they can engineer an array of crystals with varying sensitivities to create a multi-layered artificial neural network on a chip.

What happens next for vdW neuromorphic chips?

The current development is a proof-of-concept showing that a single crystal can mimic a neuronal cell. The next step involves scaling these individual cells into larger, interconnected grids. For this to become a commercial reality, engineers must find ways to mass-produce these vdW crystals with high consistency, as atomic-level defects can alter the firing threshold of the neuron.

Industry analysts suggest that if these materials can be integrated into existing CMOS (complementary metal-oxide-semiconductor) fabrication processes, they could supplement traditional silicon chips. This would create hybrid systems where silicon handles logic and vdW crystals handle the heavy lifting of pattern recognition and learning.