Light-Powered Artificial Neurons: Breakthrough Van Der Waals Crystals Mimic Biological Cells

Scientists have developed a light-controlled artificial neuron using a designable van der Waals crystal, a breakthrough that could accelerate neuromorphic computing by mimicking biological neural behavior with unprecedented precision. The advance, published in a peer-reviewed study on June 16, 2026, demonstrates how layered 2D materials can process information like a biological synapse—without the energy costs of traditional silicon-based systems.

The research, led by a team at Korea Advanced Institute of Science and Technology (KAIST), combines van der Waals heterostructures—stacks of atomically thin materials held together by weak intermolecular forces—with optical stimulation to replicate synaptic plasticity, the brain’s ability to strengthen or weaken connections between neurons. Unlike conventional neuromorphic chips, which rely on electrical signals, this approach uses light to trigger responses, potentially reducing power consumption by orders of magnitude in AI hardware.

Why it matters
Neuromorphic computing aims to replicate the brain’s efficiency by using artificial neurons and synapses. Current systems, such as Intel’s Loihi or IBM’s TrueNorth, achieve this with electrical signals but face limitations in scalability and energy use. The KAIST team’s method could overcome these barriers by leveraging photonic neuromorphic computing, where light pulses replace electrons as information carriers. According to the study, their van der Waals crystal achieved synaptic weight modulation—a key feature of learning—with 90% energy efficiency compared to silicon-based alternatives, as measured in lab tests.

The breakthrough builds on prior work in 2D materials for optoelectronics, including graphene and transition metal dichalcogenides (TMDs), which have shown promise in low-power photonics. However, previous attempts struggled with design flexibility—until now. The KAIST team’s crystal, composed of tungsten disulfide (WS₂) and molybdenum disulfide (MoS₂), allows researchers to tune its properties by adjusting layer thickness and composition, a level of control absent in bulk materials.

How it works
The artificial neuron operates by absorbing light at specific wavelengths, which alters the electrical resistance of the van der Waals layers—a process called photoresistance. When exposed to pulsed light, the crystal’s resistance changes dynamically, mimicking how a biological synapse strengthens or weakens in response to stimuli. The team demonstrated spike-timing-dependent plasticity (STDP), a learning rule critical for AI training, by exposing the crystal to paired light pulses with precise timing.

In benchmarks, the device matched the learning efficiency of biological synapses while consuming only 0.1 nanojoules per synaptic event, far below the 10–100 picojoules required by electrical neuromorphic chips. The study, published in Nature Electronics, notes that the technology could enable optical neural networks with terabit-scale throughput, a critical step toward brain-like AI.

Industry and competitive context
The development arrives as neuromorphic computing gains traction in edge AI and robotics. Companies like Sony, BrainChip, and SynSense are already commercializing neuromorphic chips, but their electrical designs limit scalability. The KAIST advance could shift the field toward photonic neuromorphics, where light-based systems leverage the ultra-low latency and high bandwidth of optics.

However, challenges remain. Manufacturing precision at the nanoscale is a hurdle, as van der Waals crystals require atomic-layer deposition techniques not yet standardized for mass production. Additionally, integrating optical neuromorphic chips with existing silicon infrastructure will demand new hybrid photonic-electronic interfaces, an area where startups like Lightmatter and Ayar Labs are already investing.

What happens next
The KAIST team is collaborating with Samsung Electronics’ Advanced Institute of Technology (SAIT) to explore large-scale fabrication of the crystals. If successful, the technology could reach commercial neuromorphic systems within 3–5 years, according to project lead Dr. Seongjun Park, a professor of electrical engineering at KAIST. Early applications may include low-power AI for IoT devices and neuromorphic accelerators for edge computing, where energy efficiency is paramount.

The study also opens doors for optical deep learning, where light-based artificial neurons could enable real-time, energy-efficient training of neural networks—a capability currently limited by the von Neumann bottleneck in traditional CPUs and GPUs. Regulators may also take note, as neuromorphic hardware could influence AI energy policies, particularly in data-center-heavy regions like the EU and California.

For developers, the breakthrough signals a shift toward materials-driven AI hardware. Researchers will need to master optical neuromorphic programming frameworks, distinct from today’s electrical-based tools like TensorFlow or PyTorch. Meanwhile, semiconductor firms may accelerate R&D in 2D material synthesis to stay competitive in the emerging photonic neuromorphic market.

Sources:

• Nature Electronics (2026) – "Designable van der Waals Crystal Realizes Artificial Neuronal Cell Mimicking with Light"
• KAIST Press Release (June 16, 2026) – "Optical Neuromorphic Computing Breakthrough"
• Samsung SAIT Collaboration Announcement (June 17, 2026) – Samsung Newsroom
• IEEE Spectrum (June 2026) – "The Race to Photonic Neuromorphics"