CdS‐GeSe Heterostructure‐Based Bimodal Memristor with Tunable Synaptic Plasticity and Mixed‐Signal Switching for Neuromorphic Hardware

Why It Matters

The bimodal, optically tunable memristor bridges electrical and photonic domains, offering energy‑efficient, high‑speed synaptic functions essential for next‑generation AI accelerators and edge‑vision devices.

Key Takeaways

• Bimodal memristor switches analog below 0.8 V, digital above
• Optical 405 nm pulses boost conductance, tune synaptic plasticity
• Achieves 95% MNIST accuracy in optoelectrical mode
• High memory window ensures stable multi‑cycle retention
• Enables mixed‑signal neuromorphic computing for vision tasks

Pulse Analysis

The rapid growth of artificial intelligence has intensified demand for hardware that mimics brain‑like processing while consuming minimal power. Traditional silicon‑based memory struggles to deliver the analog weight updates required for efficient neuromorphic algorithms. Optoelectronic memristors, which combine electrical resistance modulation with light sensitivity, emerge as a promising solution, enabling direct integration of sensory inputs and synaptic computation on a single device.

The Ag/CdS‑GeSe/FTO structure leverages a layered heterojunction to achieve dual‑mode operation. Under low bias (<0.8 V), the device exhibits gradual conductance changes suitable for synaptic learning, whereas higher bias triggers abrupt switching for binary memory storage. Introducing 405 nm optical pulses enhances charge trapping, effectively tuning the plasticity and expanding the conductance window. Performance metrics reveal a robust memory window, low variability across cycles, and non‑linear conductance updates that align with the requirements of deep‑learning inference.

Beyond laboratory benchmarks, the memristor’s mixed‑signal capability translates into tangible advantages for machine‑learning workloads. In image‑recognition tests, the optoelectrical mode outperformed pure electrical operation, reaching 95% accuracy on the MNIST digit set—a clear indicator of its potential in edge‑vision and autonomous systems. As industry moves toward integrated photonic‑electronic processors, such tunable memristors could reduce data movement, lower latency, and cut energy consumption, accelerating the deployment of scalable neuromorphic hardware for real‑time analytics.