Tellurium‐Vacancy Engineering in Ultrathin Bi2Te3 Enables Broadband Multifunctional Optoelectronic Synapse for Energy‐Efficient Neuromorphic and Optical Information Processing

Why It Matters

The breakthrough shows that simple vacancy engineering can deliver ultra‑low‑power, high‑accuracy neuromorphic hardware, accelerating the deployment of edge AI and optical computing systems.

Key Takeaways

• Te vacancies enable persistent photoconductivity
• Energy consumption 37.2 fJ per spike
• 93.3% facial recognition accuracy
• 6 × 6 array acts as artificial retina
• Optical logic gates and Morse code demonstrated

Pulse Analysis

Defect engineering in two‑dimensional chalcogenides is reshaping neuromorphic hardware. By precisely controlling annealing temperatures, researchers introduced tellurium vacancies into Bi2Te3, turning these atomic‑scale imperfections into charge‑trapping sites that sustain photoconductivity. This approach sidesteps the complex multilayer architectures traditionally required for optoelectronic synapses, offering a scalable, wafer‑compatible pathway to integrate memory and sensing functions on a single ultrathin platform.

The resulting optoelectronic synapse delivers performance metrics that rival conventional silicon‑based AI accelerators while consuming orders of magnitude less energy. With 37.2 fJ per spike and a paired‑pulse facilitation of 191.7%, the device can emulate synaptic plasticity with unprecedented efficiency. Real‑world tests—facial‑recognition at 93.3% accuracy and urban‑traffic segmentation at 86.7% after 100 training epochs—demonstrate its readiness for vision‑centric edge applications. Moreover, the 6 × 6 array operates as an artificial retina, preserving learned visual patterns with a 57.4% retention rate, highlighting its potential for compact, on‑chip perception modules.

Beyond neuromorphic computing, the Bi2Te3 synapse bridges optical communication and logic processing. Successful implementation of optical logic gates and Morse‑code transmission illustrates a versatile platform for photonic signal routing without electronic conversion losses. As data centers and autonomous systems demand ever‑lower power footprints, such broadband, multifunctional optoelectronic devices could become foundational building blocks for energy‑efficient AI, smart sensors, and next‑generation optical networks. The convergence of materials science and neuromorphic design thus paves the way for truly integrated, low‑power intelligent systems.