SinoGreenTech Academic Portal
Official PDF TranslationSCIENCE CHINA Materials

Machine Learning and High-Throughput Computation-Assisted Precise Synthesis of Quantum Dots for Reliable Neuromorphic Computing

Authors: WANG Zhiqing; CHEN Keqiang; WANG Qiao; YANG Jing; QIN Zhi; HU Yang; SHEN Jie; ZHANG Pengchao; ZHOU Jing; CHEN Wen

DOI: 10.1007/s40843-025-3507-9Status: Verified Translated Edition
Sponsored AdvertisementAd Placement Area
reCAPTCHA Bot Shield Active

Preparing Secure Academic Download

Verifying human reader & generating high-resolution document...

Verifying Document Integrity15s remaining
← Back to Article
Protected by Google reCAPTCHA v3.PrivacyTerms
Sponsored ContentAdSense In-Feed Ad Slot

Key Findings in This Report

• • 57% reduction in switching voltage: QD-based memristors achieved a 57% lower switching voltage compared to baseline devices, directly reducing power consumption in neuromorphic circuits. This matters industrially because lower operating voltages enable dense, energy-efficient compute-in-memory architectures, critical for edge AI deployment where thermal budgets are constrained. • • Two-order-of-magnitude increase in ON/OFF ratio: The ON/OFF ratio improved by 100×, enhancing signal margin for reliable state discrimination. This is essential for multi-bit storage and analog weight updates in neural networks, where high ratio ensures low bit-error rates during inference and training. • • Endurance and retention degradation as low as 0.1% over 8.4 × 10^7 s and 10^5 cycles: The devices exhibited exceptional stability, with only 0.1% degradation after 8.4 × 10^7 seconds (≈2.7 years) of continuous operation and 10^5 read cycles. This addresses the reliability bottleneck that has stalled commercialization of QD memristors in safety-critical applications such as autonomous systems and medical implants. • • 477% improvement in dynamic learning range and 27.8% accuracy gain (92.23%): The dynamic learning range expanded by 477%, and neuromorphic computing accuracy reached 92.23%, a 27.8% improvement. This directly translates to higher fidelity in pattern recognition tasks, enabling practical deployment in real-time edge computing where accuracy and adaptability are paramount.