计算机科学
航空航天
深度学习
加速度
分析
光子学
人工智能
电动汽车
认知
人机交互
超大规模集成
工程类
认知计算
模拟
量子
航空航天工程
系统工程
电气工程
嵌入式系统
计算机体系结构
航空
实时计算
大数据
电子工程
作者
Seshua Sai M,Adinath M,Joshittha K,Mohamed Shihabudeen,Sivaraman T,Ithihas Thiruvenkata Durai,Shamili E
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-02-05
卷期号:: 1-6
被引量:4
标识
DOI:10.1109/icasst68917.2026.11484247
摘要
This work proposes a Quantum Photonic Cognitive VLSI architecture that integrates ultrafast photonic tensor processing with multi-agent AI control for autonomous analytics in electric-vehicle (EV) and aerospace networks. The hardware core employs silicon photonic Mach–Zehnder interferometer arrays to perform optical matrix multiplications at light speed, dramatically surpassing transistor-based parallelism. A higher-level Agentic AI layer orchestrates multiple neural reasoning agents via reinforcement learning, enabling self-adaptive task allocation and continuous federated learning across vehicle and spacecraft subsystems[1] [2]. A sub-terahertz optical–RF transceiver fabric replaces conventional buses to maintain real-time model synchronization in a hypothetical 7G-edge network, while an adaptive energy manager balances laser power and electronics to ensure thermal stability. We fabricated a 5 nm photonic-CMOS prototype chip and demonstrated it in two environments: a live EV powertrain control twin and a rocket avionics simulator. Using tunable lasers and high-speed scopes, we achieved sub-millisecond inference at 1550 nm, with an 8.4× throughput boost and 3.8× latency reduction relative to a purely electronic accelerator. Notably, the platform sustained operation across –60 °C to +130 °C using on-chip photonic temperature control. The results show >99% decision accuracy under federated operation, even with 20% induced bandwidth loss. These outcomes establish a new benchmark for self-learning quantum-photonic VLSI, paving the way for post-silicon AI in future mobility and space networks.
科研通智能强力驱动
Strongly Powered by AbleSci AI