Perspective Chapter: Photonic AI Hardware – Emerging Optoelectronic Architectures for Neuromorphic and High-Speed Processing
Dongsu Kim, Byoung Ok Jun · IntechOpen eBooks · 2025
The growing complexity of artificial intelligence (AI) workloads has exposed the fundamental limitations of charge-based computing architectures. Conventional von Neumann systems suffer from latency, energy inefficiency, and bandwidth bottlenecks due to frequent data movement between memory and logic. While emerging electrical neuromorphic devices such as memristors and ferroelectric FETs alleviate some of these issues, they remain constrained by charge transport mechanisms and variability. Photonic hardware, by contrast, exploits the unique properties of light to achieve ultra-fast, low-latency, and high-bandwidth signal processing. This perspective chapter explores the emerging field of photonic AI hardware, with a particular focus on hybrid architectures that integrate photonics with ferroelectric materials. We discuss device-level concepts such as light-emitting synaptic devices, ferroelectric-tuned modulators, and 3D stacked LED-PD-TFT arrays, as well as system-level approaches including optical crossbars, reservoir computing, and in-memory learning. By combining the speed and parallelism of photonics with the nonvolatility and adaptability of ferroelectrics, these hybrid systems offer a promising pathway toward highly efficient and scalable neuromorphic accelerators. Key challenges such as fabrication compatibility, thermal crosstalk, and algorithm-hardware co-design are highlighted, alongside future opportunities in edge AI, secure computing, and biomedical interfaces. Ultimately, the convergence of light and matter into co-computing substrates has the potential to reshape the foundations of intelligent hardware, enabling machines that compute as fast as they sense and adapt as naturally as they learn.