ImageNet classification with deep convolutional neural networks
Alex Krizhevsky, Ilya Sutskever et al.
217
Citations
2
Influential Citations
APL Photonics
Venue
2019
Year
In the emerging Internet of things cyber-physical system-embedded society, big data analytics needs huge computing capability with better energy efficiency. Coming to the end of Moore’s law of the electronic integrated circuit and facing the throughput limitation in parallel processing governed by Amdahl’s law, there is a strong motivation behind exploring a novel frontier of data processing in post-Moore era. Optical fiber transmissions have been making a remarkable advance over the last three decades. A record aggregated transmission capacity of the wavelength division multiplexing system per a single-mode fiber has reached 115 Tbit/s over 240 km. It is time to turn our attention to data processing by photons from the data transport by photons. A photonic accelerator (PAXEL) is a special class of processor placed at the front end of a digital computer, which is optimized to perform a specific function but does so faster with less power consumption than an electronic general-purpose processor. It can process images or time-serial data either in an analog or digital fashion on a real-time basis. Having had maturing manufacturing technology of optoelectronic devices and a diverse array of computing architectures at hand, prototyping PAXEL becomes feasible by leveraging on, e.g., cutting-edge miniature and power-efficient nanostructured silicon photonic devices. In this article, first the bottleneck and the paradigm shift of digital computing are reviewed. Next, we review an array of PAXEL architectures and applications, including artificial neural networks, reservoir computing, pass-gate logic, decision making, and compressed sensing. We assess the potential advantages and challenges for each of these PAXEL approaches to highlight the scope for future work toward practical implementation.
This paper is significant because it addresses a critical bottleneck in modern computing: the end of Moore's law and the throughput limitations of parallel processing (Amdahl's law). As electronic integrated circuits approach fundamental physical limits, the need for alternative computing paradigms becomes urgent. The authors propose photonic accelerators (PAXEL) as a viable solution, leveraging the immense bandwidth and low power consumption of photonics for data processing tasks. This is particularly relevant for big data analytics in IoT and cyber-physical systems, where energy efficiency and real-time processing are paramount.
The paper's timing is crucial—it was published in 2019, when optical fiber transmission had already demonstrated 115 Tbit/s over 240 km, yet data processing remained largely electronic. By shifting focus from data transport to data processing using photons, the paper opens a new frontier. It reviews a diverse array of PAXEL architectures, from neural networks to reservoir computing, showing that the technology is maturing enough for prototyping. This makes the paper a valuable resource for researchers and engineers exploring post-Moore computing.
The paper does not present new experimental results but aggregates key performance metrics from prior work. Notably, it cites a record aggregated transmission capacity of 115 Tbit/s over 240 km for wavelength division multiplexing systems. For PAXEL, it claims that these accelerators can process images or time-serial data in real-time with better energy efficiency than electronic processors. The review assesses each architecture's potential advantages and challenges, but concrete metrics (e.g., speed, power consumption, accuracy) are not provided for individual PAXEL designs. The paper's main result is a feasibility argument: prototyping PAXEL is now possible due to advances in silicon photonics and computing architectures.
This paper has broad implications for the AI field. By proposing photonic accelerators as a front-end processor for digital computers, it offers a path to overcome the energy and speed limitations of electronic computing for data-intensive tasks like image recognition, time-series analysis, and decision making. The architectures reviewed—especially optical neural networks and reservoir computing—directly apply to machine learning workloads. The paper also highlights the potential for real-time processing, which is critical for autonomous systems and IoT. While challenges remain in integration and scalability, this work provides a foundational roadmap for hybrid photonic-electronic systems, potentially enabling a new generation of ultra-fast, low-power AI hardware.
Alex Krizhevsky, Ilya Sutskever et al.
Ashish Vaswani, Noam Shazeer et al.
Douglas M. Bates, Martin Mächler et al.
Diederik P. Kingma, Jimmy Ba