Integrated photonics new platform for implementation neural networks on a chip
Lukas Puts defended his PhD thesis at the Department of Electrical Engineering on March 11th.
Since the second half of the 20th century, scientists and engineers have been captivated by the idea of a machine that simulated human thinking and decision making. Since the early 2000s, artificial intelligence has re-emerged and continues to experience rapid growth. This, however, comes at a cost. An artificial neural network, which is the technology behind these applications, is inspired by the human brain. Implementing such networks in software and on computers requires a significant amount of energy, primarily because conventional computer processors are not well-optimized to perform neural network computations. In his research, Lukas Puts investigated the suitability to use integrated photonics as a new platform to implement neural networks on a chip, with the aim to contribute to the development of an ultra-fast and energy-efficient photonic chip for artificial intelligence computations.
In many aspects of our daily lives, artificial intelligence now plays a vital role. The application of artificial intelligence is vast, and stretches from self-driving cars, virtual assistants, computer vision, to text and image generation tools. The technology behind these advances improves at an unprecedented pace. With the development of more accurate algorithms and models, artificial intelligence is gaining a pivotal role in society. Where in the past artificial intelligence was regarded as a ‘hype’, it is safe to say that artificial intelligence is now a key technology and applied in many field of science, engineering, and in society.
Ultra-fast and energy efficient
To minimize energy consumption and increase the speed of computations, new types of networks and technologies are currently being investigated. One of the approaches is using photonics as a promising technology for certain types of neural computation. This may be done by using microscopically-small lasers on a photonic chip. Remarkably, lasers are able to mimic similar behavior as human neurons, but at a much faster speed and very low energy consumption. Thus, using lasers in a neural network on a chip may result in an ultra-fast and energy efficient neural network.
Extensive modeling
The Eindhoven area, and especially Eindhoven University of Technology, is renowned for their cutting-edge technology and research on integrated photonics. During his thesis work, has used and built upon this ecosystem by performing extensive modeling of a particular type of laser on a photonic chip. The modeling resulted in an improved laser design, which was fabricated in Eindhoven. Measurements demonstrated indeed similar properties as observed in biological neurons, pointing towards a feasible solution to implement such lasers in a neural networks on a photonic chip.
Further investigations
Lukas’ thesis work has opened the route to further investigations and improvements on laser designs and its applications in a photonic neural network. His work has strengthened the field of photonics for artificial intelligence in Eindhoven and will serve as a foundation for other PhD projects.
Funding: NWO Veni 17269 Light up the Brain NWO Zwaartekracht: Research Center for Integrated Nanophotonics.
Title of PhD-thesis: . Supervisors: prof. Daan Lenstra, prof. Kevin Williams, and prof. dr. Weiming Yao.