Sustainability

Reduced rank photonic computing accelerator

Computing for AI tasks uses tremendous amounts of energy due to inefficient CMOS architecture. Photonic computing has been demonstrated for high throughput energy efficient computing. By proposing a network with smaller number of training parameters while maintaining the computational accuracy, this work improves the efficiency of such systems, meaning a smaller number of I/O controllers are required to set weights of a neural network and thus, scales favourably in terms of energy. The material used in this work is silicon, silicon oxide and phase change material AIST.

Recent Advances and Future Prospects for Memristive Materials, Devices, and Systems

This paper is a large-scale review of memristive materials, devices and systems. As such it looks into emerging and published work in this field, and reviews the current state of the art. Some of the issues discussed include materials as well as the energy requirements of devices, in addition to their functionality, particularly in computing systems. However, the key issues related to sustainability are not the focus of this work, and thus, any future work aiming to look specifically into the sustainability aspects of this area will need a greater depth of research and understanding.

Integrated optical memristors

This paper is a review of devices that function as optical memristors, devices that serve in optics the same function as memristors do in electronics. As such the paper reviews the field broadly, but does make mention to energy requirements for each technology. The considerations for materials-related sustainabaility are not considered in detail in this paper. The energy requirements will also require further investigation, as those calculations are heavily dependent on the final application and use case.

In-memory photonic dot-product engine with electrically programmable weight banks

Computing in the emerging artificial intelligence space requires more efficient accelerators. This work progresses this area by showing the potential to combine electrical programming and optical probing to efficiently use optical computing alongside electronic computing for such applications. This work also reports a lower power consumption than previous implementations, having an energy plus. The materials used are standard silicon, and some phase change materials are used - the long term sustainability of these needs to be taken into account in any sustainability assessments.

Scalable High-Precision Trimming of Photonic Resonances by Polymer Exposure to Energetic Beams

Our paper on polymer trimming of photonic ring resonators describes a new technique to permanently modify the modal characteristics of photonic waveguides. The work takes advantage of mature fabrication technologies including electron beam lithography and photolithography to locally expose a commercial photosensitive PMMA. This approach can bring photonic components to specification thereby mitigating the energy expenditure required to transiently modify the effective index of the waveguides. The materials used are silicon, silicon nitride, silicon oxide and Poly(methyl methacrylate).

Atomically thin optomemristive feedback neurons

The group’s been pursuing new areas of computing to create more efficient computing paradigms for emerging artificial intelligence hardware. This work is one step to explore 2D materials in a new optoelectronic platform. Any implementation has to take into account life cycle of the new materials used as well as the overall system energy requirements in future.