Zero-Static-Power Reconfigurable Photonics Using Phase Change Metacoatings

Training and running large artificial intelligence (AI) and machine learning (ML) models as well as emerging large language models (LLM’s) and the push towards online learning, inference and, more broadly, artificial general intelligence (AGI) require staggering amounts of energy to train in years to come.<br/><br/>This is in large part due to current hardware architectures’ reliance on volatile transitions (for signal routing, switching and processing) such as carrier modulation and thermo-optic effects in silicon and its derivatives, transparent conducting oxides, as well as ferroelectrics. This creates an intense need for active thermal load management as volatile transitions require power to be maintained to the device at all times. This is a major obstacle to scaling up to increasingly dense chips and racks in hyperscale AI-centric data centres that would need ever more spacious rack layouts or energy-intensive cooling solutions to deal with the generated heat resulting from the volatile nature of all transitions used on the chip level. Going forward, zero-static power architectures are needed on the fundamental processing and memory component layers to enable scalable and environmentally sustainable systems.<br/><br/>The most scalable non-volatile transition available to us in nature is the phase transition commonly observed in chalcogenide semiconductors that relies on the reversible atomic rearrangement of an alloy using thermal, electrical, or optical stimuli between a reflective high conductivity crystalline and a transparent highly resistive amorphous phase. To this end, here we show that using phase change metacoatings, subwavelength thickness layers of chalcogenide phase change alloys nanostructured through bottom-up growth techniques can enable control of static optical properties as well as volatile and non-volatile transitions, with a view to photonic integrated circuits that will be the workhorse of emerging hyperscale AI-centric data centers. Notably, in these architectures, the mitigation of high insertion losses when introducing many of these alloys to the vicinity of waveguide platforms due to their inherent absorption across this band has resulted in devices with larger than necessary lateral footprints and/or poor modulation contrasts and an incessant search for low-loss/high switching contrast alloys.<br/><br/>Chalcogenide phase-change materials (PCMs) are high refractive index dielectrics across the visible/telecom band. We show that by making use of this high refractive index, mach-zender modulators (MZM’s) with built-in memory functionality and zero static power consumption, capable of 2π phase shifts, can be realized with active lateral footprints down to 10 μm. Furthermore, the patterning of these alloys on the subwavelength scale has demonstrated high-quality factor optically resonant devices under the umbrella of metamaterials and metasurfaces. In this realm, subwavelength nanostructuring can also offer non-resonant dispersion engineering of a given dielectric/plasmonic material. We show that non-resonant, lithography-free, subwavelength patterning, which enables dispersion engineering of chalcogenide glasses, paves the way to the realization of alloys with tunable static optoelectronic properties. We then go on to show that volatile and non-volatile optical transitions can be tuned and engineered in PCMs without the need for stoichiometric changes to chemical composition through glancing angle deposition and interlayer nanostructuring, enabling ultra-compact phase/intensity modulators with tunable insertion losses.