Recent developments in the field of artificial intelligence have become a significant computational burden for current electronic hardware. In order to keep up with the increasing advancements in deep learning research, novel approaches to computation are required to address the slowing progress in computing performance and efficiency of electronic hardware. We will present our experimental results on a general-purpose optical processor that takes advantage of time multiplexing in integrated photonics and can perform not only dot products but also real-time correlation detection on stochastic bitstreams. This approach to optical processing has a significant compute efficiency advantage for long-bit sequences.
Phase change materials, and other functional nanomaterials typically require energy to be applied to them in order to have tunable properties. Typically, they can be tuned either optically, or electrically. However, a fundamental issue is that the interaction size scales of optics and that of electronics are very different - electronics function efficiently (energy/speed-wise) when dimensions are smaller than the wavelengths of light; unfortunately these are smaller than the typical interaction length-scales for optics. This has meant that efficient electro-optical coupling between electronic and photonic switching has been challenging. In this talk, I will talk about recent work within our group of collaborators to integrate concepts from plasmonics to bridge this length-scale disparity in integrated photonics, and present our recent work in this area. Although applied to phase change materials, these concepts are broadly applicable to other functional materials.
Associative learning as a building block for machine learning network is a largely unexplored area. We present in this paper our results on the demonstration of an all optical associative learning element, realized on an integrated photonic platform using phase change materials combined with on-chip cascaded directional couplers. We implement the framework on our optical on-chip associative learning network, and experimentally demonstrate image classification on a publicly-accessible cat-dog dataset. The experimental implementation harnesses optical wavelength division-multiplexing, thus increasing the information channel capacity to process our machine learning task. Our unconventional approach to machine learning demonstrated experimentally on an optical platform could potentially open up new research possibilities in machine learning hardware architectures and algorithms.
The Ge2Sb2Te5 phase-change alloy (GST) is known for its dramatic complex refractive index (and electrical) contrast between its amorphous and crystalline phases. Switching between such phases is also non-volatile and can be achieved on the nanosecond timescale. The combination of GST with the widespread SiN integrated optical waveguide platform led to the proposal of the all-optical integrated phase-change memory, which exploits the interaction of the guided mode evanescent field with a thin layer of GST on the waveguide top surface. The relative simplicity of the architecture allows for its flexible application for data storage, logic gating, arithmetic and neuromorphic computing. Read operation relies on the transmitted signal optical attenuation, due to the GST extinction coefficient. Write/erase operations are performed via the same optical path, with a higher power ad-hoc pulsing scheme, which locally increases the temperature and triggers either the melt-quench process (write) or recrystallization (erase), encoding the information into the GST crystal fraction. Here we investigate the physical mechanisms involved in the write/erase and read processes via computational methods, with the view to explore novel architecture concepts that improve memory speed, energy efficiency and density. We show the achievements of the development of a 3D simulation framework, performing self-consistent calculations for wavepropagation, heat diffusion and phase-transition processes. We illustrate a viable memory optimization route, which adopts sub-wavelength plasmonic dimer nanoantenna structures to harvest the optical energy and maximize light-matter interaction. We calculate both a speed and energy efficiency improvement of around one order of magnitude, with respect to the conventional (non-plasmonic) device architecture.
Phase change materials are increasingly becoming important functional materials for applications in emerging integrated optics. Since the demonstration of a photonic phase change memory device in 2015, several new applications i this area have emerged ranging from lossless routing to on-chip photonics synapses. More recently the use of these materials in unconventional computing has seen an emerging interest, especially in the areas of optical abacuses and other forms of brain-inspired computing. There have also been advances in non-von Neumann approaches to carry out large-scale matrix multiplications. In this talk, I shall cover these topics and present a future view of these materials, not only in computation, but also in displays and holographic projections.
Black phosphorus stands out from the family of two-dimensional materials as a semiconductor with a direct, layer-dependent bandgap in energy corresponding to the spectral range from the visible to the mid-infrared (mid-IR), as well as many other attractive optoelectronic attributes. It is, therefore, a very promising material for various optoelectronic applications, particularly in the important mid-IR range. While mid-IR technology has been advancing rapidly, both photodetection and electro-optic modulation in the mid-IR rely on narrow-band compound semiconductors, which are difficult and expensive to integrate with the ubiquitous silicon photonics. For mid-IR photodetection, black phosphorus has been proven to be a viable alternative. Here, we demonstrate electro-optic modulation of mid-IR absorption in few-layer black phosphorus under field applied by an electrostatic gate. Our experimental and theoretical results find that, within the doping range obtainable in our samples, the quantum confined Franz-Keldysh effect is the dominant mechanism of electro-optic modulation. Spectroscopic study on samples with varying thickness reveals strong layer-dependence in the inter-band transition between different sub-bands. Our results show black phosphorus is a very promising material to realizing efficient mid-IR modulators.
The use of photonics in computing is a hot topic of interest, driven by the need for ever-increasing speed along with reduced power consumption. In existing computing architectures, photonic data storage would dramatically improve the performance by reducing latencies associated with electrical memories. At the same time, the rise of ‘big data’ and ‘deep learning’ is driving the quest for non-von Neumann and brain-inspired computing paradigms. To succeed in both aspects, we have demonstrated non-volatile multi-level photonic memory avoiding the von Neumann bottleneck in the existing computing paradigm and a photonic synapse resembling the biological synapses for brain-inspired computing using phase-change materials (Ge2Sb2Te5).
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.