This PDF file contains the front matter associated with SPIE Proceedings Volume 13113, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

In recent years, standard CMOS microprocessors have approached their maximum power dissipation per unit area, effectively placing a limit on computational power. This highlights the urgent need to explore alternative technologies. One promising avenue is the use of superconductors, which demonstrate zero resistivity below a critical temperature. However, circuits based on superconductors necessitate the use of cryostats to maintain low temperatures, presenting challenges in data transfer with the room temperature environment. While coaxial cables are often employed for this purpose, they suffer from limited data transfer rates and contribute significantly to heat load. On the contrary, photonics integrated circuits (PICs) coupled with optical fibers present a viable solution. They enable scalable, cost-effective, and power-efficient optical interconnections capable of supporting high data transfer rates while minimizing heat transfer. In this presentation, We will discuss the latest advancements in cryogenic PICs, focusing on their application in interfacing with cryogenic computing systems such as single-flux-quantum logic circuits and superconducting qubits.

Recent advancements in optical communications have explored the use of spatially structured beams, especially orbital angular momentum (OAM) beams, to achieve high-capacity data transmission. Traditional electronic convolutional neural networks (CNNs), while effective, face significant challenges in demultiplexing OAM beams efficiently, notably their high power consumption and substantial computational time, which can limit realtime processing capabilities in high-speed optical communication systems. In this study, we propose a hybrid optical-electronic CNN that integrates Fourier optics convolution for intensity recognition-based demultiplexing of multiplexed OAM beams under simulated atmospheric turbulence. Experimental results showed that the proposed hybrid neural network system achieves a 69% demultiplexing accuracy under strong turbulence conditions while exhibiting a three times reduction in training time compared to all-electronic CNNs. This study underscores the potential of a hybrid optical-electronic neural network to enhance both performance and efficiency in OAM-based optical communication systems.

Here, we introduce an optical computing method using free-space optics and a 4f system to enhance and integrate data processing, encryption, and machine learning. We propose a Reconfigurable Complex Convolution Module (RCCM) which enables simultaneous amplitude and phase modulation of optical signals for complex convolution operations in the Fourier domain. Utilizing spatial light modulators and interferometric techniques based on the Michelson interferometer, the RCCM achieves precise control over light properties. The system demonstrates promising applications in optical hashing, data compression, and accelerating machine learning tasks, particularly for processing encrypted data. Experimental results show the RCCM’s ability to perform complex convolutions with high accuracy, though trade-offs between compression ratios and classification accuracy are observed. This research represents a significant advancement in optical computing, addressing challenges in data security, processing speed, and computational efficiency across various fields.

Integration of chalcogenide phase-change material (PCM) with photonic circuits offers a practical route of introducing nonvolatile optical memories—a key enabling capability for optical neuromorphic computing. The prospect has motivated a surge of research efforts in this field and significant improvements in the performance of PCM-based photonic devices. These advances prompt an important question: what are the ultimate performances that can be achieved in PCM-based photonic devices? Here we address the question by quantitatively analyzing performance bounds of PCM memories on optical loss, crosstalk, energy consumption, and multi-level operation.

Developing energy-efficient optical components for computing is crucial for AI-driven hardware technologies. While previous studies primarily focused on optimizing cache-level design and managing write-intensive memory addresses, the impact of clock frequency on the energy consumption of emerging memory technologies, such as PCM, remains underexplored. In this work, through comprehensive simulation-based analysis, we reveal the complex relationship between clock frequency and the energy efficiency of OPCM, SRAM, and DRAM. The proposed memory architecture has demonstrated the potential to reduce overall energy consumption by up to 75% for the MiBench benchmark suite, a widely used set of embedded systems and IoT workloads. This work contributes to the ongoing efforts to improve the energy efficiency of optical computing systems, a critical factor in realizing the full potential of these emerging technologies.

Fast, sensitive, and compact photonic nonlinear activation devices are crucial for forming photonic neural networks with high computational efficiency. However, conventional photonic activation units lack suitable nonlinearity due to weak optical nonlinearities. Using intersubband nanostructures, we introduce the smallest possible optoelectronic approach for nonlinear processing in photonic neural networks. Here we also demonstrate the simulation and fabrication of the intersubband nanostructures. The simulation results shows a 3.5 microwatt threshold and an optical response time of 10 picoseconds within a device that is only 4 microns in length.

Inverse design has become a powerful tool in photonics for creating compact, high-performance devices. However, its application has been mostly limited to linear systems, with minimal exploration in nonlinear regimes. Additionally, the ”black box” nature of these techniques limits understanding of the optimized structures. Here, We introduce an inverse design approach to enhance on-chip photon pair generation efficiency using the opensource package EMopt. Our method employs a multi-frequency co-optimization strategy and calculates gradients via the adjoint method. Efficiency improvements result from field intensification in high-quality factor triply-resonant cavity resonances and better phase-matching conditions. We validate our method by fabricating and characterizing an optimized device that efficiently generates photon pairs, adhering to fabrication constraints and suitable for scalable quantum light sources in large-scale computing and communication applications. The design’s shape can be explained using effective potential analysis. This optimization technique can extend to other nonlinear processes for compact on-chip frequency-mixing devices.

The Vernier effect has seen extensive application in optical structures, serving to augment the free spectral range (FSR). A substantial FSR is vital in a myriad of applications including multiplexers, enabling a broad, clear band comparable to the C-band to accommodate a maximum number of channels. Nevertheless, a large FSR often conflicts with bending loss, as it necessitates a smaller resonator radius, thus increasing the insertion loss in the bending portion. To facilitate FSR expansion without amplifying bending loss, we employed cascaded and parallel racetrack resonators and ring resonators of varying radius that demonstrate the Vernier effect. In this study, we designed, fabricated, and tested multiple types of racetrack resonators to validate the Vernier effect and its FSR extension capabilities. Our investigations substantiate that the Vernier effect, based on cascaded and series-coupled micro-ring resonator (MRR) sensors, can efficiently mitigate intra-channel cross-talk at higher data rates. This is achieved by providing larger input-to-through suppression, thus paving the way for future applications.

Matrix multiplication with weights is a fundamental operation in deep neural networks, crucial for tasks reliant on data-driven modeling. However, computational limitations often hinder these operations. To address this, we propose photonics neural networks, offering faster processing speeds compared to electronic systems. Our paper aims to develop two distinct photonic neural networks. We use micror-ing resonators as neurons, exploiting their varied transmission properties for weight updates. Our experiments achieved a maximum transmission drop of approximately 0.7 dBm and a maximum spectral shift of around 1 nm per voltage change. Furthermore, we propose and explore two methods for implementing summation and non-linear activation using photonics devices with various materials, investigating the design constraints associated with each approach. Our experimental results corroborate our simulation predictions, validating the feasibility of our designs.

Neural networks, a potent data-driven method for artificial intelligence, play a crucial role in modern estimation and classification tasks. However, they often encounter computational constraints. To overcome these, we propose photonics-based neuromorphic networks, offering faster processing than electronic systems. We focus on the weight bank, crucial for matrix multiplication, utilizing parallel cascaded micro-ring resonators (MRRs). Our study on silicon on insulators (SOI) demonstrates how cascaded MRRs address cross-talk issues in wavelength division multiplexing (WDM) systems. Additionally, we design a silicon photonic accelerator for weight addition, optimized for speed and energy efficiency, providing comparable performance to electronic devices.