Back to search

Aristotle University Of Thessaloniki (AUTH)

Integrated optical memory circuits for signal processing and routing

Abstract

dc:description

The explosive growth of high-performance computing systems and high-speed communication networks has intensified the limitations of conventional electronic technologies, especially during recent years. Particularly, the continuous expansion of data volumes has led to an escalating demand for routing and address look-up operations, introducing critical performance and latency bottlenecks in modern network infrastructures and DCs. At the same time AI-driven workloads have significantly skyrocketed with LLM sizes growing at an enormous rate, requiring unprecedented computational resources. As a result, the growing number of processing elements necessitates efficient interconnection mechanisms capable of supporting high-throughput data transfer and processing, while simultaneously maintaining energy efficient characteristics. Despite the increased computing needs, the developed memory systems have indicated a limited growth rate compared to the respective processor units, with this phenomenon being known as “Memory Wall” problem. This disparity has emerged as a fundamental limitation, significantly constraining the performance of conventional electronic architectures in both computing and communication tasks. In this context, emerging photonic technologies have been identified as a promising candidate, offering ultra-high bandwidth in the order of Tb/s, reduced energy consumption and low-latency credentials, as well as the potential for inherent parallelism beyond electronic capabilities. Towards this context, photonic integrated circuits implementing especially optical memory layouts, such as RAMs and CAMs, are attracting significant attention as suitable enablers of next-generation computing and communication platforms. The current thesis investigates the design, experimental implementation, and scalability of photonic integrated memory architectures, with particular emphasis on all-optical RAM and CAM systems, for signal processing and routing of information in the optical domain. The main objective is to address and overcome the fundamental limitations and challenges of electronic memory and routing technologies, by leveraging the increased bandwidth, parallelism and energy efficient characteristics offered by modern photonic devices. To address the different challenges, the thesis initially proposes novel photonic architectures based on active Indium Phosphide platforms, enabling high-speed and fully optical memory operations. A bistable SOA-based memory cell is demonstrated, forming the basis for integrated optical RAM/cache and CAM systems. The proposed photonic memory cell has been initially characterised in terms of static and dynamic operations, allowing also for a full memory operation, at 5 Gb/s. The memory functionality is effectively validated through error-rate measurements, allowing for error values of 10-9 with a power penalty of 4.6 dB. The proposed device is additionally utilized in a CAM cell implementation where the SOA-based flip-flop cell is combined with an SOA-MZI access gate, offering high-speed CAM operations, experimentally validating its comparison operation at 5 Gb/s across different content configurations, and achieving error rate values of 10-9 with power penalties of 8 dB. Building upon this, the thesis addresses the scope of complete high-throughput, energy-efficient and all-optical memory architectures, by proposing a layout for a 16-bit RAM bank, as well as a complete cache prototype, that exploits WDM principles for memory addressing, and passive circuits for row and column decoding, while the aforementioned SOA-based devices offer storage and data access functionalities. Moreover, a complete optical 2-bit address look-up architecture is introduced, combining the respective optical active elements, allowing CAM comparison and RAM-based output selection and forwarding, achieving high-speed operations at 10 Gb/s. The specific architecture has also been validated through error-rate measurements, achieving error-free operations (10-9) with only a 9 dB power penalty. Furthermore, advanced silicon photonic approaches are explored in the present thesis, by validating novel linear operator crossbar architectures for similarity search and in-memory computing, enabling scalable and high-throughput CAM/TCAM circuits. A 4×4 silicon photonic device that utilizes EAMs for encoding the search and stored information, is demonstrated for enabling analog 2-bit CAM/TCAM computations, while due to the inherent capability of the proposed crossbar device to identify the level of mismatch between the compared contents, the similarity metric of Hamming distance can also be evaluated, at high-speed rates of up to 50 Gb/s, showing efficient error rate values at the order of 10-3. At the same time, a respective WDM‑enabled TCAM cell is also proposed, allowing for extended comparison availability and energy efficiency, utilizing wavelength division multiplexing principles. The WDM-based cell is also experimentally validated at data rates of up to 50 Gb/s,indicating efficient performance that is validated through Q factor measurements. Additional analysis signifies that the proposed schemes can efficiently scale towards higher dimensions, retaining the necessary signal integrity and non-prohibitive energy requirements, for precise comparison and vector‑distance calculations. In parallel, a novel large-scale 4×9 TCAM UV-programmable photonic crossbar architecture is presented, enabling non-volatile and energy-efficient in-memory computing functionalities, along with an increased scale layout that incorporates 36 MZI elements. The proposed design uses WDM and linear optical processing principles to realize scalable TCAM operations at 50 Gb/s with error rates that range between 10-9 and 10-3, while maintaining zero static power consumption for defining its optical states. The proposed silicon photonic approaches demonstrate high-speed operation credentials and highlight a promising pathway toward larger programmable and CMOS-compatible photonic computing systems. Overall, the presented thesis advances photonic integrated memory and computing systems, highlighting the convergence of photonic memory, search, and computation within a unified in-memory computing hardware framework for next generation high-speed and energy-efficient optical computing infrastructure, capable of overcoming key bottlenecks of electronic systems. Furthermore, the proposed research findings address also the constraints of current photonic memory implementations, such as limited capacity, large footprint, and energy inefficiency, alongside the challenges associated with photonic integration, and incomplete CMOS compatibility.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Moschos, Theodoros
  • Μόσχος, Θεόδωρος

Subjects

dc:subject × 12

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62341

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Moschos, Theodoros; Μόσχος, Θεόδωρος. Integrated optical memory circuits for signal processing and routing. Aristotle University Of Thessaloniki (AUTH), 2026. http://hdl.handle.net/10442/hedi/62341