All Positions

Research
Energy

Development of photonic integrated circuits for neuromorphic secure accelerators in LNOI platforms

DC-11
École Centrale de Lyon and RMIT
Ecully (FR) and Melbourne (AU)

Host organizations

Position Description

Download and read the full position description before applying

Proposed Projects

Option 1

Development of photonic devices for energy-efficient secure neuromorphic accelerators in LNOI platforms

Context
The rising need to guarantee security at the edge has recently fostered a large amount of research and efforts into robust and lightweight hardware security approaches towards prompt integration in accelerators for edge computing and IoT. Differently from software security layers which are prone to various attacks targeting non-volatile digital memories storing secret information e.g., secret keys, hardware security layers and primitives such as physical unclonable functions can bypass this attack vector by generating secret information ad hoc once opportunely stimulated. Photonic technologies have a disruptive potential for neuromorphic computing as they can allow to process information at high speed with ultralow energy consumption and latency, while preserving similar fabrication costs to CMOS electronics.

Thesis goal
The goal of this thesis concerns the development of devices based on LNOI and III-V materials integrated onto LNOI dies using transfer-printing approaches for building energy-efficient neuromorphic architectures. Such architecture will combine novel and robust security layers leveraging the unique performance of LNOI platforms such as low losses (< 0.5 dB/cm), high-speed modulators, and embedded phodiodes and electronics (> 50 GHz). The student will be directly involved in the fabrication of such devices, leveraging the expertise of the Integrated Photonics and Applications Centre (InPAC), as well as their design. Non-linear effects of different kinds e.g., electro-optic or opto-electronic will be investigated to find optimal trade-offs for activation functions in the case of neuromorphic architectures and randomness character in the case of security layers (e.g., physical unclonable functions).

Option 2

Investigation of photonic architectures for energy-efficient secure neuromorphic accelerators in LNOI platforms using novel security protocols

Context
The rising need to guarantee security at the edge has recently fostered a large amount of research and efforts into robust and lightweight hardware security approaches towards prompt integration in accelerators for edge computing and IoT. Differently from software security layers which are prone to various attacks targeting non-volatile digital memories storing secret information e.g., secret keys, hardware security layers and primitives such as physical unclonable functions can bypass this attack vector by generating secret information ad hoc once opportunely stimulated. Photonic technologies have a disruptive potential for neuromorphic computing as they can allow to process information at high speed with ultralow energy consumption and latency, while preserving similar fabrication costs to CMOS electronics.

Thesis goal
The goal of this thesis concerns the system-level investigation of photonic integrated circuits (PICs) for energy-efficient and low-latency neuromorphic computing as well as for robust security layers. The PICs will be based on novel photonic and opto-electronic devices leveraging the thin-film LNOI platform present at RMIT as well transfer-printed high-speed III-V devices. Novel security protocols will be investigated to relax constraints such as the need for strong non-linear effects within the PICs to prevent from e.g., modeling attacks of the security layers. The student will also have the opportunity to carry out device fabrication if interested.

Option 3

Theoretical and numerical study of photonic devices and architectures for energy-efficient secure neuromorphic accelerators in LNOI platforms

Context
The rising need to guarantee security at the edge has recently fostered a large amount of research and efforts into robust and lightweight hardware security approaches towards prompt integration in accelerators for edge computing and IoT. Differently from software security layers which are prone to various attacks targeting non-volatile digital memories storing secret information e.g., secret keys, hardware security layers and primitives such as physical unclonable functions can bypass this attack vector by generating secret information ad hoc once opportunely stimulated. Photonic technologies have a disruptive potential for neuromorphic computing as they can allow to process information at high speed with ultralow energy consumption and latency, while preserving similar fabrication costs to CMOS electronics.

Thesis goal
The goal of this thesis concerns the theoretical and numerical investigation of photonic integrated circuits (PICs) for energy-efficient neuromorphic computing and robust security layers. Various studies concerning different architectures and computing paradigms for neuromorphic computing as well as complex and chaotic architectures for security primitives such as physical unclonable functions and true random number generators are foreseen. For this type of architectures, several trade-offs are present e.g., when considering the required system complexity versus security strength of the proposed solutions. The possibility of using a high-performance platform featuring low losses (< 0.5 dB/cm), high-speed modulators, and embedded phodiodes and electronics (> 50 GHz) will provide a perfect play field to explore exotic concepts and architectures for PICs from first principles. Finally, the proposed PICs will be realized at RMIT in a thin-film Lithium Niobate on Insulator (LNOI) platform, leveraging the expertise of the Integrated Photonics and Applications Centre (InPAC). The student will also have the opportunity to carry out device fabrication, if interested.

Supervisors

Research Areas

Applied science, photonics, nonlinear optics, hardware security