Multimode Optical Interconnects on Silicon Interposer Enable Confidential Hardware-to-Hardware Communication.

Area of Science:

• Optoelectronics
• Information Security
• Materials Science

Background:

• Moore's Law drives increased integrated circuit density, necessitating advanced interconnects.
• Multimode optical interconnects on silicon interposers offer high throughput for compact 3D chip networks.
• Confidential communication is crucial for high-performance 5G infrastructure.

Purpose of the Study:

• To investigate the feasibility of physical layer security for multimode optical interconnects.
• To explore the uniqueness of optical channels for secure data transmission.
• To develop a method for enabling confidential communication in 3D chip networks.

Main Methods:

• Experimentally projecting orthogonal and non-orthogonal symbols through 380 μm multimode on-chip interconnects using wavefront shaping.
• Analyzing the uniqueness of these interconnects across multiple channels and samples.
• Training a deep neural network to recognize and decode speckle patterns caused by modal crosstalk.

Main Results:

• Speckle patterns generated by modal crosstalk in multimode interconnects were successfully recognized.
• A deep neural network effectively transformed these unique speckle patterns into readable output.
• The study confirmed the feasibility of applying physical layer security to these optical interconnects.

Conclusions:

• Physical layer security is a viable approach for enabling confidential optical communication in 3D chip networks.
• Deep neural networks can effectively decode unique channel characteristics for secure data transmission.
• This research paves the way for secure, high-throughput optical interconnects in advanced computing architectures.