Publications by authors named "George Mourgias-Alexandris"
Photonics is among the most promising emerging technologies for providing fast and energy-efficient Deep Learning (DL) implementations. Despite their advantages, these photonic DL accelerators also come with certain important limitations. For example, the majority of existing photonic accelerators do not currently support many of the activation functions that are commonly used in DL, such as the ReLU activation function.
View Article and Find Full Text PDFArticle Synopsis
- The text discusses a shift from traditional data center architectures to disaggregated systems designed for better resource use and lower costs.
- It presents a new optical switch architecture that combines broadcast-and-select methods with optical buffering, achieving high performance at 10Gb/s with minimal power loss.
- Network simulations in a 256-node setup showed very low latency of 605 nanoseconds and an 80% throughput rate using optical buffers, highlighting the effectiveness of the proposed system.
Article Synopsis
- Researchers created the first all-optical Ternary-Content Addressable Memory (T-CAM) cell that can operate at speeds of 10Gb/s, featuring two integrated flip-flops and an optical XOR gate.
- The two flip-flops store the data bit and the ternary state 'X', while the XOR gate compares these values during operation.
- Experimental results show that the T-CAM cell works without errors at 10Gb/s for data writing and addressing, suggesting it could enhance all-optical memory architectures for advanced routing uses.