Abstract | ||
---|---|---|
Instead of a single die per chip (package) multiple dies stacked vertically (3D) and placed on an interposer (2.5D), is emerging as the building block for the future in both mobile and high-performance applications. We identify that bandwidth, energy per bit, and the ability to support a large amount of memory, are the key requirements of the inter-die Network on Chip (NoC). We propose to use an interposer with optical interconnections exploiting AWGR (Arrayed Waveguide Grating Router) wavelength routing to realize a 16x16 photonic NoC with a bisection bandwidth of 16 Tb/s. We propose a baseline network, which consumes 2.81 pJ/bit assuming 100% utilization. We show that the power is dominated by the electro-optical interface of the transmitter, which can be reduced by a more aggressive design that improves the energy per bit to 0.437 pJ/bit at 100% utilization. The networks exhibit very low latency as they are based on an optical crossbar topology and are scalable to multiple chips to support 1 TB of memory or more to meet the requirements of Exascale computing. |
Year | DOI | Venue |
---|---|---|
2016 | 10.1145/2989081.2989111 | MEMSYS |
DocType | Citations | PageRank |
Conference | 2 | 0.37 |
References | Authors | |
21 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Paolo Grani | 1 | 3 | 0.73 |
Roberto Proietti | 2 | 29 | 4.60 |
Venkatesh Akella | 3 | 357 | 33.69 |
S. J. Ben Yoo | 4 | 172 | 24.55 |