Two-dimensional fiber array with integrated topology for short-distance optical interconnections Makoto Naruse 1),2), Alvaro Cassinelli 3), and Masatoshi Ishikawa 3) 1: Ultrafast Photonic Network Group Communications Research Laboratory, Japan 2: Japan Science and Technology Corporation (JST), PRESTO 3: Dept. Information Physics and Computing, University of Tokyo

Contents 1.Interconnection fabric 2.Wave-guide-base, direct implementation of interconnection topology 3.Interconnection decomposition 4.Experimental fabrication 5.Summary and future plans

Optical Interconnection fabric / switching fabric LSI Optical Interconnection fabric / Switching fabric Inter Chip, Inter-board Optical interconnection Multistage architecture

… … All optical Optoelectronic An example: Omega network OE Computation EO Optical interconnect w/o OEO Regularly interconnected multistage architecture

Wave-guide-base, direct implementation of interconnection topology … … OE Computation EO Optical interconnect w/o OEO Two-dimensional fiber array Configure the interconnection topology directly by positioning the input and output end of the wave- guides Input Output All optical Optoelectronic

Design considerations Two-dimensional (2D) parallelism Focus on Permutation network (such as perfect shuffle) Scalability Module reusability (Permutation reusability) Alignment difficulty: Both input and output end Theoretically more volume efficient than free-space equivalent Other remarks Out of scope of this paper Y.Li, et. al., “Volume-consumption comparisons of free-space and guided-wave optical interconnections”, Appl.Opt. 39 (2000), 1815

Example1: Omega network Messy topology Poor scalability Poor reusability Permutation= Perfect shuffle 2D direct implementation

Example 2:Indirect Binary n-Cube Network Permutation= Butterfly and perfect shuffle Several kinds of different interconnection topology are used

Interesting fact Perfect shuffle and butterfly permutation can be made out of the following three types of elemental permutations: Row, Column, and Diagonal permutations Column permutationRow permutation Diagonal permutation

Before decomposition Direct implementation Node assignment: Scan mapping Perfect shuffle

Interconnection decomposition Row permutation Column permutation Diagonal permutation Decompose Perfect shuffle

Interconnection decomposition Column permutation Diagonal permutation Decompose Butterfly

shuffle Processor arrays (exchange switches and more) Row permutation  90 º Overall Omega Network Column permutation Diagonal permutation

 (2)  (3)  (4)  -1 (4) Processor arrays (exchange switches and more) Row permutation  90 º Column permutation Diagonal permutation Overall Indirect Binary n-Cube Network

Two holder prototypes: Zirconium, SiO 2 Pitch: 250±5  m Multimode graded index fibers: NA=0.21 (core 50  m, cladding 126  m) Transmission loss: 3dB/km Length: 30 cm Prototype fiber module: Preliminary 4x4 array 3 mm 2 mm 5 mm Embedded interconnection topology

Pitch uniformity Zirconium Pitch (  m) Number of link 245  m-255  m Ave. 250  m Std deviation 2.0  m 246  m-254  m Ave. 250  m Std deviation 1.5  m SiO 2 Pitch (250  m)

Input Output (CCD image) No relay optics Interconnection example VCSEL array Fiber module input Input Output

 x  50  m X (microns) Exit power (a.u) xx Alignment tolerances (half peak power)  y  70  m Transmission efficiency / Alignment tolerance Transmission efficiency Max. transmittance 38.45% VCSEL driving current (mA) Transmittance (%) LED regime LASER regime

Summary and future plans Wave-guide-base, direct implementation of 2D parallel interconnection topology Interconnection decomposition for scalability and reusability 2D fiber array with interconnection topology was demonstrated Future plan: Theoretical foundation for interconnection decomposition and total system design Higher-density 2D interconnect System demonstration