I. INTRODUCTION
According to 2005 International Technology Roadmap for Semiconductor presented at Semiconductor Industry Association, it is expected that the clock rate of upcoming multi-processorchip will be rapidly increased every year and will be over 10 GHz after 2010. To meet and prepare the forecast on the roadmap, chip makers and system companies including Intel and IBM have presented the necessity of the chip-to-chip optical interconnection as an alternative of copper interconnection [1], [2]. Beyond 10 GHz, copper interconnects on a printed circuit board (PCB) made of FR4 material, will be limited by the dielectric loss from the substrate material in bandwidth and by the crosstalk between each channel. The limitation of the copper interconnection caused the necessity of an optical printed circuit board (OPCB) that is hybridly laminated with optical waveguides and electrical copper interconnect. Thus, recent studies [3]–[10] for chip-to-chip optical interconnection have focused on the development of interconnection schemes and components based on the OPCBs which have both optical layers and electrical layers in one board. One of the key issues in the chip-to-chip optical interconnection based on the OPCB is to devise the optical components to link the light between the light source (or detector) and the optical layer in the OPCB. Most recent studies have suggested -ended mirror structured schemes such as “OptoBump” interface using -ended waveguides and microlenses in NTT [3], [4], “Terabus” using -mirror and microlenses in IBM [5], [6], -ended rods in ICU[7], [8], and fully embedded -waveguides with optical sources and detectors in Texas Univ. [9], [10]. However, we suggest a passively assemblable scheme using the -bent fiber. Compared to the -ended mirror structured schemes, the -bent optical interconnection scheme is advantageous for the 2-dimensional (2-D) optical interconnection with low coupling loss and low crosstalk. Schematic of the suggested bi-directional, 2-D, and -bent optical platform.