Technologies for Optically Interconnected CPU
In the CMOS process, there are 4 materials, silicon nitride (Si3N4), silicon oxynitride (SiOxNy), silicon dioxide (SiO2), and silicon (Si) that can be used as an optical waveguide. If silicon is not used, then the waveguide should be made with the remaining 3 materials. Among the methods that can inject light into those optical waveguides, the only possible method is to use a prism. It is demonstrated that an optical network chip can be realized by our three technologies and the cavity-type waveguide made out of silicon nitride and silicon dioxide.
The average propagation loss transmitted through the cavity-type waveguide, which was designed to minimize roughness on the waveguide side walls, was measured 0.258 dB/cm. Except for the reflection loss at the prism, the net coupling loss from VCSEL to the waveguide was measured 0.855 dB. The packaged size of VCSEL or photodiode has an area of 0.4 mm2 and a height of 0.64 mm, which can be attached to the prism. The area for optical devices to be integrated on a 8 core CPU chip is much smaller, <30 mm2, than the available area, 174 mm2.
The prism used for injecting light into the waveguide can also be used as the optimal WDM filter in which two thin films with different refractive indices are alternately stacked on the bottom side. Theoretical calculations have proved the performance of a WDM filter. If anti-reflection films were coated on the two surfaces of the prism with materials, TiO2 and Ta2O5, then the reflection loss at the prism was calculated to decrease from 2.75 dB to 1.11 dB for polyimide adhesive and from 2.73 dB to 0.52 dB for optical glue.
As compared with the electrical mesh network, an optical network which interconnects 4 x 8 core CPUs optically saves the power consumption from 30 W to 0.25 W and the latency from 5120 cycles to 184 cycles. The manufacturing cost can be reduced to, approximately, 1/3 of the current 28 core single chip price.We propose a 2.5D package of CPU with the interposer chip replaced by an optical network chip and another type of optically interconnected CPU based on straight waveguides. Our three technologies and the cavity-type waveguide demonstrated in section II must be sufficient solutions to implement optically interconnected CPU and high speed devices to computers. The remaining work is to set up robotic equipment to attach the LDs and PDs to the prisms automatically and for mass production.