The increase of data throughput in today's information systems are placing stringent demands on communication bandwidths and processing needs. While there has been significant progress in the design of high-speed processing elements, high-performance interconnection network design has lagged. The primary bottleneck in today's metal-based interconnection is the very limited bandwidth, which results in limited communication speed.
Free-space optical interconnects offer the possibility of the high density transmission from chip-to-chip or board-to-board directly, without using any guided wave interconnect medium with its associated constraints of interconnection density. Several major challenges remain unsolved if free space optical interconnects are to be applied in high density interconnection systems. In addition to addressing the overall cost of the system, the power dissipation for free-space interconnects must be made sufficiently small. Furthermore, the physical size of the optical interconnects components must be minimized if free-space interconnects are to be useful in allowing greater performance for a given size of system.
To satisfy these requirements, we studied the minimization of the total power dissipation in the free-space interconnection module as function of bit error rate and bit rate. The transceiver considered in the analysis consists of VCSEL(vertical-cavity surface-emitting laser) of which threshold current is about 0.8 mA and the photodetector is PIN PD(photodiode). The transceiver circuits are designed by using 0.35-$\mum$ Si-CMOS technology provided by TSMC. The receiver power dissipation dominates compared to that of transmitter at all operating bit rates due to the high capacitance of PIN PD. Therefore the receiver power dissipation was minimized first and then the transmitter was done.
We found two important characteristics during the analysis. One is that when the bit error rate of the link and the receiver sensitivity are calculated, the optimal threshold method is not appropriate for the decision circuit having two decision levels. Thus, to calculate BER, instead of the above method, we used a new analysis method of considering the effects that the split thresholds degrade the BER of the link. The other one is that increasing the PD's aperture size facilitates aligning the link but decreases the bandwidth of the link and deteriorates the sensitivity of the receiver. Therefore, the PD's aperture size is a trade-off element between the alignment and the bandwidth of the link.
