Fiber to the chip

Researcher : Radovanovic, S.

Supervisor(s) :

Project Duration : Februari 2000 - 2004

Project Description

In the last decades, the speed of microprocessors has been increasing exponentially with time and will continue to do so for at least another decade. However, the local computing power of the microprocessor alone does not determine the overall speed of a system. Equally important to the processor's bare computing power is the speed at which data can be distributed to and from the processor. That means that the speed of the data-input and -output channel must keep pace with the processor's computing power. For the future it is expected that these data-communication channels will become the speed-bottle-neck for the whole system. For this reason a research project in high-speed data communication is started. For short- and medium distance (centimeters up to hundreds of meters) the data communication channels are usually implemented as wired electrical connections. However, at high speeds major problems occur: poor impedance matching results in distorted signals, significant Electro-Magnetic noise is generated which degrades the performance in mixed-signal (analog/digital). The problems with electrical wired communication channels could be circumvented by replacing the high-speed wired channels by high-speed optical communication channels. The electronics for the long distance channels is typically realized with expensive exotic technologies: Gallium-Arsenide High-Electron-Mobility-Transistors (GaAs HEMT) and Indium-Phosphorus Hetero-Junction-Bipolar-Transistors (InP HBT). The maximum bit-rate for these systems is around 40Gb/s per channel. The first reason for adequacy of these expensive blocks is long distance links: the cost per length of the fiber is low. The second reason for the efficiency of this solution is that a large number of users share the links: the cost per user is low. For the medium and short distances however, as well as for the small number of users (fiber-to-the-home or fiber-to-the-desk) the optical receivers and transmitters should not be expensive. This project aims at increasing the speed and lowering the costs of the fully integrated optical receivers in standard CMOS technology. These receiver chips have integrated light-sensors and are, as such, cheap and do not have wire-speed limitations. The light shines directly on the CMOS! That means that microprocessors could be connected in networks by just using light. The result could then be a low-cost and high speed fully optical data communication system for distances ranging from chip-to-chip (cm range) up to LANs (up to hundreds of meters).