Optical Computing and the Role of Photorefractive Crystals
- PDF / 969,974 Bytes
- 6 Pages / 576 x 792 pts Page_size
- 51 Downloads / 158 Views
50
electrons, and only a limited number of crossing stripline tracks can be integrated over one another on integrated circuits (ICs). The speed of electronic interconnects is limited by the dispersion of the electromagnetic pulses within the stripline, and at high bit rates the transmission lines must be terminated in their characteristic impedances. When an electronic signal is to be sent, the capacitance of the transmission lines must be charged and the energy required increases with increasing fan-out. Recent ICs take more of the silicon chip area for the circuits to drive the lines than for the intended circuit itself. When electrical transmission lines pass various components on the chip, they are affected by parasitic impedances which are often difficult to predict and which depend on circuit layout. The result is that when clock pulses are fanned out and sent down several nominally identical lines,, the pulses do not arrive simultaneously; this is known as "clock-skew." The maximum practical operating speed of the most recent ICs is limited not by the speed of the transistors, but by the limits imposed by the interconnections. Optics clearly has a role to play in performing the interconnections within computers. Optical Computing Strategies Two complementary research directions have arisen from the requirement to implement optical interconnections. The two approaches are not in direct competition and so we may see both being developed in parallel. The first approach aims to remove the electronic striplines and their driving circuits from the chip surface and replace them by optical pathways. Usually lasers
or modulators are at the transmitting end and photodetectors and minimal receiving circuitry are at the receiving end. In one interesting scenario, the light beam is guided between the source and receiver by a combination of microlenses and holograms mounted either on or above the substrate. Under the current research directions, one could envision a two-dimensional array of surface emitting lasers or modulators being simply imaged by a lens to a receiving two-dimensional array of photodiodes to perform a bus-type parallel interconnect. In fact, more complicated interconnect patterns can be realized between the two arrays, and both "crossover" and "perfect shuffle" types have already been demonstrated,2'3 similar to the "butterfly" interconnects required to realize a fast Fourier transform (FFT). The microoptical and holographic components are realized in a surface relief form which is convenient to fabricate using standard lithographic techniques already in use for forming the chip itself. In addition, they can easily be copied by pressing into a polymer blank, analogous to the technique used to fabricate compact audio disks or gramophone records. There is also interest in realizing reconfigurable interconnects so that the circuit can be rewired between or during calculations. At present the simplest technique is to hardwire the interconnect patterns optically and to electrically switch the signals from
Data Loading...
