Abstract

Optical technology has been developed for highly effective transport of information, either as very high speed temporal streams, e.g., in optical fibers or in free-space, or as in high-frame-rate two-dimensional (2D) image displays. There is, therefore, interest in performing routing, signal-processing and computing functions directly on such optical data streams. The development of various optical modulation, display, and storage techniques allows the investigation of processing concepts. The attraction of optical processing techniques is the promise for parallel routing and processing of data in the multiple dimensions of space, time, and wavelength at possible optical data rates. For example, in the temporal domain a 1 nm wide optical band at a wavelength of 1500 nm has a bandwidth of approximately 100 GHz, and temporal light modulators with 100 GHz bandwidth have also been demonstrated for optical fiber systems [1,2]. However, notional optical processing techniques can be envisioned that handle many such narrow-wavelength bands in parallel, and also operate in a combined spatiotemporal domain. Employing all domains simultaneously, it is theoretically possible to perform spatiotemporal routing and processing at an enormously high throughput in the four dimensions x, y, t, and λ. Throughput of 10¹⁴ samples/s would result from simple examples based on feasible modulation and display capabilities. In one case a 100 GHz temporal modulators can be combined with wavelength-selective devices to provide several hundred 1 nm wide channels at the operating wavelengths of existing photodetectors and light sources. A second example would consider that 2D spatial light modulator (SLM) devices can be constructed to have >10⁷ pixels/frame (see Chapter 6) and that material developments allow optical frame update rates on the order of 1 MHz [3]. Unfortunately, although an optical processor operates on data in optical form, it is presently not possible to equate these maximal modulation and display rates to the expected information-processing throughput rates of such processors. There are penalties on the throughput due to necessary data pre-processing and post-processing in any information-processing system. These include the need to format and condition the input data to a processor, to compensate for shortcomings of any analogue signals (e.g., nonuniformities in space and time), and perhaps most importantly, to examine the processor’s output data and extract the useful information. The latter is often an iterative process and requires fusion with other data processing results. An optical processor’s speed advantage could be largely negated unless all processing operations can be performed at speeds commensurate with modulation and display rates. Thus, equally important considerations are the need to identify those operations that can be effectively performed optically, and the need to develop optical processing architectures that minimize the penalties on optical throughput. Because of these considerations optical information-processing approaches have covered a wide range of topics. Therefore, we first provide a brief review of the various paradigms that have been investigated in optical processing.