Volume ICs needed for image-processing designs
Volume ICs needed for image-processing designs
Andy Wilson Editor at Large
For years, Silicon Valley analysts have been predicting the emergence of systems that will integrate all the functions of a single system onto one piece of silicon. This potential concept has spawned several new design ideas that promise to provide systems designers with the necessary tools to create such systems on a chip.
Eager to jump on the bandwagon, many application-specific-integrated-circuit (ASIC) design companies and designers have embraced new buzzwords, such as intellectual-property development, with which to describe such methods. But, despite the hope that third-party designers would embrace this technology, only a few very large IC vendors, such as Motorola (Austin, TX) and LSI Logic (Milpitas, CA), have used such ideas to develop products.
Perhaps the most limiting factor that has slowed this adoption has been the high cost of development tools and the large semiconductor volumes required to justify the development of systems on a chip. Although the cost of development tools is steadily decreasing, the volumes required to justify such systems have limited their deployment to video games, cellular telephones, and digital television. In these markets, the manufacturer can cost-justify million-piece volumes. For board-level image-processing companies selling 500 add-in boards, at $5000 each, per year, however, the cost of systems-on-a-chip development is, at present, prohibitive.
To remain competitive, board designers must instead leverage the number of increasingly available image-processing ICs that have been developed for the graphics, video-conferencing, broadcast, and multimedia markets. Fortunately, the rapid reduction in semiconductor linewidths is allowing a number of such products to be produced at relatively low cost.
Now, systems designers can choose from a number of devices to perform MPEG-2 encoding, video encoding, integrated display, and specialized functions such as fast Fourier transforms. These devices can integrate more than 3 million equivalent transistors per chip and can be purchased for less than $100, thereby reducing the chip-count, manufacturing cost, and pricing of add-in image-processing boards. Indeed, even as many image-processing vendors are embracing such integrated ICs, others are looking to lower the cost of systems even further by incorporating very-long-instruction-word (VLIW) processors originally developed for digital television.
Incorporating approximately 5 million equivalent transistors, these processors have jumped into the forefront of microprocessor design and represent some of the most-complex image processors available to date. In the future, companies will build on this technology and incorporate video digitization, processing, compression, and display into their ICs.
If Gordon Moore`s 1965 prediction that the number of transistors per square inch on ICs will double every 18 months remains true, image-processing ICs with 40 million equivalent transistors on a chip should be available by 2003. And it seems certain that IC vendors will use this real estate to cram more functionality onto ICs at lower cost.
At present, VLIW processors from Phillips (Eindhoven, The Netherlands), Texas Instruments (Dallas, TX), and Equator Technologies (Campbell, CA) seem to be the leaders in providing CPUs for system-on-a-chip designs. Incorporating such functions as image convolution, FFT processing, and video encoding onto such processors will require that these functional blocks be embedded into these multimedia processors and be "callable" from the VLIW device.
However, despite the availability of such ICs early next year, it is highly unlikely that such devices will be tailored exactly to meet the needs of specific medical, military, industrial, and scientific imaging applications. The volume just isn`t there---yet.