VLSI-Based High-Performance Raster Image System

摘要

Our research objective is to create affordable, high-performance, 3-D raster display systems that harness the highly parallel computational power of custom VLSI circuits. We want to make this power available in an open system that encourages invention of new parallel algorithms for image generation and image processing. The system has many potential applications including flight and tactical displays, computer-aided mechanical design, and medical diagnosis and therapy. As in other systems, ours includes an array of memory chips forming an image buffer from which the video screen is refreshed. The novel feature of our design is that our memory chips have been custom-designed to include processing circuitry so that each pixel (dot on display) can carry out its own image-generation calculations. Calculations are distributed throughout each chip in such a way that only a very small amount of circuitry is required for each pixel. In our current chips, this processing circuitry takes up about one-third of the total area, with the remainder devoted to pixel memory. During the three years of this work, we have achieved significant experimental results. We have developed algorithms for fast image generation, including very rapid rendering of spheres and shadow-casting (our system is the only graphics machine, to our knowledge, that can render true shadows in real time). We have also designed two custom integrated circuits for the machine; the new memory chip is the fastest memory so far designed in the university/VLSI community. These chips have been built into a running prototype. A full-scale, full-speed machine (512x512 pixels, 30,000 smooth-shaded triangles per second) is nearing completion.