Abstract
Currently, complex instruction set computers (CISCs), are used in a wide variety of applications ranging from desktop computers to aircraft guidance systems. The amount of software available for these CISCs is rather large, and exists in both high-level language and assembly language forms. However, some system designers may need CISCs that are more environmentally tolerant and/or faster than current silicon processors, and may be looking towards GaAs as a possible solution. Aerospace engineers and defense engineers are among this elite group. Some areas where the advantages of GaAs are needed over current Si technology include stabilization and control, telemetry, and command, antenna control, and failure detection. In this work, an attempt is made to draw some general conclusions on how GaAs bit-slice components can be used in a microprogrammed machine to achieve a speed increase over silicon. In particular, the MC68020 has been emulated using state-of-the-art 2900 series bit-slice components from Vitesse Semiconductor Corporation. However, before presenting the solution, we will first define the problem. It is already well known that GaAs has several advantages over silicon. Besides being faster than ECL, the fastest Si technology, GaAs is more environmentally sound. GaAs can operate in radiation levels above 100 million rads, and its operating temperature range* extends from about -200 C to about 200 C. Also, GaAs can interface directly with optical fibers, making it ideal for high-speed communications. Unfortunately, GaAs also has several drawbacks compared to Si. Besides being much more expensive, it has a higher density of dislocations, resulting in smaller VLSI area, smaller transistor count, and above all, a worse yield. Also, GaAs is more costly due to the scarcity of gallium and the fact that GaAs is a compound. Further disadvantages of GaAs include a lower noise margin, brittleness, and a limited availability of appropriate high-speed testing equipment. Despite the apparent inadequacies of GaAs, it is still used to improve system performance in selected applications. Because of the above mentioned limitations, it is not possible to implement an entire CISC (like the MC68020) on a single GaAs chip. Only multiple chip implementations are feasible. There are three basic approaches to the design of a GaAs based CISC: 1.) Divide the processor into bit slices, and fabricate each slice as a chip 2.) Functionally divide the processor and fabricate each function as a separate chip 3.) Emulate the processor using a RlSC type architecture The RISC-oriented approach was investigated by Kevin McNeley and Veljko Milutinovic at Purdue University. This approach works best for programs written in high level languages, but was not as efficient as initially anticipated for machine language CISC code. Our research centers on a combination of the bit-slice and functional approaches. Such an approach can emulate the machine language of the CISC without the need for an optimizing compiler or translator which is needed for the RISC approach.
Date of this Version
4-1-1988
Comments
This research (especially in its final phases) was partially supported by NCR Corporation, WorldTs Headquarters, Dayton, Ohio.