(1) Memory Devices
Most early spaceborne computers (e.g., Centaur, Titan 2 and 3, Minuteman 1 and 2, Saturn 1) used electromechanical drums or discs as memory devices. However, these are disappearing from space application for several reasons (ref. 14): a substantial increase in packaging density of core and film memories, the serial access nature of discs and drums, and the limited time that discs and drums can operate without maintenance. Consequently, in recent years, the ferrite coincident-current core memory has been the cornerstone of computer technology, providing fast random access at a few cents per bit in the megabit range. DRO cores used for program memory suffer the disadvantage of requiring a write cycle following each read to restore the information. More sophisticated geometries, such as the multiaperture devices (as used on the Gemini computer) or BIAX, can be used to obtain NDRO operation but at a cost in weight and volume (refs. 55 and 56). More often, NDRO memories are implemented with thin-film techniques, wired arrays, or semiconductors.
Thin-film memories have received a large research investment and significant advances have been made in their speed and bit density. However, they have not gained wide popularity, since their cost is relatively high and their capacity is more limited than core (ref. 14). The "Bicore" thin magnetic film element has been successfully used in the Titan 3-C computer and a similar device, "Quadralloy," is being used in the Phoenix missile guidance computer (ref. 55). These devices can be electrically altered, but the write time for NDRO operation is typically 1,000 times the read time. Therefore, NDRO information is usually electrically entered by external equipment and no provision is made for real-time write operations. However, DRO thin-film memories can be altered in the same time as read time. For example, the DRO-mated film memory used in the UNIVAC 1832 computer has a 750 nanosec read-write time (ref. 57).
If nonvolatility is a requirement, plated wire (refs. 55, 56, 58 to 62) is generally considered a good candidate for the next generation mainframe and mass memories. Plated wire memories can be constructed for either DRO or NDRO operation. Speed, low power, moderate density, and automated production give this device the potential advantage of use in all memory applications of future spaceborne computers. NDRO plated wire is being used for the memory of the Honeywell HDC-701 computer for Minuteman 3.
Two types of wired arrays have been used in which the information is physically contained in the memory. Missing core memories are constructed by removing or shorting specific cores. In core-rope memories, such as the Apollo guidance computer uses (ref. 14), magnetic cores are either threaded or bypassed by word lines in a manner that permits storing one or more entire words in one core. The resulting bit density is extremely high: approximately 100 bits per cubic centimeter (about 1500 bits per cubic inch) including all electronics, interconnections, and packaging hardware.
Recent developments in LSI have made the semiconductor memory extremely attractive and practical, particularly for scratchpad or high-speed control applications (refs. 55 and 56). These memories are generally constructed of MOS or bipolar devices and provide NDRO operation, but they are volatile. However, several manufacturers are presently developing nonvolatile semiconductor memories. Two potential techniques are magnetic cylindrical domain "bubble" devices (ref. 63) and the somewhat controversial Ovonics devices based on amorphous material technology (ref. 64). There is currently a definite trend toward the use of semiconductor memories, and they may become the major memory technology by the next decade (refs. 52, 55, and 62).
Plated wire and semiconductor memories are expected to be the most suitable technologies for mass data storage within the next decade (ref. 62). Serial memories using ferroacoustic (ref. 65) and magnetic domain device (ref. 66) technologies may also provide nonmechanical alternatives to drums and tape by the mid 1970s.
(2) Logic Devices
Since about 1963, bipolar silicon integrated circuits, or microcircuits, have been adopted nearly universally by spaceborne computer designers for at least the logic portion of the machines (refs. 14 and 52). Prior to the advent of microcircuits, magnetic cores and all-transistor circuitry were both strong contenders as logic elements. Core circuits were no smaller, but they were capable of operating on substantially lower power. Although special applications may favor the magnetic core, the small size, high speed, and reliability of microcircuits make them preferable to cores in nearly all instances. Moreover, the gradual reductions in power consumption of new microcircuit logic devices has enabled them to consume less power at full speed than cores. However, cores still have the advantage of reduced power consumption at low speed operation (ref. 14), an advantage which complementary MOS circuits share.
Within the last few years, spaceborne computer designers have begun to utilize LSI techniques to reduce volume and weight (e.g., ref. 67). Moreover, MOS technology is now being used to design central processing units (e.g., the Autonetics D-200), and it is likely that power, volume, and weight will continue to decrease as MOS used more widely (ref. 52). However, its susceptibility to radiation effects may limit MOS use in some critical and long-duration missions.
Home - NASA
Office of Logic Design
Last Revised: February 03, 2010
Digital Engineering Institute
Web Grunt: Richard Katz