New developments in electronic memories

I. The versatile capabilities that have made the computer the great success of our age are due to exploitation of the high speed of electronic computation by means of stored programs. This pro­cess requires that intermediate results be stored rapidly and fur­nished¹ on demand for long computations, for which high speed is worthwhile² in the first place.

Storage devices or memories³ must have capacities⁴ sufficient not only tor intermediate results but also for the input and output data and the programs.

Once prepared a program can be reused any number of times, which involves remembering.

Computers can "remember" and "recall"⁵ and virtually un­limited is the capacity of computers to remember (that is, to store information). Associated with the capacity of remembering is the capacity of recalling.

In the context of electronics "memory" (or, in British usage, "store") usually refers to a device for storing digital information. Storage ("write") and retrieval ("read") operations are completely under electronic control. The storage of auditory or visual infor­mation in analogue form is usually referred to as recording.

There is some overlap⁶ between analogue and digital recording. Described here is digital memory.

The most widely used digital memories are read/write memo­ries, the term signifying⁷ that they perform read and write oper­ations at an identical or similar rate.

Of primary importance to characteristics for memories are storage capacity, cost per bit and reliability. Other important char­acteristics are speed of operation (defined in terms of access time),cycle⁸ time and data-transfer rate. Access time is simply the time it takes to read or write at any storage location.

The demand for fast access and large capacity has grown con­stantly. Never before has man possessed a tool comparable to a computer. Today there are memories accessible in tens of nanoseconds and memories with more than a billion bits. However although the existence of computer was a reality, only in 1970s have we got a microprocessor. It is the microprocessor that helps to solve many problems.

Ideal would be a single device in which vast amounts of in­formation could be stored in non-volatile form suitable for archival record-keeping and yet be accessible at electronic speeds when called for.⁹ So far¹⁰ there is no way to realize this ideal. Fortunately, the benefits of large capacity and rapid access can be obtained by use of a hierarchy of different types of storage devices of decreasing capacity and increasing speed.

A prime distinction¹¹ between memories is the manner¹² in hich information is stored (written) and accessed (read) andom-access memories involve column¹³ and row¹⁴ matrices which allow information to be stored in any cell¹⁵ and accessed in proximately the same time. By contrast,¹⁶"serial access" means that information is stored in column order, and access time depends on the storage location selected.

The main hierarchy today comprises, on the one hand, large-capacity magnetic recording devices, which are accessed mechani­cally and serial (reels¹⁷ of tapes, disks, and drums), and on the other hand, fast electronic memories (the core¹⁸ memory and various types of transistor memories).

Random-access memories can complete read and write oper­ations in specified minimum period known as the cycle time. Serial-access and block-access memories have a variable and rela­tively large access time after which the data-transfer rate is con­stant. The data-transfer rate is the rate at which information is transferred to or from sequential storage positions.

The smallest block of information accessible in a memory system can be a single bit (represented by 0 or 1), a larger group of bits such as a byte¹⁹ or character²⁰ (usually eight or nine bits), or a word (12 to 64 bits depending on the particular system). Most memories are location-addressable,²¹which means that a desired bit, byte or word has a specified address or physical location to which it is assigned.²²

Of prime interest to a reader will be the knowledge of the de­velopment of memories.

One of the first electronic memories was a circulating delay line, a signal transmission device in which the output, properly amplified and shaped, was fed back²³ into the input. Although it was economical, it had the inherent drawback²⁴ of serial access: the greater the capacity, the longer the average access time. What was really needed was selective access to any stored data in a time that was both as short as possible and independent of the data ad­dress or any previous access. This is known as random access, so named to emphasize²⁵ the total freedom of accessing and there­fore of branching²⁶ (following one or another part of a program). The first random-access memories (RAM's) were electrostatic storage tubes.

In the early 1950's the core memory replaced these early de­vices, providing a solution to the need for random access that truly fired the emerging computer industry.

The core memory has become the main internal computer memory and was used universally until challenged²⁷ recently by semiconductor memories. Typical are memories with 1 million words of 30 to 60 bits each, randomly accessible in 1 microsecond. The core memory has also been extended to very large capacities, of the order of 100 million words.

In the 1950's and 1960's electronic memories were arrays of cores, or rings, of ferrite material a millimeter or less in diameter, strung²⁸ by thousands on a grid of wires. Ferrite-core memories have now been largely succeeded²⁹ in new designs by semicon­ductor memories that provide faster data access, smaller physical size and lower power consumption, and all at significantly lower cost.

In the early 1970's semiconductor memory cells that served the same purpose as cores were developed, and integrated mem­ory circuits began to be installed as the main computer memory.

In the 1980's new memory technologies involving magnetic bubbles, superconducting tunnel-junction devices and devices ac­cessed by laser beams or electron beams come into play.

Semiconductor memories are extremely versatile and highly compatible³⁰ with other electronic devices in both small and large systems and have much potential for further improvement in performance and cost. They are expected to dominate the elec­tronic-memory market³¹ for at least another decade.

The most widely used form of electronic memory is the ran­dom-access read/write memory (RAM) fabricated in the form of a single large-scale-integrated memory chip capable of storing as many as 65.000 bits in an area less than half a centimeter on a side. A number of individual circuits, each storing one binary bit, are organized in a rectangular³² array. Access to the location of a single bit is provided by a binary-coded address presented as an input to address decoders that select one row and one column for a read or write operation. Only the storage element at the intersection³³ of the selected row and column is thetarget³⁴ for the reading or writing of one bit of information. A read/write control signal determines which of the two operations is to be performed. The memory array can be designed with a single in­put-output line for the transfer of data or with several parallel lines for the simultaneous³⁵ input or output of four, eight or more bits.

II. Different categories of semiconductor memories and spe­cific data storage applications where they find primary use provide system engineers with a wide range of options.³⁶ In general, metal-oxide semiconductor (MOS), erasable-programmable read-­only memories (EPROMs) and dynamic random-access memo­ries (RAMs) are extensively used in micro- and minicomputer ap­plications. The slow electrically-alterable read-only memories (EAROMs) are most suitable to peripherals, at present. In addi­tion, dense dynamic MOS RAMs are used in large volume³⁷ in small and large mainframe computers, and so on and so forth. Many laboratories are looking for new options.

However, we are still far from the ideal shoe-box device with 10¹² bits accessible in nanoseconds, and still farther from the ca­pacities of 10¹⁵ bits needed for many already well-defined appli­cations. Although much can still be expected from VLSI and mag­netic techniques, these great goals (цель) may require radically new approaches.

Very high speed and very low power memories rather than large capacity may well be the benefits of some of these ap­proaches.³⁸

Thus computers today use a hierarchy of large-capacity, rela­tively slow mechanically accessed memories in conjunction³⁹ with fast electronically accessed memories of relatively small capacity. It would be highly desirable to fill the gap by some device of sufficient capacity and speed.

Candidates for gap-filling memories include metal-oxide semiconductor (MOS) random-access memories (RAMs) made by large-scale integration (LSI); magnetic bubble devices based on cylindrical domains of magnetization; electron beam-addressed memories; and optical memories based on lasers, holography, and electrooptical effects, charge-coupled devices (CCD).

One of the latest designs of a CCD serial-access memory has storage for 65.536 bits on a chip measuring about 3.5 by five mil­limeters.

Thevast⁴⁰ number of different types of semiconductor mem­ories available to the system engineer is increasing steadily.

Radically new technologies, still at an early laboratory stage, are aimed⁴¹ at a more ideal solution than today's hierarchy.

Many laboratories are looking into basic principles. Semiconductor memories based on the Josephson effect may be able to operate in picoseconds on small power. The boundaries within the walls of magnetic domains,⁴² exploited in the bubble lattice devices, are also used in a so-called cross-tie memory that may provide non-volatile storage memories on LSI chips.

One can foresee the development of cryoelectronic memories with extremely high component densities operating at speeds 10 to 100 times faster than today's fastest electronic memories.

Researchers now are looking forward to light parti­cles-photons-which will permit the performance to be made a thousand times faster. This would mean that in the future we can expect the emergence of photon computers and that computations will be done by means of light.

Any radical improvement in memory technology will ul­timately greatly affect our way of life, as previous innovations have shown.