July 1960 Electronics World
Table of Contents
Wax nostalgic about and learn from the history of early electronics. See articles
from
Electronics World, published May 1959
- December 1971. All copyrights hereby acknowledged.
|
In 1960 when this "Computer
Memory Devices" article appeared in Electronics World magazine, digital
electronic computers were still a relatively new technology. Although bulky
and power-hungry, accomplishing digital manipulations of data for logic, mathematics,
sorting, etc., was relatively easy, but unless programming instructions were fixed
and output was used real-time, an ability to store data is necessary. Memory banks
composed of vacuum tubes could - and often did - do the job, but the data was neither
permanent nor physically transportable. If power is lost, the information is lost.
We still suffer that issue even today volatile type memory. For more permanent data
storage, magnetic media was developed, solving both aforementioned down sides of
electronics memory. This two-part article by Mr. Ed Bukstein, of the Northwestern
TV & Electronics Institute, delves into the state of the art back in the day.
Computer
Memory Devices - Part 2 was in the August 1960 issue of Electronics
World.
Computer Memory Devices - Part 1
For speed and automation, storing information is important. Magnetic
tapes and drums are used.
By Ed Bukstein
Northwestern TV & Electronics Inst.
As compared to mechanical calculators such as the ordinary adding machine, the
electronic computer has two important advantages: high speed and automatic operation.
Both of these advantages result from the use of memory or storage devices. The numbers
to be used in calculation and the instructions which specify the types of operations
to be performed (addition, subtraction, etc.) are stored in the memory device before
the calculation begins. The computer then extracts this information as it is needed
during the course of the computation. As a result, the computer is able to perform
a lengthy series of calculations without the intervention of a human operator.
In this sense, the computer is automatic. By contrast, the operator of an office-type
adding machine must punch - in new information (through the key- board) before each
individual step of the calculation. Since the machine is "idle" during the time
the operator is entering new information into the keyboard, the calculation progresses
at a relatively slow rate. A full day's work might therefore be required to complete
a calculation which an electronic computer could perform in a matter of seconds.
The memory section of a computer is also used to store partial answers, obtained
in early steps of a calculation, which may be required in a later step. Final answers
are also stored in the memory and, at some convenient time, can be transferred to
a print-out device, such as an electric typewriter. Many computers are so designed
that the print -out function can be performed while the computer is simultaneously
working on another problem.
Fig. 1 - As with sound, electromagnetic heads (A) record and
(B) play back data.
Fig. 2 - Tape-tensioning vacuum columns permit rapid, harmless
stops and starts.
Fig. 3 - In return-to-zero (A) recording, binary 1 is represented
by tape magnetization in one direction and binary 0 by magnetization in the opposite
direction. In non-return-to-zero recording, direction of magnetization is changed
(B) only when switching from one digit to the other or (C) only when 1 is recorded.
Fig. 4 - Upper portion of rack-mounted Ampex FR-300 digital tape
handler with standard speed of 150 inches per second and a fast speed of 225 ips.
Fig. 5 - Detail of one of the vertical vacuum columns used in
the handler of Fig. 4. Compare this arrangement with the one highlighted in Fig.
2.
Fig. 6 - Access time of a magnetic drum is shorter than that
for a reel of tape
Fig. 7 - Although heads for magnetic drums are in parallel, they
are isolated from each other by one-way semiconductor networks.
Fig. 8 - The magnetic drum of this Royal McBee LGP computer,
which can be seen passing under the heads, stores 4096 words.
Fig. 9 - To select specific stored data, comparator produces
output only when right number is fed from timing track.
Fig. 10 - A quick-access loop makes data available in less than
one revolution by constantly repeating this data.
Magnetic Tape Storage
The magnetic tape recorders used for data storage in digital computers differ
in physical detail, but not in basic principle, from those used for audio recording.
This basic principle is illustrated in Fig. 1. As shown, the tape consists of a
plastic base coated with a thin layer of magnetic oxide. This tape is moved through
the machine by a motor-driven transport mechanism. The path of the tape is such
that the magnetic coating passes adjacent to the air gap in the write head (Fig.
1A).
If a pulse of current is passed through the coil of the write head, magnetic
lines of force will be established in the core. These lines of force bridge the
air gap by flowing through the magnetic coating of the tape. The section of tape
under the air gap therefore becomes magnetized as a consequence of the current flow
through the write coil. Playback, known as reading in computer terminology, is accomplished
by moving the tape adjacent to the air gap of a read head (Fig. 1B). As the magnetized
tape passes the gap, a magnetic field is established in the core and the lines of
force cut across the coil. The voltage induced in the coil constitutes the output.
In computer practice, the same head is used for both writing and reading. This is
permissible because the two operations do not occur simultaneously. Information
is recorded in a number of parallel tracks across the width of the tape and a separate
head is provided for each track. Binary -coded decimal, excess -3 code, and 7 -bit
code are frequently used for storing data on magnetized tape (see "Numbers Systems
and Codes," Electronics World, November 1959) . Recording is interrupted at periodic
intervals so that the information is recorded in blocks. Typically, each block of
information occupies about an inch of tape length, and the blocks are separated
by about a quarter inch of blank tape. Such grouping of the recorded material is
an aid in locating and selecting desired data for read -out. Counting circuits that
respond to the passing blocks of information activate the reading circuits when
the desired block passes under the reading heads. In audio recording, the degree
of magnetization of the tape is an important factor and varies with the audio signal
being recorded. In computer applications, however, the direction of magnetization
is the important factor and the degree of magnetization con- veys no intelligence.
For improved signal -to -noise ratio, the write currents are made sufficient in
amplitude to magnetize the tape completely to saturation. The recorded tape is therefore
fully magnetized in one direction or the other. Since data is recorded in binary
form, one direction of magnetization represents binary 1 and the opposite direction
of magnetization represents binary O. This technique, known as RZ (return -to -zero)
recording, is illustrated in Fig. 3A. In the method known as NRZ (non -return -to
-zero) recording (Fig. 3B), the direction of magnetization of the tape changes only
when the data changes from 0 to 1, or from 1 to O. In another variation of NRZ recording,
the direction of magnetization changes only when a 1 is to be re corded. As shown
in Fig. 3C, each change in the direction of magnetization represents a 1, and the
0 is recorded by maintaining the magnetization in the same direction it had during
the previous bit. As compared to RZ recording, the NRZ techniques are characterized
by less frequent changes in the direction of magnetization. This is an important
advantage because it permits more bits of information to be recorded on a given
length of tape. The mechanical requirements of a tape machine designed for computer
work are somewhat more demanding than those of a machine designed for audio applications.
This is so because the computer tape must be frequently started, stopped, and reversed
in direction. To prevent tape breakage and to permit rapid acceleration and deceleration,
the tape is allowed to hang in a loose loop on each side of the head, as shown in
Fig. 2. The loops hang down into vacuum columns, so that the atmospheric pressure
on top of the tape maintains a slight tension on the loops. This arrangement permits
the tape to be started and brought up to speed in a very short time because the
motor need accelerate only the tape in the loop. The other reel and the remaining
tape "catch up" a short time later. Start and stop times of five milliseconds are
typical. Each reel has its own motor, and a servomechanism con- trol prevents the
loops from becoming too short or too long. The servomechanism responds to signals
supplied by either pressure switches or photocells located in the vacuum columns.
The Ampex FR -300 digital tape handler, shown in Fig. 4, has a standard speed of
150 inches per second forward or reverse, and a fast speed of 225 inches per second.
Start and stop times are 1.5 milliseconds. The tape- tensioning air columns of this
machine are constructed as shown in Fig. 5. Air drawn from the exhaust ports maintains
tension on the loops, and the loops contract or expand as the tape accelerates or
decelerates. To minimize wear, the FR -300 drives on the tape backing rather than
on the oxide side. This machine handles either half - inch or one -inch tape widths,
providing 8 or 16 recording tracks. Magnetic tape, as a storage medium, offers the
advantage of very great ca- pacity. At a packing density of 200 bits per inch, a
half -inch tape 1200 feet in length can store literally millions of bits of information.
Tape lengths of 2400 and 3600 feet are also commonly used and provide even greater
storage capacities. Furthermore, a single com- puter may have at its command a large
number of tape machines. One disadvantage of tape storage. however, is its relatively
long access time. Since the information required at some particular instant may
be out at the far end of the tape, several minutes may be required to bring it under
the reading heads. For this reason, computers utilize magnetic tape for reserve
storage, but use a medium having a shorter access time for the main memory.
The magnetic drum is often used for this purpose. The Magnetic Drum The magnetic
drum consists of a motor -driven, aluminum cylinder coated with a thin layer of
magnetic oxide. Writing and reading are accom- plished in a manner similar to that
used with magnetic tape. In a sense, the drum surface is like a short, wide tape
with the ends joined to form a continuous loop. As shown in Fig. 6, many parallel
tracks can be recorded on the drum. A twelve -inch drum, for example, may hold 200
tracks of information. The number of bits that can be stored in each track is, of
course, determined by the diameter of the drum. Drum diameters from five to ten
inches are common, but smaller and larger sizes are sometimes used. Magnetic drums
are normally pro- vided with separate heads for each track. It is possible, of course,
to use fewer heads so mounted that they can be moved from one track to another.
This arrangement however, results in an increase of mechanical complexity as well
as an increase of access time. The heads of a magnetic drum may be connected in
parallel (through diodes) as shown in Fig. 7. Since the diodes are biased in the
non -conducting direction, the heads are effectively isolated from each other. Any
given track can then be selected for writing or reading by making the associated
diodes conductive. If, for example, a positive control signal is applied to the
grid of V, in Fig. 7, this tube will become con- ductive and the voltage at point
X will rise from a negative to a positive value. Since diodes DA and D, are now
forward biased, the head for track 3 is con- nected to the read and write amplifiers.
Drum storage is cyclic, because each revolution of the drum brings the same information
under the reading heads. The access time of the drum is therefore determined by
the rate of revolution. Since it may happen that the desired information has just
passed the reading heads at the time it is needed, the maximum access time is the
time required for the drum to complete one revolution. It may happen, however, that
the desired information is just ap- proaching the reading position at the time this
information is required. In such a case, the actual access time will be much shorter
than the maximum access time. For this reason, access time is often specified as
an average value, equal to the time required for the drum to make one half revolution.
A common value of drum speed is 3600 rpm, but rates in excess of 12,000 rpm have
been used. At 3600 rpm, the maximum access time would be 16.7 milliseconds and the
average access time would be approximately 8.3 milliseconds. The magnetic drum used
in Royal McBee Corporation's LGP -30 computer is shown in Fig. 8. Data on this drum
is divided into groups known as words, each of which contains 32 bits including
the sign bit (plus or minus) and the spacer bit. The drum holds 4096 such words.
The drum surface can be seen in Fig. 8 between the rows of heads. Many computers
are designed to handle numbers that are 10 to 12 digits in length. Since each of
these decimal digits can be represented in coded decimal notation by four binary
bits, 40 to 48 bits are required to represent each decimal number. In addition,
another four -bit combination is used to indicate whether the number is positive
or negative, bringing the total to 44 to 52 bits. This amount of information is
known as a word, and many such words may be recorded in each track of a magnetic
drum. To read a particular word from one of the drum tracks, it is necessary to
activate the reading circuits just before the first bit of the desired word reaches
the head, and to de- activate 8+ the circuits as soon as the last bit of the desired
word has been read. For this purpose, a timing track is recorded on the drum as
shown in Fig. 6. On this track, a series of pulses is recorded in positions such
that each pulse passes under the reading head at a time when the first bit of a
word is approaching the reading heads on the other tracks. The pulses from the timing
track are fed into a word counter, as shown in Fig. 9. This binary counter therefore
advances its count each time a new word moves into the reading position. A comparator
circuit (coincidence detector) produces an output when the number in the word counter
is equal to a number previously stored in the address selector flip -flops. If the
fifth word on the track is to be read, for example, the address selector must first
be set to 101 (decimal 5). This is done automatically by the con- trol section of
the computer. As the drum revolves and the word counter advances, the comparator
produces an output signal when the word counter reaches 101 (first and third stages
in the one condition and second stage in the zero condition). The output of the
comparator is used to activate the reading head of the track containing the desired
word. After the last bit of this word has been read, a pulse from the timing track
advances the word counter to 110 (decimal 6). Since the number in the word counter
is no longer equal to the number in the address selector, the comparator ceases
to produce an output and reading ceases. It is obvious, from the arrangement of
components shown in Fig. 9, that the word counter must be reset to 000 after each
complete revolution of the drum. This can be accomplished by means of a single pulse
recorded on a separate track of the drum. This pulse marks the starting or home
position of the drum and is used to reset the word counter. For simplicity, only
three stages are shown in the word counter and the address selector in Fig. 9. In
actual practice, a greater number of stages are required, as determined by the total
number of words contained in each track. One or more tracks of a drum may be set
aside for quick -access use, as shown in Fig. 10. Each such track requires two heads,
so that, as each bit is read, it is rewritten farther back on the drum surface.
In this manner, the same word (or several selected words) are continually read and
rewritten. Since a given word appears many times around the circumference of the
drum, the access time is much shorter than that for words written on the other tracks
of the drum. The arrangement shown in Fig. 10 is known as a quick - access loop,
a two- headed loop, a recirculating register, or a revolver. The IBM RAMAC computer
uses magnetic discs rather than a drum. These discs, each of which resembles an
over -size phonograph record, are coated with a magnetic substance and information
is recorded in circular tracks on each side of the disc. The discs are stacked vertically,
with sufficient space between for the read /write head to move freely. When a given
disc is selected for reading or writing, the head moves into position vertically
to the desired disc, and then moves inward toward the center of the disc to the
desired track.
Summary
Actually magnetic tape and the magnetic drum complement each other - one providing
the greater storage capacity and the other offering shorter access time. Thus many
computers use the two in combination, with the drum serving as the main memory and
the tape providing back-up storage of information.
In this way, blocks of information can be transferred from the tape to the drum
at some time prior to the need for this data in the computer. After the computer
has used the data, other blocks of information are transferred to the drum.
In some applications, however, even shorter access time is required than can
be provided by a drum. In these cases, there is still another type of device available.
These units, magnetic cores, and their application will be covered in the concluding
installment of this article.
Posted January 2, 2023
|