Fullscreen HTML5Med Res (??MB)HTML4

> THE HISTORY OF COMPUTER MEMORY > EARLY DATA STORAGE PART 1 BY DR DAVID MADDISON One of the most critical technological advances driving the widespread adoption of computers has been smaller, faster, higher-capacity memory chips. It didn’t start with semiconductors and ICs, though; memory has been around in various forms for a long time. This two-part series will investigate how it started and grew into what we have today. “Extended ASCII” (not an official EMORY is one of the most for temporary storage, rather than ‘secM important and commonly dis- ondary memory’ used for long-term name) uses 8 bits and has 256 characters, with extended foreign language cussed elements of a computer. These storage, such as hard disks. days, computer memory size is measured in gigabytes or even terabytes. While huge & cheap memory capacities are taken for granted, early computers had tiny memories because integrated circuit technology had not yet been developed and storing even one byte was expensive and complicated. For one byte, eight zero or one values need to be stored, so something had to be duplicated eight times. Without integrated circuits, whatever was used to store that information was expensive and big. Note that there are and have been systems that use bytes with fewer or more than eight bits, but eight is the most common number. In this two-part series, we will focus mainly on ‘primary memory’, the working memory of the computer 12 Silicon Chip However, the distinction wasn’t always clear in early computers, which also lacked convenient input and output systems. Hence, we will discuss technologies like punched cards and paper tape that were used for both primary and secondary storage. Secondary storage may be the subject of another article. Bits and bytes One byte is the unit of digital information typically used to encode a character, such as the ASCII-1977 character set members, which includes the letters and numerals A-Z, a-z, 0-9, punctuation and special characters. ASCII is a 7-bit encoding scheme that represents 128 printing and non-­printing characters. Australia's electronics magazine characters, symbols, line-drawing characters etc. The exact set of symbols depends on various proprietary implementations or standards like ISO/IEC 8859. That has largely been supplanted now by Unicode (see panel). Still, with this system, one character is stored in one byte. Five bits is the minimum amount of storage necessary to represent the alphabet; however, with just five bits, all 26 letters could be represented in one case (upper or lower), but not all numbers. So most 5-bit character code sets could switch between letters (LTRS) and numbers (FIGS), allowing 60 letters, characters and other codes to be used – see Table 1. Today, a byte usually consists of eight bits represented by 0 or 1 and siliconchip.com.au in the range 0 (decimal) or 00000000 (binary) to 255 (decimal) or 11111111 (binary). However, past computers have used fewer, such as 6-bit codes to represent 64 characters. Other byte sizes are also used for addresses and number representation in modern computers in CPUs or GPUs (graphics processing units) such as 16, 32, 64, 128, 256 bits and beyond. These architectures usually still group data in multiples of 8-bit bytes. You might have noticed a discrepancy between the stated size of a disk drive and the size reported by the computer operating system. That depends upon if the size is counted in decimal or binary. One kilobyte is 1000 bytes in decimal notation or 1024 bytes (210) in binary notation, while one gigabyte is one billion bytes (1,000,000,000) in decimal notation or 1,073,741,824 bytes (230) in binary notation. This represents a difference of 7.3% for gigabytes or 10% for terabytes. It is done this way because, for a computer, indexing into a large file is much more easily done in power-oftwo chunks (like 1024) than decimal sizes like 1000. This discrepancy has resulted in new terms such as kibibyte (KiB; 1024 bytes), mebibyte (MiB; 1,048,576 bytes), gibibyte (GiB; 1,073,741,824 bytes) etc. While it might seem more confusing at the moment, the introduction of these terms is an attempt to reduce confusion about memory sizes. Memory devices The idea of using some device to input or store data or instructions of a variable nature is not new and has its origins in the form of punched paper tape or cards, as follows: 1725 Weaving looms were controlled using paper tape ‘programs’ with punched holes, a system developed by Basile Bouchon of Lyon, France. 1804 Joseph Marie Jacquard (also of Lyon) developed a loom control system using punched cards. 1832 Semyon Korsakov (St Petersburg, Russia) proposed using punched cards for information search and retrieval. 1837 Charles Babbage (London, UK) proposed using punched cards for inputting data and instructions to his never completed (by him) “Analytical Engine”, the first ‘Turing-complete’ siliconchip.com.au Table 1: patterns represented for a given number of bits. Bits Number of patterns (2bits) 1 2 (0 or 1) 2 4 (00, 01, 10 or 11) 3 8 (000, 001, 010, 011, 100, 101, 110 or 111) 4 16 (numerals 0...9 plus some punctuation) 5 32 (26 letters plus some punctuation) 6 64 (26 letters in two cases, ten digits, space & full stop) 7 128 (all ASCII characters) 8 256 (full code page or Unicode UTF-8) 16 65,536 (UTF-16) 32 4,294,967,296 (UTF-32) 64 1.84 × 1019 128 3.40 × 1038 256 1.16 × 1077 Fig.1: punched cards were used as the memory for the first ‘Turing-complete’ computer, Charles Babbage’s Analytical Engine. The smaller cards specify the mathematical operations to be performed, while the larger cards hold numerical variables. Source: https://w.wiki/5xR7 (CC-BY-2.0). computer. It was mechanical rather than electronic since electronic technology was still in its infancy. It still contained all the elements of a modern computer – see Fig.1. IBM punch(ed) cards For over half a century, the world’s most common medium for information storage was the once-ubiquitous IBM punched card. They have a fascinating and long history, but we do not have space to cover it all here, so we will just mention the highlights. Punched cards were not developed for computers, which did not yet exist, but for machines that tabulated data. The “IBM card” originated with Herman Hollerith (New York, USA) in the 1880s and 1890s, who used them in mechanical tabulating machines. These electromechanical machines were used to summarise information encoded on punched cards, such as census data (see Figs.2 & 3). Hollerith’s company eventually became part of IBM, and the machine became a core product. The tabulating machine was not a computer, but it could perform some The Unicode Standard Unicode is an international character set with 149,186 characters and symbols (as of version 15.0) in current use. Before Unicode, every different language required a distinct ‘code page’, making mixing different languages virtually impossible and leading to much confusion. Unicode solves this by bringing all the characters needed for human languages together in one set. Clearly, you can’t encode that many characters in a single byte. Therefore, in modern computer memory systems, characters are generally encoded as variable-length byte strings, providing backward compatibility with existing single-byte character sets like ASCII. There are several valid Unicode encoding schemes. Probably the most common is UTF-8, where a Unicode character that’s also part of the ASCII set is encoded as a single byte with its top bit as 0. Other characters or symbols are encoded as multiple bytes (up to four), where the first byte has its top bit set to 1. Other schemes that are part of the standard include UTF16, UTF-32 and BOM. Australia's electronics magazine January 2023  13 Fig.2: a replica of an 1890 model Hollerith punched card tabulating machine used to process data from the 1890 US Census. Source: https://w.wiki/5xR8 (CCBY-2.0). mathematical operations, group data and print results. The first IBM card had 22 columns and eight rows (punch positions); by 1900, they had 24 columns and 10 rows; and by the late 1920s, 45 columns and 12 rows. In 1928, a new version of the card was introduced with 80 columns and 10 rows – see Fig.4 (they moved to 12 rows in 1930). Those punched cards are the likely reason that early alphanumeric computer monitors had 80 columns. The cards measured 7-⅜ inches by 3-¼ inches or 187.3mm × 82.5mm. These dimensions were that of US paper currency from 1862-1923. The IBM card had many incidental uses besides computers; they were often used for taking notes and making dot points for presentations, as they fitted the inside pocket of a suit jacket. IBM was not the only manufacturer of punched cards or equipment to read and write them, but they became known by that name. People may laugh at punched cards today but, like books, if stored correctly, the data will be readable with the naked eye far into the future. However, data stored on CDs, magnetic disks and the like may deteriorate over time (disc rot) or become unreadable due to a lack of software and hardware support. Punched paper tape Fig.4: an IBM card. The data encoded is one line of a FORTRAN program: “ 12 PIFRA=(A(JB,37)-A(JB,99))/A(JB,47)    PUX 0430” Source: https://w.wiki/4icp (CC BY 2.0). Punched paper tape is conceptually similar to cards but can be kept on long rolls (sometimes formed into a loop) rather than on individual cards. It was invented in 1725 by Basile Bouchon to control looms, but that was impractical at the time. Like punched cards, punched paper tape was used for various applications in the 19th and 20th centuries, such as programmable looms, telegraphy systems, CNC machine tools and computer data input and storage from the 1940s (including military code-­breaking during WW2; see Fig.5) through to the early 1970s. Data stored on tape was also used as read-only memory (ROM) for computers. Tougher versions of the tape for industrial use were made with Mylar. Like cards, paper tape has the advantage of being able to be read by eye and is long-lasting if used and stored correctly. Paper tape was usually 0.1mm thick and either 17.5mm wide (11/16th of Australia's electronics magazine siliconchip.com.au Fig.3: a Hollerith punched card from about 1895, the predecessor of the IBM card. Source: https://w.wiki/5xR9 (public domain). 14 Silicon Chip Beyond punched cards & tape 1918 William Eccles and Frank siliconchip.com.au CAR. RET. LINE FEED LETTERS FIGURES SPACE THRU Uppercase Lowercase BELL Fig.5: paper tape as used on the WW2 Colossus Mk2 code-breaking computer in 1943. This computer had no internal memory storage (RAM), so the program tape had to be continuously read in a loop. Source: https://w.wiki/5xRA CITY an inch) for five-bit codes, or 25.4mm wide (one inch) for 6-bit or more codes. The hole spacing was 2.54mm (1/10th of an inch) in both directions. Sprocket holes were 1.2mm (0.046 inches) apart. Paper tape could store 10 characters per inch (25.4mm). A standard teletype roll was 1000 feet long (305m), so it could store up to 120kbytes, but most tapes were much shorter than that as many contemporary computers couldn’t handle that much data. Several different encoding schemes were used, starting with Baudot’s from the 1870s. It was developed for telegraphs and used five holes (five bits). In 1901, the Baudot scheme was modified to create the Murray code that included carriage return (CR) and line feed (LF) – see Fig.6. Western Union used that until the 1950s; they modified it by adding control codes, a space and a bell (BEL) symbol to ring a bell. 1924 the Western Union code was used by the International Telecommunications Union (CCIT) as the basis of the International Telegraph Alphabet No. 2 (ITA2), a version of which was adopted by the USA and called TTY. TTY was used until 1963. All of the former systems used 5-bit codes, after which 7-bit ASCII was adopted. There were also some encoding schemes that used six bits. The IBM Selective Sequence Electronic Calculator was an electromechanical machine that operated from 1948 to 1952 – see Fig.7. It used uncut IBM card stock to create tapes that were 7.375 inches (18.73cm) wide and the length of an IBM punched card (joined end-to-end). Each of the 80 columns could contain a signed 19-digit number with parity bits plus two rows for side sprockets. The tape(s) typically contained large mathematical tables; with multiple readers and up to 36 tapes, they could be searched in about one second. There were another 30 readers for program data. The rolls could be continuous or looped; a full roll weighed 400lb (181kg). About 400,000 characters could be stored on the tapes. The machine also used IBM punched cards. It gave IBM excellent publicity and was the basis for many interpretations of what a computer looked like. - ? : $ 3 ! &# 8 ( ) . , 90 14 57 ; 2 /6 " A B C D E F G H I J K L MN O P Q R S T U VW X Y Z Feed holes Paper tape showing the five-bit Baudot Code Fig.6: the five-bit code implemented on paper tape. More common characters use fewer holes. Source: https://savzen.wordpress.com/tag/baudot/ Fig.7: a retouched version of the famous photo of the IBM Selective Sequence Electronic Calculator. The 181kg paper tape rolls on the readers in the background were made of IBM card stock. Source: www. thedigitaltransformationpeople.com/channels/enabling-technologies/ mainframes-can-be-cool/ Australia's electronics magazine January 2023  15 Fig.8: circuit diagrams of the Eccles and Jordan flip-flop from their patent application. Fig.9: the magnetic drum memory from a Swedish BESK computer, with a sample of much more compact core memory of unknown capacity above it. Source: https://w.wiki/5xRB (GNU FDL). Jordan filed a patent entitled “Improvements in Ionic Relays” and received British patent 148,582 in 1920 – see Fig.8 & siliconchip.au/link/abhs While not intended for computer memory (electronic computers had not yet been invented), it was to become the basis of later computer memory. It comprised two valves (vacuum tubes) that could exist together in one of two stable states. It was originally called the Eccles– Jordan trigger circuit, the trigger circuit or a multi-vibrator, but today it is known as a flip-flop. The ability for a flip-flop to exist in either of two stable states representing a 0 or 1 is the basis of some computer memory today, such as SRAM (static random-access memory, see the 1963 entry later) and CPU registers. 1932 Austrian Gustav Tauschek invented magnetic drum memory in 1932, which became a widely used form of primary computer memory (‘RAM’) in the 1950s and 1960s. How was this device invented before the first programmable digital computer? It was initially devised to record data from punched card machines and then was adopted for early computers. Tauschek’s original device from 1932 had a capacity of 500,000 bits or 62.5kbytes. As the name implies, drum memory consists of a drum coated with magnetic material; several read and write heads are mounted along the length of the drum. Drum memory initially displaced CRT and delay line memory (see below) because it was more reliable. Magnetic core memory gradually replaced drum memory for primary storage. Keyboard drum Decimal-to-binary conversion drum Capacitor Counter drum Decimal card reader Carry-over drum Motor Drum memory was also used for secondary (semi-permanent) storage and, in this role, drums were eventually replaced by floppy disk drives starting in the early 1970s. One of the latest known uses of drum memory is in US Minuteman ICBM launch site computers (until the mid-1990s). Fig.9 shows drum memory from the 1953 Swedish BESK computer and magnetic core memory from the same machine. The capacity of neither device is known. The BESK computer was used to create the first computer animation; see the video titled “Rendering of a planned highway (1961) First realistic computer animation” at https://youtu.be/oQMD7oufO4s 1942 John Atanasoff and Clifford Berry built the little-known ABC (Atanasoff-­ B erry computer) – see Fig.10. Some argue that this machine Memory Disk (25 capacitors per side) One-cycle switch Carry-over capacitor Drive motor Base 2 card reader Base 2 output card puncher Power supply and regulator 30 add-subtract logic circuits Electrical card-punching circuits Power supply Memory-regenerating circuit Memory-regenerating circuits Add-subtract logic circuit Fig.10: an overall view of the ABC computer (left) and details of its regenerative capacitor memory unit (right) showing only one disc of 30 and one drum of two. Source: www.researchgate.net/publication/242292661 16 Silicon Chip Australia's electronics magazine siliconchip.com.au is the first automatic electronic digital computer; others dispute that because it was not programmable and was not Turing-complete. It was at least what would today be called the first ‘arithmetic logic unit’ (ALU), now built into all computers. This was the first computer to use regenerative capacitor drum memory, not to be confused with Tauschek’s drum memory mentioned above. Regenerative capacitor memory uses individual capacitors to store memory bits. They are either charged or discharged to represent a 1 or 0. Because capacitors discharge with time, they constantly need to be ‘refreshed’, much like some other forms of memory (such as DRAM, to be discussed next month). The ABC computer had two drums that stored 1500 bits each (thirty 50-bit numbers) which rotated at 60 RPM; the capacitors were refreshed on every rotation. The ABC computer (if it is accepted as such) was the first computing machine to use flip-flop memory of the type described above by Eccles and Jordan. You can see a fascinating video about how the ABC works on YouTube – “The Atanasoff-Berry Computer In Operation” – https://youtu.be/ YyxGIbtMS9E 1943 The British Colossus code-breaking computer is regarded as the world’s first programmable digital computer (see Fig.5). It was the first device universally accepted as a computer to use the flip-flop design from Eccles and Jordan. The flip-flops were implemented with vacuum tubes as transistors had not yet been invented. They were used for counting and logical operations, as the computer had no memory except the paper tape loop mentioned earlier. 1945 The first programmable general-­purpose digital computer was ENIAC, used for artillery calculations by the US military. It started with 20 words of system memory, or about 80 bytes, in the form of accumulators. Extra data was stored on IBM punched cards; a 100-word magnetic core memory unit was added in 1953. ‘Words’ are of variable size for different computers. For ENIAC, a word was ten binary-coded decimal digits in length, at a time before eight-bit bytes were standardised. Most modern computers use 16-bit (two-byte), 32-bit (four-byte) or 64-bit (eight-byte) words. siliconchip.com.au Fig.11: a 256-bit Selectron tube. Source: https://w.wiki/5xRC (GNU FDL). Cathodes Selection Bars Collector Plate Storage Eyelets Mica Backplate Writing Plate Write Pulse Reading Plate Read Pulse Faraday Cage Output Grid Signal Out Phosphor Screen Glass Plate Fig.12: how the Selectron tube worked. The arrows near the bottom indicate the secondary emission of electrons that generate a pulse indicating a one-bit. In contrast, the arrows higher up and to the right indicate no secondary emission of electrons, indicating a zero-bit. Source: https://w.wiki/5xRD The original 20-word ENIAC memory used flip-flops in the form of a pair of triode valves. Ten flip-flops were joined to form a decade ‘ring counter’, capable of storing and adding numbers. A ring counter comprises a system of flip-flops and a shift register with the output of the last flip-flop fed to the first to make a ‘ring’. A PM (p for positive and m for negative) counter circuit was also used to store the sign of the number. One PM counter and 10 ring counters made up an accumulator. 1946 Development work on the Selectron tube (Fig.11 & Fig.12) was started by Jan A. Rajchman at RCA. This vacuum tube stored digital memory data in the form of electrostatic charges, similar to the Williams-­ Kilburn tube discussed next. The original design was for 4096 bits, but that was too difficult to build, so a 256-bit form was made. The device was never a commercial success; both it and the Williams-­ Kilburn were superseded by magnetic core memory, which was more reliable, cheaper and easier to manufacture. Australia's electronics magazine The basic principle of operation is shown in Fig.12. Electrons are emitted from the heated cathodes at the top of the diagram, like an electron gun but not a point source. Each cathode is surrounded by four selection bars, two each running in one direction and two at right angles to those. The selection bars adjacent to the cathode corresponding to the selected bit are activated to address a particular bit. Electrons move from the cathode through the collector plate and toward the storage area, which consists of eyelets (like those on some shoes but much smaller) embedded in a sheet of insulating mica with a metal backing called the writing plate. The eyelets are insulated by the mica sheet but capacitively coupled to the writing plate. By pulsing (or not) the writing plate at the same time as electrons are moving toward the selected eyelet storage location (as determined by the selection bars), the eyelet can either be charged or not, thus ‘writing’ the data to be stored. If the pulse is the same potential January 2023  17 Fig.13: a Williams-Kilburn tube from an IBM 701 at the Computer History Museum in Mountain View, California, USA. Source: https://w. wiki/5xRF (CC BY-SA 3.0). Fig.14: data in the form of dots and double dots written to a WilliamsKilburn CRT memory tube. The double dots are because a second dot has been drawn as part of the erase process. Source: https://w.wiki/5xRE 18 Silicon Chip as the collector plate, electrons will pass through the collector plate and charge the eyelet (downward-facing arrows on the left of the diagram). If the potential is the same as the cathode, electrons will be blocked and not charge the eyelet. Thus, the eyelet can be in one of two states. For reading the data out, electrons from the cathodes will either pass through an eyelet or be inhibited from passing through to the reading plate, depending on its charge state. By selecting an eyelet using the selection bars and pulsing the reading plate, the signal from the output grid will indicate whether it is charged. After passing through the reading plate, electrons go through holes in a Faraday cage and strike a phosphor screen. This causes the phosphor to glow, indicating the contents of individual memory locations (the eyelets) as well as passing secondary electrons to the output grid. For more information on how the Selectron worked, see the website: www.rcaselectron.com 1946 The Williams-Kilburn tube was patented in the UK and US in late 1946, 1947 and 1949. It was the first fully electronic (and thus high-speed) memory, using a CRT (cathode ray tube) for storage. The fact that CRTs were used this way was mentioned briefly in our article on Display Technology in the September 2022 issue (page 18, middle column; siliconchip. au/Article/15458). This type of memory was first used to run a computer program in 1948. Simply put, a Williams-Kilburn tube (Fig.13) stores memory on a CRT by writing a dot pattern representing the data to be stored (Fig.14). As with any CRT, the image has a certain persistence but eventually fades away. Therefore, it must constantly be ‘refreshed’ by each bit being periodically read and re-written (similar to DRAM). A small charge of static electricity appears above each dot which fades over a fraction of a second. It is this charge that gives the tube persistent storage. So writing a ‘one’ to the display involves steering the electron beam to a specific position and delivering electrons from the gun to allow the charge to build up. To write a zero, the charge at the dot must be neutralised. This is done by drawing a second adjacent dot (or line) Australia's electronics magazine because a negative halo is generated around each dot. This eliminates the positive charge of the first dot nearby. Reading the state of a bit is done with the aid of a thin metal plate on top of the viewing screen. The electron beam is steered to that location and energised, just like writing a ‘one’. If a ‘one’ was already present, there is no change in the charge at that location, so no current flows through that metal plate. But if there was previously a ‘zero’, writing the ‘one’ will cause a detectable current to flow. The Williams-Kilburn tube was susceptible to external influences, mainly from electric fields, so frequent adjustments were required for error-free operation. Some notable uses of the tube were the IBM 701, IBM’s first electronic digital computer from 1952. It had 72 3-inch Williams-Kilburn tubes, each having a capacity of 1024 bits, giving a total memory of 2048 words, each having 36 bits. The memory could optionally be expanded to 4096 words. Another use was MANIAC I (Mathematical Analyzer Numerical Integrator and Automatic Computer Model I; Fig.15) at the Los Alamos National Laboratory, which used 40 2-inch tubes to store 1024 40-bit numbers for hydrogen bomb calculations and it became fully operational in 1952. 1947 Frederick Viehe filed for US patent 2,992,414 for magnetic core memory (Fig.16) in 1947, although it wasn’t awarded until much later, in 1961. He filed another related patent in 1962 (US3264713), awarded in 1966. Magnetic core memory was the dominant form of computer memory from about 1955 to 1975. Incredibly, Viehe was a Los Angeles pavement inspector who played with magnetics as a hobby; he was not a professional scientist or engineer. IBM eventually purchased his patents. Core memory uses tiny toroids of magnetic material wired as simple transformers. By passing a current through wires that go through the toroid, it can be magnetised in one direction or the other, thus storing a bit of information. A sense wire passing through the core detects if the toroid has changed state. Reading the data (magnetic polarity) is a destructive process, causing the bit to be set to zero. To read a bit of data, an attempt is made to flip a siliconchip.com.au bit. Nothing happens if it is a zero; if it is a one, the toroid changes polarity, inducing a pulse in the sense line. The information is retained even when the power is turned off. A piece of magnetic core memory is one of the most desirable items in any collection of electronic ephemera. They are fine examples of delicate manual construction and are almost works of art. Other claimants to this invention were An Wang (1949; US patent 2,708,722 awarded in 1955), Jan Rajchman (1950) and Jay Forrester (1951); there were many ‘intellectual property’ disputes over it. In 1964, IBM paid MIT (where Jay Forrester worked) US$13 million for his patent, a substantial amount of money at the time. Core memory eventually obtained a volumetric density of about 900 bits per litre, and the cost went down from about $1 per bit to 1c per bit. The beginning of the end for core memory was when Intel introduced the 1103 DRAM IC in 1970, costing 1c per bit. While core memory is obsolete, computer memory is sometimes still referred to as “core”. A file containing the contents of memory from when a program was running is still often referred to as a “core dump”. 1947 J. P. Eckert and J. W. Mauchly applied for US patent 2,629,827 for the mercury delay line (and other forms of delay line) in 1947, awarded in 1953. The mercury delay line is a member of various delay-line-based memory devices. Delay line memories work by sending acoustic, electrical or light pulses, representing one bit, along a path. When a pulse gets to the end of the path, it has to be refreshed by reshaping and amplifying it. It is then recirculated. Such memory is accessed by waiting for the desired bit in the ‘pulse train’ to arrive at the read mechanism at a predictable time. The memory capacity is therefore determined by the length of the mechanism, the length of pulses and the speed of sound or similar in the medium. Mercury metal, a liquid at room temperature, was a common medium used in early computers. The resulting devices had a memory capacity of a few thousand bits. J. P. Eckert originally developed mercury delay lines to reduce clutter in radar return signals during WW2. siliconchip.com.au Fig.15: the aptly-named MANIAC I computer from 1952. The boxes on top of the main structure contain two-inch Williams-Kilburn CRTs used as memory. Source: https://permalink.lanl.gov/object/tr?what=info:lanl-repo/lareport/LAUR-83-5073 Fig.16: a 64 × 64 bit (4096 bits) array of ferrite core memory from 1961. This module measures 10.8cm × 10.8cm. The inset shows a detail of the ferrite cores with two address lines per bit. Source: https://w.wiki/5xRG (CC BY 2.5). Australia's electronics magazine January 2023  19 Fig.17: the “hot box” containing mercury delay line memory used in Australia’s CSIRAC computer. It was named that way because the delay lines had to be kept at 40°C. Source: https://collections.museumsvictoria. com.au/items/406411 (CC BY 4.0). Mercury was used in the delay lines because its acoustic impedance is similar to that of piezoelectric quartz acoustic transducers, thus minimising energy loss. The speed of sound is also very high in mercury compared to certain other media, meaning there is less time to wait for a pulse to arrive. Mercury delay lines were challenging to design due to the need to ensure there were no stray reflections. They were tricky to set up and maintain as they required very tight tolerances. The UNIVAC I computer mentioned below was an early computer that used mercury delay lines. 1949 CSIRAC was Australia’s first programmable digital computer and the fifth in the world. It is the oldest preserved first-generation computer. Its primary memory was a mercury delay line with a capacity of 768 20-bit words and a supplemental disk-like device of 4096-word capacity. Some of the delay lines were 10mm in diameter, 150cm long and a pulse took 960µs to go from one end to the other (Fig.17). You can see the computer on display at Scienceworks in Melbourne: siliconchip.au/link/abe2 1949 Jay Forrester had the idea to use core memory on the US Navy Whirlwind I computer; a 1024-word core memory was installed in 1951, replacing CRT memory. 1950 A US military version of the ERA 1101 computer (later renamed the UNIVAC 1101) was the first computer to store and run programs from electronically accessible memory, as opposed to instructions that were hard-wired or read from tape or cards. The military version was known as the ERA Atlas. 1951 Magnetic tape drives on computers were first used on the UNIVAC I computer (Fig.18). The drive unit was the Remington Rand UNISERVO I (Fig.19), which used half-inch wide metal tape (12.7mm) in 1200ft (366m) lengths. The metal tape and reels weighed 25lbs (11.3kg). The tape had six data channels plus one for parity and another for timing, and had a density of 128 bits per inch. Each tape could hold 1,440,000 seven-­bit characters. Later versions of these drives used plastic Mylar tape, which became the industry standard. The IBM standard for information formatting on tape was widely adopted. You can view an original UNIVAC I promotional video titled Fig.18: a bank of reel-to-reel tape drives (background) on a UNIVAC 1108II computer from a 1965 UNIVAC sales brochure. Source: http://s3data. computerhistory.org/brochures/sperryrand.univac1108ii.1965.102646105.pdf 20 Silicon Chip Australia's electronics magazine “Remington-­Rand Presents the Univac” at https://youtu.be/j2fURxbdIZs The Autumn 1964 issue of Martins Bank (UK) magazine reported that when data from one inch of paper tape was transferred to magnetic tape, it occupied 1/80th of an inch. The same bank reported that paper tapes were used for programming branch computer terminals as late as 1981 – see siliconchip.au/link/abht 1952 The concept of ferroelectric RAM was described in Dudley Buck’s master’s thesis. Bell Telephone Laboratories conducted some experiments on the concept in 1955, but it was not commercially available until the 1980s and 1990s (which will be described in more detail next month). Ferroelectricity is a property of certain materials with an electric polarisation state that can be reversed by applying an electric field. The state is kept even without the continued application of the electric field. The two states can be used to store binary information. 1952 The IBM 726 computer was introduced. It was the first computer to use magnetic particle coated plastic tape for storage (see Fig.20 and visit siliconchip.au/link/abhu). It could read or write 12,500 characters per second and each tape had a capacity of two million characters. The tape was about half an inch (12.7mm) wide and had six data tracks and a parity track. Fig.19: a promotional image of the UNISERVO I tape drive. Source: www.computer-history.info/Page4. dir/pages/Univac.dir/images/ MagTapeDrive.jpg siliconchip.com.au The storage density was 100 bits per inch, and tapes were up to 1200ft (366m) long. 1953 The first transistor computer originated at the University of Manchester. There were several experimental designs from 1953, culminating with a commercial design in 1956 by a Manchester company with the computer called the Metrovick 950. Only a small number were built. Early transistor computers may have used valves for the clock and other functions. Possibly the first fully-­ transistor computer was the Harwell CADET from 1955, but there were several other early claimants. Philco shipped commercial transistor computers, the S-1000 and S-2000, in 1958. The RCA 501 and the IBM 7070 are also from 1958. The TRADIC (for TRAnsistor DIgital Computer or TRansistorized Airborne DIgital Computer) was an early US transistor-based computer used on the B-52 bomber. It had 684 Bell Labs Type 1734 Type A cartridge transistors and 10,358 germanium point-contact diodes. It also used one valve in the power supply. Early transistor computers used drum memory or magnetic core memory, not transistor circuits as memory elements. However, transistors were used as registers for CPUs and amplifiers for magnetic core memory. Diodes were used in arrays as a form of ROM (read-only memory). 1955 The Konrad Zuse Z22 was the first commercial computer to use magnetic core memory (14 words of 38 bits) as well as magnetic drum memory (8192 38-bit words). It also used paper tape and had 600 vacuum tubes. 1957 Bell Labs introduced Twistor memory in 1957, first used in 1965. It comprised a piece of magnetic tape wrapped around a current-carrying wire and was similar in operation to magnetic core memory. It saw limited use; however, the ideas were incorporated into bubble memory (described next month). 1958 The Ferranti-Sirius magnetostrictive delay line was introduced (see Fig.21). It used the magnetostrictive effect whereby a material changes its shape in response to a magnetic field. A long coil of magnetostrictive material was fabricated, with an electromagnet at one end that induced a torsional wave (twist) in the wire that travelled down its length. Such torsional waves were more Fig.20: an IBM 726 magnetic tape unit, as used by the IBM 701 computer system. Source: https:// johnclaudielectronics.tumblr.com/ post/42914025003/ Fig.21: a magnetostrictive delay line. Source: https://w.wiki/5xRH (CC BY-SA 3.0). siliconchip.com.au Videos on punched tape storage ● A homemade paper tape reader: “Paper tape reader demo” at https:// youtu.be/w7_9BmthB10 ● Using paper tape with an Altair 8800, a microcomputer kit sold in 1974 and the first successful PC. The computer used in the demonstration is actually a modern clone. “Altair 8800 - Video #28 - High Speed Paper Tape Reader/Punch”: https://youtu.be/wALFrUd6Ttw Electronics Australia’s EDUC-8 was published about the same time and also supported paper tape (see siliconchip.com.au/Shop/3/1816). Australia's electronics magazine resistant to noise than the compressive waves used in mercury delay lines. A typical magnetostrictive delay line in a package about 30 × 30cm could hold about 1kbit of data. They were used through the 1960s in computers, video display terminals and some calculators. 1959 US patent 3,161,861 was filed by Kenneth Olsen, awarded in 1964, concerning magnetic core memory. 1962 CRAM (Card Random-Access Memory) was introduced by NCR – see Fig.22. It used cartridges containing 256 plastic cards with magnetic coatings, which together could hold 5.5MB. The device was mechanically complex but surprisingly successful, and was an alternative to magnetic tape until being surpassed by disk drives. 1963 Robert Norman at Fairchild patented static RAM (SRAM; US patent 3,562,721). It was faster than magnetic core memory and the logic circuitry used fewer components than January 2023  21 Table 2: generations of computers and technology used Generation Technology Approximate date range 1st Valves 1940 to 1956 2nd Transistors 1956 to 1963 3rd Integrated circuits 1964 to 1971 4th Microprocessors 1971 to present 5th Artificial intelligence Present and future for other forms of memory. It was used by IBM. According to the patent, “This invention provides a new switching circuit, particularly designed for a logic memory circuit, which achieves a substantial reduction in the number of components required.” 1964 The first 64-bit SRAM was designed by John Schmidt at Fairchild. 1965 We don’t have a precise date for the introduction of rope memory (Fig.23), but we know it was used in Apollo Guidance Computers by 1965. Rope memory was a form of core memory with its physical configuration altered to be much more compact than regular core memory (due to the woven core pattern), giving the higher storage density required for spaceborne computers, but was read-only memory. It was about 18 times more compact than regular core memory. It was Fig.22: a CRAM device from an NCR product brochure. Source: http://archive.computerhistory. org/resources/text/NCR/NCR. CRAM.1960.102646240.pdf 22 Silicon Chip used not only for storing data but also computer programs. Its operation was vastly more complicated than standard core memory, with multiple wires and bits per toroid and much larger toroids. It is described in a video titled “MIT Science Reporter—Computer for Apollo (1965)” at https://youtu.be/ ndvmFlg1WmE?t=1245 The process of making rope memory for the Apollo computers can be seen from 20:45 in that video. There is also a video about restoring an Apollo guidance computer, which has more details of its operation, titled “Apollo Guidance Computer Part 14: Bringing up fixed rope memory” at https://youtu.be/2qe4W_USweE Brek Martin has made a core rope memory simulator; the first video is at https://youtu.be/c-t2qyHOs7Y 1965 The Fixed Resistor-Card Memory was an experimental form of punched card. Information was stored by severing (or not) connections to an array of resistors on a cardboard or plastic card; it could be punched on existing punch-card machines. Next month After 1965, silicon-based memory started rapidly taking over from the technologies described so far. The second and final part of this series next month will pick up where this one left off, explaining how the semiconductor revolution radically changed computer memory up to the present day. If you haven’t already seen it, in preparation for the upcoming part two, you might want to read the series of articles on IC Fabrication technology in the June, July and August 2022 issues. They tie in with the computer memory technology revolution that SC came after 1965. Fig.23: a test sample of core rope memory for the Apollo Guidance Computer. Actual production examples were much more compact than this. Source: https://w.wiki/5xRJ (CC BY-SA 3.0). Australia's electronics magazine siliconchip.com.au