Go to On Line Documents, Go to Go to Antique Computer home page

This is G o o g l e's cache of http://www.inf.fu-berlin.de/~widiger/ICHC/papers/Huskey.html.
G o o g l e's cache is the snapshot that we took of the page as we crawled the web.
The page may have changed since that time. Click here for the current page without highlighting.
To link to or bookmark this page, use the following url: http://www.google.com/search?q=cache:W9aQeod7ETAC:www.inf.fu-berlin.de/~widiger/ICHC/papers/Huskey.html+Hardware+Components+and+Computer+Design+by+Harry+D.+Huskey,&hl=en&ie=UTF8


Google is not affiliated with the authors of this page nor responsible for its content.
These search terms have been highlighted: hardware components computer design harry huskey 


Hardware Components and Computer Design

Hardware Components and Computer Design
by
Harry D. Huskey


CONTENTS
1. Introduction
2. Available components - 1940
Triodes, Flip-flops, Dual grid tube, OR gates
3. The ENIAC design
4. New Components - 1940s
Memory development, Input/output equipment, Logic components
5. Computers based on these components
EDVAC, EDSAC, ACE, Others
6. Component Development - 1950s
Memory development, Logic components, Magnetic devices
7. Effect on Computer Designs - 1950s
8. Component Development - 1960s
9. Computer Systems - 1960s
10. Computers 1970 and Beyond

1. Introduction

This paper is an attempt to explain how available components effected the designs of electronic computers. Why were the early computers serial? Why were limited instruction sets used? What kinds of input/output equipment were used and why?

In presenting these ideas we shall take time to look in detail at the development of the ACE computers at the National Physical Laboratory [NPL] under Turing.

2. Available Components - 1940

Prior to 1940 "relays" were available to switch electricity in circuits. These involved mechanical motion which took substantial amounts of time - milliseconds.

The first electron tube with grid control, see fig. 1, was developed by Lee De Forest in about 1906. Interest in radio communication, stimulated by World war I [1914-1918], led to mass production of such tubes in the late 1930s [100 million tubes per year]. Multiple control grids were introduced in the early 1930s. These tubes promised switching times hundreds or thousands of times faster than relays. By 1940 radio broadcasting had led to tremendous production of electron tubes. The second world war increased the tempo of tube production and led to the use of pulse techniques, for example in radar.

The simplest electron tube, called a triode, is shown diagrammatically fig. 1. It has a cathode which likes to emit electrons when hot - hence, there is a heater to keep it hot. Next to the cathode is a structure called a grid. It controls the electric field near the cathode. If sufficiently negative relative to the cathode, the electrons cannot leave the cathode. If it is the same potential as the cathode then electrons do leave; some land on the grid but most pass through it. The grid must be able to maintain an electric field near the cathode but must not be such as to impede the flow of electrons. Thus, it is fragile. If it becomes positive relative to the cathode then it attracts the electrons causing it to heat and melt. That is the end of the tube!

Consider fig. 1 again. At the top is the "plate", which is connected outside the tube through a resistor R to a steady high voltage, say 200 volts. Being positive, the plate attracts any electrons which pass through the grid. They flow through the resistor to the 200 volts. However, as the electrons flow through the resistor R the voltage at the output drops. How much depends upon Ohm's law, E=I*R, where E is the potential across the resistor and I is the current produced by the electron flow in the tube. Typical values in a computer application might be 0.01 amperes [10 milliamps] through a resistor of 2000 ohms [2K]; the product I*R is 20 volts. Thus, the output drops from 200V to 180V.

Another important point about fig. 1, or any circuit, is that: although there is no explicit condenser between the output and ground there is capacity between the output wires and everything else. This capacity depends upon the surface area of the signal carrying wires, tube and socket components, etc. Thus, there is considerable incentive to use small wires and place elements close together.

Fig. 2 shows the resistor-capacitor equivalent circuit. Assume the input voltage starts at 0 volts and instantaneously changes to +10 volts. The current through the resistor I(t) and the output voltage E(t) are functions of time and obey the equation

E(t)=10*(1-e-t/R*C).

When t=R*C the voltage has reached about 63 per cent of the final value. The product RC is called the time-constant of the circuit; it takes about three time-constants before the output is near its final value [~95 per cent]. This means that any time a signal passes through an electron tube there will be a delay, up to 3 time-constants. Thus, to attain speed the circuit designer will wish to minimize the number of tubes through which the signal passes. Note that the input is the output of some other circuit and cannot change instantaneously, further contributing to the delay.

As a logical device the triode has the following properties: (1) it inverts the input [NEGATES], (2) with appropriate choice of resistance there is no loss in signal size [amplitude], (3) but, the output is nearly 200 volts above the level of the input! To cope with this change in level the ENIAC [Electronic Integrator and Computer, University of Pennsylvania, 1946] used many voltage levels ranging from large negative levels to high positive levels. Another solution to the voltage level problem is to use an output network as shown in fig. 3. As current flows from the top of R1 to the -200 volts at the bottom then Ohm's law says that the output voltage changes in proportion to the change in the plate voltage. For example, if R1=R2 and the plate is +200V then the output is 0 volts; if the plate is +180V then the output will be -10V.

What purpose does the condenser C [fig. 3] serve? Because of the time required to charge the condenser the output voltage will initially move in step with the plate voltage. Thus, a 20V decrease in the plate potential will immediately cause a drop of 20V in the output potential, which in three time-constants will come close to the value determined by the resistors R1 and R2.

Flip-flop Memory

Two circuits like fig. 3 can be connected as shown in fig. 4. The circuit [called a flip-flop] has two stable states, if one tube is conducting the other is off and it tends to stay that way. An input to the non-conducting tube will cause it to conduct turning the other tube off. This flip-flop stores one binary bit. Example voltages are shown for the left tube conducting.

If we wish the flip-flop to change its state quickly then we must minimize the stray capacities; thus, the output may be via a cathode follower and the input may use extra tubes. The output capacitance load can be reduced by using a cathode follower, fig.5. The output [cathode] is perhaps 5 to 10 volts above the input. The advantage is that the input may be a weak signal such as from the flip-flop; whereas the output signal is strong and could drive a number of circuits.

It is possible to apply a pulse to both sides of the flip-flop and cause it to change its state- a binary counter, fig. 6.

The dual grid tube or gate.

If the triode of fig. 1 has a screen and a second grid added, then the output [negative] occurs only if both grids are at cathode potential. The first grid [I1 of fig. 7] determines if electrons leave the cathode. If they leave then the second grid [I2] determines if they go to the plate or the screen. Since the screen is less robust than the plate designers favored supplying control signals to the second grid and pulses to the first grid - the arrangement was called a GATE.

We now move on to consider how these components can be used in computer design.

3. The ENIAC design.

When the ENIAC was started in 1944 the above describes the components available to build a computer.

The basic unit of the ENIAC was the accumulator, there were twenty in the ENIAC. It was to receive, store, and transmit ten digit signed numbers. The received number could replace, add to, or subtract from the contents of the accumulator.

Each decimal digit was stored in a RING of ten interconnected flip-flops. Circuitry was arranged so that each ring was stable with one flip-flop "on" and the others "off". Thus, storing a ten decimal digit number required 100 flip-flops. Numbers were transmitted as strings of 0 to 9 pulses over each of ten lines. The sign of the number required more flip-flops and another transmission line. With the control circuitry the accumulator had about 550 tubes.

There was a multiplier and a divider-square rooter. They each used a number of accumulators to accomplish their tasks. Certain accumulators communicated with punched card input and output equipment. A master-programmer initiated sequences of operations. In all there were about 18,000 tubes.

4. New Components - 1940s

Mercury delayline

The ENIAC had a very limited memory - 20 10-digit numbers. Before it was completed, several ideas for better memory were being explored. Some of the ENIAC team were experimenting storing numbers as pulse trains in mercury delay lines. Mercury lines had been used for precise measurement of time intervals in RADAR applications. Pulse trains would be applied to a piezo electric crystal at the end of a tube of mercury. The generated sound waves travel about 5 feet per millisecond. Another crystal at the receiving end would pick up the signal, it would be amplified, standardized and sent back to the input crystal. Operating at a megacycle, there would be about a thousand pulses in the line. A 32-pulse subtrain could represent a binary number which would have approximately the same range of values as 10-decimal digit numbers. Less than ten electronic tubes would provide the amplification and the gating needed. Thus, a mercury line [mercury in a tube] with crystals at the end, and a few tubes would store 32 32-bit binary numbers.

Whereas, in the ENIAC, units were operating in parallel transmitting and receiving numbers over multiple lines; the serial nature of the mercury lines led to a quite different structure. Furthermore, the complexity of arithmetic circuits using the components described above led to a single arithmetic unit shared by all the memory units.

In the summer of 1946 the ENIAC team gave a series of lectures on the ENIAC and on their plans for the EDVAC which was their delay line computer.

Cathode Ray Tube Memories

In W.W.II the military had investigated moving target detection using standard cathode ray tubes [similar to TV tubes]. The RADAR return from scanning a scene was "displayed" on the tube. Due to secondary emission the inner surface of the tube was charged varying amounts depending upon the "picture". If the scene was again scanned a short time later, there would be no difference in the charge pattern in the tube unless something had changed [moved]. This difference in charge could be detected by a screen on the outside face of the tube using capacity coupling.

Cathode ray tube memories were studied by the ENIAC team and by F. C. Williams at Manchester University in England. Williams developed a system using scan lines on the face of the tube. This led to a computer design in which numbers [words] were represented by serial trains of pulses, but access to words in memory was random. This is in contrast to delay lines where one must wait until the number appeared at the end of the line. Williams was able to have an operating computer in 1949.

At MIT special memory tubes were developed which were similar to the cathode ray tubes. These were used in Whirlwind I.

Magnetic drums

Although magnetic recording had been invented in 1898, it was 1947 before magnetic memory for computers received attention. One of the first examples, was a drum memory developed at the University of California, Berkeley. The drum consisted of an aluminum tube about eight inches in diameter by 20 inches long. The surface was sprayed with 3 mil coating of iron oxide, was vertically mounted and direct coupled to a 3600 rpm motor. 50 heads were mounted along the surface so that their magnetic gaps were about 1.5 mils above the surface. There were 10 tracks to the inch and each track stored 90 bits per inch, giving a memory of about 500,000 bits.

This memory held its information during power shut-off, and was much larger than the other proposed memories, equivalent to about 500 acoustic delay lines.

Input/output equipment

The ENIAC used punched card equipment developed for business data processing for input and output and the punched cards were used for shelf storage. Other computer projects used teletype equipment and punched paper tape.

Magnetic wire was developed by NBS [National Bureau of Standards] for the SEAC [Standards Eastern Automatic Computer]. Magnetic tape was developed by Raytheon for a Navy computer and by Eckert and Mauchly for their UNIVACs.

Logic components

Crystal diodes had been known since the 1920s. They were used as the detector in crystal radio sets. They consisted of a crystal and a "cat's whisker" wire touching the surface. The junction had a low forward resistance and a high back resistance. It converted [rectified] high frequency radio waves into low frequency sound waves.

The commercial version appeared in the late 1940s, using a germanium crystal, the same "cat's whisker", in a package smaller than a pencil and less than an inch long. It was delicate. If dropped two feet it was probably ruined. Heat sinks had to be used when soldering it or the contact would be damaged. Forward resistance was perhaps 200 ohms and back resistance 50K [50,000 ohms]. Ideally, one would like near zero forward resistance and infinite back resistance.

This component made possible more complex logic circuits. Two examples are shown in figs. 8 and 9. If inputs swing from -10V to 0V, then the grid in fig. 8 is high only if all  inputs are high - an AND circuit. The tube inverts the signal giving NOT an AND or NAND circuit. In fig. 9 the grid is high if any or all of the inputs are high - an OR circuit. The plate output gives a NOR signal. The decrease in physical size and complexity reduced stray capacity giving faster circuitry.

5. Computers based on these components

EDVAC

The team that built the ENIAC lost some of its senior people. Eckert and Mauchly started their own company to build UNIVACS. Goldstine and Burks went to Princeton to work on a computer for von Neumann. I went to the National Physical Laboratory [NPL] in England on a one year appointment to work on the Automatic Computing Engine [ACE] under Turing. Sharpless became the leader and continued the work on the EDVAC using mercury delay lines for memory. Then he went into business for himself,

obtained a lucrative patent on magnetic disks and built mercury delay-line memories for sale.

The EDVAC was divided into 1) a memory, 2) an arithmetic unit, 3)a control unit, and 4) an input/output unit, see fig. 10. Von Neumann wrote a report "First Draft of a Report

[ *missing image* ]

the EDVAC" [June 1945, University of Pennsylvania] describing this structure. Thus, it is now called the "von Neumann computer"; perhaps a better name would be "Eckert-von Neumann Computer".

The delay lines were 384 microseconds long and operated at a clock frequency of 1MHz. Words were 44 bits and an instruction consisted of a 4 bit OP code with four 10 bit addresses. Sixteen instructions kept the control simple and the 384 microsecond lines reduced the delay waiting for operands or instructions.

EDSAC

M. V. Wilkes of Cambridge University attended lectures at the University of Pennsylvania given by the ENIAC team. Following their EDVAC plans he returned to England and built the EDSAC. Using half of the clock frequency of the proposed EDVAC, 500KHz, he was able to have his computer operational by May, 1949.

Turing and the ACE

Turing used a general logical symbol shown in fig. 11.

[ *missing image* ]

There was an output "1" if at least m inputs were "1" and no inhibiting input was "1", otherwise the output was "0". The motivation for this symbol comes from the study of neurons in animal nervous systems. It is interesting to note that von Neumann used similar symbols and ideas in his 1945 EDVAC report. Turing's notation had little relation to possible hardware realizations of the circuits. It is a challenge to use the components described above to accomplish the effect of this symbol. Compare the flip-flop of fig. 4 with that of fig. 12.

We describe the work at NPL under Turing in greater detail since it represents a major effort to cope with the delays caused by waiting for operands or instructions in cyclic memories.

[ *missing image* ]

D. W. Davies reports in a 1972 reprint [NPL, Com. Sci. 57] entitled "A. M. Turing's original proposal for the development of an electronic computer" that after the end of W.W.II J. R. Womersley became head of the new Mathematics Division at NPL and, having met Turing, encouraged him to propose to Sir Charles Darwin [grandson of the famous Charles Darwin], director of NPL, the building of a stored program computer. His proposal consisted of 51 pages of single spaced text and 52 figures, and was presented to the NPL Executive Committee in February, 1946.

He proposed an "electronic calculator" with (1) a memory of 50 to 500 mercury tanks holding about 1000 digits [binary bits] each, (2) 50 quick reference storage units each holding about 32 binary digits, (3) an input organ capable of transferring "instructions and other material" into the computer from the outside world, (4) an output organ to transfer results out of the calculator, (5) a logical control to interpret the instructions, and (6) a central arithmetic part. He also mentions switching "trees", a megacycle clock, temperature control for the mercury lines, binary-decimal converters, a starting device and a power supply. The input and output was to be to and from punched cards.

Problems suitable for the computer might involve 5000 real numbers, would not take more than 100,000 times that of a human to solve, and the list of instructions would be comparable to an ordinary novel. He included ten sample problems, one of which was to compute "winning combinations" in chess to a depth of three moves on either side.

In January of 1947 there were three people working on the project, they were Turing, J. R. Wilkinson and Michael Woodger. I joined them on a one year visiting appointment.

The group was working on the logical design of a delay line computer. The distinctive feature of the design was the locating instructions and data so as to minimize wait times. In contrast, other computers with cyclic memories usually obtained their instructions from consecutive addresses, which meant that any instruction took at least one memory cycle.

Another distinctive feature of the ACE design was the dispersion of arithmetic and logical operations in the memory switching.

There was a source switch which connected any memory tank to a "highway" or "bus", which connected to a destination switch connecting, in turn, to the same array of memory tanks. As we shall see, some of the sources were logical combinations of two memory lines, and some destinations performed arithmetic operations. Also, some source destination combinations controlled input and output actions. This made control simple: set up the connections and decide when to transfer. In 1953 using the ACE principles,

I designed a magnetic drum computer. The result was manufactured by Bendix and called the G15. Delay lines were replaced by recirculating tracks on a drum. Each track held more than 100 words so the speed improvement by optimal location of data and instructions was more significant.

There are three ACE computers and the Bendix G15 that we shall consider. The three ACE computers are:

1. The ACE Test Assembly [ACE-TA] which I proposed that the group design and build to acquire experience and, perhaps, to have a usable computer. The project was approved in April, 1947.

2. In November, 1947, the computer design and construction activity was moved from the Mathematics Division to the Radio Division and the Test Assembly was re-designed to become the ACE Pilot Model [ACE-PM].

3. The ACE was the full scale computer that Turing wished to have built. The Mathematics group was working on this in January, 1947.
It was to have two source switches supplying data to a function box, which, in turn, connected to a destination switch.
The function box had 64 functions including logical, addition/subtraction, shifting, and discrimination. After the success of the ACE-PM and its commercial cousin the DEUCE, NPL went on to build the ACE.

The instructions for the four computers were structured as follows:

ACE-TA   ACE-PM   ACE   Bendix G15  
Bits Function Bits Function Bits Function Bits Function
1-2 Spare 1 Spare 1-5 Wait time W 1 Single/double
3-7 Source S 2-4 NI Source 6-11 A Source A 2-6 Destination
8-12 Destination D 5-9 Source 12-17 B Source B 7-11 Source
13-15 NI Source N 0-14 Destination 18-23 Function 12,13 Characteristic
16-18 Spare 15 Charcteristic 24-29 Destination D 14-20 NI Time N
19-24 Trans time T 16 Spare 30 Stop bit 21 Breakpoint
25 Spare 17-21 Wait 31-35 NI Source 22-28 Trans time T
26-31 Wait time W 22-24 Spare 36-40 No effect J 29 Block/item
32 Spare 25-29 Timing No. 41-45 NI Time    
    30-31 Spare 46-47 Charcteristic    
    32 GO digit 48 Not used    

Trans: Transfer, NI: Next Instruction

The TA control circuitry was quite simple, consisting of a means to set the S, D, and N switches, and a one word line with a half-adder which controlled the timing.

The four computers are essentially characterized by their source and destination tables, which are partially given below:

  ACE-TA ACE-PM ACE Bendix G15
SOURCES        
Delay lines/tracks 1-7,26 1-7 1-18 0-19
4 word lines     32-36 20-23
2 word lines DS   12,14,26 30-31,40,44 24-26
1 word lines TS 8-12,14 8-10 25-29,39,43 28
AND TS9&TS10 TS9&TS10   27=(20&21)v
        (~20&AR)
NOT ~TS11 ~TS9    
Pulses 1,16,32 1,17,32 1,6,12,24,36,48  
[Units: 1,W,Sign 1,W,Sign W,A, B, D, J,Sign]  
Precession 1via11 26via11 17,18 delayed char:Via AR
SPECIAL DESTINATIONS        
ADD +TS12 +TS16,+DS14 +DS40,+DS44 29,30
SUBTRACT   -TS16 -DS40,-DS44 ~char

The TA and PM are very similar, differing mainly in the addition of two word lines, DS. In the ACE the logic and most of the arithmetic features have been moved to the function box. All four computers have a precession capability wherein the items in a list in a line may be delayed one word time relative to the rest of the memory.

Other Computers

F. C. Williams at Manchester University was developing a computer using cathode ray tubes [CRT]. His work lead to the Ferranti I.

The Institute for Advanced Study [IAS] under von Neumann was designing a parallel computer expecting to use a memory tube developed by RCA.

The National Bureau of Standards was building the SEAC [similar to the EDVAC] in Washington and the SWAC, a parallel CRT computer, at the Institute for Numerical Analysis [INA] on the campus at UCLA.

MIT was building Whirlwind, a 16 bit parallel computer using special CRTs.

Others were designing and building computers with magnetic drum memory.

The following table [derived from data in the Encyclopedia of Computer Science, 3rd Edition, 1993] summarizes the characteristics of the delayline and CRT computers.

Computer Location Memory[wds] Wd Length Address Max Access No. Diodes
ACE-PM NPL DL 360 32 bits - 1.0 ms -
EDVAC Army DL 1024 44 4 0.38 ms 10K
EDSAC Cambridge DL 512 35 1 1.1 0
Ferranti I Manchester Wms 256 40 1 0.64 0
IAS Princeton Wms1024 40 1 0.025 0
SEAC NBS DL 512 45 3 0.38 15.8K
SWAC INA-LA Wms 256 36 4 0.016 3K
UNIVAC E&M DL 1000 12 char 1 0.40 18K
Whirlwind MIT ES 256 16 1 0.016 22K

This table shows: 1) that the CRT memories were faster than the delay line memories, and 2) there was some doubts about using germanium diodes. The small memories led to magnetic drum auxiliary memories.

6. Component Development - 1950s

Memory development

The magnetic core became the memory element of choice in the 1950s. It was a very small ring of ferrite which had a hysteresis curve similar to that shown in fig. 13. If a

current of I flows through the horizontal and vertical select wires the remanance moves to H [fig. 13] and to F when the current ceases. Similarly, if both currents are -I then it moves to D then to B when current ceases. With pulse current I in one of the select wires the remanence moves from F to G and back to F, or from B to A and back to B, depending upon the initial state of the core. The point F way correspond to a "1" and point B to a "0".

[ *missing image* ]

The cores may be arranged in a square array, say 32 by 32, with such an array for each bit in the computer word. If a selected core receives current I in both select lines it will move from F to H to F or from B-A-H-F. The movement F-H-F represents little change in magnetism so there is no induced signal in the sense wire, on the other hand B-A-H-F

gives an induced signal in the sense wire. Thus, sending current I through particular horizontal and vertical select wires in all the planes produces a representation on the sense wires of the binary number at that location. Unfortunately, it erases the word, so circuitry must be provided to restore it.

A nice feature: memory is not lost when the power is turned off. Initially, the hysteresis loops are not as square as one would like. This lead to schemes requiring more than the three wires through each core. Speed depended upon the core being small, but this increased fabrication difficulties. Materials were improved, the loops became more square, but fabrication remained a problem.

Logic components

Soon the germanium in diodes was replaced by silicon, the junction was made more rugged, back and forward resistances were improved. For example, in a junction diode the reverse current was less than 20 microamperes and the forward current at 0.8 volts bias was, perhaps, 50 milliamperes. All computers made in the 1950s used silicon diodes.

The first transistor, a point contact using germanium, was developed in 1947. As in diodes the technology soon moved to silicon and to junction devices. Originally, transistors more or less replaced electronic tubes in similar circuit configurations, see fig. 14. This reduced voltage swings to, perhaps, five volts and there was no heater structure.

Thus, there was a tremendous reduction in power, components could be closer together reducing stray capacity and increasing speed, and reducing the need for air conditioning.

Printed Circuits

The cost of connecting circuit components [wiring] and the cost of checking the result, led to the development of printed circuit boards. A circuit diagram was printed on a copper clad substrate. The board was then etched to remove the copper not protected by the printing. Components were placed on the board and it was dip soldered. The wiring on one side of the board must be one-layer [no cross overs]. With through-the-board connections both sides of the board was used giving two layers of wiring. If actual cross-overs were required than a wire jumper was used.

Magnetic tapes, magnetic drums and magnetic disks

The typical magnetic tape in the 1950s was 2400 feet long and 1/2 inch wide. Information was recorded 9 bits across the tape, perhaps an 8 bit byte and a parity bit. At 1600 bits to the inch a tape could hold over 40 million bytes, in practice, inter-record gaps would substantially reduce this number. Locally, the tape could be started and stopped very quickly. There was a slack arrangement and servo system driving the reels to match [catch up] with the fast movement at the read/write station. Data transfer rates were about a megabyte per second.

A drum might be 8 to 20 inches in diameter and up to 4 foot long. Data was recorded on a track around the drum, there was a read/write head for each track. At 3600 rpm the access time for an item might be as much as 17 milliseconds.

A typical disk had a number of platters, there was a pair of read/write heads for each platter, and the assembly of heads for all platters was moved in or out by a track seeking servo. Access might involve track seek time plus time until the data passed the heads.

Furthermore, disks rotated slower than drums but had much larger capacity.

7. Effect on Computer Design - 1950s

The low cost of magnetic drum memories led to "small" computers whose cost was in the $50,000 range.

Magnetic cores were more reliable than CRTs or mercury delay lines. There was no waiting for data as in the delay lines. Physically, they occupied less space, were much less sensitive to temperature than delay lines, and more reliable. Cores quickly became the memory component of choice for large computers.

Magnetic tape replaced punched cards for shelf storage of data. Punched cards remained the primary input medium. Tape drives became the primary auxiliary memory for computer systems.

Magnetic core memories and transistor logic brought significant improvement in reliability. People no longer talked about "mean time to failure".

Two classes of computers were designed, those oriented toward scientific computation and those toward data processing.

Since the main computer was expensive, organizational systems [computer centers] were developed to keep it busy. The user was no longer allowed to laboriously debug his program one instruction at a time.

The user was separated from the computer, he submitted his program and data to a queue, computer center staff moved it to the computer and placed printed output in the output area. The user picked it up and, if he was lucky, had his results. If unlucky, he hoped he had enough information to correct any errors, so he could resubmit it.

So called "assembly" languages were developed which made it easier for the user to write a program. Particularly, programs in assembly language were easier to correct. Labels for points in the program were introduced and more-or-less mnemonic names could be used for variables. The language processor translated the program into machine language.

User oriented languages such as FORTRAN were developed, the user was further removed from the hardware details.

8. Component Development - 1960s

Impurities could be introduced [doping] into a sheet [substrate] of polycrystalline silicon so as to change its electrical characteristics. For example, arsenic introduced in the silicon matrix produces an excess of electrons, boron produces a deficiency of electrons [called holes]. These are called, respectively, n-type and p-type silicon.

Next came the field effect transistor, see fig. 15. Usually the gate is metal and the

[ *missing image* ]

insulator is silicon dioxide, which gives the name metal oxide semiconductor or MOS. Control of electron flow between the source and drain depended upon the field established by the gate terminal. No control current was required beyond that needed to charge the condenser represented by the gate terminal. This reduced heat allowing components to be even closer together, giving more and faster logic units on a chip.

The many-step manufacturing process involves ion implantation, pattern transfer, etching, etc. Size of the individual elements depends upon purity of materials, cleanliness of environment, and stability of equipment.

9. Computer Systems - 1960s

The late 1960s saw memory chips storing 1000 bits. This was the beginning of the end for magnetic cores. The repetitive pattern of a memory chip simplified the mask design, naturally leading to its popularity with chip manufacturers.

The availability of memory chips and transistor circuitry led to cathode ray tube [CRT] displays. These were a welcomed replacement for printers and their stacks of paper.

The new hardware capability opened the way for new operating systems. Instead of the computer center with its queues, several CRTs were connected to the central computer and the operating system gave time slots to the active CRTs in order. This allowed the user to do small problems quickly. On-line storage using magnetic tapes or disks allowed him to submit large problems and to debug them on line. Outputs could be sampled and printing ordered when results were thought to be correct.

The success of FORTRAN led to other languages. COBOL [Common Business Oriented Language] for data processing applications was developed. BASIC, a small, highly portable [easily moved to new computers] language was developed on a time-sharing system. ALGOL [Algorithmic Language] was defined by an international committee. It was not commercially successful but is significant for introducing important ideas such as block structure and explicit declaration of variable types. Other languages with more specific areas of application were developed.

10. Computers 1970 and Beyond

The 1970s was the beginning of refinement: the number of tracks and the number of bits per inch on magnetic disks grew dramatically and the number of transistors on a chip doubled every two years.

The increasing number of transistors on a chip gave the possibility of building a CPU on a chip. The first of these was the 4004 microprocessor. It had 2300 MOS transistors on a tenth inch square chip. Since then there has been continuous improvement. Clock rates have gone up and up, main memory size has increased dramatically. Processor speeds, memory, and transmission rates have been increasing 60% per year [Moore's law]. Dudley Buck in the 1950s talked of vacuum triode components produced by photolithography [like integrated circuits]. This would make possible microcomputers like we have today. They would cost 25 cents. If they failed you threw them away! He was right about the result but wrong about the technology. Last year's $2000 personal computer costs less than $1000 this year.

Networking started in the 1970s. Now the majority of my correspondence is via e-mail and I buy my London theatre tickets via the World Wide Web [WWW]. There are more computers in my car than was in my state a generation ago!