• of" ...

• of" ...
"'"'
"
621.31'.'72 : 621.385.132 :
-~I'.
-10 .........
5.'.5
•
.'
(1) IN'I1lODUCDON
Propoeak for the coaitructioo of ~ diaital computiq
macbiMI haw nsu1ted ill a ~ 'for • DOW type bl atorqe
system. 'In order to, eltabJilb a ~ aaainst wbidl the
particular stora_ syttem dclcribediD the p&.ent paper may be
let, the introductory SectiGo of die .....
a delcriptioo
of the ayMem, oIllumbara to .. ..ct • propoeed computina
machiDea, tbe·,~ . . . . . . . - C"6 .... aystem. and a
statlme.tlt
dpea~:I!IMI~ ." .........'.,...'
•
•
,-=--
of"...
.', (1.1)' 'III!t.
"
."
' 1 : ' . ' .',
,
....., ',"" ., PI•• , ..
Tbe~'ol. . . . tIiIiIal~""the.,mecr­
~priinariIJ. ia the ~ of 8uitable
. electronic devices .haviaa tho lime' IlUIftber ()t·lta. as. the.'
lif)atimber or P<*l~ values .r a dilit. 10 that • oao-to-oDe com- .
iDa ItaDdpoinf,
~ spoo.deoQC
,
s_
..
. 1•. /1 .
may be ~blished between each
of the device
or die diait. The' number of values which a
aDd eadl value
::1...•,.-
;~3"
digit may take depends, of COUlICt Gl the system of numbers
tiled in the machine, and it· follows that it is advantageous to
dIooIe a system wbic:bcan.be~Dted, elcc;trically .with ease
and e(:ODomy. For theIe reasons the binary ,system of numbers
bas ~ popular in reoeat plans fqr. ~Jectronic digital computina mach ina. I, 1 although in the past the decimal system has
bOea \IIed.J.4«
'
~
'~
- 0
-.;
o
,
..
Systems of numbers may be derived from the common series:
a._.IJra-1
+ ... ,+ .... + a,bl + tJobO
(h)
'C
which repreaents all intelen with " siaamcant' figures. The
FII~ l;~;~.~llCli~.
dccinial systaD, (or example, is obtained jf b = 10, and -the a's
(a) SedeI: ~ .....;~9i;~:" ebaaDeI •.
are allowed any one of the vaI~ ,between, ,and including, 0
(.)PuaIId: pu1Ie 00'. . .· . . . . . . '
"
and 9. In tbebinarysyst.elJl b == 2 and" is.either 0 or l. The
decimal number 19. say, is d1cn1 X 24 + 0' x -2l + 0 x 2l + 1 , equivalea.t of the Dumber 19' being used as All example. In tbC
X 21 + 1 x 20 ~ the binary scale, which can ,be writtc.. 10011
Figure. video pulses are'uaed for. digital repreIeIltah9n. and at
with the least sianificant figure placed on the. right.
Fig. j (a) .the least signifiant figure, is plaQed on the left, so that
The decimal, or binary, POint is OD the immediate right of the time can be shown increasing from left to right in the conventiOnal:
term tJobO. The series can be CQIltinued to the right as follows: JDaJU1CI'.
.
Information may be represented ~'dynamicallyU by pulses~: ,i
which' only exist transiently, or "statically'· by d,c. ~~-, 'flip-flop circuits. which retain the information until tbeyale "~
Prof. WiWama &ad Mr. KJlbUl'll are at the UaiversitJ ot Mancbestcr.
p
( 81
.
.t
1
. r '. ' .
.!
. '.
,
T\.~$~'l\)£.R,
U,~\..~,
.
,;;
.........'.. ~d:f.i
~~~~__"~~_'_'_w~L'_\:'_)~'f'_'~~&~"~-_M__~N~t~t~·~~~\~~(~~t~~···.~it~;_H·~)j~~~'~**·~t*~b~~j~·C___
~'~*,__*__~'~~-~;m~;·~+wt~·~~'**.·:_'~·*~HM.L~·~~~$~"~~~¥~:'5~~~~~;b~~~'4~~;~~~~'<~*~';~~.h¥~;·~~·~*.·(~q;[email protected]~~1
WILLIAMS AND KILBURN: A STORAGE SYSTEM FOR
82
•
purposely reset to a standard condition. Dynamic information
may be converted into static information by, for example,
applying the pulses shown at (b) to five d.c. coupled. flip-flop
," circuits. The set of flip-flop circuits is called a "staticisor."
•
-.
\
( 1. 3) Required Properties of a Storage System
It should be stated at once that a computing machine cannot
·'think." It follows that the first step in setting up a problem
on a machine is to sub-divide it into a sequence of simple
arithmetic or logical operations externally (i.e. outside the
machine), and construct a ~ "table of instructions." Each
instruction in the table will, in general, require that an elementary
operation be performed on, or by, a number, i.e. a number will
be moved from one "address" of the machine to another. To
'every address a digit combination will be assigned •. so that an
instruction consists of two digit combinations, aRd is indistinguishable from a number in appearance. Instructions and
numbers, which are collectively termed "words." are therefore
similar, the only difference between them being their function
in the machine.
Since all the words applicable to a probl~m cannot be introduced into the machine simultaneously, they must be '~remem­
bered" during the loading period, and. until used, during the
solution. Further, temporary "memory" of some type must be
provided during each elementary computing operation. The
storage system provides this memory property of the machine.
The general opinion of mathematicians is that it will be
necessary to store approximately 3·2 X 105 binary digits, in the
form of lC4 words, with 32 digits per word.
If the two-valve flip-flop circuit. were used, 6' 4 x lOs thermionic valves would be required, which is clearly' impracticable
from the points of view of size. cost and probable reliability of the
equipment. Even in smaller machines the use of flip-flop circuits
would defeat, to sOme extent. the purpose of the change from
decimal to binary representation, since decimal representation
by ring counters in a machine of similar capacity would require
only three times more valves; and against this would have to be
set the expense of the conversion from the decimal to the binary
system, and vice versa when a binary machine is used.
Recently developed two-state devices, which are far less complex than existing two- or ten-state devices. are the main justification for the change from decimal to binary representation.
Further, they make digital computing machines with large
storage capacity a practical proposition.
Sufficient attention has been given to the memory property
to indicate that it is of primary importance, but in order to make
practical use of a store it must also be possible to insert, extract
or erase the remembered information. The insertion of information into a store has been ~alled ··writing.·' The extraction of
information from a store. "reading." does not imply that the
information is erased from a store, since it may be required at a
latcr time. "Erasing," of I:ourse, implies that informati,on is
era~d from a store. but in its prefcrahle form it is really a supersedtng process in that a word may be written into an occupied
address. deleting the word already there. This property increases
the etfective capa~lty of a storc. since new inlormation, such as
panial answers. may he '\\rincn ovcr information which has been
used, "ithout an intermedlah! crasurc proccs~.
T \) summarilc. a st(lrc mu"';{ ha\c the following propert:es:
(a) It mll ... t tx: pOSSible tll \\ rile a \hlrd qtl1\:kly into any
addrc!\s, ~Lh;h \\ riling superseding any word alread) present at
that address.
(/I) 1 he \\tll\.!s at all adJrc"~,,, not hcing \\filten in must he
rcmemhcreJ inddinitcly, ch.tnges occurring only as the result ( ) f
a detinite writmg prcx:ess: errors of I digit per million would be
fatal.
,
(c) 11 must'be possible to read the word in any address quickly
without erasing it, or .any other word.
(d) It must be possible to write into or read from any address
with absolute certainty. Reading from or writing into an
adjacent address in error, even if it occurred only once in a
million times, would be a serious disadvantage,
(e) The sfore must be capable of holding a very large number
of words (about 1(4) each comprising a number of digits. which
are either O's or I's.
For (0) and (c) the significance of "quickly" is related to the
time-scale upon which the machine as a whole is to work. The
longest operation of frequent occurrence is multiplication. If
this process occupies, say, 5 millisec wr~ting and reading should
occupy less than, say. 1 millisec, otherwise the computation will
be seriously retarded.
The paper describes an attempt to meet these requirement,
using charge storage on a cathode-ray-tube (c.r.t.) screen as the
memory mechanism.
./
(2) PHYSICAL BASIS OF THE STORAGE SYSTEM
Before describing the mechanism of digit storage, the arrangement of digits on the storage surface will first be mentioned
The digits are reprtsented 'by charge distributions which exi~l
o~ small areas of a c.r.t. screen. the charge distributions bein~
arranged in the form of a two-dimensional array. This array IS
produced by a television type of raster, in which the digits of
a line. and the lines of the raster, are scanned sequentially, each
digit corresponding with a "picture element. to Typical displays
are shown in Fig. 2, which illustrates the appearance of the C,T.t.
(a)
'k'·
(h)
(,I)
If"
I "~I ,1 ..':01'
: 0":."" dl~IL'"
face \\hen 'itm'age is in rlllgrl.':-.... In Fig. 21n) there an! J2 lir.
of,12 digits, ILKh dIgit Il1LlY have one of two t,,'rnt-indicatcd hythe r;lttcrn Slll'\\1l )toreJ. A "signal" or "pick-t:
plate, nlOsi~tm~ of.t ~hl.'('{ llf In.:t;!1 toil, fir gau7e, e\ternal h.) t:
c.r. tube is d~)sdv attached III the t'a~:c of the tube isl.'l.' Fig.
Fach arca of thl.' ...aCt-'1l I" therefor\.' Capa(ltance courkd mt('
c~lch
i.-/3
IJ
100 MO. The voltase output from the amplifier is. then. I volt
per hundredth of a microamp of current ftowina
or from, the
pick-up plate. There is no plwe reversal in the ~fier, and
conventional current ftowing" from the pick-up plate to tbe
amplifier sives a positiVe output voltap.
It should be noted that this equipment can only detect rates of
C'hanp of surface chariC on the c.r.t. screen, so that thefoUowiaa
daaiptions of potential distribution on the screen are qualitati~
The absolute value of these distributions is not of primary
. importaDcc to the final ~toraae system.
to:
PICk - up plate
1 - - - - - 0 Output
' - -_ _...J
FII. 3.-DetJtion or lianaJa.
common duumel, as in the iconoscope. This method of detectina changes of charge on an insulatina surface has also been UIed
to determine the ICCOIJdary emission ratio of insulaton usina
pulse technique.'
.
Havin. formed a general picture of the rep...,.tation of digits
by a two-dimensional array of suitably cbarlCd aras, attention
wiD now be confined to the small area of the screen correspondinl
with a sinJle digit. The potential distributions existin. on thia
area with different types of electron bomhaJ"dlnent. and the
resuitin. video sipals that ate obtained from the -'p~filf'plaelt-'
are dac:ribed below. Referma:s to literatuJ'e6. 7 on the subject
of charp ·storage have been .included.
(2.1) FAaafpalcat
'J'be voltap level of the video sisnaIs is increased by connectina
the pick-up plate to the input of a suitable amplifier as shown in
9
Fia. 3. The equivalent input circuit of the amplifier is shown in
Fia. 4(41) where I. represents the sianal current due to electrons
•
•
To am !sfler
c
(2.2) Pot6iIdaI I>JIbiIJutbI db Steady SiIIIIe Spot
In a c.r. tube which has its deftector plates; internal con-
ductive coatiq, and first and third anodes all connected to earth
potential. and its grid, cathode and focus electrodes connected
in a normal manner with respect to a neptiyc poteDtial (say
- 2 000 volts), the inner surface of the ICReD will also be a'
earth potential, because it is in contact by Ieaka,e resistance witb~­
the internal conductive coating. ~ia ~~ &bat no beam
current has been preSent for some tlrne.·~ In the typea of
co~ c.r. tube invcsti~~ (CV~097 and CVl13I).~
relation between secondary cmtsslon ratio of the tcrecn material
and primary electron velocfty is of the form shown in Fil· S.·
ur~
.€
I>nmary
eltctr~n
Wloclty
'tc. 5.-5econdary cmlMion ratio
as a function of primary eIeruoa
velocity.
At points of operation such as A, the secondary emission nahO
80 is arcater than unity. This is true for all primary ~tia
in the ranae 1 000 volts to 3 000 volts at least. It follows that if.
r
opcratina under such conditions. the electron beam is switdlod
on and falls steadily on a single spot on the c.r. t. acreaj,the
number of secondary electrons lcavilll the spot and movina
towards the electrode ~mbly, will exceed the number of primary
electrons arriving at the spot. The resultina net loss ofneptivo
To am"ltf..'r
charlC causes the potential of the bombarded spot to become
positive, and its potential IS then higher than that of any electrode
in
the tube. Later secondary electrons will therefore be ejected
r
into a .retarding electric field. and those which have emissioD
velocities below that CQrrespondinl with the potential of the spoJ
will be returned to the screen. The electrons with low emisSioIi'
velocities will, in fact, return to the spot; those with hisher'
velocities, repelled by other electrons, will have time to acquire
Fla. 4.-Amplifier input circuit.
an additional component of velocity parallel to the screen surface
arrivilll at and leaving the screen surface; C,. the capacitance of and wiD return to the immediate vicinity of the spot. Expcrithe bombarded area of the screen to the pick-up plate; C, the ments indicate that. for times of bombardment le5S than 400 p.sec.
capacitance of the bombarded area oth~r than that to the pick-up the screen is substantially unaffected at distances greater than a
plate, C,t the remaining stray capacitances to earth; '. the input. spot diameter from the centre of the spot. If tbe effective
resistance of the amplifier; and R, the ohmic resistance due to secondary current is defined as that caused by secondary electrons
the fact tbat the screen material is not a perfect insulator. The which leave the spot, and are not returned to it by the retarding
leakage time constant (ep + C)R is known to be of the order of field, the effect of the retarding field will be to reduce the effective
O· 2 sec, while very approximate values for (CL + C) and R are secondary emission ratio S. The potential of the spot will, in
0·002 p.p.F and lOS megohms respectively. The time constant fact, rise to a value Eo. thought to be about three volts. 6 such
C,r is less than 0·1 p.sec and r is approximately 1 000 ohms. that the retarding field causes the effective secondary emission
Since R > '. the signal voltage developed across, issobstantially ratio to be unity. Eo can be interpreted in terms l")f the '¥clocity
unaffected by R, which is therefore neglected. The pick-up plate distribution of the secondary electrons. 9 typically. indi~ated in
current appropriate to i, and ,flow'ing through ep is very nearly Fig. 6 as that point to the right of which the number of sec,}nJ.lry
C,;J(C -t Cp '. so that thetinJ"ut circuit may be reduced to·that electrons per unit time equals the primary current Ip. The
shown in Fig. 4(h). the final signal voltage being C"r;J(C +'Cp )' potential of the spot will now remain constant at Eo. but the
The amplifier. which is fuUy described in Appendix 9.1 bas a longer the spot is bombardeJ the larger IS the affected .lrl!a around
bandWidth of 2 Mc/s, and may be regarded as a resistance of it. The potential distnbution on the screen is ~ummJrizcJ 10
(0)
'';'
£. -13
•
•
j
•
Pnmary
wloClty
.... 6.-Velocity distribution of secondary electroua.
Fig. 7 in which increasing positive potential .is plotted in the
direction of the arrow, so that, us~ the analogy of gravitational
fie~d. electrons may be said to '~faW towards regions of positive
o
+1
-~~~-~-ov
U~ ---Eo
.... 'I.-Potential distribution with
•
'
a sinale spot-a "well."
potential. The depression in the distribution has been termed
a "well."
TIiC time taken to elItablish the potential Eo depends on the
, capacitance per unit area of the screeD, the current density of the
beam, the secondary emission ratio and the velocity distributiOn
of the aecondao' electrons. It follows that, with a given c.r. tube,
the time taken is inversely proportional to the current density.
Defocusing at constant beanJ..cunent, to double the spot dia·
meter, will increase the time scale by four, whilst doubling the
beam current, with constant spot size wiD hal\'C the time scale.
1be spot capacitance appean to be charged exponentially towards
Eo as shown in Fig. 8(a), and the electron beam may be regarded
(2.3) Effect
or lDtaiiaptina
the Beam
OD •
SiIIIIe Spot
With 'the spot held stationary as before, let the beam current
be switched on and off by applying a square waveform of fre ..
. Quency I kc/s to ~control grid of the c.r. tube. When the
beam is switched on for the first time, the potential distribution
shown in Fig. 7 will be established on the screen surface; but, sub..
sequentlyat instants of switching on the beam, substantially no
change will ha\'C occurred in this distribution, bccaU!e the
leakage tiJne..constant of the screen (C, + C)R is large compared
with a cycle of the grid modulating waveform. It follows that
. only a small change in surface charge is required at theae instants
to maintain the potential distribution, and conJequently the
output voltage of the amplifier in Fig. 3.' due to this dlanae. is
negligible. Howe\'Cl', when the beam is switched on, • cloud of
electrons in the secondary current, and in the beam itself: is
suddenly introduced in the vicinity of the pick-up plate. Thil is
equivalent to bringing a negative charae near to the pick-up
plate, and a transient current flows to the pla~ to supply the
required induced positive charge. 1be electron cloud is introduced extremely rapidly if the grid modulating square waw It
sharp, and the shape and time scale of the resulting amplifier
output pulle, whidl is negati\'C going, will be defined entirety by
the transient response of the amplifier. When the beam i •
switclled otT by the'square wave, the electr~n cloud is suddenl)
removed and an equal and opposite positive pulse appears at the
F'II. 9.-Electron cloud pUlses.
.
(G) Grid modlliatina waveform.
(11) Amplifier output .
amplifier output, as shown in Fig. 9. The amplitude of these
pulses increases with the beam current. TIle pulse waveforms
are completely independent of spot size.
(2.4) Interrupted Doable Spot
Two spots, as shown at 1 and 2 in Fig. JO(p), may be obtained
on ~ tube screen by applying to a deftector plate' a square
~
Fla. I.-Owaina of a bombfrded spot to equilibrium potential.
•
as an ohmic resistance to the first order of approximation, the
time constant formed by the spot capacitance and beam resistance
being of the order of I p.sec or less. The net current; flowing
to the spot is therefore of the form shown in Fig. 8(b), rising to
an initial value Ip(So - I) corresponding with the secondary
emission ratio So, and falling approximately in an exponential
manner to zero as the effective secondary emission ratio, 8,
approaches unity. The area under the curve is the charge
required to raise the spot capacitance through Eo volts, and
is therefore proportional to spot area. Since the capacitance of
the spot is almost entirely that to the pick-up plate, this current;
wilt also flow from the pick-up plate to supply the required bound
negative charge. The pid-up plate meaSures the rate of change
of charge over the whole screen surface, and this means that the
electrons which return to the screen around the spot will cause a
slight reduction in the plate current. A. further reduction. due
to another effect, will now be described.
,
CntKAII~
Rt6
~11JiuL
! I I~'I""'·
~
q
j
I'
(C)
MID
1
,-,
'1.+
~'... ~.~
tAv
Fig. to.-Interrupted double-spOt potential distributions.
waveform having half the frequency of the grid modulating
waveform, and phased relative to it as shown in Figs. J 1(0) and
11 (b).
If the spot is initially at I, the potential distribution will be as
previously described and is shown by the full line in Fig. JO(b).
The beam is now ~witched off, and then switched on again in
position 2, causing this spot to move rapidly positive and
generating the well shown dotted .
If the separation between the spot centres is greater than a
critical value (about l' 33 spot diameters), no other effect will
•
4- /3
•
USB WI1H BlNAltY-DlGlTAL COMPUfING MAOIINBS
_--f~}-~--r
1-- f-l_.~c:
'-' : ' '_I , :
W-P",..:--u... i 'J!
'('lR1
.
:
I
I
Ie)
tIJ
,
I
I
,
,
I
f'
I
I .
I
rRlsiticn I
:4-'...J~2
I
I
•
•
I
I
,
,
•
t
I ' , f
I
~
•
I
•
r
I
,
(i'J~J-L-~~~
Y'
i
~
: ,
.<f)-
I
I
I
,
:
I
I
l
:
~.x;
:
~
:
:
'
~
I
~
•
Pig. I I.-Interrupted double-spot wavefonns.
(a) Grid modulatiq waveform.
(~ Shift waveform.'
Output pulse due to i •.
Output pulse due to i r •
(~ Outl"UI pulse due to i~.
( ) Amplifier output voltale.
(c
(d
•
•
oocur, and at subsequent instants when the beam is switched on
in the positions 1,2, 1 and so on. this double well distribution will
be maintained by insignificant changes in surface charge, making
aood the small leakage, as. was the single well distribution
described in the previous Section. ConseqUently the amplifier
output waveform will again be as shown in Fig. 9(h).
If, however, the separation is less than the critical value, as
shown in Fig; 100e), some of the secondary electrons emitted
during the excavation of well 2 'will be attracted to well 1 and
begin to "refnr' it as at Fig. 10(d). 1be extent to which well 1
is refilled depends on the separation between the spots. and the
time for which weD 2 is bombarded, but it is probabiy never
completely refilled with the times of bombardment used in
practice, since' the fuller it gets ~be less likely are secondaries
from well 2 to reach it. The panell refilling of well 1 causes a
potential distribution in position 1 corresponding to an effective
secondary emission ratio which is greater than unity, since for
unity ratio a well must be excavated to the depth Eo. Therefore
when the beam is switched off, moved back to position I. and
switched on again, well I is rapidly re-excavated to full depth,
whilst well 2 is panially refilled, producing the distribution shown
at Fig. 100e). This process of excavating one well and panially
filling the other can be repeated indefinitely. and. if the system is
symmetrical. the charge ejected from one well will equal in
magnitude that deposited in the other. since the charge ejected
was deposited during the previous half-cycle of operation. In
fact, if the precise electrons emitted in excavating one well went
immediately to the refilling of the other, no signal due to changes
in surface charge would be obtained. However, the excavation
process is much more rapid than the refilling process, as would be
expected from t~ fact that whereas all' emitted secondaries
emerge with velocities away from the well being excavated, less
than half or them have a component of velocity in the direction
9f the well being refilled, and many of these have velocities too
great to be attracted to the well, or to any part of the screen.
The amplifier output pulse~ at the instant of switching on the
beam, under th~ conditions will therefore be the sum of three
pulses. namely that due to excavating a partially filled well to
full depth. that due to partially filling the adjacent well, and the
negative pulse induced by the introduction of the electron cloud.
These effects will be considered separately.
The excavation of a panially tilled well to full depth establishes
an additional positive charge on the screen surface, which binds
an equal and(oPposite negative charge on the pick-Up plate.
This negative charge is produced by a current i.flowing from
the pick-up plate into the amplifier input circuit, and a positive
pulse is obtained at the amplifier output. The current i~ is simflar
to the current ; of Section 2.2, but its initial value corresponds
.5
to an effective secondary emisSion ratio less than 80, since the
excavated well was only pa~ially filled. Thil current is abo
, slightly modified by electrons retuminl to the 'ICfeCfl around the
well. With a perfect amplifier the output pulse, which is
indicated at Fig. 11 (e), would be a replica of i•.
The partial refilling of the adjacent well reduces the positive
~surface charge in that position. and releases an equal and opposite
bound charge from the pick-up plate. Heoce a current i, ftows
to the pick-up plate and a negative pulse is obtained at the output
of the amplifier. The areas under the i. and i, wavefonns are
equal, because the charges involved are equal. but i, has a lonaer
time scale and a smaller amplitude. 1be I, waveform is shown
in Fig. J 1(d), and its shape chanaes with beam cu.rrent and spot
size in the same manner as ; or i, (see Section 2.2).
The induced current ie which ftows because of the presence of
the electron cloud has been described previously, and produces
the output waveform shown again at Fig. I I(e).
The net output voltage of the amplifier is the sum of the three
waveforms (e), (d) and (e), Fig. 1I; and is typically as shown at (f),
though many variations are possible by adjusting brilliance and
focus. The net pulse at the instant of switchina on the beam
. can, in fact, be, made negative if the brilliance is sufficiently
increased, but it is not proposed to run the C.T. tube in this
condition.
(2. S) Separadoa. or DCMlbJe-Spot
It follows from the previous section, that both the amplitude
and sign of the pulse obtained when the beam is switchesl on.
with fixed beam current and focus, depend upon the separation
of the spots. This pulse is that marked X in Fig. 11 (1), and its
amplitude change as a function of the separation between spot
centres is summarized in Fig. 12. The pulse shown in Fig. 1..1(.)
j
~!l
~~
~ ~ llP'---:=---~-
fa~
~~40
2
~~~d~rlrr.-d~~~
Separation ,In spct diameters
Fig. 12.-Output pulse at beam switch-on as a function of double-spot
separation.
occurs about O' 2 (Lsec before those shown at (e) and (d). because
the current pulse in the pick.-up plate. which produces the
.' output waveform (e), is much larger and narrower than the
current pulses which produce waveforms (e) and (d), and the
output of the amplifier therefore responds to it with less delay
time. Hence, the pulse X is never quite zero at a'ny value of
spot separation, but may be a small negative pulse followed by
a small positive pulse. For this reason the positive and negative
amplitudes are plotted separately. In plotting Fig. II oldy the
amplitude of the pulses immediately following the instant of
beam switch-on is considered. The negative overshoot of the
pulse X due to partially retllling the adjacent well is ignored. as
is the positive pulse contributed by waveform (e) when the beam
. is switched off.
Referring to Fig. 12. if the separation is 7.ero, the conditions
are identical with those for a single spot and the pulse is negative,
as shown in Fig. ] 3(a). The pulses in the latter fi~ure are tracings
taken during the experiment from the face of a tuhe monitoring
the amplifier output. As the separation is gradually incrc:l,ed.
the negative pulse decrcases in amplitude :lnd is 31most zero with
a separation between centres oro· flYd. d being the diameter of the
spot. During this time. the po..;itive p.plsc amplitude is increasing,
the waveform being as inJic.lted in Fig. I J(h). The positive
L.-/3
WILLIAMS AND XD...BtJRN: A STORAGE SYSTEM FOR
•
(0)
6 ,
4
~
lIme.f'S
!I
i<:3
:~b ~
8
TIme. #AS
'.
(b)
Item (a) above indicate! a mechanism of writing a digit on a
storage spot, and item (c) a mechanism of reading it. From (b) it
is clear that the ~rent storage time is inadequate. This can' be
(c)
overcome by ananging that each stored digit is read and ~
20 O}46A~
written well within the inherent storage time, thus giving a new
Ttmt.J.IS
. start to the stored charge with all the leakage compensated. This
FII. 13.-Output pulac at beam switch-on.
. "reaeneration" process completely eliminates the commonly
(Cll Separation - zero or peater &baD 1 . 3Jd (critical).
quoted objection to digit storage by charge distribution, thaI
(6 Separation - O·l5tl.
leakaae will lead to "spreading" and mutual interference betweaI
(c SeparatioD - d to 1 . If•.
digits, it also has other advantages described later.
.
. puJIe continues to increase Up to a separation d, i.e. no overlap
Item (d) is important in that it sets a limit to the cl~ness With
betvtwD the spots, and then passes through a flat maximum which individual storage spots can be packed on the stOlql
between d and l' J6d. The pulse during this stage is shown in surface and hence influences the digit holding capacity of the
Fia. 13(e). The amplitude of the positive pulse falls oft' quite store; this factor is discussed next.
lharply towards zero with increased separation beyond this
point, while the negative pulse amplitude increases rapi~ly from (~1) Estimated Separation and Ammgemeot of Storage ~
ao until at l' 33d the output pulse is entirely. negative apin as
Since each storage unit of the type outlined requires a c.r. tube,
at Fig. 13(a). Separations greater than l' 33d, which has been amplifier and regenerating Ihcchanism, it is important ec0called critical, give no further change.
.
nomically to store as many digits as possible in each unit. It bas
. Six curves of which Fig. 12 is typical, were taken for various been shown that bombardment of the screen at a distaooe or
fiMd values' of beam current and focus. The negative pulse, more than 1· 33d from a storage spot has no influence on that
the amplitude of which gives a measure of beam current, was spot. It follows that the single storage spot considered so rar
varied in amplitude from 16 volts to 42 volts; and the spot can be surrounded by other storage spots, provided that the
diameter was varied from 1 mm to 2t mm. For all the curves, separation t.etween spot centres is greater than 1· 33d. For
the aitical separation was within the limits l' 28d to 1. 3&1. with present purposes a separation of 2d will be assumed (ace So.;.
the mean value I· 33d. The diameter of the spot was deduced tion 5-.3). Furthermore, each storage spot must also have
from the amount of shift required to move the spots from reserved an adjacent spot which can be bombarded to perform
coincidence to just touching. The difficulty of setting up the the "filling" process. 1be form and magnitude of this additional
latter condition visually may involve errors up to about ± S %. area depends on the detailed arrangement of the system, but in •
The critical distance has, then, no a~lute value. In the case simple case it may be considered as a second spot spaced d from
of two spots of equal area, it is equal to kd, where k = 1'33 for the storage spot. The whole storage element is then contained
the particular SCMen ma~rial investigated. (The separation in a rectangle d x 2d. Associated with each such element is a
experiment was performed with a CVI097 type of tUbe.) The "separation area" to provide clearance from other element'
fact that the critical distance increases linearly with spot diameter, The boundary of this separation area must, for safety, be at
indicates, as might be expected, that increasing linear dimensions least !d away from the boundary of the storage element proper.
has no effect. For the increased separation is compensated by The whole rectangle occupied by each digit is therefore 2d x 3d
the 'increased attraction on secondary electrons by the adjacent and it follows that an estimated area of 6(/2 is required per digit.
well, due to the increased spot area.
.
or 0'06 cm2 if d = I nun.
The constant k is determined by the screen material, and will
The screen of a c.r. tube is circular and a circular array of digits
depend on the velocity distribution of the secondary electrons, would give optimum use of the available area. This is difficult
for this determines the depth Eo of a well, and therefore influences to arrange and in any case assumes complete absence of plate
the attractive force due to the adjacent well. The secondary shadow. Accordingly a rectangular array has been. chosen. A
emission ratio, like the beam current, influences only the times ~in tube, which has an available area of 8 cm x 12 cm, should
taken to excavate one well and fill the adjacent'w~ll.·
therefore have accommodation for I 600 digits with the estimated
allocation of 0·06 cm2 per digit. Greater numbers should be(3) APPUCATION TO DIGIT STORAGE
come possible by improvement of focus, or by increase of tube
From the phenomena described above, the following state~nts screen area with given focus.
may be- made:
Bearing in mind that each digit is to be regenerated at frequent
(a) Either of two states of charge may be left at will at a given intervals, necessitating continuous scan of the whole array. the
spot on the c.r.t. face. These states are
method of setting out the array is to set up the digit~ in a series of
(i) a well of full depth, by bombarding the storage spot, ceasing spaced horizontal lines· as in television rasters, as mentioned
the bombardment and not bombarding any other spot in earlier [see Fig.2(a)]. Willi this arrangement the basic requin:the vicinity, or
ment is regenerative storage of a number of digits on a sin(EK
(ii) a partially filled well, by bombarding first a storage spot. . horizontallir..e, the array bemg developed from the line by shiftillt!
and then another spot in the vicinity before ceasing the line perpendicular to its length through 2d after cadl
horizontal sweep.
bombardment.
!f~
•
8
(b) Charge distributions will be maintained for a timo-a few
tenths of a second-depending on. surface leakage.
(c) Renewed bombardment [within the few tenths of a second
,noted under (b)) of the storage spot will give, at the instant of
recommencing bombardment, a negative signal from the amplifier
in case (a) (i), or a positive signal in Case (a) (U).
(d) Bombardment of spots displaced by more than 1, 33 spot
diameters from a given spot has no influence on the potential
distribution at that spot.
f'l '
\
I
i
L-/~
17
USE WI1H BlNARY-DlGlTAL COMPUI'ING MACHINES
•
Many sys~ deriwd from the properties stated in Section 3,
may be used· to regenerate the s~ information. Five such
oriJiDal systems have been tested, and although, for reasons given
later, it has been decided that one of these is outstaodin. at
present, a brief outline is given of the alternatives, because, with
further. development, this decision will be reconsidered. The
first four of these systems operate on the pulse obtained from
the amplifier when the electron beam is switched on to a spot, and
all use the principle that the sign of this pulse is positive or
negative depending on whether an adjacent spot within the critical
diStance has or has not been bombarded since the storage spot
was last bombarded. The fifth system operates on a slightly
different principle explained later.
(3.2) System I-Dot-Dash Display
Fig. 14(h) is a segment of a horizontal time-base waveform in
which short periods of constant voltage alternate with lonser
•
3d so that the dashes may be accommodated. The amplifier
output waveform is now as shown at Fig. 14(d). 1be pRdae
nature of this wavefonn will be explained later, and it is necessary
here to note only that the initial pulse when the beam is switched
on is positive. This is in accordance with note (a)(ii) of Scction 3.
because now there has been bombardment of spots in 'the vicinity
of the storage spot since the latter was last bombarded. This
bombardment took place during the previous sweep as the spot
moved away from position (i) towards position (ii) [Fig. l4(c)].
Dots and dashes thus correspond with states (a)(i) and (a)tii) of
Section ~ respectively and give rise to characteristic signals as
defined under (c) in that Section. Either dot or dash may be
written in at will by using either wavefonn (f) or (g) of Fia. 14
as an intensifying wavefonn. On a subsequent sweep dots will
be "read" as negative output pulses and dashes as positive output
pulses. irrelevant parts of the amplifier output waveform beiDa
discarded by using the "gating" waveform or strobe shown at
Fig. 14(1").
In order to make the system regenerative it is necessary to
cause dots to be re-written wherever dots are read, and dashes to
be re-written wherever dashes are read. That this procedure is
possible may be seen from the fact that whether a dot or a dash
is to be written the intensifying waveform is the same in the time
interval '0 to 'l' whereas the amplifier output waveform has
indicated which should be written well before 'l' during the
strobing interval 'I to '2' Hence if the intensifying sianal to
the c.r.t. grid is fed through a gate circuit which is controlled
by the strobed amplifier output in such a way that the intensificationis turned off at I) if the control signal is negative (or zero),
but is maintained until 14 if the control signal is positive, the
system will be regenerative in that it will immediately. re-write
everything it reads. This arrangement is shown in outline in
Fig. IS. Details ofa suitable gate cir~uit operated by the positi~'
control signal appear in Appendix 9.1 .
Itt bnHl<lllCt
• .Jwfonn
U:I>l1 hnnkll1C<.'
lI,wef0l'm
Fig. lS.-A regenerative storage system.
•
In practice it has been found possible to replace the special
time-base waveform, Fig. 14(11). by a simple linear time-bue,
provided the duration ratio of waveforms (g) and (f) is not less
than about 2·4 to I, and provided also that the sweep speed is
(a) Dot display.
(d Strobe.
(/) Dot brilliance waveform.
such that not more than o· 7 of a spot diameter is traversed during
(b~ Dot waveform.
(c Dash display.
(g) Dub brilliaDce waveform.
the short intensification period of I· 9~. The dots then
(d Dub waveform.
(It) Time-b... waveform.
appear as very short lines instead of true dots.
periods of constant rate of change. If a repetitive waveform of
The waveform of Fig. 14(d) iso'f considerable interest and will
this kind, containing, say, 32 such segments is used to deflect a now be analysed in some detail.
c.r.t. spot, which is intensified only during the periods of constant
Let a horizontal line on the c.r.t. screen be produced by
voltage, by applying wavefonn of Fig. 14(/) to the control grid applying the wavefonn (a) of Fig. 16 and its paraphased form to
of the tube, then a row of 32 dots will appear on the screen. the X plates of the tube. the grid modulating waveform being
Two of these are shown at Fig. 14(a). If the separation of thC phased as at Fig. l6(b). Electron cloud pulses shown at Fig. 16(c)
dots is in excess of 1 . 33d, each can be used independently as a will of course provide a part of the amplifier output. The
,storage spot, the beam being used to operate on each one in remainder of the output. shown at Fig. 16(d), is due to the
turn. TIle corresponding amplifier output wavefonn shown at following causes. When the beam is switched on initially, the
Fig. 14(b) goes negative at the instant of switching on the beam, positive well which is formed. is partially filled as the spot moves
as stated in Section 3[a(i)], since there has been no bombardment away from the beginning of the Jine. This happens in all
of spots in the vicinity between successive bombardments of the positions previously occupied by the spot. as the spot leaves them
ltorap spot. If the intensifying wavefonn is changed to behind, and moving trail of positive charge is formed beneath
Fig: 14(g) the dots on the C.r. tube will change into short lines or and behind the spot as indicated in Fig. J7(b). When the spot
"dashes" (see Fig. l4(c»). The initial dots are spaced by about reaches the end of the line, the beam is switched off, the trail of .
7
II
WJLI...IAMS AND m.BVRN: A STORAGE SYST£M FOR
•
*' +---...,....-.v----i'",
maximum amplitude of the initial positive pulse is aucb that the
trail of charae is completely established ~fore secoudary
electrooa..beain to destroy the remanent charae at the e:od of the
line. The currentsflowiDg from and to the pick-up plate to
produce the po5itive and anticipation pubea, respecti\'eJy, are
then entirely separated in time. In practice, it is found that little
1011 in amplitude 'of the pulse occurs if the lenlth of a dash i.
made IUCb tbat the aepuation between centres of the initial and
final spotl, which fonn tbe lateral boundaries of the dub, it
not_ than 1·7d. With this value the positi\'C and anticipation
puIIes are hePmioa to.'-'OI.Iesce as shown in Fia. 18.
.... l'.-AmpIifier output-pullc with a line dilPlay.
(~
n..........ont.
(6 Oriel -0+': t,. .."'-.
(c EJer:uoe cloud .,....
due 10 ~ 08 c.r.t. 1CNeIl.
..-.. .................
( rw..
(
U
.--.
Spot
-
FIe.• I.-Strobina 'of lianals.
(II) Sipab.
(') Strobe.
(c) SvobecI output.
(b)
By way of example, the display, and amplifier and Itrobed
outpUts appropriate to the decimal number 19, are shown in
Fi,. 19 (b, c and d). In a two-state device either state may be
(c)
•
oVdlSl +
(d)
.... 17.~PotentiaJ distributions with a liDa display.
. ~:~ Poteodal
::-J~I diatributioa: beem OD.
dilUtbution:
beuI~.
(d
(4) Subeequaat potCDtial clilcributioa: . . . . OD.
cbarae iI
left on the screen, and the potentia) distribution is
tben u indicated at Fig. 17(c). Now, when the beam is switched
OIl apin at the .beginning of the line. a trail of charge has to be
recreated, and this causes the initial positive pulle of Fig. 16(d).
Once tbe trail of clwp iI created, there is no net change of
duup on the c.r.l. ICReIl until the remanent charge at the' end
Fig. It.-Electronic binary representation of decimal number IY.
of the tiDe ia approached. Durina this period the amplifier output
(.) Amplifier output.
(II) Binary.
is zero, and the potential distribution is of the form shown in
(b) C.R.T. display.
(d) Strobed output.
Fig. 17(d). As the spot approaches within the critical distance
of the remanent charge, low-velocitY secondary 'electrons with
\ defined as representing a "0," the other state tepresenting a "I .
component velocities along the line begin to destroy the remanent
In the present paper the digit will be said to be "0:' when the
charge. Since, when the beam is Iwitched off the potential
potential distribution on the c.r.t. screen is the same as it wouk
distribution must again be as in Fig. 17(c), a quantity of surface·
, be if the" amplifier gave zero output and the gate circuit acte,,'
charge, equal in mapitude to the created trail of charge, . is
appropriately.
destroyed during this period. Hence a negative pulse, equal in
area to the initial positive pulse, appears in the amplifier output
(3.3) System 2: Dash-Dot Display
voltage, as shown in Fia. 16(d). This negative p~. which
This systdb is identical with sys~m'), except that the negati';'
anticipates the cause to which it is due, namely, the switching off
of the beam, has been called the anticipation" pulse. llMt·ReI.. pulse at beam switch-on operates a suitable gate circuit jnste;",
output of the amplifier is the sum of the waveforms (c) and (d) of the positive pulse. The positive pulse now corresponds h
of Fig. 16 and is shown at (e). If the length of the Jine is the digit "0" and the display is a dash as shown in Hg. 20Cc
dcc.raIed, the wavefonn of Fig. 16(e) becomes the waveform of The negative pulse shortens the dash to a dot, and correspond.
(
Fig. 14(d). ~ The theoretical minimum length of the li~ for to the digit "1."
66
•
..
L-/3
.,.
•
USE WI11I BINARY-DIGITAL COMPUI'ING MAaDNBS
~ lll~
t:)
1
~
I
f
t
t
f
~
'
j .1
Fig. lO.-Storaae-system displays.
(d) Dcfocul-roc:ua.
(to) Foc:us-def'ocus.
Ca) Binary..
(b) Dot-dash.
(C') Dash-dot.
(f) Anticipation.
(3.4) System 3: Defocus-Foras Display
An alternative method of achieving' the choice between a
positive or negative indication at beam switch-on i5 to apply the
waveform of Fi,. 21(b) to the focus el~trode Al of the c.r. tube.
t7) ~l.SU8I'fght.
. '
(b)
r.--, ; ~ :
I
8I.1ck
I
I
.L!-J :
f;x-U$
up
n : r.- o.rXU1
..J : ~ ; L;-J :
•
I
" ,
I
•
OUl
t
Fla. ll.--C.R.T. electrode waveforms for dcfocua-(ocus display.
•
(_) Grid JDOduI.tin• •aveform.
(b) A.2 ••wlorm.
If waveform of Fig. 21 (b) is phased relative to the grid modulating
waveform as shown, the result will be a defocuse~ spot which
suddenly becomes focused, as shown in Fig. 22(cl).
IbIV-
"'T
Fla. ll.-Potential distributions 'with focus-defocus display.
(a) Display.
•
(b) Well I.
If the system is operatc4 on the positive pulse, the gate circuit
of Section 3.2 is used. The only modification is to make the.
time-base pause from I =
to I = '. (Fig. 14). Horizontal
separation of the digits is achieved by allowing the timc-basc to
run down linearly from I = I. to I = I" when the beam current.
is always off. TIle spot is defocuscd from I == '0 to I == IJ and
focused (or blacked out) from I ;.. t) to I = I •.
The display appropriate to the decimal number 19 is shown
in Fia. 20(d).
'0
.~t-'r
!
89
(c) Well 2.
When the beam is switched on for the first time, well I shown
at Fig. 22(b) will, of course, be excavated by the defoe-used spot.
However~ when the spot is focused, the shaded area at Fig. 21(a)
will be partially filled by secondary electrons, producing the'
potential distribution well 2 shown at Fig. 22(c). At subsequent
instants of beam switch-on it will always be necessary to convert
well 2 into well I, and a net positive pulse will be obtained at
the amplifier output. If the c.r.t. beam is switched off before
it is focused, the focused spot will never be present, antt the
potential distribution is always well I. Once this distribution is
established. the output from the amplifier at beam SWitch-on
will be the negative pulse due only to the introduction of the
• electron cloud near to the pick-up plate. The sign of the output
pulse at beam switch-on is therefore po!\itive or negative, depend- .
ing on whether the spot is allowed to focus or not .
(3. S) System 4: Focus-Defocus Display
If the system in the previous section is operated by the nepti~
pulse at beam switch-on, in conjunction with the gate circwt
required in Section 3.3, the display will be as shown in
Fig. 20(£».
(3.6) System 5: Anticipation
Whenever the beam current is switched-off, a remanent du.rae
is left on the screen, and with a moving spot, an anticipatIon pul~ .
is obtained during tho next time-base sweep. This gives a
wamin'g that at some "later" instant during the previous sweep,
the beam was switched off. If the possible instants of swiichina
off the beam are predetermined by a square wave applied 10 •
gate circuit. which allows the beam to be switched off once only
after an anticipation pulse has been received, then the system is
regenerative. For, 01lClC established, a remanent charae w.D
cause the beam to be switched off at the same instant of C<Kh
successive sweep, and the charge will be reinstated each time.
The display is indicated in Fig. 20([).
(4) A COMPLETE STORAGE UNIT
Attention will now be confined to system I, which is summatized in Figs. 14-17 and 19. The remaining systems. which
operate satisfactorily on a single line of digits, have bec-n rcjcc1N;
systems 3, 4 and 5, because of the difficulty of maintaining similar
conditions of focus over the whole c.r.t. screen. when many linn
of digits are used; and system 2 because operation on the ~tfatl\oC:
amplifier output is not as satisfactory as operation 011 -the
positive pulse.
Horizontal spacing between the digits on the tube screen is
achieved by using a linear X-time-base waveform, generated a~
described in Appendix 9.4. Vertical spacing is achieved ~) U\lI\tc
a specialized Y shift generator described in outline below, anJ
in more detail in Appendix 9.5. Each horizontal line contains
32 digits (i.e. one word), occupies a distance of 10 cm on the c.r.t \
screen. and lasts for 272 ",sec. The blackout period is J4IL'Cl.
The raster has 32 lines, and at present occupies a vertical distance
of 8 em. A 10 em by 8 cm rectangle on the tube face therefore
contains I 024 digits or 32 words, Fig. 2 (see Section 5.3).
The type of YMshift generation used is intimately conne~tcJ
with the writing, reading and timing properties of the storage.
For not only is it necessary to scan the raster lines sequentially
with the object of regenerating the stored information; but it is
also essential to arrange that any line may be written in or read 011"
as soon as possible after the machine has given that instruction,
without waiting for that line to be scanned in the regeneration
sequence. A suitable circuit is now outlined.
(4.1) Y-Shift Generator
Fig. 23 is a schematic diagram of a Y -shift generator. which
. produces 32 ~qual-step changes of potential followed by a rapid
flyback. Along the top of this figure is a five-stage Sl.:ale-of-two
counter. each stage being triggered from the prevIous onc: the
first stage is ~nggered by the X time base blackout wavefoml.
The counter output waveforms are as shown in Fig. '!.~(b·fJ. If
theSe waveforms are added in the form 1 x (b) -+. 2 . (n + 4
·L-13
Jf
WILLIAMS AND JaLBl1IIN: A S'I'OIlAGB SYS'IEM FOR
•
TiJ1l('-be5e
black-out·
waw-form
[24(aj
.
PII. 23.-5imp1e Y-Ihift paeratDr.
n. Dumbera ia . . . . . lbaa
1lA<-)J ref" to abe "!3P
____ ~
o·~
_.............
•..••• -
••
- -•• - - -
. . ~ 01 ••~ ia Fi8. 24 .
., 111111111111111111' III 1.11 Hf nf n---' .-.--------.-oft' or write in • given tiDe 0DIy wheD its
tum came in the IC8Il
Tbia cliIability can be eliminated by alJocatinl alternate
lweepleof the time bale to 66scan" and ..action" pbaaea of thesystem. • Durina ICaD pbaaes, the raster linea are IC8IUled
scquendauy with the .ole object of reaenentiq the stored
information. Dwinc action phases any line in the ruter may
be ctw.n at wiD, and information read oft" or written in that
line. With this arranaement the lines of the raster are DOt
scanned sequentially but in the order 0·, n, I, n, 2. n ... 31, n•
,~.
cyde.
\If'
t)
d
~J
•
(f)
J
• ii1i1lifiliiiiilill
lbI JlIlIUUUU1JU1f
tI
dI
..,
•
FII. 14.-Waveforma of simple Y-shift generator.
~~
_ _out wa_Oral.
(6 Counter O.
.
(,.) Counter 3.
(.f) Counter 4.
(c
(6)
~
•
Couat«..
Aftlcn
V-shift (paraphue).
Coaater 2.
X (4) + 8 X (e) + 16 x (/), the desired waveform, Fig. 24(g)
results. With this waveform and its paraphased version applied
to theY-plates of the c.r. tube the single line display of Section 3.1
becomes a 32-line raster. Addition in the appropriate ratios is
performed by the circuit shown in Fig. 23, the operation of which
is described in Appendix 9.5. It should be noted, however, that
the counter waveforms, Fig. 24, (b)-(/),are used to switch
careful "weighted" component shifts in and out of circuit, so
that the amplitude of the counter waveforms is -relatively
unimportant.
A scan of ~is type is entirely adequate for the purpose of
regenerating stored information, provided that the vertical
acparation of lines is adequate and the interval between scans
of a given line is sufficiently short compared with the inherent
memory time of the ~n. It would, however, be possible to read
I
~
I
Fla. 1S.-Waveforms of improved Y-shift generator.
(a) Blackout.
(b) Halver.
(d Counter 0;
(d) Counter I.
(to)
Counter 2.
(I) Counter 3.
td Counter 4.
(h)
Y -shift (parapbued).
~
0, n 1 ... where n is any chosen line not usually constant b\,(
varying as the computation proceeds. It is. however, she,,",
constant in the waveform of Fig. 25(h), which illustrates the rwt
• The fint liDe
or the raster i. called IiDc 0 fOl' convenielra.
L-I.3
01
-.
USE WITH IIINMY-DIGlTAL COMPUTING MACHINBS
Jomi
of ~ ¥-Ibif\ ~Yeform. To produce this waveform,
the blackout
or Fil. 2S(.) reeds a balver, wbidl is a
~orm
acaJo.of-two counter yiddina the waveform of Fil. 2S(b)~ 1bia
wawiorm then operates the counter chain of Fia. 23 which
cWiven the wa~orma of Fil. 2S(c-g). The balver waveform
abo performs electronic switdliDI operations audl that durina
the an pbue of the balver. waveforms of Fil. 2S(c-i) are
added, appropriately uwcijbted," as before, but durina the action
pbaae they are superseded by five control voltqes, which,
operatina tbroup the same "weiahtina" circuits as the counter
waveforms, take over ~ntrbl of ~ line saumcd. Technical
details of a suitable circUit will be found in Appendix 9.S. Here
, we need only note the importance of usina a common "weightinltt
circuit so that liDe II during action phases exActly coincides with
.
Though it miaht appear at first sight that this action/scan
aequence will halve the speed of Qllculation, in fact such is not
the case, because siDee it applies only to the store, computation
can proceed ,unhindered, and further, some time must be set
uide for chaDgina the fiw potentials seJcdinI the ~ion line.
1beIe potentials must not be chanacd whilst they are controlling
the shift,
diagonal ·traces will be Jenerated, but they can
conveniently be cbanaed during the scan pbuc.
In a machine the fiw control voltqes would be derived from
a statidlOr operated by a sequential five-pulse code. In the
experimental ltore tbey are in fact derived from a staticisor, but
the operatinl pulses are coincident pulaea on five separate lines
and may be made positiw or negative on each line by setting five
switches appropriately. The pulses are obtained from the balver
waveform and ocatt at the end of each action phase, so that the
staticisor can only chanae its state during scan phases when it
is not controllina the Y-shift. .
'.
An amplified discussion will be found in Appendix 9.S.
It should be noted at this point that although the Y -raster
aenerator, X-time-bue aenerator, and the ciraJitJ aenerating
such waveforms as the atrobe for the amplifier output, are
euential to the operation of a single storage unit. these circuits
ape common to aD further units. It is only necessary to repeat
the c.r. tube, amplifier and pte circuit, since all the c.r. tubes and
gate circuits will be operated in parallel to a common time scale.
For Jhis reason, a c.r. tube and its associated amplifier and gate
circuit will be called a "stOJ1!F unit."
_ Line It durinl scaa phaacs.
or
•
(4.2) Esperimeiigllaput Ullit
Infonnation will eventually be introduced into the store
•
via an input unit, which may take' many forms. An experimental method of input, far too laborioW! to be used in
practice and designed with the sole object of testing the storage
unit, is as follows.
..
1be bam is swiiched on 32 times during one X-time-base
• sweep, and vtith an empty store a negative pulse is obtained from
the amplifier each time, the display being 32 lines of 32 dots.
If a negative pulse of dash width is inserted into the gate circuit,
in such a manner as ·to give the same effect as a positive pulse
output from the amplifier (see "write'; input in Fig. 32) and
timed coincident with one of the instants at which the beam is
switched on, then a dash appears ~t the corresponding point of
the display and a 661" is inserted into the store. If. further, the
pulse is gaaeratcd during action phases only then the dash--will
appear only on the action line. 'The circuit supplying the pulses
isartanged in such a manner that pressing one of 32 keys,
arranged in the form of a typewnier, causes a pulse -to be
generated at the corresponding instant of switching on the beam.
A "I" is inserted in position k ofJine 1, by operating the switches
controlling the Y -shift staticisor so that line I is the action line,
and then pressing key k of the typewriter. Once inserte4 the
91
diait (t, t) is reaenerated, and nmembend by the store
iDdcfinjtely.
A "I" may be erased from the store by interruptiRa the •
repnerative loop during the time that the "1" would normaUy
be reaeoerated. 'Ibis is conveniently done by switchinl the
action line into the appropriate position and applyiDl a, nep.ti...e
pulse to the suppressoqpid of VI in the pte circuit (sec' Fig. 32).
Thus, although the control arid of this valve receives the positive
pulse from the amplifier, appropriate to a dasb, no anode currmt
will flOw in the valve, and the display will be converted into a dot.
It follows that the typewriter unit may either write or erase a
&61," and that a sin&le pole, double·throw switch may be WJCd to
lelect either of these alternatives.
If the input to the gate circuit from the amplifier is connected· to earth potential for a period longer than a raster period, then
the store will be filled (dashes everywhere), for pulses will be
applied to the control grid of V I durina ~ strobe Period.
If, on the other band, the output of the amplifier is disa:mDeCted
from the pte circuit for a period longer than the raster period.
the store will be emptied.
AND OPERATING DATA
It is now possible to fill or empty the store, and, by mama
three control systems, namely, the iD8ert-erase switch, the type-,
writer and the Y-shift staticisor, to change the sta'" or'di~l
(k, 1). Specific patterns. such as the one shown in Fig. 2(11), . ,
therefore be written into the store. in order to test the IIleI1I\iy
of the storaae unit. The pick-up plate prevents direct pboco..
graphy, and a monitor c.r. tube wired in paratJel, with tile
storage c.r. tube was photographed. During the initial .....
many memory periods of between oI)C and two howl wac
(S) EXPERIMENTAL RESULTS
or
recorded.,
By adding a sixth unit to the Y -shift generator a 64-line raster - was produced, and the storage capacity doubled (:! 048 diaits).
The scan frequency was then about 2S cIs. 'The pattern shOwn
in Fig. 2 (b) was written into the store. and was "remembered"
for four hours before the equipment was switcbcd off. it is
interesting to note that in this time approximately 7· S x lOt
opportunities occurred for a change in the pattern to take 'place,
due to possible spurious pulses occurring during strobe periods.
The existence of this multitude of opportunities for change may
be regarded as a disadvantage of frequent regeneration. There
is, however, a very great corresponding advantage. In any
system where the position of a storage element is defined in
terms of two multi· valued deflecting potentials (or currents),
subsequent recovery of the stored infonnation depends on aCcurate correspondence between geometrical position and deflecting,potentials and also on accurate reproduction of the deftectini
potentials themselves. Both these requirements are much more
easily met when the interval between storage and recovery is very
short, since long-period drifts in supply voltages. or~. tube
sensitivity, result only in a drift of the pattern as a whole without
damage to it, provided the drift between regenerations is small
compared with the spot size.
The factors influencing the operating conditions used to obtain
these results will now be described. It will be seen that a compromise has been made between many conflicting factors.
(S. 1) Primary Electron Velocity
From Section 2:2 it follows that the third-anode voltage 9f q.e
c.r. tube should. be such that the primary electron velocity
corresponds with a screen secondary emissian ratio greater than
unity. With the screens used in co~1 ~.r. tubes a further
limitation is irnp6sed, and this is now discussed •
The signal output from the amplifier normally produced by
scanning aline on the c:r.t. screen is shown in Fig. J6(e). How-
/1
...... ~,..
.
",,-
_ ·_·...,L~
. £. -13
"
•
•
.AcceIInl. . . . '•
..... 2I.-Vari8daa
or ICI'em voltap.
~ ....... WiLda.
.
, •
.
7•
the iIP"ed~ aiaDals due to aIuI or ........ ..... • .....
tbe ICCIIeratiDa voltqe ia reduced to ........ COIl..,. UltIna
with tile point B in Fi,. 27. . . . . . . . dMI wi.... t .... ",!I . '
potcotial barrier. The .....,ailude 01 Chc ~ due CD ..,~..
~ w.riy as the ~tiaa voIta_ is deerE - tends to aro u the acceIeratinl voltaae approada zero. ,-.
results wen obtained with a CVl131 type 0( C.r. tube. wbidt -.... .a
soft iJasa eDveiope, and the particular imperfect_ ........ ".4tId
weM the Jarpat which occuned at &II) poeM 01 tht ~'
~
c.r. tube baa yet been found with more than OM \.4IIh1f\ ' ..... ,f
. imperfection. The number of aWl or impurity amper1a tKmS
with ratios areate!' than 0'1 (but less than o· 2) is about twenty
at 1 «JO volts.
With tbeIe restdts in mind it was decided to ope...............
c.r. tube at the comparatively low acceleral.R, Vi." ..... ut
1400 volts. The tube was not specially IC_~.
Nt
it had a hard alau- en~)ope, and there was a cubon trnprtf«..tioo in the screen. 'Ibis c.r. tube, which stored the infornwl ..",
for Fi.. 2, ~ been uaed for a period of three month. .• ,-, 00
difliculties ba~ arisen due to acn:en imperfections.
C''''
•
<5.2) Etl'ed 01 Spot Size (F--)
1'be ltoraae capaeity of a sinale c.r. tube is determined pnmarily
by tho ac:curacy of focus, and its uniformity over the UIed area ",'.
the tube tcreeD, it beinl neoctS&I'Y to choose Jess than opltJDUm
fOCUl at the centre' if bad defocusing at the corners IS to be
avoided. FOI' maximum capacity. the tube should be operated
at the mabest possible aaeleratina voltage. W it" present
~, die imperfections limit this voltaae to I 400, and it is
apparent that an attempt should be made to produce a more
• IawnioII poi8t anwoaiaalCfy •• I 2100 yolu. -
"
/~.
/
.
L -1.$
VIiB WD1I BlNARY-DlGrrAL COMPUTING MAQIINES
•
pcrf'ect eaeen, with the cooaequmt iDaeue in atorqe capacity.
TbiI is beina attempted.
'
Althouah, for CODItaDt beam density, the amplitude of the
positive sipalsobtaincd wIleD the beam is turDcd on will
dec:reue u the focus b improYed, DO dif6culty is anticipated,
because, u Ibown iftFia. 18, exc:eUeot IiaPW/noile ratio is
obtaiDed with the Pl'CIIDt spot . . (about I mm diameter).
u. of Jlipal current may therefore be c:ompcDIated by iDa'cued
amplifier pin. I
M shown in Fi&- 2(b), 2048 cliaits have been stored. the atea
occupiedonthestoralCtubebeina lS4anl(Han x, •• an). In
Fia- 2(a), the area occupied by I 024 diaits is 64 anl (10 em
x 6,4 em). The deterioration in the uriifonnity or focus over
the Iaraer area can be aeen by the fact that for satisfactory
operation-nealigible interfereoce between digit areu-the ara
ocwpied by 2 048 digits is more than twice that occupied by
I 024 diaits.
(S.3) Adioa LIIIe , ......... 01 Stonae
The conception
•
ea.-:ItJ
or critical distance between two adjacent spots
93
write the information shown in Fig. 2 into the store, and provides
a close approximation to haviq no action line whatever. The
control of black-out by halwr waveform was removed and the
action liDo scanned during every action period. The extra
vertica11epar&UOD was then detennined. It was found that when
this was done the area occupied by the raster bad to be increased
by 20% (10 an x 8 an) for a 32-line raster. This is a very
stringent test, because it is incooceivable that it will be necessary
in practice to read or write into one particular line durina
32 CODICCUtive action periods. An increase of 20% in the area
of the 64-line raster (II em x 17 an) allowed any particular line
to be scanned eight times only durina one I'8$ter period. mainly because rapid deterioration in uniformity of focus oCcurred with
this large raster area.
VelodtJ ., tile FJectro. . . .
The time period assisned to each digit. whidl at present is
8· S p~, determines the final speed at which the computin.
machine will operate. This time period, which should therefore
be made as small as possible. depends upon the maximum rate
(S.4) I.atenI
cannot in fact be applied without reservation as it was in Sec- at which the electron beam can be allowed to move acroD the
'lion 3.1, to determine. the separation required between any two c.r.t. screen. It is apparent that the faster the electron beam
charpd areas of a raster for nealigible mutual interference. For moves, the less will be the amplitude of the positive pullC
the net electric field at a spot under electron bombardment is Ilt) obtained when the beam is switched on at the beginning of a dash.
lonaer due to a sinale adjacent charged area, but to all the For the amplitude of this pulse is determined by the amount of
remainin, cbarpd an:as of tbe raster. It follows that the refilling of the well which occurred durina the previous sweeP.
required separation is greater than the critical distance previously and the depth to which the weU is excavated during the present
defined, and this is particularly true for the ~ comprising the sweep. Since. as was explained in Section 2.4. refilJina rcqulm
edaes of the raster, whe~ the net lateral electric fields are more time than excavation. the speed of operation is limited
primarily by the refilling pr<X.'nS. Less refiOin, wiD occur n
patest.
The method adopted to determine the required separation the speed is increased, because the electron beam bomn.-rJs
'. experimentally is to adjust the focus, and separation between adjacent spots for less time. and lhe amplitude of the ~itive
areas, until the relevant positive signals obtained at any point pulse therefore decreases.
" of the raster, are not decreased by more than 5 % by mutual
An estimate of the speed at which some decrease in amphlude
may bcexpccted. can be derived from the double-spot exp.:rimenl.
._ terference, when adjacent dots are converted to dashes. It is
found to be sufficient to examine areas in the comers and centre From Fig. 12. the limits of separation between which maximum
r the 'raster only.
a:efilling of the adjacent well occun. are seen to be d and I - I bJ.
" Since a small amount of mutual interference between adjacent From Fig. 13(c). refilling of the adjacent well is seen to be almost
cbaracd areas is allowed to occur, the time of bombardment complete in 4 p~ at the operative value of beam current If
ot the areas becomes important. particularly when this time is therefore, the'diameter of the spot is 1 mm and the spot travels
only of the order of I~. (The time of bombardment of the at speeds greater than 0·04 mm/p.sec some loss of po.)itivc pulse
areas is determined by the velocity with which the spot travels amplitude is expected.
~ the ·c.r.t. screen. This velocity is discussed in the next
As the speed is increased, the pulse obtained when the
SectipP.) It may be seen, for instance, from ·Flg. 13 that, in the beam is switched on· at the beginning of a dash ch41naes
double-spot experiment, bombardment times less than 2 p.sec from a positive pulse to a negative pulse via stages similar
to those shown in Fig. 13 (c, b, and a). Fig. 29 10 in which
prod~ but little filling of the adjacent well. If, then, some
areas f the raster are scanned more frequently than others. both positive and ru!gative amplitudes- are plotted, indicale\
more- utual interference will occur in these regions. This will the change in the amplitude of the pulse as a fU.1ction
happen ~hen the ~tore is part, of a computing machine and it is of the reciprocal of the speed with which the electron
necessary to read or write into the store frequently during one beam moves across the c.r.t. screen. The point A ~orr\!s~nds
raster period, for under these circumstances cenain lines of the with the speed estimated above. If the speed is increas~ by a
raster will be scanned more often than other lines. _With a factor of ten, approximately, the point B is obtained. This IS
32-line raster, and the method of raster generation described in the operating point used at present and it corresponds with
Section 4 and Appendix 9.S, it is never possible to scan a 8' 5 ,...sec and 3' I-mm digit separation. If the speed is increased
particular line more than 33 times as often as the adjacent lines. by a further factor of ten (4 mm/j.LSCC), the positive pulse ceases
For in one raster-period this line would be scanned once during to exist. and theroutput pulse is the negative pulse due to the
electron cloud' effect. It is. of course, impossible h) operate the
a scan period, and 32 'times during action periods at most.
To determine the extra degree of separation between digit, storage system at this last speed because the pulse obtained when
areas required to nullify this effect, the following procedure ~as the beam is switched on will always be negative ind~p.:nd.!nt of
adopted. It was arranged for the c.r. tube to be normally whether a dot or a dash is being read. However, it mJY prove
blacked out - by halver waveform during action periods. A possible to operate the system at some point between Band
circuit triggered by the waveform of the last counter in the C, the speed corr~sponding with C being twice the present
Y -shift generator. and re-triggered by, halver waveform pro- speed.
It should be noted that Fig. 29 holds for particular values of
hibited this black-out for one action period during each raster
period. The action line was therefore scanned only twice as brilliance and focus only, the values chosen b.!ing the ones
often as the remaiqing lines. This arrangement was used to normally used in operation.
7
VOL. 96, PART ~II\
~
\
,
L.-13
•
AJtemaUWiIy, tile time sweep ..., . . .boIiabed aDd replaced
"by a deflectioo lianal .-rator wIaidl it at'I'aDIId to aweep ...
IPOt ·dilcoDtinUOUlly from IPICC to II*» OIl the c.r.l &oe by
means of appropriate X and Y voltqea. Provided the deIec:tDa
..-..or could be IWitched to any cIeIiIed co-ordinatel rapkIIJ
- any diait could be RICO\'eRd at .., time. T'be approprialt
co-ordinatea could probably be ____ted with the requiNd
8CIanCJ in about 20 fItICC.
•
(7) CONCLUSION AND ACKNOWLItDGMKNTS
.... 2f.-Variation with•sweep w:Iocity of ~ anduced wbaa the
...... illWitcbcd 011 . . eM .......... oIa dIIb.
.
U.5) ...... kill
Recapitulating, it mutt be possible to diatiqu.iIb' between
two typeI of sipal, wbidl COIiespond to two types
atond
charte dilCributioD. The area oa:upiId by the dicit &Na, and
the time . .iped to each dJaft are IDIde • smaU. u possible
CODIistent with die maiDteDaIM:e of this distinction.
·In the clot..... .,..aIm the c:boia: is between a poIitift or a
neptiYe sianaJ wbeD the c.r.t. beam is awitcbed 00, and the
amplitude of there lipalt . . . hdl upoD flld.ors dilculed ~
and in Section 2.
The liDe of immediate future deYelopment is dear. ImproYoment of focus aID' be attacked from the oormaI standpoints.
Final raults wfU depcDd 00 the production of a more perfect
1CRIeII, and the . . . . .t IDaeue in aaderatina wimp. If
this prowa difficult. KNCIII
the preaent quality, but free from
carboa, depoIieecl OD lOme Dulator (such u mica) with a higher
iDWftioD point than glass, may be 1UCCeIIfuI. Hisher storqe
capacity per c.r. tube will then be possible. When the maximum
ItOrqe per tube hal been oblaiDed the required storap capacity
can be built up by UliDI the appropriate number of tubes. AU
tobel would be KaJU'1ed syncbronously. but the actioo line
would be intensified only on the appropriate tube 10 that the
co-ordina", of an individual stored digit becomes k, I, t, where
t it the tube DUmber.
or
or
(6) ALTlRNA11VE SCANNING SYSTEMS
The acannina sys1lem So far described is most suitable for series
machina in wh.idl one complete word is stored 00 each line.
Machines deaiancd for the parallel mode of operation require
aU the digits of a word to be simultaneously available on different
wires. Workinl on a basis of 32 digits per word. 32 c.r. tubes
might be used, one digit of each word occupying one space on
each tube. With Uti! arrangement. the diaits occupying the
32nd space in each line would not be available for 272 p.tcc.
This tirnecan be reduced by splitting the scan into 8 colunms of
Fla. 30.-Alternative sCannina arranaemcnt.
•
short lines. each containing 4 digits, as shown in Fig. 30, arrangements being made to re,d any digit in any line at any time.
.. This would reduce the time to obtain. any digit to 34 ~.
It baa t..l demonstrated tbat .... II1JIIIt.. 01 diP" caD
be stored OIl the screen of an ordinary co..._c:iaI c.r. tube and
that the deYelopment of special tut. for .... purpose it wortb
punuiq and abouId lead to an iar ..III IIOrqe caPKitJ per
. c.r. tubc.
.
1'be authon wish to actn~ their iDdeMed _ to till
Olief Superintendent, T.R.E., 'for facilities pro~ided durina the
researdl, and to Prof. M. H. A. NewmaD. F.R.S.• and Mr. A. M.
Turin& O.B.B., for much helpful ~ or the matMa . '
requirement. of diaital computer storeI.
,(8) RQI'I'll&Naa!I
(I) VON NIVNANN, J.: Unpublished wdrt.
(2) TuaIMo, A. M.: Unpublished wort.
(3) WILKIB, M. V.:
F..niac-Hiab-SPeed EIecuoIuc: (.•
culatiDa ~ .. &ctrolrk ~, April , . ..,
19, No. 230.
(4) BuJta, A. W.:· ..Electronic ComPUtiaa CtrcWtI oIUW
Eniac," ~6 0/11w hull". of .RAtIio ~
1947, 35, p. 756.
(s) HIDtANN. W., and G!YIIt, K.: EkkttVcJw AIel .......
Tecltnik, 1940. 17, p. 1.
(6) McGa, J. G.: ··EIectronic Generation m TeIeviID
Sianals," "FJcctronics" edited by B. lo¥aI (............
Ltd.).
.
(7Y McCoMNEu.. R. A.: "Video Storage by Secondary em...
from Simple Mosaics." ho«141f1J (»{ ,,.
Rodio EngiM~I. 1947. 35, p. 12S.
(8) BaUINING, H.: "Sccondary Electron Emillion," ",."..
Teclurlcal RnIew, March 193', 3, p. Ikl.
(9) RUDBDO. E.: PItY$IC6 Rni~. 1936.50, p. 138.
(10) BoARDMAN. E.: Unpublished work.
(II) WILI.1AMS, F. C.: "Introduction to Ci.mJlt TechRiQ'" t
Radiolocation~" Journal I. E.E. , 1946, 9.~. Pw\ tilp. 303. (See Section~ 9.2-9.4.)
.-n.e
la,,;,. ."
(9) t-PPEND1CES
(9.1) The ADlpIifter
Each stage of the amplifier is separately screened. and -.
beater and h.t. supplies are fed to each staF throuah knt....
filtered leads to prevent h.f, oscillations. Tbese arnm...... are omitted from the circuit shown in Fia. 31.
The first tb.n=e stages. which are id~tical, are fed bad b
resistor of SOO kO connected between the anodc of V3' and f.,.
pid of VI' The feedback, reduces the cffect of micro,""",
voltages in Vito negligible proponions, and defines the '''''P''
voltage of V3 as O· 5;, volts, if i, is the signal current in n."roamperes provided by the pick-up plate. The fourth stage I" ,')ntrolled by its screen-grid voltage to give manual galn-COlifrol.
The anode lOad of the fifth stage is very much greater than the
, input impedance of the sixth stage, Consequently. almc". "n
the pulse current, delivered by V5, flows in the J3-kU t'cedh.Jd
resistor of stage 6. Stage 6 is d .c. fed back. as shown, to provkk
outputs with approximate d,c. le,'els of either + 5 'oIolla or
- IS volts.
1'1
£..
-''':>
USB WI11f BlNARY-DlGITAL COMPlJTlNG MAaDNa
•
-}&Ov
..... 31.-The amplifier circuit.
va-v, ~ CV1091.
V. - CV113.
The voltqe outpu& of the amplifia' is UJOI.wbal the ~ua1
pin is set 10 that the voltaac pia or tile last three Itqes is 200.
The dou~ aperi.meotl . . . paf'0IIDDd with tbia letting.
. (9.2) G.-. CIraIIt
Referencee wiD be made to Fia. 32.
•
The eWeet of this cin:uit is to provide the arid of the c.r. tubc
with narrow positiw pula. td ai~ a standard display of doll
•
300 V
•
Input from
. amplifier VI
\ <-ISV)1 O-O}L.'-'~:':':~:4-~~~"-"
Dl~
Fla. 33.---Waveforms appropriate to pte circuit.
D4~'~~~
D3
.
__4-_'"
0
Strobe Erase Dash Read Dot Writ.e
input input iRput input. input. input
(-lOY) (OV) tOY) (+4Y) (+4V) (+4Y)
• Fla. 31.-The pte circuit.
VI-V. - CVI09J.
conaponding to the digit '-0"; these pulses are made wider,
prodocina dash corresponding to the digit "I" if, and only
if, the circuit receives a positive pulse from the amplifier at
specified instants, the instants at which the beam is switched on.
The standard display is provided by narrow negative dQt pulses
[Fia- 33(d)] applied to the cathode of the diode 06, the cathode
beina biased positive with respect to its anode. The~pulses
cut off the control grid of V and the anode of VJ. w.hich was
bottomed, rises Quickly. in voltage until caught by the diode 07
at about + SO volts. The resultant anode waveform shown
dotted at Fig. 33(f) is cathode fonowed by V4' and applied to the
c.r.t. arid via a d.c. restoring circuit, whic:h defines the highest
voltage reached by that grid, as the voltage set up on the brilliance
control of the c.r. tube. Black-out of the X-time-ba~ recovery
sweeps is provided by the fact that the dot pulses .Are IllhJbited at
a
3:
•
\
\
\
(d) A,mplifter OUCIIUI.
(0)
Strobe.
(c) VI anode.
~
-,
(d) 'Dot wavefone.
(,) DbIa . . ~•.
(f) (Mpur
'e u.t.trid.
their source during the black-out ~iod This i! all() tr\ll of the
dash and strobe pulses.
~ valves VI and Vlt and their anociate4 diode!. are t'-'
true gate circuit. 1be amplifier output, ... JI, J~\~J. bla!Cd 10
- 15 volts, is fed to the grid of VJ only during the strobe period.
At all other times this is provcoWd by cond~ion of 0, 11k
strobe waveform is shown at Fig. 33(b). the strobe period being
a short period immediately after the beam is switched on. There
is normally no anode current in VI' Olnd the anode vo.taae IS
defined as + SO volts by the diode D 2• The anode waveform
Fig. 33(c) has a negative pulse for every pc:xiti\'C pulse c1ehverw
hy the ainplifier during a strobe period. "The negatl\e pulses are
cathode followed by V 2 via the diode 0 •• and applied to the
control grid of V 3' The upper \oltagc limit of the control gnd
of V2 is defined as 0 volts by conductJOn of D. and OJ. and its
lower limit is defined a!-· 15 \olts, by conduction of Ds. . The
cathode orv2 wiU therefore ~wing In voltage between the approxImate limits +- 3 volts and - J ~ volts, which are sufficient to cause
full anode current, or no anode current. respectively, in V J . The
condenser taken from the control grid of V1 to earth prevents
the grid changing its voltage unJess it is driven. The grid will
therefore remain at - 15 volts for a period. the dash period,
determined by the waveform of Fig. 33(e) applied to the anode
of Ds. It will then be driven to 0 volts and will remain there
L-/3
"
UDtil it naiwI aDOtber aepti\le puIIe (rom tho anode of V ••
•
•
18 IboM the practical equiYaJea1l or tho idealized wa~
forma or Fia. 33 (G. b"and c).
The action of the dn:uit iI IIJIIIIIIatizec as follows. If the
diapIay at a certain &pot 00 the Cor. tube was previously a dot, a
__dYe puIIe will be deIiwred by the amplifier duriq the Itrobe
period, wbal tbe IpOt II bombar~ apin. smce the controllrid
olV. is normaJJy c:ut-oJr, the.,..tiYe pul8e has n'o"oct. aDd the
. pte drc:uit it iDoperati\le. A dot is therefore produced ap¥l
by the dot wawfonn. D, and V J. The comspondini wawforms
ant Ibown by the dotted lines. in FiJ. 33. If the display wu
~ a dub, a poIitive pulIe from the ampijfier aiw. rill
to anode cuneIlt in VI. The resuItiDa neptiw pulse at the
bode orv•• takcI the arid 0( V2 to - IS volts when it rcmaiDa
until dri. . bKt to 0 volts by the dub wawform IICtina throuIh
D,. The arid
J it then:(ore cut off Initially by the dot wa\'eo
faim and beId off Cor a dub period by the cathode
V2'
npoduciDa the dub display.
A conWllieot "read" output for the storqe unit II dcriwd
hID the cathode V2- and it takes the form a aepti¥e puIIe
dash width for cadl Itored "I:' ExtcroaI information.
fepleMDted in this 1DIUIDeI', can be written into tho ItOrqe unit
by appIyiDa it to the c:atbode of D.. Each ucptivepulle extmds
a dot into a dub. When writina DeW iDf'ormation 0\IeI' old
information, it is abo DeCeSIaI'Y to conw:rt a dub into • dot.
ThiI II KbieYed by applyina a nepti\le waveform to the ~
lOr arid of V.' which cuts 08' the anode currcmt in VI duriq the
writiDa period, breaks the re,eaerati\le loop, and allows c0mpletely DeW information to be inIerted via D •.
PiI-
paUleS duriq the dot period. 'Ibis is achieYed by retumina the
tiJDe..bue arid leak to the dot waveform and d.c. restorina the
wawCorm with an inverted diode to a potential equal to the mean·
arid-poteotial during the sweep (Fia. 34). I>urioa the dot periOd
DO cumm 80Wl in R and the rate of sweep is tbaef'ore zero.
If E is the amplitude of the dot . .wform the rate of sweep at all
ott. times is EIRe.
Tbe sweep requinK1 for focus-defocus storaae pauses duriDa
the . . period, and il achieYed as above by usina the dash
wawiorm iDs&ead of the dot wa~.
. (9.5) Y-s1118 G . . . . . .
R.elei'aas
wiD be made to Fia. 23, which is a achematic:
diqram of the silnple Y -shift ____tor. A10na tho top of thiI
fipR is the fi...taae scaIo-of-two COUIlta', each I . beiDa
orv
trigered from the previous one: the first slap is trigered by the
or
X time-bue bIack-out wawform~ Each c:ouoter D (n =- 0, 1,2, 1
or 4) is associated with a triode Tn wb.ic:b bas a resistaDcc RJ2rt
in its cathode lead, and the cathode or the 'triode is connected
or
or
via a diode On to the arid of a pentode called the Y -shift val....
or
1be output of the Y-shift valw and its parapbascd version- are
applied to the Y-plates of the c.r. tube. 1be circuit is complCltid
by triodeI T. whOle cathodes are also connected to the raiItin
RJ2a. For the ~t it win be assumed that the CUl'I'alts ill
tbeae triodes are cut off by ueptiw voltaIC'S E,,: The outputs of
the counters 0 to 4 areu shown in Fig. 24 (~/), and if thae are
added toptber in the proportions I, 2, 4, 8,. 16 (Le. 2-- 1)
~Iy, then the resultant output is the step waveform,
FiJ. 24(r)~ Thia step wawform is therefore the output vol. . .
0( the parapbue of the Y-shift valve, since eadi time a triode
(9.3) 'Be Clock CIradt
T. is cut off by the ueptiw aOina half-cyclc oftbC waveform of
The clock circuit, wbidl produces the S-p.sec digit cycle, counter D, a CUI'l'eDt proportional to 2"/R flows into R throuab
comprises an LC OICiUator. squan:r, and cathode follower. The' D.. The V-shift valw operates as a fed-back addinacin:uit,t
ItrObe and dot and dash waveforms are produced from this square addiq contributioaa from R/2" whenever D. conducts. R' it
.:
..w aDd fed to the pte circuits from low impedance sources. chosen to giw suitable Y-shift. .
It follows Cram the above discussion that if the grids of the
Two pbantastron circuits in aeries, dividiDa four and nine
~,aretrigered by the dock waveform. Tbe outputs 0( triodes T. have neptive voltqes applied to them, Which. aR
the pbaDtutrous are used to produce a square wavefonn, which suflicient to cut off the valve currents, then the line of the ,.....
is positive for 4 dock periods. and negative for 32 dock periods. scanned by the time-bue can be mo.en at will by applYm.
'Ibis it the X-time-buc bllck-out waveform, and it is used to suitable voltqes E" to the triodes T;. For Ell can be chosen to
that Do either does or does not conduct (25 possibilities),.aad It
control tho X·tiD»-bue circuit and Y-shift pnerator.
The circuits UIed are well known, and JequUe no detailed D. conducts a contribution 2" is made to the line number. If
for example, with the convention that the first line in the ~
delcription.
is called line 0, it is desired to scan line 21, then E J and E) aft
. (9.4) X Tiaae-Bue QaIt
made positive and Eo, E" ·and E.. are made negative. Only thr
1bia circuit is a Miller time-bue followed by an anode- diodes DOt D" and D .. conduct and a Y-shift of21 (20 + 21 + 2.,.
foltower ciJaUt to provide the parapbue. A linear sweep is units is produced. It will.be observed that the line chosen b:-- '
produced startina at a potential defined by a diode.
operating the triodes T~ and the corresponding line of the raster
The alternative sweep for dot-dash storage, shown in Fia. l4(h) produced by the triodes T a are accurately the same, since thc'Y ..
both depend on the ~jstors R/2D and not on the triodes involved.
provided the triodes are actuated by sufficiently large potential"
O·Ol#,lP
1be requirement for prompt execution of an instruction h~
(!-u-_.........--....
reading or writing leads to the division of the raster operation
into the two phases called "scan" and "action," control beinl
.' Dot,
~
exercised by wavefonns applied to the grids of the triodes T.
w6'leform
and To.
1be modifications necessary to make the circuit of Fig. 2.\
confonn to this requirement are shown in Fig. l5.. Here
black-out wavefonn triggers a halver circuit, wfuch, in t
triggers the five-stage scale-of-two counter, the wavefo
Dot. waveform
volved being as shown in Fig. 25 (a-g). The halver ci it is
applied to I\. E~'
itself a scalc-of-two counter. The halver waveform is (led to
-4V f
each of the counter waveforms, and the' resulting w eforms.
Fig. 36, (a-e). are applied _ the .. ids of the triodes T fier being
rate
I·
a
•
Rat.elc~
.... 34.-GeDeration of time bale with pa~ durina dot period•
• Parapbue is ~..w or uodo-follower circuit.1
t RdCreoce II.·
.
9.2.
.
/6
.... -'-'
....
VSE WI'I1I BlNARY-DIGrrAL COMPVTING MAaDNBS
•
To V-shift,
------?-~~------~~~~~~~~~~~~~
vAlve grid
Negat:.ive
pip valve
Fla. 35.-1mproved V-shift ItMrator.
The IlUJllben ia brKkets thus
{~{G)J
rerer to tbe waveforms a.bowa ill tbe c:orr.poadi...., _bercd Fipra.
~.c. restored to earth potential. In other words. the greatest
achieved by any of the wavefonns. Fig. 36. (a-e). is zero
/volts. Further. the wavefonns have sufficient amplitude to
~oltage
i
PftVent current flowing in the triodes T D except during those half
I cycles of the halver waveform during which they are at zero volts.
•
between scans of adjacent linea. The V-shift waw(onn will be
as shown at Fia. 25(11) except that here line 10 it belDl !CIccaed.
In order to seJect line 10, say, appropriate positive IT; conductina)
and negative (T~ non-ronducting) voltages, ~,.. IbouJJ be applild
I
Now. if it is assumed for the moment that the potentials ell
, applied to the triodes T~ are sufficiently negative to prevent
current flowing in T~. it will be seen that during the first scan
period (Fig. 36) current flows in all the triodes Ta. so that the
diodes D. do not conduct. the Y shift is zero and the electron
beam of the c.r. tube scans line O. But during the first action
period no current flows in any Ta , so all Do conduct, the Y shift
is at its maximum value and line 31 js select~. During the
second scan period only Do conducts, so that unit shift occurs
and line 1 is scanned. During the following action period all
the diodes Dn conduct again, so that line 31 is again selected.
It will be clear from such consideratipns that the whole raster of
32 lines will be scanned sequentially, line 31 being the action line
o
ScaT" I;
Scan 2
action I", {action 2
OYJ .
(0) p,J9j-- SCAn 3; aet,jon3 -, Gnd base
OY..;
(b) JUl-vUlt'Lut[
-,
OY..;
(c)~
t
(d)~
OYJ
-,
qy,
(Q)~-'.
l50kO
-150V
(i)mnnnn.nft+0n~n#lnl#ntL0_ _ _~0Y...t
UUUUTIULJuuliuu
(iVl.L ..
(f) { (ii~1fU1JlIUl
_oTto
- - t - - - - ! t - . - - - - +100 V
••
J(i)~
()
g \(ii) rrrnrrrrY
•
+ counter O.
+ counter I.
Halver + counter 2.
(a) Halve,
.LIt} Halver
(~
Fig. 36'
+ count"" 3.
(,) Halver + counter 4 .
(f) HaJvcr, phase 2 (}fa)'
(d) Halver
(6) Staticisor triuer pips.
Fig. 37.-Circuit of stage "of Y-shift generator.
L-/3
'WJLUAMS AND m.auRN: A SI'ORAGE SYSTEM FOR
II
•
to the grids of T,. during the action periods only, since these
voltqes must not interfere with the scan, i.e. with the voltages
applied to the grids of Ta. To select line 10 requires a shift of
10 units durina the action periods ooIy, 10 that during those
periods eo, e2 and e" must be positM and e1 and e3 negative.
Henc:e if the waveform (i) of Fia. 36(f) iI used for eo. e2 and e"
and waveform. (0)· of Fig. 36(J) for eJ and e3~ line 10 will be
selected. Here the opposite pbale of the hal\Ier to the one
previously considered is used, ana is arranged alternately to
switch on and cut off current in T~ at (i), or is biased well beyond
cut-off at (iO, Fig. 36(j). One cycle of the shift waveform under .
these conditions is shown at Fig. 2~(h).
In order to change the action line, the d.c.lema of the wavefonDS ell must be changed from 0 volts to beyond cut-off of
or vice versa. These volta. . must not be changed during an
action period, because, if they ate, a diagonal line will be traced
across the scnen by the electron beam. and stored information
will be wiped out. 'J'hey may, however, be changed at any time
r.
during a scan Wiod, since they only affect T~ which plays no
in the operation during a scan period. It is convenient to
arrange that a change in voltage can only occur at the beaiDnina
of the scan period immediately following the throw of a switch.
To achieve this. either the positive or the negative pips shown
in Fia. 36(g), which occur only at the beginning of scan periOcII.
are applied to the input grids of five flip-flops by means of ftw
switcJa. This arrangement is shown at the bottom of Fig. 3~.
When a switch is thrown. the' corresponding flip-flop QDQOl
change its state until it receives a pip. This ensures that c:hIaIt
of state can neYa' oc:cur during an action period. The positiw
or negative voltaaes produced by the ftip--ftops are added to the
balver waveform by anode follO'Nen to produce the waveforn
of Fig. 36(j).
StalC II of the schematic diaaram of the V-shaft generator
shown in Fig. 35 is reproduced in schematic form on the left 01
Fig. 37. Details are shown in corresponding positions on thr
right of the fiaure.
part
DISCUSSION BEFORE DIE MEASUREMENTS AND RADIO SECI10NS, lND NOVEMBER, 1948
Dr. A. M. Uttley: Prof. Williams started this work a few
months before he left the T.R.E., and I should like" to refer to
developments carried out at the Establishment since his departure. It is stated ill the -paper that there
five different
ways of using the principle of the dug and partially-filled well.
and I believe I am right in saying that before the work went from
T.R.E: to Manchester the anticipawry-pulJe method 'of storalC
was being used. TIle dot-and-dash method was later adopted
in Manchester. At the T.R.E., we built a store based on the
same principle as the author's, but having certain differencel .
]n May, 1948, we completed ,a serial store containing 1024
digits. The positive- and ocptivc-going waveforms can be Been
quite ckarly. I believe, however, that it is wrong to represent
oby the absence of a pulse. We are hoping to do our computiq
Work with a positive pulse for 1 and a negative pulse for .0.
Many relay computers use both .• 0 relay and a 1 relay. rather
than an UDoperated relay to mean O. Checking of all digits
then becomes possible. To this end of a three-state computer,
we have mod.ifiedtbe pting Qrcuits, so that the positive-going
wave detected at the moment of switching on of the ~ causes
one trigger cin:uit to SO over, and a ncptivc-going wave triggers
anotl1er circuit; the combined output of these two trigger circuits
gives· positive ....miaosec pulse .for 1, and a negative-soing
pulse for O.
This has not converted storage into a complete three-state
system. We have to switch on the beam to find whether there
only the two states. the excavated
is a 1 or a 0 there, and there
or the partially-excavated well. For three-state computing.
therefore, this is only a temporary measure.. We hope and
believe that truly three-state storage will one day be achieved.
Needless to say. our routing, adding and trigger circuits are all
three-state.
.
Another way in which our work is differing from that of the
authors, and deliberately so, is thai we are hoping to complete
a parallel ari~tica1 machlnc rather than a . .ial machine.
This results in an interesting change in the use of the cathoderay tube. If one cathode-ray .tube is used to store the least
significast digit of 1 024 different numbers. and the next tube
to store the next most significant digit of I 024 different numbers,
and so on. then, in order to read one number, all the cathode-ray
tubes in paralleillave to be switched to corresponding points
of the Same X-V co-ordinates on all the tubes; it is then possible
to take out simultaneously all the digits of a number. In
between the actions of reading from, or writing into. the store, ,
were
••
a
arc
-.
•
"
we interleave moments of rel)lCDCration. Oilits are regenerated
sequentially as in a television scan. but between th~ momeftt,
the store can be used at any point. It is possible to ·lCMp frOf'
point to point in a quite random manner; our e'tperiments su....
that the time taken to move from one point to an), other is hUt·
to be about 10 miaosee.
) am sure that it is generally realized that the authoa. .,. . .
tint to have sua:ecdcd in mak.ina a practical ,tora.., system f ....
electrOD.ic computers. Such variations as those t have u.
doocd only emphasize the fundamental aclUevemcnt ~retIIft"'"
by their work.
".
Dr. F. AIIIIatIe: How did the authors discover. the
property described in the paper; was the Jltur.-.r" IU.... .ac.'X:.1
dental, or was it based on any theoretical reasoruna? J'J.
have prodUQCd what seems to be the tnt ~... -,'t:'",~JlI! ."tr~_
digital storage device.
,";
'The authors suggest that there are five methods ofusina ,_
basic storage property. but it seems to !''\Iot lI,;,' tlv .~ ....
and dash-dot methods are essentially the same, '\I~ (n, ,...,
in which of the characters is used to represent 1. This ,,-.
tinction is not a fundamental one, becaUK, U' ~ !Mae ~~.
the respective representation of 1 and 0 may t.. .HHelen. 10
different parts of the machine. The Ymt' comment appl~ to
the two focusing methods, so that it seems to me that .M"Y
three different methods are descnbed 10 the parer
Do the authors confirm that in Fig. L\ Ule mpu\ /I nltM.lnce ....
of the amplifier is Jow compared with any of the resastan..a ..
R/16, etc.? If this interpretation ''I Uti 10..:1, ,ht'v have used
current and not the more common voltage swrundtuUl. What
was their reason for this choice?
Mr. W. P. Anderson: Devices depend. n~ for their operation
on secoodary emission have been used 10 the radio tield for
many years, but they have always had rather .1_ bad reputation.
mamly owing to the large and unpredi\.:table variations in their
characteristics which are liable to occur. It appears. however.
that the storage tube described by the authors should escape
these difficulties as, being an on/ofl device, its operation should
be unaffected by large variations an the secondary emitting
properties of the screen material.
This equipment is a good example of the value of a realistic
engineering approach to a problem. undeterred by the complexity involved, when this complexity is due only to the extensive
application of known techniques. In a comparatively short time
it has been brought to the point where it can be used as a part
,lor.
If
.... L-J3
"'
lJSB Wl'I1I BINARY-DIGITAL
•
.
•
D~JON
99
of a laraar project, while simpler and apparently much more '0· S mm/microsec and SOO such elements were stored on 20 anl
elegant solutions, such as the Haeft" tube. are still in the early or the available l00cm 2 of the storage-screen area.]
stqes of d e v e J o p m c o t . ,
Mr. R. Benjambl: I believe that the technique described in the
There are many possible uses for oJectronic storage devices, pa,er has applications of great value not only to eomputors, but
outside the computina field, which do Dot require the storage also to any other electronic devices which deal with the sorting
of such a Iarae number of elements. At what point would it and handling of classifiable information. The apParatus
be worth while to cbanae over' from a bank of simple flip-flop described has the panicular merit
giving a storage capacity
comparable to' that of many highly-specialized storage tubes,
circuits to the stora(le tube?
I should like to know whether the si,nal/noise ratio in the although it WICS only simple and readily available components.
amplifiec which handles the signal from the pick-up plate sets
The authors state that "an electronic device cannot think." • '
the limit to the accuracy of the system. It would be expected This appears to be a practical statement rather than a fundathat a signal/noise ratio of 12 db would result in an error rate of mental ODe. It should be possible to construct a device which
about I, in I 000 000.
could extrapolate from past experience in dealing with present
Mr. M. V. Needham: At Borehamwood we have carried out problems, provided it had sufficient memory capacity; thus, the
some experiments that had been suggested by Prof. Williams's authors' work might tum out to be a step towards the provision
earlier work; they have been confined to the anticipation-pulse of a limited amount of automatic th61ught.
method referred to in the ,paper as system S. The authors"
It seems that the cloud effect sets a limit to the 'speed of
indicate that, becaUle of non-uniformity of focus over the whole operation of the apparatus described. It might be possible to
storalC surface, they have rejected this method in favour of reduce this effect either by using less steep-fronted grid-modusystem 1. We have used a CV960 cathode-ray tube and have lating .puises, or by feeding the differential of these pulses in a
not found this non-uniformity of focus to be the limiting feature suitable manner into the video circuits. It also appears tha.t the
in the use of system S. The CV960 tube, however, does enable • variations of the electrostatic field towards the edges of the
• more uniformly focused raster to be obtained than does the raster are one of the limitations of the total storage capacity
VCR97 tube. I should therefore like to know in what way of the tube. Might it be possible, by means of a guard rin,
Don-uniformity of focus is more serious with system S than with external to the tube envelope, to reduce these effects? If special
system l.
tubes are developed for this application, should attention be
T'here seem to be several fundamental differences between the given to keeping the thickness of the glass envelfJlSe constant
methods desCribed in the paper. First, in system I the over the ra~ter area, and to minimizing variations of the angle of
beam is normally switched off, and indication of the digit values incidence of the electron .beam?
•
o and 1 is made by switching on the beam to make dots and
With reference to the relative merits of the alternative systerm.
dashes respectively. This means that it is possible to write into at tint si&ht it appears that the anticipation-pulse method would
or read out of the store at any digit position without loss of offer a neat way of utilizing the storage phenomena. In partime. In system S, however, the beam is normally switched on ticular. it seems to be independent of the c!oud effect. and.pemaps
to produce Ii continuous trace or raster. A transient signal- also to penn it some reduction of both the time and the'-space
plate output occurs at each end of a line which must die away occupied by one digit. I should be grateful if the authorscouJd
before the trace can be used for storage. This involves a waste provi\k more information on this point.
" :
of time and space, which can, however, be reduced to a minimum
Dr. R. A. Smith: It seems at first sight to be very re~able
of one digit period per line if a suitable raster, such as the that the first material chosen for making ~se experiments on
Z-scan, can be used.
electronic storage has proved to be so very successful. The
Secondly. if system 1 is used, what is in effect a digital scan commercial cathode-ray tube has the very great advantage that.
can be used, and digit positions can be defined by means of preset when something is written on it, it can be seen as a luminescent
voltage increments. In system 5, digit positions must be defined trace; but it is by no means obvious that the screen. materi~lof
in a more indirect way by means of the scanning rate and digit a cathode-ray tube is the most suitable material for electronic
periods. .
,storage. A moment's thought, on the other hand, convince<i
The third point of difference concerns speed of operation, and one that it is at least not unreasonabh: to suppose that such
in this connection some measurements we have made may be material would be suitable. It has a fairly high' secondary
of interest. A single trace was produced 00 the storage tube, electron emission, and it has a fairly high. but not too high,
and a small element of this trace was blacked out to obtain an resistance; sO that, although fairly good short-time storage is~
anticipation pulse followed by an excavation pulse. This latter ~btained, one does not get the paralysing effect to be expected
p~ is identical with that produced by the dash of system I.
from a very good insulator. From that point of view, therefore.
It was found that the excavation-pulse amplitude falls with the materials nonnally used in cathode-ray tubes have the characincreasing writing speed, whereas the anticipation pulse increases teristics likely to be required for such a storage system.
to a maximum and is still of usable magnitude when the other
It is obvious from the method of operation which the authors
is quite small. These results suggest that the anticipation-pulse I\OW use that it is no longer necessary to look at the tube; one
method might be usable for shorter digit times than is poaible examines what has been written on a separate monitor tube, so
with system 1. A possible explanation of these curves may be that there is no need to use a phosphor for the screen material.
that the anticipation pulse is unaffected by the cloud-effect pulse It would therefore be interesting to learn whether any other types
~rring at the commencement of the break. whereas the
of material have been used for storage;
Mr. P. G. Redgment: I should like to enlarge on Mr. Ander-,
excavation pulse is always offset by the negative-going cloudeffect pulse produced at beam switCh-on.
son's reference to the signal/noise ratio. If a true random
[The speaker showed a film illustrating the use of the distribution of noise voltages is accepted. it can be only a matter
anticipatjon~pulse method for the storage of pulses on a circular
of time before a false pulse is injected into the system, so that its
s6ln, which was foundcanvenient in the initial work, and then practical use in a computing machine dealing with lengthy
on a conventional television-type raster. The film showed problems will depend on the tim~ in which a given probability
stored patterns made up of elements of 2-microsec duration with of the occurrence of a mistake will be reacheJ. In any circu2-microsec intervals between them. The writing speed was latory storage system this time appears to depend on the
or
"anoos
•
coMPU'I'ING 'MAOIINES:
'lIIf.;';'-- L-"13
~ .. '.I-
•
.,.
.
,
100 WILIJAMS ANi> JW..B1.JRN: ~'S"YS1ZM FOR USB WD1I BlNARY-DIGITAL COMPU11NG
siana1/noise ratio. In the ~ of the men:ury-delay line type
it acems that the output puIIe may poIIeII qualitiel-«lcb u an
accurately controUed .wid~wbicb enable it to be distinauilhed,
fiom noise and tbemore improve the' sipaJ/noiJe diIcrimiDation. The cathode-ray tube storap systems delcribed provide
a rather odd-ibaped puJse,and it aecms that not much cin:uit
dilcriminatioD from noile, other'thaD that liven by the time at
whidl the pulse occurs, can be obCaiDed.Whether dle "antici.,.tory puIIe." or the "weD aDd weI1..fi11ina" type deiCribed by the
authors. is u.d, it eema ,that some importance must au.dl to
the liPaIInoile ratio, as it may determiDe the ultimate accuncy
whicb the computor can adlieve with any 8iwn Iia 01 problem.
Mr. D. M. MKKa,: Is anytbina known 01 the iDftuence of
~OIINES: DISCVSSION
temperann on the shape and penisteace 0( the potential well?
Althouah anaIope computation is not Itricdy within the terms
of the paper, I should like to mention u.t "" bave been experimeotina ad' a store for anaIoaue information, using a modification
Of the priDcipie delcribed. Briefty. a trail of char8e is depoIited
on a catbodo-ray ICI'een by a '-writing" spot. It is subiequeody
scanned by a raster, and the pulaes. detected in the nwmer die
authors have deacribed, are used in the obvious way to reproduce
the written waveform via a flip-ftop or a strobe circuit. With
this, anaqement it is perhaps more important than with the
dilital method to reach the limits of definition, and it would be
of interest to know what.relevant cft'ect. if uy, is produced' by
cM.nJCS in the temperature of the pboIpbor.
11IE AUIlIORS' REPLY TO 11IE ABOVE DISCUSSION
PNI. F. C.W....... and Mr. T. KJIINrn (i" "ply): We
shall not attempt to answer speakers individually, but will
attempt rather to COYer 'most of the points railed in the
di8cussion.
Our knowledae of the history 01 this type 01 ltorap, apart
from our own development, is JaraeIy . . . on beanay. As
far • we know the discovery that.1ipaIs symptomatic of
previous aamnina could be obeerw.d 011 an ordinary cathode-ray
tUbe wu 'made accidentally at the Radiation Laboratories,
Boatoo. U.s.A.• durioa some expcrimmta with a special atoraae
tube. ~ experiments did DOt expote the vital fact that
sipals wcreavailable from the pick~up plate in4*mty of time
to permit 0( ~on on a siDaIe tube.
AI rqpuda .the taeeIl surface material. it is
coune moat
CODWlliait to UIO COIJUDCI'Cial tuba, but it ... CODfideady
expected that a f~ experiments would ~ • prefcrabk
surface. Experimen1a have been made iD a continuouslyevacuated tube with· various kinds of ..... aeveral fluorescent
and other powders, ~d with mica; DOGe of the materials tested
was u satisfactory as the ordinary ..... or blue ICI'eeIl, but it
may well be that the technique of testina ia DOt yet adequately
developed, extreme cleanliness bein. of peat importance ill
studyinl lClCiODdary-emission effects.
·.A
The number of Iystems described in the paper is five or three.
dependinl OIl whether or not the pte. dn:uit is reprded as part
of the l)'ltem. It is qreed that only iluee charae CODfiaurations
are delcribed. The choice between tbeao was made at a fairly
early stqe in favour of the dot-dash ~t, because the
timina of the indicatina pulles was cloaely controlJed, because
the shape of thcac pulses was leu dependent on focus. and
becauSe thil system was felt t(J" be IDOI'e flexible in that it could
also be applied to parallelltoMS.
We arc very interested in, and now agree with, the statement
that in seria-typemadlines the anticipation 'pulse method it
essentially' fater. It 1CeIIlI, however. that since the waveform
shown in Fi•• J4(d) is !devant to either systern;-~ the space
per diait will be the same for both systemS.·.. It aeems
or
•
-.
--------~-----.--
•
•
likely that botbtime and space per diait can be improved
by the use of the fOCUl-defocus method.The sianal/noile ratio of the systems we have operated haS been
10 lood that we have paid scant attention to calculatin.· the
probability of enor from this callie. The caJculation is a difficult one since allowance must be made for strobinl and for the
fad that a aoile pulac must exceed the datum level for a certain
period. before it becomes effettive. Some rough calculations
bave DOW been . . . . and indicate that failwe from this cause
should not oa:ur more than'orx:e in thirty centuries o(-continuous runniDa. 10 that the scant reprd paid to this pomt
therefore appears to be justified.
It may weU. be that it is sounder enaineerina to ~t zero .
." a sianal rather than by the absence of a sipaJ, linc:e lJeIO it
ewry bit a .vital a ~picce of information as l. Our attitude.
boweYer-and this is relevant to the two previous items as welt
-hal been that we would proceed for the time beina with the
limplesteaentiala of the machine, since f t feel that some
experience of the actual operation of a madline is quite 'Utaent
at this stqe of the development.
We should like to thank Dr. Aughtie for the co~tions, communicated to us privately. which have been incorporated ip ·the
paper, and confirm that we have, in f~ct. used current summAtion.
the IUIOft beinl that in general we reprd this procedure as both
. simpler and more elegant than voltage summation, whicb usuaU),
calls for sublcquent amplification.
.
It is interatinl to see from Mr. Mackay's remarks that the
storaae property has some application also to analogue computen. We rearet that we have no information about tty
dfect of lemperature on definition,
It is difficult to state at precisely what point it becomes eco~
mical to use c.r.t. storage instead of storage on flip-ftops. If a"','
cathode-ray tube is counted as equivalent to 10 valves, then in
a series system taking valve numbers alone as a criterion, the
c.r.t. system is preferable for any number of digits in excess. of'
3~, but it is recognized that the flip-flop system might be m()re
reliable and might therefore be used for numbers not too greatt~
in excess of 32 digits.
---------- -"------. - --- ._-
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement