Design Considerations for a Photooptical Storage Device

Gilbert W. King

The design considerations and performance of a photooptical information storage device are developed.A figure of merit of a class of such devices is given in terms of fundamental parameters expressing thecurrent state of technology or the laws of physics.

Introduction

The Photostorel is a device constructed to meetcurrent demand for a memory of very large capacitywith a reasonable random access to its contents. It isbased on the observation that much information in verylarge quantities is relatively permanent (like printedbooks in a library). Performance of interest in thisregime is only practical at the present time by the useof optical and photographic technology.

The equipments built2 have the general configurationof a glass or film disk on which the information is re-corded photographically in binary form, as opaque andtransparent squares (marks) arranged in circulartracks forming an annulus. The disk is rotated totransport the store by a reading head, which can moveradially (see last citation in ref. 2).

The reading head consists of a microscope objectivefocusing a spot of a cathode-ray tube onto the marks.A photomultiplier tube placed immediately behind thedisk catches the light passing through the transparentmarks. Its output is a current modulated by the in-formation on the track passing the reading head.This signal is converted into binary digits and passedto a processor.

Three such units have been built, and one has beenoperating for five years in a data processing system.This paper describes the general principles of design ofthe Photostore memory and the considerations takeninto account to make it a practical and workable de-vice.

Morphology

The basic concept of the Photostore is that digitalinformation can be stored on photographic emulsion atvery high densities (of the order of 1.1 million bits/cm2). Because of the compactness of the stor-age, the scanning process, necessary in any store, can be

The author is with the Itek Corporation, Lexington, Mas-sachusetts.

Received 5 January 1965.

accomplished with a short random access (15 msec)and high reading rate (4 million bits/sec). In order toattain a high total capacity (at least 100 million bits/disk), an area of film 200 cm2 must be scanned. Tomeet the time requirements, the scanning is done withtwo motions: a mechanical transport of the wholestore past a reading head, which in turn allows a scanperpendicular to the first motion.

After consideration of many forms of equipment inwhich these two concepts could be effected, it wasdecided that the simplest practical scheme was to havethe emulsion on a rotating disk, the information beingstored in a ring or annulus. The annulus is scannedradially by an image of a spot caused by an electronbeam of a cathode-ray tube. The low inertia of thebeam permits rapid shifting of the spot.

Even with this basic morphology, the pattern of re-cording the information in the annulus could assumevarious forms. The simple sequential readout seemsto be as adequate as any. Information is naturallygrouped in entries which are recorded as segments alongconcentric circular tracks. Thus, if the spot staysstill, it illuminates in sequence all the binary digits in atrack. Shifting the spot radially allows other tracks tobe read.

System Specifications

In order to see what is feasible in an optical-photo-graphic memory system, it is necessary to develop aformal analysis of possible configurations. Such ananalysis is a set of interlocking formulas relating allthe parameters of the design of a memory system, com-bined with the operational, physical, and economicalconstraints. These formulas show the interactionsbetween the various parts of the system and permitover-all optimization. It is also used to clarify certainissues by separating the parameters into classes. Forexample, this analysis identifies areas of critical im-portance, it reveals the limiting performance of thesystem, and indicates what the potential and ultimateperformance can be. Is the limit something funda-mental, such as diffraction theory or the statistical

April 1965 / Vol. 4, No. 4 / APPLIED OPTICS 429

nature of light, or is it a state of technology, such asresolution of photographic emulsions?

In this analysis, parameters are classified into thefollowing categories:

a. performance or functional requirements,b. fundamental, from laws of physics,c. empirical, from the state of the art,d. constrained, by existing technology, economics,

or ease of use,e. design, consequences of the analysis.

The performance parameters are those which wewish to optimize in the memory system to meet practicalrequirements. The most important are capacity, andaverage, random access time. The rate of readingbits for scanning or readout is secondary but importantpractically.

The fundamental parameters are determined by thelaws of physics. A primary one is the angular resolu-tion of the optical system, determined by its absoluteaperture. Another is the statistical nature of light,which says that, if photons pass through a mark on theaverage, the fluctuation is n' /2 and the consequent sig-nal to noise is n'

The empirical parameters are fixed by the currentstate of art or technology. The major parameter is theresolution of photographic emulsions, e.g., resolvablebits per unit area. Another is the brightness of mov-able light sources, such as the spot on a cathode-raytube. The bit rate is limited by the current state ofelectronics. Basically, the empirical parameters alsodepend on fundamental laws of physics and chemistry(such as the heat conductivity of the crystals of a CRTphosphor), and it is important to try to find these de-pendencies in order to learn the ultimate limitations ofthe system. At the same time, it is equally valuable toappreciate what materials with better parameters canbe obtained by a determined experimental effort.

Certain parameters are constrained by technology.For example, there is a limit to the numerical apertureof a lens. Economics and inconvenience restrict thediameter of the disks which can be considered.

The design parameters are those whose values aredetermined as a consequence of the fundamental andempirical parameters by optimization. In other words,the designer should not start by choosing a radius forthe disk, but should have this determined by optimiza-tion, including the restraints. It is important toidentify design parameters, otherwise attention anddevelopment are directed toward the wrong factors.

Capacity

Capacity C is measured by the number of bits of stor-age space available, called marks. For several reasons,discussed elsewhere, two storage bits, or marks, are usedto record one bit of actual information. In addition,an area equal to the mark is used as a border to guidethe spot. The tangential width of a mark, along atrack, is a; the radial height is b. Since the bordersare shared by alternate tracks, the area used per bit of

storage is 1.5 ab. The area on the disk used for storageis 2 rwr, where r is the average radius of the annulusand w is its width. Thus C = 4 rwr/3ab storage bits.

Access TimeInformation on the disk is usually in the form of

groups of digits, or entries. The functional require-ment is to get out a desired entry as fast as possible. Ex-haustive search is too long so the system has to providerandom access, a search procedure leading to the de-sired entry in a minimum time T, which is one of thetwo prime functional parameters.

The search procedure used in the Photostore isessentially that of looking up a number in a telephonedirectory, or a word in a dictionary. By a more or lessrandom process, the pages are turned until samplematchings show the desired entry will be on or near aselected page. Then this page is read exhaustively.*

The access time to a randomly selected entry is com-posed of two parts: the time t to arrive at the desiredtrack and the time t2 in which a particular track must beread before the arrival of the desire of entry. As thetwo times are uncorrelated, the mean access time T isthe sum of the averages of t and t2 is T = (ti)av +(t2 )av. The first factor is determined by the rate ofrotation of the disk.

Rate of RotationThe peripheral linear velocity of a disk of radius r is

then 2 7rwr. The only limit on r from this considerationis the breaking strength of the material of the disk, sothat 2 7rwr < R is a mechanical limitation. For designparameters in the regime ultimately defined by thecurrent analysis, the product 2 7rwcor is well below thislimitation. However, in the general parametric studywhere we might be interested in faster speeds, the aboveinequality must be remembered.

An important design parameter, dependent on co, isthe reading or scanning rate p. This is the number ofmarks passing the reading station per second: p =27rwr/(a marks-sec). An upper limit to this is set byelectronic considerations. Pulses faster than 109/secget smeared out in transit through the stages of thePMT. The pulses have to be decoded in electroniccircuits, and it is very difficult to go beyond 107 pulses/sec. Thus p < po, where po - 107 sec' reflects thestate of the electronic art for handling fast pulses.

AnnulusIn order to maximize the capacity, d, clearly, we wish

to make the width of the annulus, w, as large as possible.The upper limit is determined by the range of the lensin the search compatible with a good access time. Theflexibility of a large memory is enhanced if an integraladdress method of search is used. The spot is alwaysreading something, and when a new search is started

* In the Photostore, the sampling is done by turning pagesuntil one is reached where the entries are beyond the desired one;then the entries are read exhaustively backwards.

430 APPLIED OPTICS / Vol. 4, No. 4 / April 1965

the desired address is compared with what is being seen.The result of the comparison is used as an instruction tomove the spot to the next track, either inside or out-side. This signal to move is sent to the lens arm. Itsinertia is so large, however, that it could not move thespot at the desired speeds. The slack is detected andtaken up by a parallel signal to the electron beam in theCRT, which moves the spot at a high rate of speed.Figure 1 shows the desired motion of the spot during asearch, and the actual.

Lens Arm

In the reading process the disk is rotated under aspot of light from a cathode-ray tube source. Thelight is focused on the information marks and, where itcan pass, is caught by a photomultiplier tube. Asecond photomultiplier tube looking at the spot servesthe light intensity for tracking purposes.

In this system, the capacity of the store depends onthe area of emulsion capable of being scanned, and thisis a product of the diameter of the disk and the radialfield of view of the lens. The field in this applicationis somewhat different from that normally considered inoptics. Optically, the field is only one mark, of theorder of a few millions, which is a minute fraction ofthe optical field of a true lens. However, the lens mustbe able to focus the searching spot of light on marks inany track, spreading over the width of the annulus.This can be done in two ways, either by moving thesource of light in the object space and allowing theimage to traverse the optical field of the lens, or byactually moving the lens. The latter involves a me-chanical motion which cannot be made fast enough forthe access time desired. The former does not giveenough field, i.e., wide enough annulus, and also resultsin a very large field in the object space when the lightsource is to be moved. In the present system, bothmethods are used. The lens is moved parallel to itsoptic axis as fast as possible, and the sluggishness ofthis motion is overcome by moving the source of lightas well.

Moving the lens actually serves three purposes inthe system by:

a. increasing the width of annulus scanned, thusincreasing the capacity;b. compensating for a 30-cycle motion of the spoton the CRT face, as the spot traverses a circular raster(allowing greater brightness);c. allowing a standard CRT to be used as a sourceof light; without the motion of the lens, a muchlarger diameter CRT, with the same spot diameter,would be required.

Lens Acceleration

The condition in this system is that the lens mustmove with sufficient acceleration so that the spot, intaking up the slack, does not run off the face of theCRT. The acceleration can be determined by thechange in velocity in the time available. The formeris the maximum slope in the diagram, namely, the rate

J

o

0

Final position

Ideal path

Ar, lag in lens motion

-At, tine lag

Sampling path

Initial position

Time

Fig. 1. Motion of spot during search.

of traverse of the annulus. This is nominally thewidth, w, divided by the time of one revolution, 1/w.The time available to reach maximum acceleration is,according to the diagram, At, which is 1/ww times thedistance available Ar. This in turn is equal to -, themaximum possible displacement of the spot on theCRT, divided by the magnification.

The design parameter magnification is determined byconsideration of available optics. Since the spot has aGaussian distribution of light of half-width -, for thepresent we can say the optimum magnification is ap-proximately 3 -/a. The effect of lens aberrations anddiffraction also should be included, but in the regimehere established this is a minor modification. Thus,A = 3w2c2 u/a&

The maximum lens acceleration depends on theproperties of the drive motor and of the lens mount.The basic motor characteristics are maximum force Fand maximum displacement X. By the choice of suit-able linkages, other forces or displacements can be ob-tained which arrive at the same product F X X. Abasic lens drive parameter is then b = F X X.

We should evidently choose X = w, then the maxi-mum lens acceleration is A = /2Mw, where M repre-sents the effective mass of the lens and the factor 2takes into account the inertia of the lens mounting anddrive mechanism. The force 1D is limited by thefundamental nature of electromagnetic coils. We havethen

A 3wcola-2Mw 2aa

from which we may solve for the design parameter, thewidth of the annulus.

Lens

The lens was chosen on the basis of the followingconsiderations:

a. high ratio of field to definition,b. excellent definition,c. low mass,d. reasonable depth of focus,

April 1965 / Vol. 4, No. 4 / APPLIED OPTICS 431

e. high aperture,f. reasonable working distances.

The type of lens satisfying these demands is a mi-croscope objective. The actual lens used is a 15-mmobjective, designed for tenfold magnification. Thenumerical aperture is 0.25 (F = 2). This lens wascarefully selected and has excellent definition. It iscapable of making photographs of small marks 300 . in.wide with edges sharp within 30 u in. The limitationsseem to be the theoretical limit imposed by the aper-ture, namely, X/2 sinG = 40 ,u in. To realize this highdefinition, the working distance must be controlled to100 ,u in. As will be seen below, in this application,300 M in. variation in working distance is permissible.The lens mount is such that this distance is held to10 /i in. The main source of variation lies in the varia-tion of emulsion thickness.

Restrictions Imposed by Amount of LightThe error rate of the system will be determined by the

number of photons passed through a white mark. Thenumber desired, no, is determined from the desiredfrequency of unsuccessful searches, the effective con-trast, and fluctuations in transmission due to mechani-cal and photographic imperfections.

The number of photons passing through a mark de-pends on the characteristics of the CRT, lens, andPIT. If s(X) is the efficiency of the photosensitivesurface, the number of photoelectrons produced by aphoton of wavelength X is n = f dx s(X) p(X), wherep(X) is the number of photons arriving at the PMIT.The latter depends on the optics of the system and thebrightness of the CRT spot:

7 X i m ab T(X) B(X),

where

T(X) = transmission of the optical system,F = F-value of the lens,m = magnification,B(X) = number of photons of wavelength X emitted per

second per unit area by the phosphor,a, b = dimensions of the spot, andp = reading rate.

(Because the illumination of the spot is not uniform,the proper expression is more complicated but this isadequate for these purposes.)

The brightness is proportional to the current densityof the beam and the accelerating voltage V.

Now the electron current not only produces light butalso heat, which has to be dissipated in order not tohave the phosphor rise to a temperature Tmax at whichit is damaged. Elsewhere, we have shown that, if thebeam is on a spot for a time r but off for a time of theorder of the cooling time constant for the layer of phos-phor, the power in the electron beam iV is limited byiV < CTm/r, where C is the heat capacity of the phos-phor (per unit area) and T,, is the maximum tempera-ture without damage to the phosphor. In the present

system the beam is kept from sitting on a spot by havingit move in a circular raster (somewhat disturbed by thesearch routine). The rate of circulation is chosen tomatch the cooling time constant of the phosphor,namely, kck where k is the heat capacity of the phosphor.If D' is the diameter of the circular raster, and d thebeam diameter, then r = dc/7rDk. The diameter ofthe raster is the diameter of the CRT, less the leeway6, D' = D -28. The diameter of the beam is somefraction, , of that of the visible spot, which we definefor convenience to be 3a, a being its half-width. Thusd = 3a. However, owing to diffusion of the beamelectrons and scattering of photons by the crystals ofthe phosphor, this electrical energy in the beam ofdiameter 3a, is converted into light in an area ofdiameter 3a-. Thus, the brightness will be only 4)2 thatof the power density.

The efficiency of the phosphor is a(X) photons perelectron volt. The over-all efficiency of conversion andtransmission will be represented by E = f dX a(X)s(X) T(X), where the integral is over quantities given inthe technical literature. The factor n/rn 1 isnearly unity.

Thus we have

w'2EkT 1 D ab12 e F2

(1 p

The minimum number of electrons is determined by afunctional parameter, the error rate, so we set n = no.

The last equation can then be viewed as the onedetermining the reading rate p,

7r2 Ek T,5k 4 _ Dp = - - -(1 - [25/D]) ab.

12 eno F a

System Parameters

Combination of FormulasWe may now combine the formulas developed in

arriving at the system specifications to express thefunctional parameters capacity, C, and frequency ofaccess, co, in terms of the fundamental and empiricalparameters, eliminating the design parameters.

The quantity /D is the fraction of the CRT faceused for leeway in the high speed part of the search ofthe annulus. Since the quantities on the left both areto be maximized, we may maximize the right-hand sideby differentiation with respect to 6/D. The maximumvalue is /D = '/4, so the factors (6/D)'1/3(1 - 2/D) become simply (1/2)"/'.

The factor 8/D is a property of the CRT; D/2a isthe number of spot diameters in the face. (In general,this is independent of the dimensions of CRT.)

We are left then with the single design parameter, a,the width of the marks. (Note that b, the height ofthe marks, has cancelled out.) Now a occurs with thesame exponent as D/3a-, and we note Da/3- = A,where A is the flat field of the lens on the emulsion(i.e., in the image space). The factor 3a- is the mag-nification in, and this relation assumes that the face ofCRT is not larger than mA, or else it could not be used.

432 APPLIED OPTICS / Vol. 4, No. 4 / April 1965

Obviously, it should not be smaller, otherwise the fullamount of leeway would not be available. Thus thefield of the lens and the CRT face are matched. (An-other way of stating this match is that the number ofspot diameters -D/3a- equals the number of marks inthe field A/a.)

Thus, finally

C2- = 1/3 E Tmk A4'/eno /3F2

where

the quantity Cu"/ is to be maximized;the quantity b is the force available to move the lens;the factor VI'/I(M'/'F2) depends on the lens;the factor p is the ratio of electron beam to visible spot dia-

meter, and is a property of the phosphor; andthe quantity kTm is also a property of the phosphor.

The factor E is an integral over wavelength of theefficiency of transmission through the optical com-ponents, of the conversion of electrons to photons inthe phosphor, and of the conversion of photons to elec-trons in the photosensitive surface of the PMT.

The remaining factor, no, the number of photoelec-trons necessary for the desired error rate, is discussed

-I

w

later. It is in a sense remarkable that the size of themarks does not enter in this expression at all. There is,however, the implied matching of the CRT with lensdiscussed above.

Conclusions

The last equation giving a figure of merit in terms offundamental parameters expresses the current and po-tential capabilities expected from a photostore-likedevice. The parameters can be grouped as individualfigures of merit for the components, namely, CRT, lensand mechanical devices.

References1. This machine resulted from contract support for USAF,

RADC, AF30(602) 1566 and 1823. Preliminary results werereported by Gilbert W. King, Louis N. Ridenour, and GeorgeW. Brown, Proc. Inst. Radio Engrs. 41, 1421 (1953).

2. Gilbert W. King, Control Engr. 2, 48 (1955). A briefdescription of the system was presented by Gilbert W. King,in "Data Processing with the Photostore", Chapter 19 ofLarge Capacity Memory Techniques for Computing Systems(Macmillan, New York, 1962), and by Robert W. Potter inChapter 13, Optical Processing of Information, D. K. Pol-lock, C. J. Koester, and J. T. Tippett, eds. (Spartan Books,Baltimore, 1963).

I L~~i! Q..itics an a u ~p uc s

This column is compiled partly from information sent by our Reporters in various centers of opticsacross the continent, but the Managing Editor welcomes news from any source

A Case History in Selective Eliminationas a Solution to the Information Crises

William Tinker, Princeton UniversityBenjamin Evers, Yale UniversityPaul N. Chance, Smith College

In this month's column we reprint by permission of AmericanDocumentation, a publication of the American DocumentationInstitute, 2000 P Street N.W., Washington, D.C., a letter whichappeared in the October 1964 issue of that journal:

C.O.D. 13 February 1966'

Published studies on the so-called information explosionin science generally proceed from the unstated premise thatnewly published research results constitute a direct incre-ment to the total store of knowledge. Since the number ofresearch workers and the number of published papers are bothdemonstrably increasing (Perish's Law), this premise leads toexponential growth rate projections for such variables as annualjournal pages, costs of society publication programs, linear feetof library shelving required, gross weight of abstract journals,and information-transfer study budgets.

One of us (1), in the course of a research project on tastedifferentiation mechanisms in vermicula exlibris (a subject ofintense interest to the library paste industry), was recently the

l C.O.D. = Calculated Obsolescence Date.

beneficiary of one of those fortunate accidental discoveries thatunexpectedly illuminate a hitherto unnoticed bypath on the mainroad of scientific advance. Comparative thickness measure-ments on two chronologically adjacent editions of the Encyclo-pedia Britannica revealed a difference of less than 5%, althoughthe time-lapse was 22 years, and the predicted change (Price)should have totaled more than 103 pages (273.4% by weight).

Subsequent analysis showed, moreover, that this anomalycould not be accounted for by compensatory variation in any ofthe common physical variables of paper thickness, type size,leading, number and size of illustrations, outside dimensions ofeither volume or bookcase, or marginal considerations. Aradically new hypothesis was thus required to account for thediscrepancy.

Thus was born the concept of Historic Cancellation of Fact,and with instantaneous and understanding assistance from theNational Science Foundation we proceeded to turn the undividedattention of our laboratory at Witte's End to testing it.

Taking a sample population (A-Mc) from each edition, one ofus (2) assigned to each fact a coded identification number cate-gorizing that fact in terms of its location on the McGeever spec-trum of human knowledge. Upon computer comparison of thefacts from each edition it was determined that the total numberof facts in the two editions was nearly identical (Table I).

Price's equations were obviously omitting some significantfactor. Although the trend had been predicted to 100% accuracy,the magnitude had been grossly overestimated.

Using the lab's 14'/2-33 computer and a program originallywritten by one of our (3) graduate students (Lajoie), one of us(4) examined the 7,384,622,141 items from the second edition

April 1965 / Vol. 4, No. 4 / APPLIED OPTICS 433