274 Philips tech. Rev. 34, 274-287, 1974, No.IO

The Phototitus optical converter

F. Dumont, J. P. Hazan and D. Rossier

In the last ten years there has been increasing interest in the optical processing of imagesand other data. One result of this has been an increased demand for devices that canfulfil the function of real-time re-usable film. In recent years several such 'image conver-ters' have been developed; 'Phototitus' is one of them. Phototitus will store images, add orsubtract them, and process them in other ways. An important feature that distinguishesPhototitus from most other image converters, and couidwell be ofinterest in the field ofcoherent-optical processing, is that a stored image can be read out with coherent lightwithout destroying the coherence of the light. In Phototitus the read-out light is veryefficiently used and only moderate voltages are required for operation. The attainablecontrast, image resolution andwrite-in sensitivity compare well with those of other devices.

If a thousand pencils are scattered across a table, andone is then removed, an observer who sees the scene'before' and 'afterwards' would find it hard to tell thedifference. Even if pictures of the two scenes arepresented to him for comparison, it might take himquite a while to find out that they differed by one pencil.But if a 'negative' of the second picture is superimposedon the positive of the first, the difference shows upimmediately.This 'picture subtraction' is an example of 'image

processing'. An essential part of the processing is thestorage of an image until it has been dealt with. Imagescould be processed by using photographic film. How-ever, while film is ideal for storage, such processing istedious, complicated and often unsatisfactory.Phototitus [1] is a device that can subtract images -

or perform other operations on them - instantane-ously. Its multilayered structure is shown schematicallyin fig. 1. It consists of a single crystal C of deuteratedpotassium dihydrogen phosphate (DKDP), a dielectricmirror M, a layer L of amorphous selenium and twoconducting transparent electrodes Al and A2. The sele-nium is photoconductive. If an optical image is pro-jected on it, electron-hole pairs are created near Al. IfAi is made positive with respect to A2, some of theholes drift towards the interface with the dielectricmirror M, where they are stored, presumably in surfacetraps. At any point of the interface, the accumulatedcharge is proportional to the intensity of the incidentlight. The optical image is thus converted into a latent

Dr Ing. F. Dumont, Dr J. P. Hazan and Dr D. Rossier are withLaboratoires d'Electronique et de Physique Appltquée (LEP),Limeil-Brévannes (Val-de-Marne), France.

charge image, as in xerography. When a new image isprojected, the corresponding latent image adds to thefirst. If an image is projected while Al is negative withrespect to A2, electrons instead of holes drift towardsthe interface, and the corresponding charge patternsubtracts from the latent image already there.

The DKDP plate allows the latent image to be readout. This operation depends on the Pockels effect ofDKDP: linearly polarized light passing through thecrystal emerges elliptically polarized if there is an

---+---- PIM A2

-- 1/-- / I-- / I//M' I-- 0/ I-- /

/ ,I-- / I/--P2

Fig. I. Principles of Phototitus. L photoconducting layer ofamorphous selenium. C single crystal of DKDP. M dielectricmirror. AI, A2 conducting transparent electrodes. PI, P2 crossedpolarizers. M' semitransparent mirror. An optical image projectedon L from the left produces a latent charge image, which is storedat the interface with M. The latent image is positive, consistingof holes if Al was positive during the projection, and negative,consisting of electrons if Al was negative. The latent imagepolarizes the DKDP. Owing to the Pockels effect in the DKDPthe beam of polarized light is modulated in accordance with thelatent image after it has passed twice through the DKDP andhad also passed P2. _

Philips tech. Rev. 34, No. 10 PHOTOTITUS· 275

electric field across it. The larger the field, the larger thedeviation from linear polarization, and therefore thelarger the amount of light transmitted by a crossedanalyser. Thus the light beam in fig. 1, after havingpassed the analyser P2, is modulated by the latentimage: bright patterns correspond to high-field areasin the DKDP plate, i.e. areas of high charge density inthe latent image.Phototitus is derived from the TITUS tube [21. In

this tube, the main element is also a plate of DKDPmaterial, but here the image is written into the deviceelectronically, by means of a swept electron beam anda video signal across the DKDP plate. In both casesthe DKDP plate is cooled to just above the Curie tem-perature (about -50°C). In this way a large Pockelseffect is obtained: the effect is proportional to thedielectric constant Bc along the optical axis, whichincreases strongly near the Curie temperature.Phototitus belongs to a large family of devices devel-

oped recently for image storage and processing, allbased on the idea of combining a photoconductor withan electro-optic material. In a later section we shallshortly compare Phototitus with other devices of thisfamily. Here we want to stress one feature that it shareswith only a few of these: it allows a stored image to beread out with a laser beam, without destroying thecoherence of the laser light. It can thus convert printon paper into images that can be analysed by coherentlight. This may open the way for real-time coherentoptical processing- such as the recognition ofpatternsor characters using holographic filters - of imagesprinted on paper. This feature is present because theDKDP is used in the form of a single non-scatteringcrystal, and above the Curie temperature.In this article we shall review the physical principles

of Phototitus, with emphasis on the special propertiesof the photoconductor. The properties of DKDP andthe Pockels effect will only be mentioned briefly, asthese have been treated before in articles in this journalon the TITUS tube [21. We shall discuss the 'image-transfer characteristic', relating it to the photoconduc-tor properties, and the spatial resolution. The articleends with a short review of possible applications, anda short comparison with related devices.

The Pockels effect of the DKDP plate

In a zero field a single crystal of DKDP is opticallyuniaxial, the optical axis z coinciding with the crystallo-graphic axis c. It is therefore birefringent for lightpropagating in the a,b-plane, but light propagating inthe c-direction is unaffected in polarization. An electricfieldE along the e-axis,however, induces a birefringencefor light propagating in this direction proportional

to the polarization BeE of the crystal. This is the Pockelseffect. The optical axes x,y in the a,b-plane bisect theaxes a and b (fig. 2). In TITUS and Phototitus thec-axis is perpendicular to the plane of the plate. Thelight propagates in this direction, and is polarized alongthe a-axis by the polarizer PI (see fig. I). After passingthrough the crystal (twice), the two components po-larized parallel to x and y have acquired a phase differ-ence cp proportional to BcElc, where le is the thicknessof the crystal. The light thus emerges elliptically polar-

Fig. 2. Pockels effect in DKDP. An electric field E parallel to thec-axis induces birefringence in this direction, with optical axesx,y bisecting the axes a,b. Light polarized parallel to a by PI(see fig. I) emerges elliptically polarized, as the componentsparallel to x and y have acquired a phase difference '" (propor-tional to eeE). The light thus acquires a component polarizedparallel to b. This component is selected by P2 (fig. I).

ized. The amplitude of the light transmitted by thecrossed analyser P2 is proportional to sin tcp. For lightpolarized by PI, the transmittance T of the systemof DKDP plate plus polarizer P2 is given by

(1)

where V = Elc is the voltage between the faces of theDKDP plate. The factor K is proportional to Bc, andindependent of the thickness of the crystal [31.

[1] J. Donjon, F. Dumont, M. Grenot, J. P. Hazan, G. Marieand J. Pergrale, A Pockels-effect light valve: Phototitus.Applications to· optical image processing, IEEE Trans.ED-20, 1037-1042, 1973 (No. 11). See also: .G. Marie and J. Donjon, Single-crystal ferroelectrics andtheir application in light-valve display devices, Proc. IEEE61,942-958, 1973 (No. 7).G. Marie, J. Donjon and J. P. Hazan, in: Advances in ImagePick-up and Display 1, 226, 1974.F. Dumont and D. Rossier, Le convertisseur optique Photo-titus, and J. P. Hazan, Traitement des images, to be publishedin 1975 in Acta Electronica.

[2] G. Marie, Philips Res. Repts. 22,110,1967.G. Marie, Philips tech. Rev. 30, 292, 1969.'TITUS' is an acronym for 'Tube Image à TransparenceVariable Spatio-temporelle'.

[3] More details are given in the article by G. Marie and J. Don-jon, note [I J.

276 F. DUMONT et al. Philips tech. Rev. 34, No. 10

Design and operation of Phototitus

In the experimental models of Phototitus that havebeen made in our laboratory, the active area of themultilayer structure is about 30 x 40 mrn-. The DKDPcrystal is about 250 um, the selenium layer about 10 [Lmin thickness. The dielectric mirror (M in fig. I), theselenium layer CL) and the semitransparent electrodes(AI and A2) are vacuum deposited upon the DKDPcrystal. As in TITUS, the DKDP plate is attached byadhesive to a disc of calcium ftuoride, to minimize theinteraction between neighbouring points due to thepiezoelectricity of the crystal. The entire unit is placedinside an evacuated container (fig. 3), where it can becooled to about -50 oe by Peltier elements.

Fig. 4 shows an experimental arrangement for op-erating Phototitus. The polarizing beam splitter P,made of two calcite crystals, performs the functions ofthe polarizer PI, the semitransparent mirror M' andthe analyser P2 in fig. 1.

In fig. 5 the different stages of operation are repre-sented by the potential across the device at some imagepoint. The potential is given by curve a after a d.c.voltage, usually 150-200 V, has been applied betweenthe electrodes A 1 and A 2. Most of the voltage dropappears across the selenium, because of the high valueof Ee in the DKDP. After writing-in curve b is obtained.The image is stored or read out with the electrodesshort-circuited (curve c). For a sufficiently low level ofthe write-in light the voltage VI at the interface in stage

l'If!'!'I'r'i1'l"~""'f1II'Irm'om1t";II!illr"iliiir"ft'i,,'ll'J!;iitMIl!'RI~'(;t!'.il~'~.. • 1\, -'W"W

Fig.3. Experimental model of Phototitus. The multilayerstructure is mounted in an evacuated container; the seleniumlayer can be seen through the window. Two Peltier elements arealso mounted in the container; these cool the system to - SO -c.

Sc

Fig.4. Diagram of experimental arrangement for Phototitus. PT Phototitus. Writing-in takesplace to the left of PT, read-out to its right. The polarizing beam-splitter P is made of twocalcite crystals. LI, L2 lenses. F colour Alter. CL condenser lens. S source of read-out light.M" spherical mirror. Se projection screen. Sw switch for reversing the voltage on PT, orshort-circuiting it.

c is proportional to the exposure at the image pointconsidered (the 'exposure' is the product of the lightintensity and the writing time). Finally, to erase thelatent image, the whole area of the selenium is ftoodedwith light while the electrodes are short-circuited(curve d). A selected area alone can also be erased byftooding that particular area. In fig.5details such asthe presence of a dielectric mirror have been neglected;we shall return to these later on.

The photoconducting layer

Photoconducting amorphous selenium combines anumber of properties that make it remarkably suitablefor Phototitus. In the first place, vacuum deposition ofa selenium layer is compatible with DKDP. This is notthe case with CdS-type photoconductors, because ofthe high deposition or annealing temperature requiredto obtain reasonable sensitivity and low dark current.(CdS-type, i.e. U-VI compound, photoconductors are

Philips tech. Rev. 34, No.IO

used in several of the related devices, because of theirhigh sensitivity in the visible range.) Secondly, as hasbeen found in xerography, charge storage in amorphousselenium is very good, much better than in CdS-typephotoconductors. In certain devices this may not beimportant, particularly when the images are not storedin the photoconductor (see p. 284). Thirdly, at the

Se DKDP

Fig.5. Potential variation in the selenium and the DKDP, atsome image point of the structure, for different stages of op-eration. After a voltage Vt has been applied (a), the optical imageis written in (b). The electrodes are then short-circuited (c). Inthis stage the latent image is stored or read out. It can be erasedby flooding the write-in side with light, with Al and A2 short-circuited (d).

100 ~----\L_ 100%

PS oe

f f\oe\

10 \ 10%\\\

1~----~------L--L--~--L-~--L-_a~1%300 400 500 600 700 800nm

-ÀFig. 6. Photosensitivity (PS) and optical absorption (0:) of a1O-gmthick amorphous selenium layer coated by a gold electrode,as a function of' the wavelength Ä of the incident light. The left-hand part of the photosensitivity curve shows an intrinsic prop-erty ofthe selenium, and holds for both polarities of the voltageacross the layer. The bump at the right is only present- for apositive polarity (AI positive), and is due to holes photoemittedfrom the electrode Al Into the selenium. The PS scale on the leftis in arbitrary units.

PHOTOTITUS 277

temperature of operation.i--öû oe, the photoconductoris ambipolar; the p,r: products for both electrons andholes are approximately equal and their value is high(p, is the mobility and r: the lifetime). This makes thedevice electrically symmetrical, and allows the reversaland subtraction of latent images. Finally, selenium hasa sharp threshold for photoconduction in the short-wavelength part of the visible spectrum. This allows theinput and output to be optically well separated.In the following, we shall go a little further into the

properties of amorphous selenium, when examiningsuccessively the photoelectric conversion of an opticalimage into a charge pattern (taking place in a thinlayer at the selenium surface) the transfer of the chargepattern towards the interface and its storage at theinterface.

Photoelectric conversion

The generation of free charge carriers in selenium byincident photons is a rather complicated process. Theyield of this process, the 'quantum efficiency' 1], is notonly sharply dependent upon the wavelength of thelight but also on the applied field.Let us first consider the dependence on the wave-

length. Fig. 6 shows the photoconductivity spectrumand the optical absorption spectrum of a 10-urn layerof amorphous selenium. These results were obtainedwith a small fixed voltage across the layer, and underconditions similar to the operating conditions forPhototitus; with the layer at a temperature of -50°C,and covered with a semitransparent electrode thatpartly absorbed the photons. The photoconductivitycurve has a sharp edge near 550 nm, the absorptioncurve has one near 700 nm. Yellow light is completelyabsorbed in the layer, but it is ineffective in producingfree carriers. In fact, there are photon-absorptionprocesses in selenium that do not produce free carriers.From the data used for fig. 6 it can be deduced that thephotons that do produce carriers efficiently are allabsorbed within a layer less than one micron thickbehind the electrode. The drop in photoconductivitynear 550 nm isjust a drop in quantum efficiency [4].

As mentioned above, the presence of an edge in thephotoconductivity curve is rather fortunate: it permitsa high optical isolation between input and output. Thewrite-in light can be partly or wholly in the blue. Ifyellow light (wavelengths larger than 550 nm) is usedfor reading out, the latent image-is little affected by it.This-point will be further discussed later.In- selenium the quantum efficiency is rather low at

-50°C, as long as the applied field is weak. In fieldshigher than 104 V/cm, however, 1] increases strongly[4) J. L. Hartke and P. J. Regénsburger, Phys, Rev. 139:A 970,

1965.

278 F. DUMONT et al. Philips tech. Rev. 34, No: 10

with field-strength. In Phototitus, at low light levels, thefield-strength in' the selenium layer is of the order of105V/cm (100 V across a 10-(Lmlayer). As we shall seebelow, our results indicate an efficiency' of about 20%at such a field-strength. Of course, when the field in thelayer decreases as charge builds up at the interface (seefig. 5, curve b), 1] decreases again. The field dependenceof 1] is thus an important factor in the saturation behav-iour of Phototitus. In our discussion of saturation,which we shall return to later, we have assumed that 1]

depends on field-strength E in accordance with therelation:

1]'= 1]0exp(-Eo/ke + j3Et/ke), (2)

proposed by D. M. Pai and S. W. Ing [5] and by M. D.Tabak and P. J. Warter [6]. In eq. (2), e is the absolutetemperature, k Boltzmann's constant, and Eo an ac-tivation energy at zero field. The factors 'YJoand 13, andalso Ee, may depend on the wavelength A of the light.

In the theoretical model on which eq. (2) is based, the prob-ability that photons will generate a pair of free carriers at zerotemperature and zero field is zero; 1)0 is the probability of creat-ing a pair of one hole and one electron, bound together by Cou-lomb forces. The hole and the electron may separate throughthermal excitation, and this process is assisted by the electricfield. The situation closely resembles that of jig. 7, representingthe situation for an electron bound to a jixed positive pointcharge. The activation energy required to liberate the electrondecreases with increasing field-strength. This is the 'Poole-Frenkel effect'. Elaboration of this model (5] (6] leads easilyto eq. (2).

Field-induced transfer of the charge pattem

The efficiency of the transfer of the charge patternacross the bulk of the selenium layer is given by theratio R of the lifetime l' to the transit time 1'tr of thecharge carriers. If f.l is the drift mobility of the carriers,ls. the thickness of the layer and V the voltage across it,the drift velocity is f.l Vlls: Thus 1'tr = lL2/ f.l V, andR = wr:V/h2• Therefore, as far as the material is con-cerned, it is the product f.l1' that counts.In selenium f" is the product of a microscopie mobility

and a thermal activation Boltzmann factor associatedwith trapping by shallow centres [6] [7]. Thus f" is smal-ler at -50 oe than it is at room temperature. From themobility data [8] for -50 oe, we find, with V = 100 V,IL = 10 (Lm,a transit time of about 0.5 (Lsfor holes and100 (Ls'for electrons. In practice, high-quality imagescan be formed by using a 10-(Lswrite-in flash.

The lifetime ~ is determined by trapping by deepcentres. When a charge carrier is trapped by a deepcentre, this becomes a recombination centre for carriersof the opposite sign.

The f.l1' product at room temperature, as derivedfrom the values of f.l and l' that have been found [9], is

about 3 X 10-7 c,m2/V for electrons, and ranges fromabout 3 X 10-7 to 10-6 cm2/V for holes. From the im-age-transfer characteristics of Phototitus, to be discuss-ed later, wefind values for f"1' of 3 to 10x 10-7 cm2/Vat -50 oe, for both holes and electrons. To indicate thepractical meaning of these values, we note that, withV = 100 V, ls,= 10 (Lm, they correspond to lifetime-to-transit-time ratios R between 30 and 100.

Q

Fig.7. Potential energy of an electron bound to a fixed positivecharge, (a) without and (b) with an applied external field. Theenergy required to free the electron from the lowest bound state(Eo in a, El in b) is lowered when a field is applied. This is thePoole-Frenkel effect.

TO-ljf'

50 TOO W/cm2

TmW/cm2

-ljf

Fig. 8. Storage time om as a function of the intensity tp of theread-out light. The upper curve (om+) is for a positive latentimage, the lower curve (om-) for the negative one. tp' is the resid-ual intensity of the light that has passed through the dielectricmirror and is absorbed by the selenium. Wavelength ofthe read-out light 579 nm.

Philips tech. Rev. 34, No. 10 PHOTOTITUS ' 27~

Image storage

The persistenee of a charge pattern built up near theinterface between selenium and the dielectric mirror islimited by two processes. In the first place, charges maybe thermally activated from the surface traps and dif-fuse laterally, which leads to progressive erasure andloss of definition. Secondly trapped charges may beannihilated by capture and recombination of freecharges of opposite sign, leading to decrease in contrastwithout degradation of the definition. This is thedominant process in practice.In the dark (without read-out light) thermal detrap-

ping is very slow. This is a subsidiary advantage of op-eration at a low temperature. The storage time is of theorder of an hour. On the other hand, when the DKDPis illuminated by the read-out beam, the residual trans-mission of the dielectric mirror (> 5% near thereflection maximum) allows some light to reach theselenium layer. Although the efficiency is low at theusual wavelength for observations (A ~ 560 nm), thislight generates free carriers, some of which recombinewith charges of the latent image, thus erasing it (thesecond process mentioned above). Fig.8 shows thestorage time om as a function of the intensity 7jJ of theread-out light, with logarithmic scales. The curve for apositive latent image (the upper curve) has a slope lessthan unity. This can be explained by assuming thatsome of the holes generated by the read-out light dis-appear by a process of bimolecular recombination; ifthis were the only process, the carrier density would beproportional to the square root of the luminousintensity [lOl and the slope would be 0.5. For a negativelatent image erasing is much faster (lower curve). Thiscan be explained by an additional recombination pro-cess, due to holes injected into the selenium by photo-emission from the electrode Al [lIl. In the case of agold electrode, only holes are photoemitted. The photo-sensitivity spectrum of this process is represented bythe bump in the red region in fig. 6. The photoinjectedholes are attracted by the negative charge of the latentimage. Their density is proportional to the read-outlight intensity, leading to a mean storage time inverselyproportional to this intensity. In both cases there is nosign of loss of definition during the erasure.

Image-transfer characteristic

It is important in practice to know the transmittanceT of the system for the read-out light as a function ofthe exposure H on the write-in side. This functionenables the exposure necessary to obtain a certain con"trast to be determined. Such an 'image-transfer charac-teristic' for exposure. wavelengths, centred around401 nm and an applied voltage of 150V (AI positive),

is shown injig. 9. The characteristic is rather similar forthe opposite polarity, which demonstrates the ambi-polar nature of the selenium layer at the temperatureof operation.The experimental characteristics are generally S-

shaped, and three zones can be distinguished: a resid-ual-transmittance zone, an intermediate zone and asaturation zone. When the exposure approaches zero,the transmittance tends to a constant level, owing tostray light scattered or reflected byelements of theoptical system. The curve in fig. 9 was obtained with avery low level of stray light, and does not show thiseffect. When this level is not so low, the same curve isobtained after the residual transmittance is subtractedfrom the experimental T-values. After correction forstray light, T is proportional to the square of the expo-sure, for small exposures (intermediate zone). Thiscorresponds to the first-order expansion of eq. (1),where V, the voltage across the DKDP, is proportionalto the exposure for small exposures. For large expo-sures the transmittance saturates to a constant value

0.5T

r0.1

p.T,;!0-3cm'/V---~ 10-'/_--.../

s-: ----3xl0-7

_---'10-7./ ./ :

/ I./ _-3xl0-8/ "'/ //

!J.V //, / _--10-8

I. / »"~ / /.~/ /1',/ /

//'1 /II /I! /I /I

0.2

0.05

0.02

2 5 50fJJ/cm2

-HFig. 9. Image-transfer characteristic. Transmittance T for theread-out light is plotted against the write-in exposure H. Write-inwavelength 401 nm. Applied voltage 150V. Solid curve: exper-imental results; dashed curves: theoretical predictions for dif-ferent values of In: arid a value ofO.20 for 1) in afield of 105V/cm.

[5) D. M. Pai and S. W. IngJr., Phys. Rev. 173, 729, 1968.(6) M. D. Tabak and P. J. Warter Jr., Phys. Rev.173, 899,1968.[7) W. E:Spear, Proc. Phys. Soc. 'B 70, 669; 1957. . ' ''-'(8) J. L. Hartke, Phys, Rev. 125, 1177, 1962. . .•... ,

J. Schottmiller, M. Tabak, G. Lucovsky and A. Wárd,J. non-cryst. Solids 4, 80, 1970. '. .,See also the article by W. E. Spear, note [7].

(9) See the articles of note [8], and also M. D. Tabak and W. J.Hillegas, J. Vac. Sci. Technol, 9, 387, 1971. .

[10) See for instance R. H. Bube, Photoconductivity of solids,. Wiley, New York 1960.

[11) 'J. Mort, F. W. Schmidlin and A.1. Lakatos, J. appI. Phys. 42,5761} 1971. .. ~ . . ..' ,',

280 F. DUMONT et al. Philips tech. Rev. 34, No. 10

Ts, depending upon the. applied voltage. The presenceof stray light together with the saturation of the trans-mittance limit the maximum contrast (ratio of trans-mittances) obtainable at a given voltage.

Stray light can be reduced to a levellower than 1%of the read-out intensity by careful design and adjust-ment of the optical equipment and by depositing anantireflection coating on the optical components of thedevice. Furthermore, when a very low residual trans-mittance is required, it is necessary to use a read-outbeam accurately perpendicular to the DKDP singlecrystal, in order to avoid unwanted depolarization bythe natural birefringence of the material [121.With thisprecaution, we have obtained a maximum contrastmuch higher than 100 to 1.Saturation occurs, first of all, because the potential

drop across the selenium (see fig. 5, curve b) eventuallydisappears as charge accumulates at the interface, sothat no more carriers will drift across the layer. If thatwere all, the relation between Ts and Vtwould be givendirectly by eq. (1). The matter is however complicatedby several factors. First, the dielectric mirror reducesthe voltage across the DKDP to 2/3 of the voltage atthe mirror-selenium interface (see fig. la). Secondly,charge carriers caught in deep traps in the bulk of theselenium give a space charge, and thus a curvature ofthe potential characteristic (also indicated in fig. 10).Finally, in accordance with eq. (2), the quantum effi-ciency 1') falls off rapidly as saturation is approached.We have worked out a theoretical model that takes

these factors into account, and we have calculatedT-H curves and Ts-Vt curves from it. The main param-eters in the calculation are /ho and the value of 1') ata particular field-strength. Fig.9 shows, besides theexperimental curve, some calculated T-H curves. Achange in the value of 1') is equivalent to a change inexposure and therefore shifts the curves in the horizon-tal direction. Fitting the experimental and calculatedcurves in this way, for a wavelength of 401 nm and atemperature of -50 oe, yields a value of 0.2 ± 0.03for 1') at a field-strength of 105V/cm. With the sameconditions of field and temperature Pai and Ing founda value of 0.3 [51.

The position of the saturation plateau in fig. 9 yieldsdirectly an estimate for /ho of about 10-6 cm 2/V. Amore accurate way of determining /hoconsists in fittingthe calculated T« Vt curve to the experimental data, asinfig.ll. The best value thus found is 3X 10-7 cm 2/V.The set of calculated curves in fig. 11 is useful for

evaluating the minimum value of /ho compatible withthe contrast required for a given application, This prob-lem is of practical importance in the technology of theconverter, since for long-term reliability of the devicethe selenium layer has to be stabilized by doping it with

glass-forming additives. However, as tlo is usuallyaffected byeven a small quantity of impurities [131,acompromise has to be found which depends on whetherhigh contrast Ol long life is most important.

O~----_L--_L------------~Fig. 10. The saturated transmittance T« is determined via eq. (1)by the saturation value Vs of the voltage across the DKDPcrystal. Vs is lower than the tube voltage Vt because: a) Vs isonly t Vs', where Vs' is the voltage at the selenium/mirror inter-face; b) trapped charge carriers give a space charge, bending thepotential curve in the selenium; c) 11 depends strongly on thefield. (This last effect is not taken into account in the figure.)

0.9

7: 0.8s

0.6

0.5

0.4

0.3

0.2

0.1

-\.'tFig. 11. Saturation transmittance TB plotted against tube voltageVt. Experimental points: + electrode Al positive, 0 electrodeAl negative. The solid lines are theoretical curves for differentvalues of ut:

Philips tech. Rev. 34, No. IQ PHOTOTITUS 281

In many applications high definition of the storedimage is a crucial requirement. Calculations made byJ. Donjon of this laboratory have shown that thetheoretical resolution is 40 line pairs per mm with amodulation of 10%, or 80 I.p.jmm with a modulationof 5%. As in related devices, the main source of pointspreading is the fringing of the stored electric fields [14].

By using incoherent read-out light at 590 nm, we haveachieved a resolution of more than 75 I.p./mm onPhototitus (see fig. /2). Th is value corresponds to 5700elements per horizontal line, since the DKDP crystalshave a width of 38 mill. This kind of high-quality imagecan be stored and displayed during several tens of sec-onds without degradation of the resolution with a totalincident luminous flux ofabout I lumen on the DKDP.This enables an output illuminance of about 60 lux on Applications

Resolution

UlO JHID iSO 700

11111 _11 III IIII75D liDO lI51J

300 550 4001111 1111 Ill"450

600 !iSO lOaIJllllllmllll1

Fig.12. Photographs made from images stored in Phototitus.Above: standard television test pattern. Below: high-definitiontest pattern. The pitches of the vertical lines are 42 fLll1, 35 fLm,24 fLm and 16 fLm, from left to right in the lower photograph.The 16 fLm pitch is still clearly discernible: the limiting resolutionattained in practice is 75 I.p.(mm.

the surface of a 9 x 1J cm screen corresponding to stan-dard photographic film. It has not previously beenpossible to store an image with such a high resolutionat higher luminous flux (e.g. more than 10 lumens) fora long time, because of the existence of the variousrecombination processes in the storage layer discussedabove. We have however managed to store and displayhigh-illuminance images of 100 lux on a t m2 screenfor 30 seconds by depositing an 'optical shield' betweenthe dielectric mirror and the photoconducting layer.But in this kind of arrangement the resolution wasnever better than about 10 I.p.rrnm because of thefinite - although very small - lateral photoconduc-tivity of the optical shield layer.

Some examples of image processing by Phototitusare shown injigs. 13, 14, 15 and 16. In these examplesthe read-out was performed with ordinary incoherentlight.

Fig. 13b shows the result of subtracting a uniformcharge distribution from the latent image of fig. l3a.This is done by simply making a uniform exposure withreversed voltage. Contours of a particular grey level inthe original show up as black contours in the treatedpicture. A similar result can be obtained by subtractingan 'optical bias' (i.e. a phase shift) by means of aBravais compensator or a rotabie quarter-wave platein series with the device. Fig. 13e and d indicate howcharge variations along a line in the latent image aretranslated into variations of transmittance. In fig. J 3ethe original variation is reproduced only slightly dis-torted. In fig. 13d, however, where a 'uniform image'has been subtracted, lines of zero charge (lines of aparticular grey in the original) become black lines.Itwill be clear from fig. l3dthat a complete 'positive-

to-negative image conversion' can also be obtained, bysimply subtracting more charge than in the case offig. 13d.

Writing a picture into Phototitus and subtrac.ting itagain after a slight shift in a particular directionamounts to determining the spatial derivative of thegreyness in that direction. Differentiation can be usedto enhance contours or small details in pictures (fig. 14aand b); in such cases the subtraction is incomplete topreserve part of the original image. As indicated infig. 14e, positive and negative derivatives will all showup as white lines because of the quadratic nature of thephotographic plates. However, as is seen in fig. 14d,

[12] See the article by G. Marie and J. Donjon, note [I].[13] See the article by J. SchottmilJer et al., note [8].[14] D. S. Oliver and W. R. Buchan, IEEE Trans. ED-1S, 769,

197 J.

282 F. DUMONT et al. Philips tech. Rev. 34, No. ID

a

-p

T(x}

tT

t

-x

-p(x}

b

T

tT(x}

t

-x

d

----- p(x}

Fig. 13. Obtaining lines ofa particular grey level. (I) Original picture, b) picture obtained aftersubtracting a 'uniform image' (i.e. a uniform exposure). The diagrams show how a chargevariation e(x) in the latent image is translated into a transminanee T(x), for the original (c)and for the treated image (d).

negative derivatives can be made to show up as blackand the positive ones in white, by adding a uniformimage to the differentiated image.

Fig. 15 shows the effect of subtracting a slightlyblurred picture from a sharp original. The result is acontour enhancement in all directions. This procedureamounts to an approximate detennination of thesecond derivative (fig. 15c), or, more precisely, theLaplacian 02/0X2 + 02/oy2 of the greyness (where xand y are the coordinates in the plane of the picture).

These almost instantaneous methods of determiningconstant grey-level contours and of contour enhance-ment may be useful where pictures have to be scru-tinized for particular details, as in medical X-raydiagnosis.

In cases where a scene is almost but not completely

stationary, changes in the situation may be detected bysubtracting two successive pictures. In the limit thisamounts to determining the time derivative. This feat-ure could be applied in security monitoring, or in theanalysis of pictures from weather or Earth-resourcessatellites.

The addition of pictures is the basis of integrationand averaging. For instance, restoration of photo-graphs or television images affected by opticalorelectrical noise during the pick-up or transmissionstages can be made in this way. This process is illus-trated in fig. 16.

The most important application of Phototitus willprobably be in image processing with coherent light.Since the advent of the laser many schemes for coher-ent optical processing of images and data have been

Philips tech. Rev. 34, No. IO PHOTOTITUS

Fig. 14. Small-detail enhancement by differentiation with part of the original preserved.a) Original. b) Result obtained by (partial) subtraction, from the original, of the same picture.c) The principle of first-order differentiation. The charge distributions of the original (1), theshifted (2) and difference (3) images are indicated schernatically. Both leading and trailingedges show up as white because the photographic film is a quadratic detector. d) Adding auniform image will make one kind of edge show up as white, the other black.

a

p

t

0------x

P T

t t 4

0 0-x

c

devised [15]. ln seismology, for instance, the spatialfrequencies present in a particular pattern are of inter-est and these can be identified by the method indicatedin fig. 17. The Fourier transform of the transparencyT will appear in the focal plane F of the lens L if thebeam B is coherent and T is in the other focal plane of

b

p

t v-x --x

T

to OL_----------~--~

d

L. Thus gridlike patterns in T can be identified by spotsin F. Another of the many applications of coherent op-tical processing is in character recognition, as indicated

[15) See for instance A. Yander Lugt, Optica Acta 15, 1, 1968,and also J. W. Goodrnan, Introduetion to Fourier optics,McGraw-Hill, New York 1968.

284 F. DUMONT el al. Philips tech. Rev. 34, No. 10

a

b

c

Fig. IS. Omnidirectional con-tour enhancement. a) Originalpicture and b) treated picture,obtained by subtracting ablurred picture from the origin-al. c) Charge distributions of theoriginal (1), the blurred picture(2), and the difference (3); thecurve 3 is approximately thesecond derivative of 1.

inftg.18. When a holographic filter of a character isprepared as in fig. 18a, tbis character ifpresent in sometext on a transparency T, will be identified in thearrangement of fig. 18b by a spot on the screen S. Thismethod can be extended for fast automatic reading [16].

Until now there has been one great difficulty withsuch methods. Because of the speckle effect observedon laser illumination of a stationary scattering object,coherent optical processing of images on data dis-played on a scattering medium (e.g. paper) is notin general possible. In practice high-quality photo-graphic film is used. This, however, excludes suchapplications as fast reading of ordinary print on paper,which would be very useful for feeding printed infor-mation into a computer. Making a film of such aprint, to be read perhaps only once, is a slowandwasteful process. In short, the method is inhibited bylack of 'real-time, re-usable film'.It will be clear from the foregoing that Phototitus

offers a solution to this problem. Print projected on thewrite-in side can be read out by coherent light. Fig. 19shows the result of an early character-recognitionexperiment in which Photo titus was used in this way.We estimate that more than 20000 characters per sec-ond could be read with a developed version of themethod. This is about ten times faster than the fastestoptical reader generally available.

Finally, we would like to mention the possibility ofusing Phototitus to eliminate the 'zero-order diffractionterm'. In many coherent optical processi ng ex peri ments,e.g. those of figs. 17 and 18, the zero-order term of theoptical Fourier-transforrn, representing the averagepicture amplitude, carries no interesting information:it yields a much brighter spot than the useful informa-tion, and scattering by the filter plate of a small frac-tion of this spot could be a serious source of noise.With Phototitus, the zero-order term can be eliminatedby subtracting from the latent image a 'uniform' imageof amplitude equal to the average amplitude of theoriginal, or by using an optical bias. A reduction ofthe zero-order diffraction term by several hundredtimes has been achieved.

Comparison with related optical devices

In this final section we shall briefly compare Photo-titus with other devices that have been developed inrecent years for image processing, which are also basedon the idea of combining photoconductivity with anelectro-optical property.

TnTable J we have collected the most important datafor different types of converters, including Phototitus.Comparison of these data should be made with somereserve, because the conditions of measurement were

Philips tech. Rev. 34, No. 10 PHOTOTlTUS 285

Fig. 16. Signal-to-noise enhancement by addition, on Phototitus, of images affected by un-correlated noise. Left: image obtained by placing a diffusor in front of a picture and projecting iton Phototitus. Right: imageobtained with the diffusor rotating duringtheexposure. The rotatingdiffusor simulates uncorrelated noise, which is added during the exposure in the right-handcase. The total exposure is the same for both cases.

L

T

B

f f

Fig. 17. Analyses of spatial frequencies by coherent optical pro-cessing. Tsubject, in the form of a transparency, to be analysed.B coherent beam of light. L lens. If Tand F are in the front focalplane of L, the two-dimensional spatial Fourier transferm of Twill appear in the back focal plane F. Grid-like srruciurcs in Tgive dots in F.

not always reported in detail, and also because of dif-ferences in the physical principles employed.

Devices I and 2 in Table I are of the 'Fe-Pc' type inwhich a photoconductor (Pc) is combined with a ferro-electric material (Fe). (By analogy Phototitus would beof the 'Pa-Pc' type.) In these devices images can bestored in the ferroelectric material because of the op-position of the coercive force to the relaxation of'flipped' domains. They are therefore operated at atemperature below the Curie point. Phototitus, how-ever, is operated above the Curie point, so there is nodanger of coherent light being scattered by domainwalls. Devices based on ceramics [17], which are mostuseful in other applications, do of course scatter coher-

[16] See for instance M. Treheux, in: Applications de l'holo-graphie, Proc. int. Syrnp., Besançon 1970, paper No. 13.3.

[17] J. R. Maldonado and A. H. Meitzler, Proc. IEEE 59, 368,1971.D. W. Chapman, J. Vac. Sci. Techno!. 9, 425,1972.W. D. Smith and C. E. Land, Appl. Phys. Letters 20, 169,1972.

A F(A)

B

Q

5T F(A)

bB

Fig.8. Character recognition by autocorrelation. a) Preparationof a holographic filter F(A) of the letter A, presented as a tram-parency. Beoherent beam of light. M', and M2mirrors (MI semi-transparent). b) Searching through a text on the transparency Tfor the letter A. F(A) holographic filter, as prepared in (a). L1,L2 lenses. S screen. Each letter A in T will give a separate auto-correlation spot in S, related to its position.

Fig.19. Automatic reading of the letter A with the method offig. 18, where, instead of A in fig. 180, and T in fig. l8b, Photo-titus was used, translating print on paper into a transparency.Above: text. Below: dots identifying the letter A.

286 F. DUMONT et al. Philips tech.i Rev. 34, No. 10

ent lightcompletely; for this reason they are not consid-ered here.

In the devices 3 and 4, the 'Ruticon' and 'Fericon',the electro-optical effect is not a bulk effect, as in Photo-titus and the other devices, but a surface effect. Theoperation of the Ruticon is based on the surface de-formation of a membrane, whereas the operation of thePericon is based on the deformation of a metallic layerdeposited on the surface of a ferroelectric material. Adisadvantage of using a surface electro-optical effect isthat only a relatively low contrast can be obtained.

hundred volts, whereas the Bh2Si02o PROM needsvoltages of the order of a thousand volts. The maximumread-out transfer ratio of Phototitus is higher than inthe PROMs, because of the high transparency of theDKDP single crystal and the electrode A2 in fig. 1, andalso because of the possibility of approaching a 900

phase retardation, on account of the high electro-optical sensitivity of DKDP and the high transferefficiency of selenium.The DKDP crystals used in Phototitus have an

area five times that of the Bh2Si02o crystal used in

Table J. Data for several types of optical converters. VJ./2 applies only to devices using thePockels effect (the PROMs and Phototitus). It is the voltage across the electro-optic crystalnecessary to obtain a phase difference of n between the x- and j-components (see fig. 2), i.ea 100% transmittance; VJ./2 equals nf2K, where K is the factor used in eq. (1).The 'writing sensitivity' is the exposure (energy per cm 2) required on the write-in side to

obtain a 'read-out transfer ratio' of 10 or 20 % of its maximum value. 'Read-out transfer ratio'is the intensity of the read-out light leaving the device divided by the intensity of the incidentread-out light. The term 'grating transfer ratio' is used when the only part of the read-out lightcounted is that present in a beam diffracted by a latent image in the form of a grating.

Active Sensitivity Write-in Max. Max. Max. read- Storage GreyType of device area VJ./2 for writing wavelength resolution contrast out transfer time scaleratio (dark)

1. Fe-Pc 1000 [lJfcm2 514.5 nm 70 Lp.fmm 5 : I 0.01%["] days limitedBi4Ti3012 (binary)+ ZnSe [18] ----

2. Fe-Pc limitedBi4Tia012 0.16 cm2 2500 [lJfcm2 114Lp·fmm (binary)+ PVK

3. Metal-platedRuticon > 20 cmê 30 [lJfcm2 632.8 nm 40 Lp.fmm 1.8 % [**] very good+PVK [19]

4. Fericon [20] I I I 57Lp·fmm I 2.2: I

5. PROM [14] 2 crn'' I 13 kV I 10 [lJfcm2 340 nm

I 85 I.p.fmm I__I~I 2% I lOOh goodZnS or ZnSe

6. PROM [21] I 2.25 cm'' 3.9 kV I 12[lJfcm21 400 nm I 80 Lp.fmm 1000 : I 8% 2h goodBh2Si020

7. Phototitus 12 cm- ISOV I 10 [lJfcm2 401 nm 70Lp·fmm > lOO : I 50% th very good

r-: Grating transfer ratio at 632.8 nm.[**] First-order grating transfer ratio for a latent image of 20 Lp.jrnm.[18]S. A. Keneman, G. W. Taylor, A. Miller and W. H. Fonger, Appl. Phys, Letters 17, 173, 1970.[19]N. K. Sheridon, IEEE Trans. ED-19, 1003, 1972. .[20]C. E. Land and W. D. Smith, Appl. Phys, Letters 23,57, 1973.[21]P. Vohl, P. Nisenson and D. S. Oliver, IEEE Trans. ED-20, 1032, 1973.

Devices 5 and 6 are 'PROMs' (Pockels Read-outOptical Memories). In these the two functions are per-formed by one material, which is photoconductive aswell as electro-optic.

Inspection of Table I shows that only the PROMbased on a single crystal of Bh2Si02o could be a seriouscompetitor to Phototitus for incoherent-tc-coherentoptical conversion. Phototitus, however, has the ad-vantage of a much higher electro-optical sensitivity. Itcan be operated by switching voltages of the order of a

PROMs; they can nevertheless be made sufficientlyhomogeneous to givegood images over the whole area.Finally, we should note the importance of the phys-

ical separation ofthe photoconductivity and the electro-optical effect, which permits both to be optimized. Asan example, we have seen that, in the case of Photo-titus, it is possible to select an ambipolar photo-conductor, enabling us to manipulate both positive andnegative charge images, and to subtract them accur-ately. This is not possible with the PROM based on

Philips tech-Rev. 34, No. 10 PHOTOTITUS 287

ZnS [141.Image subtraction was not reported in the caseof the Bh2Si02o PROM [211.Function separation is also essential for preserving

the image while it is read out. As explained in the fore-going, read-out light can produce free carriers in thephotoconductor. This unwanted effect becomes seriouswhen the read-out beam passes through the photo-

Summary. Phototitus, a device for image processing, consists ofa 30 x 40 mm single-crystal slice ofDKDP, 250 fLmthick, coveredwith a dielectric mirror and a layer of amorphous selenium,10 fLmthick. Semitransparent electrodes cover both faces of thestructure. With a voltage of 150-200 V between the electrodes, anoptical image projected on to the selenium is converted into alatent charge image at the interface. The Pockels effect in theDKDP permits the latent image to be read out with polarizedlight. A large Pockels effect is obtained by cooling the DKDPto just above its Curie point (-50 0c) by Peltier elements. Thestorage time of the latent image in the dark is tens of minutes.Images can be added by writing them in one after the other.

conducting layer itself as in PROM, whereas in Photo-titus it is reduced by two orders of magnitude by thedielectric mirror. As a result, Phototitus allows animage to be observed for a few minutes with yellow-orange light (590 nm), where the eye is very sensitive,whereas images in the Bh2Si02o PROM are veryrapidly erased if the read-out light is not red.

An image is subtracted by writing it in with the voltage reversed.This possibility is due to the ambipolarity of the selenium. Thusimages can be integrated, averaged or differentiated with respectto position or time, leading to such applications as the eliminationof uncorrelated noise, contour enhancement and the detection ofchanges in an image. An important application could be as 'real-time re-usable film', i.e. the instantaneous conversion ofprint intoa 'transparency' that can be processed by coherent light, e.g. forcharacter recognition. A brief comparison is made with otheroptical converters. The relation between the characteristics of thedevice and the intrinsic properties of the photoconductive sele-nium layer is discussed.