207 SIMULATION OF SEMICONDUCTOR DEVICES AND PROCESSES Vol. 3 Edted by G. Baccaran, M. Rudan - Bologna (Italy) September 26-28,988 - Tecnoprnt Computer Smulatons of Parallel-to-Seres Converson n Sold State Frame Transfer Image Sensors J. Bsschop Phlps Research Laboratores P.O. Box 80.000, 5600 JA Endhoven, The Nehterlands Abstract In developng a sold state mage sensor wthout the problems of ncomplete charge transfer or nsuffcent charge handlng capacty t s necessary to smulate the electrcal behavour of the devce. We perform these smulatons usng a 3D off-state and a 2D on-state semconductor smulaton package. The use of a 3D package s necessary because of the complcated structure of the devce. As an example of the applcaton of these smulatons the calculaton of charge transfer effcency n the parallel-to-seres converson n a frame transfer mage sensor s explaned. Introducton The frame transfer mage sensor s a type of sensor n whch the whole mage area s senstve to lght. It conssts bascally of three sectons: mage, storage, and read-out secton (see Fg. ). The prncple of operaton s as follows. Photons ncdent on the mage secton generate electron-hole pars n the bulk of the slcon. The electrons are collected n each pxel under postve electrodes, whle the holes are draned off. The mage and storage secton consst of parallel runnng vertcal CCD regsters. Charge packets ntegrated n the pxels of the mage secton are transported quckly through these CCD regsters to the storage secton. After that a new feld s ntegrated, whle the storage secton s read out lne by lne. The horzontal read-out regster conssts of three parallel CCD regsters to allow a hgh pxel densty at relaxed desgn rules. The lnes are read out from the horzontal regsters at the proper vdeo frequency. All regsters are bulk n-channel CCD regsters.
208 WnckWl ' n-d J!CG pntc-rr?68 nou pet ted!!j'!;r! -!!!-- -,- "j~r_t * J' r J TG, 7G?! 73 TG, ^ ) ^ -JTL L ;'" " ' /./' xo! : : : 2c] c 3c' XC j xd,- r be A lc XC] XO J j L ~^v~ r 2C! C XC j XD '! XC, ' XC; < ', 2C c] 2C XR ' XC / f IC XC; J - " / 2C* IC j XC '.-,.. f! XD Fg. J, Schematc representaton of the frame transfer mage sensor. Fg. 2. Output regster of the mage sensor. The shaded areas are channel stop (SP) regons. IC, 2C ndcate the three phases of the horzontal read-out regster. XA,, XC are the frst, second, thrd poly-s layer respectvely. Vertcal charge transport durng parallel-to-seres converson occurs preferentally along the lne A-A'. The dashed lnes ndcate the area used n the smulaton of Fg. 3. The charge packets contan at most about 20U,000 electrons. The senstvty s lmted by nose whch amounts typcally to 20 to 70 electrons. So, the sensor s an analog devce whch must be able to handle very large as well as very small charge packets. Ths places hgh demands upon the accuracy of devce smulatons. All sensors are desgned wth the help of computer smulatons n order to prevent problems wth charge storage and transfer of charge packets. Due to the small scale of the detals of the devce 3D calculatons are necessary to obtan relable results. Furthermore, t s necessary to have the possblty of performng on-state (transent) calculatons n order to study charge transport. We have at our dsposal the packages Paddy (3D off-state) and Curry (2D on-state) (Polak, den Hcyer, Schlders. 987). In the next secton the use of these packages wll be demonstrated on the bass of a problem n the parallcl-lo-seres converson at the transton from the storage secton to the horzontal read-out regster.
209 Parallel-to-seres converson A part of the transton between the storage secton and the horzontal readout regster s shown n detal n Fg. 2. In the mage and storage sectons the vertcal CCD regsters are shfted n parallel. In the horzontal read-out regster the pxels are read out serally. Charge packets n vertcal channels debouchng nto C are transported down to the bottom regster (OB) through OT, TG2, OM, and TG3 regsters. The packets above 2C are transported to OM and the packets above stay n the top regster OT. After that the packets are transported n parallel n the three regsters. The structure whch provdes ths converson s rather complcated. The onzed dopants together wth the voltage dfferences between the electrodes create electrc felds for the transportaton of charge n the horzontal as well as n the vertcal drecton. Moreover, the onzed dopant atoms create a potental dstrbuton perpendcular to the surface of the devce. Therefore 3D smulatons are necessary. I 6 50 - l,,. l. ',»' *?'"*:' ;] ;: :; a u, -0.6 I 0 5 20 25 x (ym) Fg. 3. Contour map of the channel potental under the C electrode. There s a large potental well under C and a small one under the TG lp. Fg. 4. Channel potental along the drecton of charge transport from C to TG3. One of the problems encountered n parallel-to-seres converson s mperfect charge transfer from C to TG3 (or TG2). The effect of such an mperfect-on s that charge from OB stays behnd n OM or OT, leadng to dark and brght vertcal strpes n the dsplayed mage. The problem s due to a potental barrer near the TG entrance. The barrer s caused by the p-type channel stop regons. Fg. 3 shows a contour map of the channel potental (= potental at depth of charge transport) under the C electrode. The map was obtaned from 3D Paddy calculatons. The fgure shows a potental barrer near the TG entrance. The barrer can only be found wth 3D calculatons. Fg. 4 shows a plot of the channel potental along the lne of charge transfer A - A' n Fg. 2. It s not possble to make the barrer dsappear by rasng the TG voltage.
20 On-state calculatons There are several ways lo solve the problem. We shall menton and elucdate a few possbltes, () Shorten the C electrodes. Ths ncreases the drvng electrc feld from C to TG, and hence lowers the barrer, () Enlarge the lp of the TG electrodes. The lp serves to reduce the potental barrer and to draw electrons under TG. Both methods lmt the charge handlng capacty of the horzontal regsters. Therefore, the maxmum amount of charge whch can be held under the C electrodes has been calculated wth Paddy, () Reducton of the SP mplant reduces the repulsve acton of the channel stop regons and consequently results n reducton of the barrer. In order to determne what demands a soluton must satsfy we must know what potental barrer heght stll yelds undsturbed charge transfer. It s clear that no barrer s the best case, but ths turned out to be hardly feasble n practce f we wsh to allow a certan degree of msalgnment. Therefore we studed charge transport over such a barrer wth the 2D on-state package Curry. From 3D calculatons the channel potental along the lne ol charge translcr (A - A' <-. J_\ss, TG3, TGM n TG33 C dp C,3 S.O? Sj ) LlL co -" s, f / n V,,.35V TG3=6V c «3V c?.c3.3v p 3 V tje" N u= 8v f a ^ _ K, s 2 ^ n $ to _ Z n!m «5! ^\ v.. z s, - Fg. 5. Input structure for the 2D on-state calculatons. The thck lnes ndcate metal electrodes. Tle electrodes TG3 and C have been dvded nto three parts. Fg. 6. Channel potental along A-A' (Fg. 2) from 3D ( ) and 2D (- -) off-state calculatons.
2 n Fg. 2) was determned. Ths regon was then used as nput regon for Curry (see Fg. 5). In the 2D model n-contacts have been added on both sdes, whle a p-contact s on one sde only. Furthermore, two swtchng electrodes SI and S2 have been added to solate the C - TG regon from the contacts. In order to obtan the rght channel potental everywhere, the electrodes C and TG were devded nto three parts each. The potentals on these electrodes were adjusted such that the 2D off-state calculatons produced a good ft to the results from Paddy (see Fg. 6). The heght of the potental barrer can be adjusted wth the electrodes TG33 and Cll. The 2D on-state calculatons are rather complcated. In the frst run all potentals are set to 0 V. In the second (transent) run the holes have to be removed by drvng them to the p-contact. After that the electrons are removed by the n-contacts, except for a certan amount of charge whch stays under C. Ths s the startng stuaton for our on-state calculatons. The swtches SI and S2, and TG are negatve, whle C s postve. The next transent takes TG to a postve voltage. Charge flows over the potental barrer from C to TG. Dependng on the barrer heght a certan amount of charge stays under C. C goes negatve n a transent after 200 ns. Most of the rest charge s drven to TG. The results are shown n Fg. 7. I 'I V,,, _l L 30 0 50 00 50 200 250 0.0 0. 0.2 0.3 04 0.5 I!nj) _, y (volt) Fg. 7. Charge transfer from 2D smulatons for dfferent barrer heghts. At t = 0 TG3 goes postve. At t = 200 ns C goes negatve wthn 20 ns. Fg. 8. Rest charge n C as a functon of barrer heght from 2D calculatons, (a) start of C transent (t = 200 ns). (o) end of C transent (t = 220 ns). In Fg. 8 we show the rest charge at the onset and at the end of the last transent. The electrons whch are stll under C at the moment ths electrode becomes negatve, can go to the neghbourng 2C and electrodes and cause
22 the vertcal strpes. From Fg. 8 wc fnd that less than 0 electrons stay under \C f the barrer heght s less than 0. V. In vew of the nose level of 20-70 electrons wc state that a barrer heght < 4 kt s acceptable n ths case. Fg. 9 shows typcal expermental curves found n a sensor whch exhbted vertcal strpes. Devatons from the form of the curves n Fg. 7 are thought to result from the 2D character of the on-stale smulatons, n whch the effect of a narrow entrance s neelectcd. -30 0 50 00 350 200 250 Fg. 9. Expermental number of electrons n C as a functon of tme for four dfferent C cells. Here C does not go negatve at t = 200 ns. Solutons The solutons whch were tred n practce are shortenng of the C electrodes and reducton of the shallow p-mplant. Smulatons predcted a slghtly belter performance for the latter soluton. Ths was also found n practce. Sensors wth a reduced p-mplantaton were free of vertcal strpes. Conclusons In vew of our results we can state that () the complcated structure of the parallel-to-seres converson n frame transfer mage sensors necesstates the use of 3D smulatons. () n order to smulate charge transport n CCD regsters wth potental barrers n the channel t s necessary to use an on-state smulaton package. () the essental propertes of parallel-to-scres converson n frame transfer sensors can be modelled properly wth 3D off-state and 2D on-state smulaton packages. Reference Polak S.J., C. den Hejcr and W.H. A. Schlders (987). Semconductor devce modellng from the numercal pont of vew. Internatonal Journal for Numercal Methods n Engneerng, Vol. 24, pp. 763-838.