Variation of diffusion length with processing in dielectrically isolated π-silicon tubs

1678 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-33, NO. 11, NOVEMBER 1986

Variation of Diffusion Length with Processing in Dielectrically Isolated r-Silicon Tubs

DOROTHEA E. BURK, MEMBER, ].FEE, G. M. FLOWER, AND SANG-SUN LEE

Abstract-The variation of minority-electron diffusion length with processing in dielectrically isolated n-silicon tubs is investigated wing electron-beam-induced current measurements. The analysis of :Illese measurements is complicated by the six boundary conditions oil the n-silicon. The diffusion lengths measured in virgin n-tub silicon a 71: 5 250 pm while those in n-silicon undergoing a phosphorus gettet. are >> 250 pm, The determination of exact values for the diffusion lex@hs as a function of processing are not possible without a more complete understanding of the parasitic effects of the inversion region at th 3 tub walls. Trends in the diffusion lengths are indicative of trends i l l the carrier lifetime, which are critical in the high-voltage applicationis for silicon tubs.

D I. INTRODUCTION

IELECTRICALLY isolated (DI) silicon tubs are currently used for high-voltage and radiation-hardenled

devices in integrated circuits. The high-temperature processing and chemical and mechanical lapping that ;are necessary in the formation of these tubs introduce urde- sirable defects and strain into the silicon lattice [ 11. Time defects lower the carrier lifetime and, hence, diffu4on length in the DI silicon. In high-voltage applications, because the tub center usually becomes the active portio 1 of a device, the lifetime in the silicon tub is critical to the device performance. For example, in the case of the ga :ad- diode-switch, very low-doped n-silicon becomes the (h-ift region of the high-voltage device [2] in which the lifet m e must be controlled. When the device is off, the n-silj :on must support a large reverse voltage with low leakage '1:ur- rent, while, when the device is on, the n-silicon must 16eip- port a plasma of both holes and electrons with a low on- resistance.

Optimal lifetimes in high voltage devices in n-sili-on are not necessarily the longest lifetimes [3]. An opti nal lifetime is one that is long enough to ensure low on-resis- tances (7, 2 10 ps), but short enough to ensure Fast switching times (7, < 1 ms) [3]. For the best device re- liability and performance, the optimal lifetimes muslt be well controlled in the fabrication process. Phosphorus lif- fusions during high-voltage device fabrication help getlter

Manuscript received January 13, 1986; revised July 10, 1986. This lvork was supported by AT&T Bell Laboratories, Reading, PA, and by the Na- tional Science Foundation under Grant ECS-8419427.

D. E. Burk and S . S. Lee are with the Department of Electrical I ngi- neering, University of Florida, Gainesville, FL 3261 1.

G. M. Flower was with the Department of Electrical Engineering, Jni- versity of Florida, Gainesville, FL 3261 1. He is now with Hewlett-Pac card Laboratories, Palo Alto, CA 94304.

IEEE Log Number 8610581.

impurities from the active n-region and control the lifetime [4]. Therefore, a knowledge of the lifetime or diffusion length variation in the n-silicon as a function of different phosphorus diffusions could facilitate the opti- mization of processing.

11. THE EXPERIMENT

A set of preliminary experiments was devised to measure diffusion length variations directly in virgin and phosphorus-gettered r-silicon tubs using electron-beam- induced current (EBIC) measurements. The experimental setup and a schematic of the test structure is shown in Fig. 1. A Keithley 480 picoammeter is in series with the test structure and power supply. The incident electron beam is the source of the JOEL 35-C-JSM scanning electron microscope in a stationary mode [5]. Data was taken at 2-pm intervals by moving beam position toward the junction.

In EBIC, the electron-beam excitation volume used to probe the minority-carrier transport is a tear-drop-shaped volume with a radius, typically, of about 10 pm at an incident beam energy of 39 keV. Because the volume radius is two orders of magnitude smaller than the tub di- mensions, it can be treated as a point source. However, the EBIC analysis in the DI tub is complicated by the six boundary conditions on the n-silicon region, i.e., at the junction, the back contact, the top and bottom surfaces, and the two sidewalls. The need to consider recombination of electron-hole pairs at the sidewalls was eliminated by choosing large-area tubs (1750 pm X 1350 pm). (The sidewalls are in the third dimension perpendicular to the page in Fig. 1.) The EBIC data as a function of distance from the junction was recorded at the tub center, about 650 pm from the tub's sidewalls.

Because the n-silicon at the tub sides is inverted, the collection at the tub bottom, approximately 50 pm from the center of the electron-beam excitation, was not neg- ligible [ 11. The EBIC current ZEBIC lost to this surface by collection was accounted for by reverse biasing the junction and correcting the collected current to obtain the EBIC, i.e.,

IEBIC = ZCOLLECTED + ID(VR) (1) where ID( V,) is the diode current in reverse bias, approximately 10 nA. The EBIC at the junction is approximately 1 pA.

0018-9383/86/1 lo( 1- 1678$01 .OO 0 1986 IEEE

BURK et al.: VARIATION OF DIFFUSION LENGTH

ConJact Back

+----1~00 pm-4 Oxide Field

x-SI Tub

Isolation Oxide

"R

Fig. 1. The experimental setup and equivalent circuit representation. The current associated with the MIS junction is represented by I,. In the cross section of the test structure in a DI ?r-silicon tub, the electron beam (Ib) is incident on the top surface.

The remaining three boundary conditions were incor- porated into the analysis by assuming that the junction and back contact were surfaces that extended infinitely in depth. The analysis of the EBIC with the beam incident on the top surface of finite surface recombination velocity, bounded on one side by a semi-infinite junction and on the other by an ohmic contact, was taken from [5]. This finite-region analysis uses an infinite number of images of the original electron-hole pair excitation volume to obtain an analytic solution of the EBIC problem. In the normal semi-infinite analysis, the original excitation volume and three images are necessary to solve the EBIC problem for a junction, semi-infinite in extent, perpendicular to a surface. Incorporating the third boundary condition, i.e., the ohmic contact within several diffusion lengths from the junction, results in increasing the magnitude of the slope of the EBIC obtained from the two- boundary-condition analysis.

Theoretical plots of log ZEBIC [5] as a function of distance from the junction are given in Fig. 2 for an incident beam energy, Eo = 39 keV. The top surface is assumed to have a low surface recombination velocity of 10°-103 cm/s because of the passivating oxide. Higher beam ener- gies between 20-39 keV were used to ensure the electron- beam excitation volume was in the n-tub and not in the top oxide layer. The theoretical plots for L, = 100-600 pm are in reasonable agreement to those presented by Kuiken and van Opdorp ([6] in Fig. 3) for a planar, collecting junction with the smallest normalized surface recombination velocity, S, = 0-2. Therefore, the assump- tion that the junction is infinite in extent is valid.

Two test structures were devised to measure diffusion length in the virgin and phosphorus-gettered tubs. All test structures were fabricated in 1700 pm X 1350 pm DI tubs, doped NA - 7 X 1013/cm3, and having a passivating layer of LPCVD oxide deposited and densified at 900°C for 15 min or a thermal oxide grown at 1100°C for 15 min. All collecting junctions were 1000 pm ( f 2 0 pm) from the back contact.

For the data to be collected on the virgin n-silicon, an

10 O

10 -1

0 10 - 2

z w

m - I

'

1 o - ~ 0 200 400 600 800 1000 1200

x ( 1 p - t )

Fig. 2. Theoretical normalized EBIC plots as a function of distance from the junction for the three-boundary-condition EB[C problem given in [ 5 ] . The EBIC are plotted for diffusion lengths observed in n-silicon tubs. Incident beam energy is 39 keV.

MIS diode, fabricated by low-temperature processing, served as the collecting junction in order to minimize additional high-temperature processing. Pin aperture was opened for the junction and back-contact formation, and aluminum deposited and patterned for the back contact using a liftoff process. A tunneling oxide over the junction aperture was grown while sintering the back contact at 450°C for 20 min [7]. Aluminum was them deposited over the tunneling oxide to form the MIS inversion region. Thus the only additional heat treatment that the tub under- went, aside from the initial oxidation, was a 450°C an- neal.

For the data to be collected on the processed n-silicon, the phosphorus getter served also as the (n'-T) junction diffusion. An aperture for the diffusion was opened and a phosphorus predeposition at 900°C for 30 min was made, resulting in a-shallow junction, xj - 0.5 pm. To simulate a longer getter, the phosphorus predeplosition was then driven in at 1000°C for 60 min resulting in a junction depth xj - 2 pm. The windows for the emitter and ba'ck contact were then opened and the aluminum was patterned. A back contact sinter was not done because of concern for aluminum spiking through the junction, but care was taken to ensure good ohmic contacts.

111. EXPERIMENTAL RESULTS AND DISCUSSION Some representative results for EBIC, normalized to

EBIC collected of the junction Z&C, are given in Figs. 3 and 4 for virgin r-Si using the MIS junction, and phosphorus-gettered n-Si using a n+-n junctilon for collection, respectively. In Fig. 3, two plots of normalized EBIC, the upper one with an applied reverse bias V R = -0.5 V and the lower are at V, 9 0 V, demonstrate the importance of taking into consideration the inversion region at the tub bottom. The diffusion lengths for the upper and lower plots are approximately 250 and 100 pm, respectively.

1680 IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. ED-33, NO. 11, NOVEMBER 1986

t *. I L . *. L .. I

10 - -1 -

- 0

‘2 m -1

io-*:

t

. . ... .. .-.

*a. e.. . ....

.. ”. ... .... .

1 1 I I I I , , I L.. -A 0 200 400 600 800 10011

x(pm)

Fig. 3 . Normalized EBIC plots as a function of distance from the j u ~ ction for virgin a-silicon tubs using MIS junction for collection, The ppe r plot ( 0 ) is for VR = -0.5 V and the lower (m) for VR = 0 V. Imident beam energy is 39 keV.

Reverse biasing the MIS junction effectively cuts off collection from the inversion region along the tub bottzlm. As shown in Fig. 5, this inversion region at the tub blot- tom is connected to the MIS junction via the inversion region along the sidewalls up to the surface, which if; lde- pleted under the thermally grown oxide and furthel: inverted by the electron-beam excitation at the top surf‘xe.

In Fig. 3 , the collection under reverse bias is incre8;li;ed over that at zero bias. With no applied bias, carriers are collected by the inversion region at the tub bottom, which, at 50 pm (or less) from the excitation volume, is cl11ser than a diffusion length to the excess electron-hole pairs generated. However, the inversion region at the tub ?lot- tom is connected to the top surface through the serie:: resistance of the inversion region along the sidewalls. LJn- der zero bias (shown in Fig. 5(a)), the drop across this series resistance due to collected carriers can result ia-1 rhe inverted region at the tub bottom being forward bi;:.e;ed similar to the open circuit condition in a solar cell. The net result is that less current is collected by the inver!ion at the tub bottom and therefore, less appears at the: lop contact. With reverse bias (shown in Fig. 5(b)), the: inversion region at the tub bottom and sidewalls act itis a barrier to carrier recombination there, thus enhancing, /:he number of carriers being collected at the MIS junct on. Based on this argument, the diffusion length of the vi:gin n-Si is L, 5 250 pm (7, = Li/(35 cm2/s) I 18 ps), de- pendent on the new boundary condition at the tub bottllm.

In Fig. 4, plots of normalized EBIC taken under :!era

bias for a phosphorus-gettered tub are given. The sern Jog plot represents an unrealistically long diffusion len;:th, which, upon further examination, is not exponential I see Fig. 4(b)). This linear behavior has also been observed for the MIS data taken at zero bias at distances greater than 2*L, from the junction. In the case of the MIS, 1 is

1 I I 1

0 200 400 600

.95

2; .90 0 -

0 200 400 600 800

x (pm )

(b) Fig. 4. Normalized EBIC plots as a function of distance from the junction

for phosphorus-gettered a-silicon tubs using the p-n junction formed for collection. (a) Semilog plot for VR = 0 V, and (b) linear plot for V, = 0. Incident beam energy is 39 keV.

an indication of the minority-carrier collection by the inversion region because, upon reverse biasing the MIS, the EBIC reverts to an exponential dependence. A parameter, derived from the linear plots as a function of distance from the junction, is the resistance per unit length RIZ of this

BURK et al.: VARIATION OF DIFFUSION LENGTH 1681

\’ Zinverdlon Region

Contact Back

vR # o (b)

Fig. 5. A schematic of the test structure including the parasitic inversion region around the perimeter of the DI tub for bias condition (a) VR = 0 and (b) VR # 0. In the inset equivalent circuit, I, represents the diode current through the original MIS junction, and IF” represents the diode current through the resistor Rs associated with the inversion region.

inversion region. The observation of this resistance arises from a small self-bias recorded during the EBIC measurements of - 0.04 mV. The relationship of RIZ is

where ZEBI~(X) is the EBIC collected with the beam incident at a distance x from the junction. The RIZ for both the MIS and nf-p junction varied from 2-5 Qlpm, de- creasing in value with increasing x. The fact that the normalized EBIC is almost independent of x over the entire range of x for the phosphorus getter suggests that the diffusion length is longer than that for the MIS. Applying reverse bias did not yield definitive results, but L, >> 250 pm (7, >> 18 ps).

These results for diffusion length are consistent with lifetimes deduced from current-voltage measurements of gated-diode switches in smaller a-silicon tubs [2]. In these devices, the gate is a phosphorus-diffused junction (xj - 20 pm), resulting from a long drive-in. The lifetimes deduced from a fit to the on-resistance of the finished device

are - 10 ps. Because there are other processing steps in the more complicated switch that could lower the lifetime, the longer lifetime for the phosphorus getters re- ported here is not unreasonable.

In order to measure the diffusion lengths after the phosphorus getter, further refinements in the: data analysis and experiment have to include: 1) placing guard rings around the phosphorus-diffused junction and 2) adding the boundary condition at the tub bottom in a two-dimen- sional numerical computer simulation of the EBIC problem. This is left for future work.

IV. CONCLUSIONS Preliminary results demonstrate that trhe virgin a-Si tub

has a diffusion length 5250 pm that phosphorus-getter tubs have diffusion lengths >> 250 p m . The EBIC analysis has to account for collection by the inversion region at the tub bottom. Reverse biasing the junction reduces the collection by the tub bottom, but additional improve- ments in the experiment and data anallysis are required before definitive values for the diffusion lengths after phosphorus getters can be known.

ACKNOWLEDGMENT The authors would like to thank M. A Shibib at AT&T

Bell Laboratories, Reading, PA, for unprocessed DI tub material, and both him and J. G . Fossunn at University of Florida for helpful discussions.

REFERENCES [ l ] K. E. Bean and W. R. Runyan, “Dielectric isolation: Comprehensive,

current and future,” J . Electrochem. Soc., vol. 124, pp. 5C-l2C, 1977.

[2] R. J . McDonald, J . G. Fossum, and M. A. Shibib, “A physical model for the conductance of gated p-i-n switches,” IEEE Trans. Electron Devices, vol. ED-32, pp. 1314-1321, 1985.

[3] S . K. Ghandi, VLSIFabrication Principles. New York: Wiley-lnter- science, 1983.

[4] S . P. Murarka, “A study of phosphorus gettering of gold in silicon by use of neutron activation analysis,” J . Elect,wchem. SOC., vol. 123,

[5] D. E. Burk and R. Sundaresan, “Diffusion length and surface recombination velocity measurements with the scanning electron microscope: The highly-doped emitter of a p-n junction,” Solid-State Elec- tron., vol. 27, pp. 59-67, 1984.

[6] H. K. Kuiken and C. Van Opdorp, “Evaluation of diffusion length and surface-recombination velocity from a plenar-collector-geometry elec-

2090,1985. tron-beam-induced current scan,” J . Appl. Phys. , vol. 57, pp. 2077-

[7] P. E. Russell, D. E. Burk, and P. H. Holloway, Grain Boundaries in Semiconductors. New York, Elsevier, 1982, p. 185.

[8] G. M. Flower, M.S. thesis, Univ. of Florida, 1984.

pp. 765-767, 1976.

* Dorothea E. Burk (S’77-M’81) received the B.A. degree in physics from Aldelphi University, Garden City, NY, in 19’70 and the Ph.D. degree in electrical engineering from Brown University, Providence, RI, in 1981.

She is presently an Associate Professor in the Department of Electrical Engineering, University of Florida, Gainesville. Her research activities include the electrical characterization of semiconductor devices for new technologies such as poly- silicon self-aligned, silicon-on-insulator, and

dielectrically isolated-tub technologies.

1682 IEEE TRiANSACTIONS ON ELECTRON DEVICES, VOL. ED-33, NO. 11, NOVEMBER 1986

Dr. Burk is a member of Sigma Xi, the American Physical Tau Beta Pi.

%

G . E. Flower, photograph and biography not available at the lication.

Society, t . n.d

time of p , .I)-

Sang-Sun Lee was born in Kyungkido, Korea, in 1955. He received the B.S. and M.S. degrees in electronics engineering from Hanyang Univer- sity, Seoul, Korea, in 1978 and 1983, respectively. He is currently working toward the Ph.D. degree in electrical engineering at the University of Florida. His interests include development of surface passivation technology of 111-V compound semiconductor devices.

Variation of diffusion length with processing in dielectrically isolated π-silicon tubs

Documents

Transcript of Variation of diffusion length with processing in dielectrically isolated π-silicon tubs