(PDF) Trench depth measurement method for wafer patterning

Trench depth measurement method for wafer patterning

K. Takada and J. Noda °

Various types of three-dimensional vertical capacitor cells have been proposed for realizing very-large-scale integration direct random access memories (VLSl DRAMs). With vertical processing of the VLSl DRAMs, the need for a non- contacting method of measuring trench depth has greatly increased. This paper describes trench depth measurement systems with an optical interference technique. Experimental results are described in addition to the measurement principle and configurations of the system. With these systems, the trench depth can be quickly and precisely measured without cleaving the wafers. Therefore, they are appficable to in-line monitoring systems in the VLSl DRAMs fabrication process.

Keywords: trench depth measurement, silicon wafers, interference

The evolution of DRAMs with higher level integration has been achieved by decreasing memory cell areas. This has been accompanied by a reduction in the capacitance of stored electrical charges. The usual approach to increasing capacitance charge is to reduce the thickness of the storage capacitor insulator. However, this increases the possibil i ty in megabit DRAMs of catastrophic dielectric breakdown 1 .

Various types of three-dimensional cells, such as the corrugated capacitor cell (CCC) 2 and the isolation-merged vertical capacitor (IVEC) cell 3 have been proposed to reduce cell capacitance. With vertical processing, the need for a novel method to measure quickly and precisely trench depth has greatly increased. The conventional measurement method uses a scanning electron-beam microscope (SEM). The advantage of this method is that trench shape, including depth, can be precisely measured. However, the wafer must be cleaved to observe the shape, and must therefore be measured in a vacuum chamber. Consequently, the SEM method is not applicable to in- l ine monitor ing systems.

This paper presents new methods for trench depth measurement. These methods use the optical technique, and are based on white l ight interference. The measurement systems differ depending on whether the interference signal is detected in the time domain or the frequency domain. Experimental results are presented, first in the time domain and then in the frequency domain.

T i m e d o m a i n m e a s u r e m e n t

Fig 1 is a schematic diagram of the experimental set-up for measuring in the time domain 4. The system consists of the optical head, a Michelson interferometer, and a controller. The optical source in the optical head is a halogen lamp whose output l ight coherence length is about 1 /~m. When the

• NTT Opto-Electronics Laboratories, Nippon Telegraph and Telephone Corporation, TokaL Ibaraki-ken, 319-11, Japan.

Multimode fibre r -

Optical head

4:~< .[~j , BS 2 Lamp

Holder "" '~, / , Wafer

Stage+ I

Fig 7

CRT

-I Host computer

I \

0 x oY

xl l Optical input . . ~

Interferometer

Schematic diagram of the trench depth measurement system using a Michelson interferometer. BS 1 and BS 2 are beam-splitters

output light from the lamp irradiates a sil icon wafer containing trenches, two kinds of the reflected l ight are produced. One is from the sil icon wafer surface, and the other is from the trench's bottom surface. The optical path difference between these two reflected light waves is 2d, where d is the trench depth of the wafer.

Takada and Noda- - Trench depth measurement method for wafer patterning

The reflected light from the wafer is passed into the Michelson interferometer through a mult imode optical fibre. The configuration of the Michelson interferometer is shown in Fig 2. It comprises two prisms M1 and M 2, and a beam-splitt ing cube BS. Prism M 2 moves whi le prism M1 is fixed. The interferogram shown in Fig 3 (a ) is produced during the prism M 2 scan. The ac signal output from the detector obtained during the prism scan is represented by

/ ( T ) = ( I + B K / A ) S ( ~ ) + B x / ~ x

[ S ( ~ - 2d / c ) + S(~ + 2 d / c ) ] (1)

where A is the area of the sil icon upper surface, B is the area of the trench bottom surface. K is the factor of penetration into the trench, ~ is the group delay time difference given by the prism M 2 scan, c is the velocity of l ight in a vacuum and S(~) is an auto- correlation function of the incident electrical field.

Since the individual terms in Eq (1) are independent of each other for white l ight interference, three peaks, one central peak, and two side-band peaks, are produced as shown in Fig 3 (a). The optical path difference between these two side-band peaks is 4d so that the translation of the prism from one side band to the other is 2d. Thus, the trench depth can be estimated by measuring the translation of the prism. Usually, the reflected light

Optical input

M 1 • . • , ° . " . " . °

PD~

HeNe Laser

PZT driven stage /

PBS

Ref I

PD2 Ref2

Signal

Signal processor

Fig 2 Arrangement of the Michelson interferometer QWP is a quarter wave plate, PD I and PD z are silicon photodiodes, Ref 1 and Ref 2 are reference signals proportional to sin(~p) and cos(r~) with

= 47txl2

Fig 3 Demonstration of the acquired interferogram (a) and the calculated envelope (b)

power from the trench bottom surface is very small compared with that from the sil icon wafer surface and thus the signal-to-noise ratio ( S / N ) is relatively low. In this case, the resolution of the A / D converter used for interferogram digit ization is insufficient to produce side bands unless the amplifier gain is increased until the central band is deformed. However, in the present system, it is only necessary to measure the distance between the two side bands.

S /N improvement is accomplished by averaging the individual interferograms obtained during periodic scanning of prism M 2. Mechanical instability or vibration causes random changes in optical path difference between the two prisms M 1 and M 2. The optical path difference is therefore monitored during the prism scan by launching the HeNe laser output l ight into the Michelson i nterferometer.

As shown in Fig 2, a quarter-wave plate is placed in the optical path of one arm of the interferometer to transform the linearly-polarized light to circularly-polarized light. The beam-split t ing cube BS combines this circularly-polarized light with the linearly-polarized light propagated through the other arm of the interferometer. The polarization beam-splitter PBS divides the combined light beams so that the resulting two beams are detected by the reference detectors. The outputs from the detectors are proportional to sin (~ ) and cos (b ), where r~ = 4~x/2 is the phase difference between the optical paths of the two arms of the interferometer, x is the displacement of prism M 2 relative to prism M1 and ). is the HeNe laser output wavelength. These two

146 JULY 1989 VOL 11 NO 3

Takada and Noda--Trench depth measurement method for wafer patterning

outputs are used to produce the sampling pulses for digitizing the interferogram every 2 /8 interval. After averaging the individual interferograms, the envelope of the averaged interferogram is calculated to estimate the distance between the two side-band peaks as shown in Fig 3(b) .

The developed system is illustrated in Fig 4. The measurement error of the system is within +_ 0.1 /~m for a 2 pm to 7 l~m trench depth and a 1 /~m trench width. The feature of the developed system is that it uses a multimode fibre for optical connection between the optical head and the Michelson ir~terferometer. Use of the multimode fibre places the optical head part only in the in-line process, while the interferometer itself is fixed far from the VLSl fabrication line. In this way, mechanical vibration or noise produced in the line is not induced into the i nterferometer.

Since less than ten seconds is required to complete one trench depth measurement, trench depth distribution of the silicon wafers can be measured easily. Fig 5(a) shows the trench depth distribution of a silicon wafer with 0.9/~m wide trenches. The contour line map of the trench depth at 0.1 l~m pitch is shown in Fig 5(b) . The contour lines are symmetrical about the centre of the wafer. Figs 5(a) and (b) show that the trench depth in the periphery is 0.3 #m shallower than that near the centre, thus demonstrating the non-uniformity of the dry etching process.

The trench depth measurement system using a Michelson interferometer has the advantages that the system is very simple and requires a short time to accomplish one measurement. However, a trench depth less than the coherence length ( ~ 1 /~m) of the source cannot be measured precisely, and the depth of very narrow ( ~ 0.2 Fm) trenches also cannot be measured because the central wavelength of the light is fixed at 0.8 lzm. If the central wavelength is reduced to 0.4/~m or less by using a Xe lamp instead of the halogen lamp, the coherence length cannot be reduced to less than 3 #m because of the modulated spectral shape of the Xe lamp. That is to say, it is relatively hard to measure a trench depth of 3 Izm when the Xe lamp is used. Furthermore, as the wavelength becomes shorter, the shot noise

b Pfl F4 Fig 5 (a) Distribution of trench depth on a silicon wafer (b) Contour map of the trench depth of the wafer whose trench depth distribution is illustrated in (a)

becomes a serious problem in the trench depth measurement using the Michelson interferometer. The frequency domain or spectroscopic method solves such problems and this wil l be discussed in the next section.

Fig 4 Illustration of the developed system with a Michelson interferometer

Frequency domain measurement

The reflected light from the test wafer with trenches is composed of light reflected from the wafer surface and of light reflected from the trench bottom surface. The intensity of light reflected from the silicon wafer with trenches at a particular wavelength )~ is given by

1(2) = C1/0(2) r l + c2 cos (47zd/2) ] (2)

where/0(2) is the intensity of the incident light, and C1 and C2 are constants. From Eq (2), it can be seen that the reflected light intensity has maximal and minimal values at wavelengths of ). = 2d/n and 2 = 2 d / ( n + 1 /2 ) (n=O, 1, 2, 3 . . . ) , respectively. Therefore, the trench depth can be estimated by measuring these peak wavelengths.

PRECISION ENGINEERING 147

Takada and Noda-- Trench depth measurement method for wafer patterning

The frequency domain (or spectroscopic) trench depth measurement system is shown schematically in Fig 6. The system comprises a light source, a wafer set-up, and a signal processor. The light source is a Xe lamp. The acousto-optic TeO2 tunable filter (AO filter) 5 is used to tune the Xe lamp output light. Rapid wavelength tuning from 0.70 to 0.35/~m can be easily achieved by applying an rf frequency of 40 MHz to 110 MHz to the AO filter. The tuned light is passed into the wafer set-up part after propagating through the multimode fibre. The wafer set-up part comprises a beam-splitting cube BS, a test wafer, and a reference silicon wafer without trenohes. The two wafers are placed symmetrically with respect to the BS. The beam transmitted through the BS is normally incident to the test wafer. It is reflected from the test wafer back to the BS and is finally led into a photo- multiplier PM1. Here, the reference wafer is inclined at 1 ; at the incident beam. The beam reflected by the reference wafer is transmitted through the BS,

and is led to the other photo-multiplier PM2. Output signals V s and V, from PM 1 and PM 2 are transmitted to the waveform recorder.

The wide and rapid tuning range of the AO filter from 0.7/~m to 0.35/~m has several advantages: (1) A trench depth shallower than 1 /~m can be measured easily because a few peaks are produced in the reflected light intensity during one wavelength scan even for such a shallow trench, (2) depth of narrow (~0.2 /~m) trenches can be measured by using violet light. (3) High efficiency and rapid tuning can be executed by the AO filter so that the S/N ratio improvement is easily made by fast averaging the individual waveforms. Furthermore, by using the symmetrical set-up configuration of the test and reference wafers with respect to the beam-splitting cube BS, (4) output power fluctuations of the Xe lamp can be compensated for by dividing the signal V s by V, at individual tuning wavelengths.

The calibrated signal waveform Vs/V, is shown

Xe Lamp

F~g6

Light source

A_ \ /

Tunable optical f i l ter

Fibre

Acousto-optic tunable f i l ter

Wafer I set-up

[ RF Sweeper

l ' ADC

Test wafer

/ Amplifier

PM2 ! i t PM1

- BS

Reference wafer

Experimental set-up of spectroscopic trench depth measurement system

Multimode fibre

148 JULY 1989 VOL 11 NO 3

Takada and Noda--Trench depth measurement method for wafer patterning

Wavelength

0.711pm

Spectra o f

tuned A 4 nm

0.550um 0.445~m

~ 4nm

a / b/ c /

..?--

I J I I I I I 0.0 0.1 0.2 0.3 0.36

Time, s

Fig 7 Demonstration of calibrated signal waveform V, /V , Figs (a), (b) and (c) are the spectrums of tuned lights at the peak positions of the calibrated signal waveform

L390/~m

.37 #m

SEM photograph

Wavelength, #m

0.48 0.45 0.40 0.37 I I 1 I I I I I I I I I

~ .~. Vp, I V , c = 0.09

0.454 .m ~ 0.417 pm . . . . . r--- % ~ .

I I I I 70 80 90 100 110

b Sweep frequency, MHz

Fig 8 (a) Cross-sectional view of 0.26 i~m-wide and 1.37 iJm-deep trench, and (b) its calibrated signal waveform

in Fig 7. The trench is 1.3 #m wide and 1.15 #m deep. The horizontal axis in the figure is rf sweep time, and rf sweep start is chosen as the origin. Although the output fluctuations of the Xe lamp are in the order of ___5%, the fluctuation effects cannot be observed in the calibrated waveform as shown in Fig 7. The figure shows three maxima at 43.3 MHz, 59.2 MHz, and 79.5 MHz which correspond to tuned light wavelengths of 0.71 p.m, 0.55/~m, and 0.45/Jm, respectively. Therefore, the trench depth is estimated at d-- 1.2/Jm. This value agrees fairly well with those measured by the SEM.

Next the wavelength is scanned in the range from 0.48/~m to 0.37/Jm by sweeping the rf frequency from 70 MHz to 110 MHz. The trench cross- sectional view of the test DRAM wafer is shown in Fig 8(a) . The trench depth is 1.37/Jm. The trench bottom width is 0.26 #m and its surface is flat. The calibrated signal waveform obtained for this test wafer is also shown in Fig 8 (b ) , where the oxide thin film is removed by a hydrofluoric acid. The waveform has two maxima and one minima and its periodic changes are superimposed on a linear trend. The ratio of the peak-to-peak value Vnp of the changes to the dc value Vdc in the calibrated signal waveform Vs/V , is Vpn/Vdc = 0.09. Peak wavelength values are obtained after subtracting the linear trend by calculation. The trench depth is estimated at 1.36 l~m whose value agrees very well with the directly measured value of 1.37 izm as shown in the SEM photograph of Fig 8.

The trench cross-sectional view of a 2.6 Flm- deep test wafer is shown in Fig 9(a). When the deeper trench is formed, the trench becomes a taper as clearly seen in the figure so that the trench bottom width is about 0.09 l~m. The calibrated signal waveform Vs/V r, obtained for the test wafer is also shown in Fig 9(b) . The peak-to-peak value of the ac component in this signal waveform is Vpp/Vac= 6 x 10 -3 and decreases by a factor of 15 compared with that for 1.37/~m-deep trench shown in Fig 8. The periodic change is observed even for such a taper trench. However, the signal waveform does not change sinusoidally with wavelength. By using the least squares fitting procedure, the trench depth is estimated at 2.7/~m with _+ 20% error. This kind of waveform deformation cannot be avoided because the light wavefront launched into the tapered trench is highly deformed. To obtain a more precise trench value, it is only necessary to use a light source with shorter wavelength.

Measured comparison between the SEM (horizontal line) and the present (vertical line) methods are shown in Fig 10. Three kinds of trenches with W= 0.5, 0.26, and 0.09 llm width are used as the test wafer. The waveforms obtained for W = 0.26 l~m and W = 0.09 l~m trenches have already been shown in Figs 8 and 9, respectively. At W= 0.5 and 0.26 l lm-wide trenches, the values measured by the present method agrees well with those measured by the SEM all within _+10% error. Even for W = 0.09/Jm-wide trench with 2.6 l~m depth, the measured value variations are only within + 20%. Therefore, the present method is useful for in-l ine monitoring systems for future DRAMs such as 16 Mbit DRAMs.

PRECISION ENGINEERING 149

Takada and Noda-- Trench depth measurement method for wafer patterning

!.6/~m

SEM photograph

Wavelength,/Jm

0.42 0 .41 0,40 0.39 0.38 0.37

T , , , , I ,

I- Yea J r 0 c = 0 × 10 3 j j l I

I I I 1 85 90 95 100 105 110

b Sweep frequency, MHz

Fig 9 (a) Cross-sectional view of 0.09 ~ro-wido and 2.6 llm-deep trench, and (b) its calibrated signal waveform

Conclusions Non-contacting trench depth monitoring is required because the three-dimensional trench structure is now widely used for realizing VLSI DRAMs. As shown in this paper, the trench depth can be measured by optical interference techniques without cleaving the silicon wafers. Thus, trench depth measurement systems with optical interference techniques will be introduced into in-line processes for fabricating VLSI DRAMs. The method using the Michelson interferometer is applicable to relatively wide ( > t /zm) trenches. On the other hand, the frequency-domain and spectroscopic method are applicable to future VLSI or ULSI with very narrow ( ~ 0.2/~m or less) trenches. Trench depth measurements for silicon wafers with silicon oxide films have not been discussed here. When the

E 7 ~ / / / = 5

5 w == g

E O W:0.50pm == • W=0.26pm

o. A W=0.09~m >, 3

r- E

"o 0J L

0 I I I 2 3 4 5 6

Measured depth by SEM, um

Fig 10 Measured trench depth value comparison between SEM (horizontal line) and present (vertical line) methods of trenches with 0.5, 0.26, and 0.09 l~m width

effective thickness of the oxide film is almost equal to trench depth, it is rather difficult to estimate the trench depth. Therefore, a new signal processing procedure for estimating such a trench depth for wafers with silicon oxide films is strongly required for a spectroscopic trench depth measurement system.

Acknowledgement

The authors would like to express their sincere thanks to S. Nakajima for his useful discussions and for supplying us with the test DRAM wafers. They would also like to thank N. Uchida for his useful discussions and continuous encouragement.

R e f e r e n c e s

1 Sunami, H. Cell Structure for Future DRAMs, in Prec. IEEE Int. Electron Devices Meeting, December 1985, 694

2 Sunami, H., Kume, T., Hash imoto , N., I toh, K., Toyabe, T. and Asai, S. A Corrugated Capacitor Cell (CCC) for Megabit Dynamic MOS Memories, Prec. IEEE Int. Electron Devices Meeting, December 1982, 806

3 Nakaj ima, S., Miura , K., Mineg ish i , K. and Mar ie , T. An Isolation-Merged Vertical Capacitor Cell for Large Capacity DRAM, Prec. IEEE Int. Electron Devices Meeting, December 1985, 240

4 Takada, K.. Chida, K., Noda, J. and Nakaj ima, S. Trench Depth Measurement System for VLSI DRAM Capacitor Cells using Optical Fiber and Michelson Interferometer, IEEEJ Lightwave Technol, 1987, LT-5, 7, 881

5 Uchida, N. and Sai to, D. Acousto-optic Tunable Filter Using TeO2, Proc IEEE, 1974, 62, 1379

150 JULY1989 VOL11 NO3

(PDF) Trench depth measurement method for wafer patterning - DOKUMEN.TIPS (2024)