Three-level charge pumping on submicronic MOS transistors

~

Solid State Communications, Vol. 84, No. 6, pp. 607-611, 1992. Printed in Great Britain.

0038-1098/9255.00+.00 Pergamon Press Ltd

THREE-LEVEL CHARGE PUMPING ON SUBMICRONIC MOS TRANSISTORS J.L. Autran (1), F. Djahli (1), B. Balland (1), C. Plossu (1) et L.M. Gaborieau (2) (I)Laboratoirede Physique de la Mad6re Associ6 au Centre Nationalde laRecherche Scientifique,U R A n° 358 Institut National des Sciences Appliques de Lyon 20 avenue Albert Einstein, 69621 Villeurbanne Cedex, France

(2) Compagnie IBM FRANCE, Usine de Corbeil-Essonnes, BP n° 58 224 Boulevard J.F. Kennedy, 91105 CorbeiI-Essonnes Cedex, France (Received July 31,1992 by P. Burlet)

A modified three-level charge pumping technique on submicronic MOS transistors is used to determine the energetic distribution of capture cross sections of electron Si/SiO2 interface states and their density on an energy scale in the silicon bandgap including the midgap. The results are compared with values obtained by standard charge pumping technique.

I - Introduction

applied to the transistor gate. When the semiconductor surface is inverted, the system is assumed to be in quasiequilibrium. Interface states whose energy is below the Fermi level EF are filled by substratc minority carriers (electrons for a n.channel transistor), while interface states above EF are empty. When the gate potential is switched to Vc, many eleclrons are emitted back to the conduction band by interface states. The occupancy factor of these states tends towards the one corresponding to thermodynamic equilibrium. But interface states above El: emit their electrons only if their electron emission time ~c is less than re. Switching the transistor to strong accumulation drifts holes towards the Si/SiO2 interface. These holes recombine with electrons remained trapped in interface states, giving rise to the charge pumped into the substrate. For a pchannel transistor, complementary gate voltage pulses are applied to the gate (obtained by changing all the signs of the biases). By varying the intermediate gate voltage parameters, i.e. bias Ve and lime re, one can select interface states which are involved at each charge pumping cycle. One of the energetic limits of the domain scanned in the silicon bandgap depends on the value of Ve whereas the value of te determines which are the traps, into this energetic window, that can emit their electrons. Only interface states having a lime constant xe verifying trf < Xe< te are detected [I0]. The emission time window can be adjusted by varying trf and re. This modulation, in bias and/or in emission time is the basic principle of all threelevel charge pumping techniques recently developed [8]. If Vel is changing to Ve2 for a fLxed tel (figure Ib), the equilibrium occupancy level of the traps will change. The electron interface states density, between EFI and EF2 (respectively determined by Vel and Ve2) with a time constant ?.e verifying trfl < Xe< tel can be deduced from the corresponding variation of the charge pumped into the substrate:

Originally developed by Brugler and Jespers [1], the charge pumping technique is currently the only technique that allows a direct determination, with a great sensitivity, of the Si/SiO2 interface quality on submicronic MOS transistors. In ~ddldon to the mean value of interface states density, its energetic distribution on a large energy scale in silicon bandgap and its sp~tiA! distribution along the channel of the transistor can be obtained with the standard charge pumping technique [2-7]. New developments of this technique have been pro.posed in the previous years. These electrical spnctroscoptc methods allow to characterize interface states in the almost full silicon bandgap while taking into account the variation of capture cross sections versus energy. Some methods modify the periodic gate voltage pulses applied to the gate o f the transistor and develop a three-level or staircase charge pumping [8-13]; others make the transistor temperature vary and propose a spectroscopic technique, similar to DLTS (Deep Level Transient Spectroscopy) [14-15]. Recently, Saks and Ancona have proposed a technique that is the most complete synthesis of the three-level charge pumping procedures already published [11]. One originality of our work was to apply, for the first time, this charge pumping method to submicronic MOS transistors in order to characterize, with a high sensitivity, the electron interface states. In this paper, we show that 1) the sensitivity of this three-level technique can be increased, 2) this method can be applied to transistors with low interface states densities 3) electron capture cross sections of these devices depend on energy, 4) a simple calculation allows to take into account this variation with energy without requiring a particular formalism and to obtain the energetic distribution on a domain including the silicon bandgap. 2 - Three-level charge pumping theory In three-level charge pumping technique, a threelevel gate voltage pulse, as represented in figure la, is

Dit(E0 = 607

] dQit dVt qAeff dVe d~/e

(1).

SUBMICRONIC MOS TRANSISTORS

608

Vol. ~4, No. 6

sections of electron traps by monitoring the charge pumped per cycle as a function of te and Ve. The characteristic Qit vs te (Ve fixed) saturates for a value of te equal to tsar corresponding to the emission time constant of traps having an energy Et equal to the energy of Fermi level. The value of the corresponding capture cross section can be written

Vinv

VT Ve

as"

I I I

V FB

i I

V acc

I

t•hnt,== trd

=

t~_.J

_ _t--e

/ . I

1

fin = Vth ni tsa't

, tact

I ta,

I

To

la V inv

exp(- ~ TE')

(3).

By selecting a duration of the third gate voltage level long enough that the quasi-totality of interface traps are indeed emitting their electrical charge, it is possible to calculate the energetic distribution of interface states density using the relation (1) and considering the variation of the charge pumped per cycle in saturation.rate, as a function of te and for different values of Ve. 3 - Resulis and Discussion

VT

v~2 Vel - _ _ j

:

V FB

V acc

i

//

I

tel I

t,,I I

r"-

tel

-*"

lb

Fig. la. - Gate voltage signal used in three-level charge pumping for the determination of emission times of electron interface states on N-channel MOS transistor, lb. Variations of the parameters of the three-level gate voltage signal allowing the energy sweeping (variations of Ve,) and the emission time sweeping (variations of re).

where Qit is the charge pumped per cycle and Aeff is the channel effective area. Qit depends on the source and drain bias applied [3--4]. If tel is changing to re2 for a fixed Vel (figure lb), the interface traps above EFI with a time constant 're verifying tel < Xe < re2 will participate to the emission process. As te becomes higher than the greatest time constant of the traps, the emission process stops and these traps return in thermodynamic equilibrium with the energy bands. A saturation rate is reached for te=tsat. For values of te above tsat, the charge pumped during a cycle, when the transistor is switched to accumulation, remains constant and equal to the charge trapped in all interface states below EF (EF is determined by Ve). According to the Schockley-Hall-Read model (SHR theory) [16], the time constant of a trap occupied by an electron is an exponential function of its energetic position Et in the silicon bandgap: '~e

=

| VthOn ni ex~(- Ek-~T")

(2)

where Vth is the thermal velocity of electrons in silicon, ni the intrinsic carrier concentration and Ei the intrinsic energy level. Then, it is possible to determine the energetic distribution of emission times and capture cross

The test devices were 0.75 I.tm CMOS transistors. They presented an interdigital structure: the source and the drain form a single comb-shaped juncdon with the substrate that interpenetrates the "gate comb." Consequently, these devices require a very reduced area on the silicon wafer in spite of their great gate area (about 32000 I.tm2). We have especially tested these submicronic transistors at the beginning of the elaboration process, which explains the relatively high values of measured interface states densities. Nevertheless, the sensitivity of our technique (109 states eV -l cm -2) allows the characterization of these devices during the entire process and after the different passivation annealings. The knowledge of the relation between the gate bias Vg and the surface potential V/s is required, in order to calculate the energetic distribution of emission times, capture cross sections and interface states density. The characteristic Ws(Vg) has been obtained by quasi-static C-V measurements on the MOS capacity associated to the test device [17]. The gate voltage pulse, described in figure la, was applied to the transistor gate. The different rise and fall times were fixed to 50 ns. The time te varies between 200 ns and 30 ms. Our acquisition process is different than those proposed by Salts and Ancona in so far as the pulse frequency f is not fixed [l I]. By using a frequency of I00 Hz, the resulting charge pumping current is low because Icp is proportional to f (Icp= f Qi0. A low sensitivity results. In order to improve the sensitivity of the technique, we have directly measured the charge Qit pumped per cycle, equal to the value of charge pumping current divided by the frequency. For each duration of the third level, we can choose the highest possible frequency that still maintains a quasi-equilibrium during accumulation and inversion times. Practically, frequency is not adjusted for each measurement but only three times for te varying between 200 ns and 30 ms. A I0 kI-Iz frequency is chosen for te < 20 ms, then for 20 < te < 800 ms frequency is fixed to I kHz and for te > 800 ms, frequency is reduced to I00 Hz. Times tiny and tact are adjusted to be compatible with the gate voltage pulse period. The characteristic Q/t vs log(re), shown on figure 2, is recorded for different values of the third-level voltage Ve; these values are chosen from the curve U/s(Vg ) so that the different positions of the Fermi level corresponding to Ve are compatible with the silicon bandgap energies. From the previous network of curves for each voltage V¢, the value of tsar corresponding to the beginning of the saturation of the pumped charge (Qsat) can be easily determined. But the saturation times have not been determined like Saks and Ancona [I 1]. These authors define tsat as the abscissa of

Vol. 84, No. 6

SUBMICRONIC MOS TRANSISTORS

609

Qi! (pC) L0

B

~

iu 0

0 O

8

9

0

0

g

Z o

I1

8

V e (V)

o

o

°

o

0

O

0

0

O

O

0

0

0

O

O

D

0

a

0

0

0

D

0

0

O 0

7 0

8

6

0

8

$

D

0

0

0

£2

0

O

O

0

0

O

0

~8

Z

0

0

0

D

0

C]

D

0

O

O

0

0

O

O

0

0

0

0

O

O

0 D

O

D

O D ~

D

O o O 0

0

O

o

O

O

t3

O O

O D

D

O O o

0

O

o

0

0

O

0

o

O O

D

4

D

O O

D

O

0

° O 0

D o

D O

O t~

o

[3

O

O

D

O

o c~

1,00 0,95 0,90

=

0,85

D

0,80

o

0,75

o

0,70

o o

0,65 0,60

J-

0,55 0,50

D O

0

0

3 2

tsat(Ve=O,85 V) 1

tsat(Ve=O,50 V)

\

\

!

0 10-7

10-~

10-S

!

!

10-4

10-3

u (s) 10-2

Fig. 2. - Charge pumped per cycle as a fonction of the voltage and the duration of the third gate voltage leveL

the intersection of the linear part of the Qit vs log(re) saturation level characteristic with the linear part of this curve for short re. This method is not correct because, for this value of tsar, the charge Qit has not still reached its saturation value. Therefcre, we choose tsar as the time at the effective beginning of the saturation, as shown figure 2. These values of the saturation times obtained with our process arc sometimes more cliffcrent than those obtained by Saks and Ancona [11]. Owing to the "Ps(Vg) values, tsat(~Ps) et Qsm(xPs) or tsar(E) and Qsat(E) curves can be easily determined. Interface states density are calculated using relation (1) for the successive values of the pumped charge in saturation as a function of Ve. If Qi (respectively Qi+l) is the pumped charge in saturation for a Vi voltage (respectively Vi+l), then the mean interface state density between Ei and Ei+l is given by : Dit(Em ) =

AQ

q Adt AE

_ ~Qi+1 - Q~ - [Ei+l " F.~

with : Em = Ei + Ei+~-Ei

energy domain of the capture cross sections, represented figure 5, is limited by the position of Fermi level in inversion (upper limit = +0,22 eV) and by the minimum voltage Ve giving a saturation characteristic still measurable (lower limit = -0,08 eV). This energetic distribution shows that electron capture cross sections vary as a function of energy and that this variation is large: this result is in good agreement with works already published. Obtained capture cross sections vary from 3.10 -16 cm 2 for E - Ei = +0,20

10.2

10"3

(4) ~-~ .V

i0 ~

(5).

The network of Qit(log(te),Ve) curves, shown figure 2, has been obtained on a n-channel transistor. The energetic distribution of emission times of electron interface traps (figure 3) and the variations of the pumped charge in saturation as a function of energy have been calculated (figure 4). Figure 3 shows that the distribution of experimental (square dots) emission times of interface states and theoretic (continuous line) emission times of traps which would have a constant capture cross section (o = 2.10-15 cm 2, value obtained by the standard two-level charge pumping technique) are not quite different. The

10"s

a -- 2.10-15 cm2 I0-6

.

,

-0,15 -0,10-0,05

.

,

.

,

.

. ,

.

,

.

x

.

0,00 0,05 0 , 1 0 0 , 1 5 0,20 0,25

E - Ei (eV) Fig. 3. - Emission time of electron interface states as a function of their energed~ position in silicon bandgap.

610

SUBMICRONIC MOS TRANSISTORS

Vol. 84, No. 6

l0 t`

10'

1013

E

&

~" 10

t2

c~



II

i ~TI4

|

,

10

zt

10

t°

i

I

/ ! t t tt

3

.

,

-

,

.

i

•

,

-

,

-

,

-

,

-

,

-

-0,15-0,10-0,05 0,000,05 0,10 0,15 0,20 0,25 0,30

E - Ei (eV)

,

.

,

-

,

-

E - Ei (eV)

Fig. 4. - Charge pumped per cycle as a function of the energetic position of the Fermi level.

eV to 6.10-14 cm 2 for E - Ei = -0,07 eV. Comparatively, Saks and Ancona have obtained values ranging from 3.1015 cm2 to 10-13 cm 2 for the same energetic limits. Values of a and do/dE are similar: the capture cross section of an electron trap decreases when E drift towards Ee. Nevertheless, the nature of these traps is probably dependent of the test device. The sharpness and the good reproducibility of measurements reveal a peak in the a distribution for E - Ei-- +0,05 eV. This peak could be associated with the presence of a speciftc interface defect, having a capture cross section of about 10-14 cm 2. By deriving the curve of figure 4 according to relation (4), the energetic distribution of the interface states density is obtained (figure 6). The mean value of the interface states 10

.

-0,15 -0,10 -0,05 -0,00 0,05 0,10 0,15 0,20 0,25 0,30

"13

lO-I~

Fig. 6. - Energetic distribution of the interface states density in silicon bandgap corresponding to electron traps.

density, obtained by standard two-level charge pumping, is in good agreement wi.th our results. The distribution presents a peak for E - Ei = 0,05 eV: it corresponds to the peak observed in figure 5. We suppose that this peak is the signature of an electrical defect at the Si/SiO2 interface. The method presented above shows several interests. First, the energetic distribution of interface states density is obtained very easily, by deriving the curve Qsat vs E. This curve is a monotonous function: the miscalculation of this derived curve is weaker than for the case of a curve admitting many extrema. Furthermore, no electron or hole trap modei is needed. In particular, the knowledge capture cross sections values is not necessary since the total charge trapped in the states is directly measured as a function of the Fermi level position. Last but not least, it is possible to determine the interface states density on an energetic domain including the midgap; domain that can not be explored by the other charge pumping techniques. The first two points make this technique a powerful method to calculate correctly the energetic distribution of interface states density. The third point is very interesting because density distributions obtained by this method are complementary of those calculated by other charge pumping techniques. 4 - Conclusion

lO-lS

10-j6

10

"17

•

,

.

,

-

,

•

i

-

,

-

,

•

,

-

,

-

-0,15 -0,10-o,05 -0,00 0,05 0,10 0,15 0,20 0,25 0,30 E - Ei (eV) Fig. 5. - Energetic distribution of capture cross sections of electron interface states as a function of their energetic position in silicon bandgap.

We have set up, for the first time, a three-level charge pumping technique on industrial submicronic transistors. Although the nature of defects depends on the technological process, we have shown that capture cross sections of electron interface traps arc strongly depending on energy. The strict calculation of the energetic distribution of interface states density must take into account this variation. The proposed method allows to obtain very easily these distributions on an energetic domain including the silicon midgap. The value of the mean density, obtained by the standard two-level charge pumping is in good agreement with it. This spectroscopic method for interface states characterization would allow to: i) specify the nature of the physical defects associated with these electrical states, ii) show the influence of a technological CMOS process on electrical properties of the Si/SiO2 interface.

Nol. 84, No. 6

SUBMICRONIC MOS TRANSISTORS

Acknowledgement. The authors would like to thank the IC manufacturing plant of Corbeil-Essones, of the Company IBM FRANCE for the logistical support of this work. REFERENCES [1] BRUGLER J.S., JESPERS G.A. IEEE Trans. Electron Devices, 1969, Vol. 16, N°3, p. 297. [2] GROESENEKEN G., MAES H.E., BELTRAN N., DE KEERSMAECKER R.F. IEEE Trans. Electron Devices, 1984, Vol. 31, N°l, p. 42. [3] BALLAND B., PLOSSU C., CHOQUET C., LUBOWIECKI V., LEDYS J-L. Revue de Physique Appliqu~e, 1988, Vol 23, N°l 1, p. 1837. [4] PLOSSU C., CHOQUET C., LUBOWIECKI V., BALLAND B. Solid-State Commun., 1988, Vol. 65, N°10, p. 1231. [5] M A H N K O P F R., PRZYREMBEL G., WAGEMANN H.G. Journal de Physique, 1988, Vol. 49, N°9, p. C4-775.

611

[8] TSENG W.L.J. Appl. Phys., 1987, Vol. 62, N°2, p. 591. [9] HOFMANN F., KRAUTSCHNEIDER W . H . J . Appl. Phys., 1989, Vol. 65, N°3, p. 1358. [10] CHUNG J.E., MULLER R.S. Electronics, 1989, Vol. 32, N°10, p. 867.

Solid-State

[11] SAKS N.S., ANCONA M.G. IEEE Electron Device Lett., 1990, Vol. l 1, N°8, p. 339. [12] ANCONA M.G., SAKS N.S.J. Appl. Phys., 1992, Vol. 71, N°9, p. 4415. [13] SAKS N.S., ANCONA M.G. Appl. Phys. Lett., 1992, Vol. 60, N°lS, p. 2261. [14] VAN DEN BOSCH G., GROESENEKEN G., HEREMANS P., MAES H.E. ESSDERC 90. 20th European Solid State-Device Research Conference, p. 579. [15] VAN DEN BOSCH G., GROESENEKEN G., HEREMANS P., MAES H.E. IEEE Trans. Electron Devices, 1991, Vol. 38, N°8, p. 1820.

SHAW J-J., WU K. IEDM Tech. Dig., 1989, p. 83.

[16] SZE S.M. Physics of Semiconductor Devices. NewYork : Wiley & Sons, 2nd edition,1981.868p.

[7] HENNING A.K., DIMAURO J.A. Electronics Letters, 1991, Vol. 27, N°16, p. 1445.

[17] NICOLLIAN E.H., BREWS J.R. MOS Physics and Technology. New-York : Wiley & Sons,1982. 890p.

[6]