Influence of the preparing conditions on the

Influence of the preparing conditions on the physicochemical characteristics of glasses for thick film hybrid microelectronics M. Prudenziati, B. Morten, and P. Savigni Department of Physics, University of Modena, Via G. Campi 213/A, 41100 Modena, Italy

G. Guizzetti Department of Physics, "A. Volta", University ofPavia, Via A. Bassi 6, 27100, Pavia, Italy (Received 26 January 1993; accepted 18 April 1994)

Seven batches of a high-lead glass were used for the preparation of RuO2-based thick film resistors. Investigation of their electrical properties showed a lack of reproducibility of results whose origin was related to changes of the physicochemical properties of the glassy matrix. A systematic investigation of the glass batches, both in form of frit powders and screen printed and fired layers, was carried out with several spectroscopies to detect changes in composition and structure. The spectroscopic methods included x-ray Energy Dispersive Fluorescence (EDS), Scanning Electron Microscopy (SEM), Atomic Absorption (AA), diffuse optical reflection of the powders and specular reflection of the layers, optical transmission, and other complementary methods. The dissolution of Al, due to interaction between the glasses and the alumina substrate, as well as the diffusivity and solubility of Ag due to interaction with the Ag-bearing terminations were measured. The results demonstrated that, apart from small compositional differences, the various batches were characterized by differences in residual stresses, redox reactions, and "microstructure." The latter was responsible for very notable differences in the optical properties of the glasses, which in turn are closely related with the difference in atomic solubility and diffusivity. Optical spectroscopies have been found to be a suitable means for testing reproducible preparation methods of glass frits for thick-film hybrid microelectronics.

I. INTRODUCTION Thick film materials are prepared by screen printing and firing on a suitable substrate. Generally, the pastes (or inks) contain a glass constituent in the form of frit, consisting of micrometric glass particles.1 This constituent accomplishes different functions, according to the nature of the layer. Glass is present in minor quantities in conductive layers where it promotes adhesion with the ceramic substrate and the sintering of the conductive grains. In dielectrics it improves density, dielectric strength, adhesion, and provides passivation, while in resistive pastes it plays a vital role as dielectric matrix of the cermet structure and determines the electrical properties (mainly resistivity and its temperature coefficient).2'3 Reproducible properties of these components require, of course, reproducibility in the properties of the glass frit which consists of a high lead glass for most of the air-fireable thick-films.4'5 This work was motivated by problems faced during studies of model resistive systems, where we observed that pastes prepared from different batches resulted in resistors with sheet 2304 http://journals.cambridge.org

J. Mater. Res,, Vol. 9, No. 9, Sep 1994 Downloaded: 12 Dec 2012

resistance values different up to 100% and temperature coefficient of resistance (TCR) not only different in values but also in sign. After accurate tests which ruled out any responsibility for this behavior of the RuO 2 powders and of the resistor firing processes, we undertook a systematic investigation of the properties of the glass component. A first outstanding result of this study is the observation that the diffusivity and solubility of aluminum, due to interactions with the alumina substrate, and silver, due to interactions with silver-bearing terminations, differed quite substantially in experiments made with various batches of the same glass. These findings prompted us to reconsider carefully every step of the preparation of the glass frits and look for spectroscopic methods capable of discriminating even fine differences in composition and structures of the glass in frit and layer forms, respectively. The comparative analysis of the results of the various techniques suggests that the elemental diffusivity and dissolution in the glass layers are affected by their composition and even more so by structural characteristics. Moreover, this study shows that each step of the frit preparation © 1994 Materials Research Society IP address: 155.185.12.195

M. Prudenziati e( a/.: Influence of the preparing conditions

(batch melting, fritting, and milling) must be carefully designed and carefully controlled in order to get reliable and reproducible properties of glass containing thickfilm materials. II. MATERIALS AND METHODS A. Materials

Table I summarizes the main characteristics of the glass batches, prepared with a mixture of PbO, SiO2, and A12O3 in the ratio of 68:31:1 by weight (36.68:62.14:1.18 M%). The oxide powders were obtained with purity higher than 99.99% by Merck. The mixture was transferred to an open platinum crucible and inserted in a vertical oven (with free circulation of air) preset at the temperature Tt and heated up to the temperature Tm, for 1 h. Then the crucible was very rapidly lowered and the fluid glass dropped (fritted) in de-ionized water. The gross particulate so obtained was milled in an alumina jar with water or chlorothene for several hours till particles of 5-8 fim mean size (Fisher measurements) were obtained. Pastes were prepared with 75% wt. of a glass frit, 25% of organic vehicle consisting of a-therpineol, butylcarbitol, and ethyl-cellulose. The pastes were screen printed, dried, and fired in a conveyor belt furnace. Most of the firing experiments were performed on samples prepared on 96%-alumina substrates (Hoechst, CeramTec 708), but some layers were printed on beryllia (99.5% BeO, Thermalox, Brush & Wellmann) and sapphire (Kyocera). The heating cycles lasted 45 min with 10 min of dwell time at the peak temperature Tf. Values of Tf were chosen in the range from 850 °C to 1000 °C. A disk of the glass was also prepared by melting the same oxide mixture in a small alumina crucible heated at 1300 °C, periodically removed from the oven for rinsing and agitation of the fluid glass. Finally the crucible was slowly cooled down and sectioned in order to separate a chunk of glass, ground, and polished for optical experiments. B. Analytical techniques

Glass frits have been analyzed with x-ray Energy Dispersive Spectroscopy (EDS), Scanning Electron Microscopy (SEM), Fourier Transform Infrared SpecTABLE I. Main characteristics of the glass batches. Batch

T, (°

F8(l) F8(2) F8(3) F8(N) F8(6) F8(V)

200 400 400 500 500 500

Fluid 1400 1400 1300 NA 1200 1200

Water Water Water Water Chlorothene Water

troscopy (FTIR), UV-VIS optical spectroscopy, Differential Thermal Analysis (DTA), and Atomic Absorption (AA). Powders were also observed under a microscope (transmission mode, polarized light). FTIR transmission measurements (T) were performed with a Bruker spectrometer model IFS 113 V, using the technique of KBr pellets (2 mg glass/200 mg KBr), at a resolution of 0.5 cm"1. The absorbance A was calculated as A = — In T. A Prais Mantis attachment to a Cary 5 spectrometer allowed us diffusive reflection measurements in the range 200-800 nm, using as a reference an aluminum mirror. DTA measurements were performed with a STA 409 Netzsch analyzer in an open alumina crucible, at a heating rate of 10 °C/min. For AA analyses, performed with a Philips PU9900 spectrophotometer, glass frits were leached in aqua regia, which dissolves free platinum (i.e., metal not pertaining to the glass network), whose concentration was determined through direct readings. At least 10 g of frit was needed for each analysis. The high lead content in the concentrated solution was responsible for a significant matrix effect. Hence a relatively high measurement error was observed, about 10-15% of the reading. Birefringence of a small fraction of the glass particles was revealed with an optical microscope (in transmission mode) provided with a rotating stage and optical polarizers. The efficiency of the method was tested with observations of a quartz powder, whose trigonal structure is optically active. Pressed pellets of glass frits were analyzed in EDS (EDAX 9900 system) equipped with a Si(Li) detector with a parylene window (detectable elements with Z > 5). The same apparatus was used for measurements of elemental concentration and profiles in glass layers prepared at various peak temperatures Tf. In particular Al on top offilmsfiredon alumina and Ag concentration profiles near the interface between the glass and Agbearing terminations were measured. Finally, transmission and reflection spectra of layers prepared on sapphire or ceramic substrates, respectively, were collected by means of a double beam spectrophotometer, Perkin-Elmer 330. The resolution of the instrument was 0.07 nm and the wavelength accuracy was 0.01 nm in the range from 200 to 800 nm. Absolute reflectance was derived from the reflectivity of the aluminum mirror used as a reference. III. RESULTS A. SEM-EDS

Figure 1 shows the SEM image of backscattered electrons from the F8(l) pellet. The arrows point to two particles which are clearly brighter than the surrounding 2305

J. Mater. Res., Vol. 9, No. 9, Sep 1994

http://journals.cambridge.org

Downloaded: 12 Dec 2012

IP address: 155.185.12.195

M. Prudenziati et a/.: Influence of the preparing conditions

TABLE II. Relative concentration of constituent oxides in the glass batches as measured in EDS on pressed pellets (wt. %). Batch

PbO/(PbO + SiO 2 ) 66.76 67.34 67.06 67.26 67.40 68.49

F8(l) F8(2) F8(3) F8(N) F8(6) F8(V)

± ± ± ± ± ±

A12O3 2.29 3.06 3.22 2.38 2.71 2.07

0.02 0.30 0.04 0.02 0.03 0.01

± ± ± ± ± ±

0.2 0.4 0.2 0.2 0.2 0.07

B. FTIR transmittance

FIG. 1. SEM micrograph of a pressed pellet prepared with F8(l) glass batch, backscattered electrons.

material; their composition is rich in platinum, according to EDS analyses. Presence of Pt can be related to a reaction between the crucible and the melted glass batch; this result induced the systematic study using AA analysis to obtain quantitative figures of Pt content, as described later. Relative concentration of components was measured on the pressed pellets. It should be remembered that a quantitative analysis on the basis of EDS spectra requires corrections of the peak heights for x-ray fluorescence yield, auto-absorption, and internal fluorescence (known as ZAF corrections).6 This procedure involves the rough evaluation of the sample composition on the basis of the peak heights and then the computation of the sample density (p) and mean atomic number (Z) values which are used in an iterative procedure to obtain the correct composition. This procedure is not completely adequate for samples of density different from the theoretical density, which is the case of pressed pellets. In fact, the voids inevitably present between the compacted grains make the actual values of p and Z lower than those assumed in the program. Consequently the accuracy of the absolute values of concentration is relatively low. Using a pellet of the glass F8(V) as a standard sample we obtained a deviation of 1.5% between the measured EDS values and the data of a plasma analysis (performed by VIOX Corporation, Seattle, Washington) on the same glass. However, this error should be the same on pellets prepared with powders of the same mean size and size distribution, under the same pressure, and with very similar compositions. Accordingly, we evaluated a relative error of the measurement less than 0.3%. With these considerations in mind, the differences in composition obtained in EDS analyses (Table II) appear larger than the relative error, even if at the boundary of the performance capability of this technique. 2306


The absorbance spectra of the glass powders F8(l) to F8(N) in the range of wave numbers from 1300 to 800 cm" 1 are reported in Fig. 2. A wide band is observed between 1250 and 800 cm" 1 , with an intensity notably changing in the glasses of the various batches: the main peak in this band shifts from 1030, 5 cm" 1 in sample F8(l) to 1017 cm" 1 in sample F8(6). Even if small, this shift is much greater than the spectrometric resolution (0.5 cm" 1 ) and, more important, it is systematically related with the molar concentration of PbO in the samples as observed from EDS measurements (Table II). This result qualitatively agrees with previous findings reported in literature7 for high lead silicate glasses (alumina free) where the absorption peak changes from about 1030 to 1015 cm" 1 by increasing the molar ratio of PbO from 36% to 39%. C. Optical microscopy Observations of the powders in transmission mode with polarized light revealed in every batch the presence

1300

1200

1100

1000

900

800

WAVE NUMBER (cm 1 )

FIG. 2. Infrared absorbance of glass frits.



IP address: 155.185.12.195

M. Prudenziati et al.: Influence of the preparing conditions

of a small fraction of optically active particles. In other words, the rotation of the microscope stage, with the polarizing filter held at a fixed position, gave extinction of the transmitted light through a few glass particles. This finding may be related either to grains devitrified in a crystalline structure having a polar axis, or to vitreous grains with optical anisotropy due to consistent residual stresses.8 The relative fraction of the active grains is a few percent in all the powders except in F8(V) batch, where the effect was seldom observable. Other investigations should be performed to clarify the origin of this interesting effect. 400

200 D. Diffuse reflection Figure 3(a) shows diffuse reflectance spectra of the frit in batch F8(2) in repeated measurements; the reproducibility of the results is quite satisfactory even when the sample is removed and loaded again in the sample holder or after realignment of the optical system. The shape of the spectrum closely resembles a transmission spectrum with a threshold wavelength near 300 nm.9>1° We note that the typical values of the refractive index (n = 1.7-1.9) for these glasses at higher wavelengths (500-800 nm) should determine a bulk reflectivity of 8-10%, which is much lower than the measured reflectance. This is due to the very low absorption coefficient in this region (see below) and to the mean size of the glass grains (5 /Am), which yield a small optical density ( < 0.1). So the incident light is not only reflected by the facets of the glass grains but undergoes multiple reflections and refractions at every interface between glass and air [see inset in Fig. 3(a)]. On the other hand, at wavelengths lower than 300 nm the glass becomes absorbing so that multiple reflections are hindered and the diffuse reflectance decreases. These considerations show that the notably higher diffuse reflectance of sample F8(N) than that of sample F8(2) at wavelengths greater than 350 nm is due to the lower absorption of the former, rather than to a different value of the refraction index. This contention is supported by the transmission spectra of the glass layers which show consistent differences in transmission (see Fig. 6 and Sec. IV. A below) as well as by the reflectance spectra (curves a in Fig. 8 and Fig. 9) which imply negligible differences in the refraction index of the same glasses in this optical range. E. Differential thermal analysis (DTA) DTA experiments are frequently used to investigate the glass transformation range11'12 as well as possible fields of crystallization.12'13 In our experiments the most interesting results concern the identification of remarkable residual stresses present in the frits and the

800

600

WAVELENGTH (a)

(nmi

40 F8(N)

30 -

F8(2)

•

o z < 20 o LU _1 UL

111

at

f

^r

•J

-

10 -

200

600

400 WAVELENGTH

(nm)

(b) FIG. 3. Diffuse reflectance spectra of glass powders, (a) Comparison of the spectra of the same batch in repeated measurements; (b) two different batches of F8 glass.

determination of a relative scale of fictive temperatures for the various samples. It is worthwhile to remember that the fictive temperature r fict is that value of temperature inside the transformation range corresponding to the atomicmolecular network (or glass configuration) of pseudoequilibrium, frozen by fast cooling processes of the glass.12 So the Tf\a value is notably affected by the thermal history of the glass, particularly during the preparation processes: the higher the cooling rate, the larger the r fict value and, accordingly, the higher the content of excess energy "frozen" in the glass in the form of disorder or stressed network. Figure 4 14 compares the typical behaviors of DTA measurements performed on two samples of the same glass, one of which (a) was characterized by a higher fictive temperature than the second one, (b) which was very slowly cooled from the transformation temperature. 2307




800

IP address: 155.185.12.195

M. Prudenziati etal.: Influence of the preparing conditions

i

I

i

+2 LU

+1

.Evolution | of h e a t ^ - -

O Z

ill

"~\ \

High fictive sample

0

oc

| Absorption ' o f heat

LU

LU OS

a

-2

_o

on LU Q.

-3

100

-

200

300

400

500

600 700

TEMPERATURE OF GLASS (°C)

-4

Very low fictive sample

LU

FIG. 5. Plots of differential thermal analysis measurements of four different batches of the glass under investigation.

-5 i

J

i

300

400

500

TEMPERATURE

OF GLASS/°C

FIG. 4. Typical plots of differential thermal analysis of glass samples with high (a) and low (b) fictive temperatures.

The former sample gives up, during the heating process of the DTA measurement, the excess energy, which is observed as an exothermic peak. On the other hand, in both cases the endothermic peak is associated with heat absorption required by the glass transformation. It can be noted that the minimum of the endothermic peak is much lower in sample (b) because of the higher preexisting content of excess energy of this sample which has a higher fictive temperature. DTA curves of our glass frits are shown in Fig. 5. A comparison between Figs. 4 and 5 suggests high values of the fictive temperatures of the frits, which is not surprising considering the processes (frittage in water and milling) used for the powder preparation. The shapes of the various curves are quite similar, but the exothermic peak is shifted and increases in intensity as well as in area according to the sequence F8(3), F8(N), F8(2), and F8(l). Similarly, the same sequence is observed regarding the intensity of the endothermic peak, whose position around 560 °C is related to the temperature of glass transformation typical of these glasses. This picture is consistent with a value of the fictive temperature which is comparatively low for glass F8(3) and increases going through samples F8(N), F8(2), and F8(l). F. Atomic absorption (AA) The atomic concentration of platinum was measured with this technique after the observation of Pt parti2308


cles in SEM/EDS analyses. Unfortunately, not enough powder (minimum quantity of 10 g) was available for this analysis of the batch F8(N). Table III summarizes the results. The Pt concentration is relatively low except in batch F8(l). The presence of Pt can be related to deficiency of oxygen in the glass-forming melt. In fact, according to well-known reports of the literature15 and the Pt-Pb binary phase diagram,16 when high lead glasses are melted in platinum crucibles, there is a slight possibility that glass is partially reduced and Pb metal is formed, which reacts with Pt, giving rise to several compounds which consume and make the crucible fragile. Our data show that the F8(l) batch suffered particularly as a result of this problem. IV. GLASS LAYERS A. Optical transmission Figure 6 shows the transmittance spectra, in the range from 200 to 800 nm, of films (20 /xm thick) prepared from two glass batches on sapphire substrates at a peak temperature of 850 °C, for 10 min. It is interesting to note the difference in transmittance at large wavelengths: films prepared with F8(N) frit exhibit a transmittance higher than that of F8(2) samples. The values of absorption coefficient a are quite low in this region; for example, in glass F8(N) the following values were obtained: a = 1.11 X 102 cm" 1 at 400 nm and a = TABLE III. Concentration of Pt in glass batches (ppm at.). Batch F8(l) F8(2) F8(3) F8(6)

Pt concentration 128 3.60 6.95 8.00



IP address: 155.185.12.195


i

i

.u

i

i

i

20

i

1

0 .8 -

-

LU

{

0 .6

0 .4 -

-

15 -

I

- -

-

^-

ECT

cc

F 8 (N) r n (2)

All/

ANSMI TTAN

O

0 .2 0 .0

i

200

l

i

i

400

WAVELENGTH

• •

600

800

LL.

10 -

LU

(nm)

FIG. 6. Transmittance spectra of two glassy layers prepared on sapphire.

0.96 X 102 cm" 1 at 800 nm. The transmittance sharply drops by decreasing the wavelength below 350 nm, and a becomes lower than 10~3 cm" 1 at 320 nm for all the samples. Accordingly, at low wavelengths the incident radiation "sees" only the front sample surface; this information will be useful in the interpretation of results of specular reflection.

B. Specular reflectivity Measurements of specular reflectivity at near-normal incidence have been performed on glass layers prepared on various substrates (BeO, 96% alumina, sapphire) at different peak temperatures as well as on a bulk sample. Effects related to the firing conditions were observed, mainly in the near UV-region, e.g., the range involving transitions in the Pb 2+ ion.17 Figure 7 shows the reflectivity spectrum of the bulk sample, which has been assumed as the reference spectrum, since preparation of the glass is free from the possible defects inherently associated with the film preparation, such as fritting and milling (causes of stress and strain), blending with organic vehicle (possible cause of chemical reduction), sintering of glass particles (inhomogeneity), and interactions with the substrate (change of composition and strain). The main feature of the spectrum in Fig. 7 is the wide peak around 275 nm (4.5 eV) where the absolute reflectivity is about 13.5%. This peak is related to the 6s-6p Pb 2 + transition,17 and its width has been observed and discussed in connection with the configuration of the ground state and excited states of Pb 2 + . We were unable to observe the peak (190 nm, 6.5 eV) associated with the same transition,17 because it is out of the spectrophotometer range. Figure 8 shows the reflectivity spectrum of glass layers F8(N) prepared on alumina and beryllia at two

800

600

400

200

WAVELENGTH

(nm)

FIG. 7. Specular reflectivity of a slide of the glass F8; this spectrum is assumed as a reference for comparison with the reflectance spectra of the glassy layers on various substrates.

different peak temperatures (Tf = 850 QC and 1000 °C, respectively). The peak located at about 255 nm is notably sharper than that observed in bulk samples; moreover the reflectivity increases by increasing the firing temperature.

F8 (N)

on BeO

o

200

300

400

500

WAVELENGTH

600

(nm)

FIG. 8. Specular reflectivity of glassy layers prepared with frit of the batch F8(N), on beryllia and alumina at two peak temperatures.

2309




IP address: 155.185.12.195


20

1

20

i

F8

(2) ^

: 8 50 °C : 1000 °c

a :850 °C b :1000 °C

F S ( 3 )

15 -

15

> i-

v\\ b

O

O

LU

UJ

u!

i

10

»

^ —

200

1

I

300

400

WAVELENGTH

10

UJ

500

200

(rim)

400

300 WAVELENGTH

500

(nm)

FIG. 9. Specular reflectivity of layers prepared with the batch F8(2) on alumina; two samples fired at Tf = 850 °C and 1000 °C are compared.

FIG. 10. Specular reflectivity of layers prepared with the batch F8(3) on alumina; two samples fired at Tf = 850 °C and 1000 °C, respectively, are compared.

Figures 9 and 10 show the relative reflectivity of samples F8(2) and F8(3) prepared on beryllia at Tf = 850 °C and Tf = 1000 °C, respectively. In both cases the reflectivity at the peak maximum, located at 255 nm, is higher than that measured on layers F8(N). In F8(2) films the peak is well shaped already in samples prepared at Tf = 850 °C, whereas it increases and becomes considerably broader by increasing the firing temperature to Tf = 1000 °C. Samples F8(3) exhibit a high shoulder in samples prepared at 850 °C. The reflectivity of samples prepared from batches F8(6) and F8(V) is very similar to that of layers F8(3). From these findings it is evident that some samples differ quite notably from the bulk, but a tendency to mimic its reflectivity is noted by increasing the Tf value. This latter, in fact (and not the changes of composition induced by interactions between the films and their substrate), seems to be the parameter that more strongly affects the specular reflectivity of the various glass batches.

A comparison of the data in Tables II and IV shows that the values of Caium in samples prepared at Tf = 850 °C scales in the same sequence as those measured on the pressed pellets, and a quite constant ratio between the two sets of data is found. The higher concentration obtained in the case of pressed powders should be ascribed to the systematic error associated with the porous character of the pellets themselves, already mentioned and discussed. These observations indicate that the interactions between the glassy layers and alumina substrate have negligibly affected their composition where the material at a distance of 10-12 fim from the alumina interface is concerned, in agreement with previous results.5 Much more interesting is the comparison of the concentration Caium measured on samples prepared at Tf = 1000 °C. In fact, the measured values differ up to 40%. Diffusion of silver from the Ag-bearing terminations has been widely studied in a previous work where diffusivity D and solubility Cs of this element have been determined for a series of high-lead glasses.18'19

C. Diffusivity of the elements Table IV reports the values of the A12O3 content Caium measured on top of the layers prepared at Tf = 850 °C and 7) = 1000 °C. The EDS analyses have been performed with a 25 KeV electron beam impinging on the surface of the films. Consequently, the reported values represent the composition of a glass layer about 3 /xm thick adjacent to the sample surface. 2310


TABLE IV. Relative concentration of AI2O3 (C, wt. %) on top of glass layers prepared with various glass batches. Batch

F8(2)

F8(3)

F8(N)

F8(6)

F8(V)

C (Tf = 850 °C) C (Tf = 1000 °C)

1.9 7.51

2.15 8.97

1.57 7.19

8.75

10.7



IP address: 155.185.12.195


200

DISTANCE (urn) FIG. 11. Concentration profiles of silver near the interface between glassy layers and Ag-bearing terminations. Samples prepared at Tf = 1000 °C. Samples F8(2), F8(3), and F8(N).

8

S

(6) (3)

00

150

1

00

100

—i

r

CO