Ceramics International 43 (2017) 5441–5449

Contents lists available at ScienceDirect

Ceramics International
journal homepage: www.elsevier.com/locate/ceramint

Effects of microstructure and intergranular glassy phases on thermal MARK

conductivity of silicon nitride
Jin-Myung Kim, Sin-Il Ko, Ha-Neul Kim, Jae-Woong Ko, Jae-Wook Lee, Hai-Doo Kim, Young-
⁎
Jo Park
Engineering Ceramics Research Group, Korea Institute of Materials Science, 797 Changwondaero, Changwon, Gyeongnam 641-831, Republic of Korea

A R T I C L E I N F O A BS T RAC T

Keywords: In this study, the binary sintering additives Y2O3-Sc2O3, were first applied to the Si3N4 system to investigate
Si3N4 their effects on microstructure and thermal conductivity. The microstructure and thermal conductivity of both
Thermal conductivity sintered silicon nitride (SSN) and sintered reaction-bonded silicon nitride (SRBSN) were found to be
Grain boundaries significantly dependent on the additive composition. Among various combinations of Y2O3 and Sc2O3, 1 mol
Sintering
% Y2O3−3 mol% Sc2O3 prominently enhanced thermal conductivity, and the enhancement could not be
attributed to any difference in microstructure or lattice defects. TEM observation revealed that this composition
was more liable to devitrify the glassy phase with a lower degree of stress accumulation, and to possibly produce
a grain boundary that was cleaner or with a higher order of atomic arrangement. A microstructure model for
thermal conductivity was proposed which took the thermal resistance of the grain boundaries into account. The
grain boundary state exerted a remarkable influence on the thermal conductivity of fine microstructures, and
the experimentally measured thermal conductivity values were consistent with those given by the proposed
model.

1. Introduction In contrast to the optimal intrinsic thermal properties of single

crystal Si3N4, polycrystalline Si3N4 fabricated by liquid phase sintering
Recent advances in power-conversion electronics and high power usually exhibits quite lower thermal conductivity values. One of the
devices have caused a dramatic increase in heat density on the circuit major reasons for this degraded thermal conductivity are the defects in
boards. That enormous concentration of heat is detrimental to the the fabricated Si3N4 lattice, such as oxygen, metal impurities, vacancies
devices’ reliability and durability, and many researchers have sought to and dislocations [6–9]. In particular, to ensure sufficient thermal
develop ever more efficient methods to dissipate heat from semicon- conductivity, oxygen content must be dealt with in a careful manner,
ductor chips to metal heat sinks [1,2]. The thermal resistance between since oxygen is inherent and ubiquitous in SiO2 layers on Si3N4 raw
an electronic component and heat sink is mainly due to the low thermal powder and in oxide sintering aids, and therefore can be easily
conductivity of the surrounding or intervening materials, such as incorporated into the Si3N4 lattice. [7] In order to inhibit oxygen
ceramic substrates, and the interface that joins metal to ceramic. As dissolution into the lattice, the selection of proper sintering aids such
a result of these conditions, ceramic substrates must have the capacity as those having high oxygen affinity were selected. These are favorable
to maintain electrical insulation at high operating temperatures and to because lattice oxygen can be expelled out of the lattice by the
endure the accumulation of severe stress, which is triggered by difference in chemical potential between the lattice and glassy phase.
numerous thermal cycles and a large temperature gap. Among the [10] Lowering the oxygen concentration in the glassy phase by using
candidate materials for such a ceramic substrate, silicon nitride (Si3N4) (oxy)nitride or fluoride compounds instead of the typical oxide
has attracted interest as a promising material on account of its superior additives has also been recommended [11,12]. In terms of the
mechanical properties (i.e., fracture toughness, flexural strength, wear processing route, sintered reaction-bonded silicon nitride (SRBSN) is
resistance, thermal shock resistance) and electric breakdown resistance known to be more effective at decreasing total oxygen content as well as
at both ambient and elevated temperatures, as well as high thermal lattice oxygen content, compared to conventional sintered silicon
conductivity according to theoretical and experimental evaluations [3– nitride (SSN). [13–15] Unlike SSN, which begins as ready-made silicon
5]. nitride powder, SRBSN works with silicon as a starting material, and

⁎
Corresponding author.
E-mail address: [email protected] (Y.-J. Park).

http://dx.doi.org/10.1016/j.ceramint.2017.01.037
Received 3 November 2016; Received in revised form 8 January 2017; Accepted 8 January 2017
Available online 09 January 2017
0272-8842/ © 2017 Elsevier Ltd and Techna Group S.r.l. All rights reserved.
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

the silicon is converted to silicon nitride by a nitridation process. 2. Experimental procedure

Afterwards, the porous nitrided material goes through post-sintering
for densification near to the theoretical density. SRBSN exhibits high The starting materials for sintered silicon nitride (SSN) were Si3N4
thermal conductivity because the silicon powder contains a lower (95% α-Si3N4, SN-E10, Ube, Japan), Y2O3 (99.99%, grade C, H.C.
amount of oxygen compared to the typical Si3N4 powder, and the Starck, Germany), Sc2O3 (99.99%, Alfa Aesar, USA) and MgO (99.99%,
oxygen content is further reduced during the nitridation process in a Sigma Aldrich, USA). The two categories of composition were prepared
reducing atmosphere, due to the reaction between Si and SiO2, which as (1) with an equimolar Y2O3/Sc2O3 ratio with differing total additive
produces gaseous SiO. contents by 2, 4, 8 mol%; (2) a total 4, 8 mol% additives with varying
Apart from lattice defects, a second factor that lowers thermal Y2O3/Sc2O3 ratio. The notation for Y2O3- Sc2O3 in the SSN experiment
conductivity is the sintering additives used for liquid-phase sintering. is Sxy, where the integers x and y indicate the mole percent of Y2O3 and
They form residual glassy phases at multigrain junctions and grain Sc2O3 in the (1-x-y)Si3N4-xY2O3-ySc2O3 system, respectively. In addi-
boundaries, resulting in very low thermal conductivities of about 1 W/ tion, 92 mol%Si3N4–2 mol%Y2O3–5 mol%MgO was prepared, as the
mK [16]. The detrimental effect of the glassy phase at a grain boundary most common sintering additive for high thermal conductivity Si3N4,
is reported to be negligible when the grain growth proceeds to a size abbreviated as SYM.
larger than several microns, which reduces the total volume fraction of The raw powders were weighed according to each stoichiometric
the intergranular glassy phase [16]. A coarse microstructure also composition and mixed by low energy ball-milling at 200 rpm for 24 h
contributes to a drop in lattice defects because of lattice purification with anhydrous ethanol and milling media in a PTFE bottle. Zirconia
via dissolution-reprecipitation [9]. Despite the positive effects of coarse balls were used for the ball milling in order to inhibit Al contamination.
microstructure on thermal conductivity, excessive grain growth is not The mixed slurry was then dried via rotary evaporator at 80 °C and in
suitable from the viewpoint of material reliability, because abnormal an oven at 105 °C overnight. The powder mixture was sieved by 150
grains can themselves act as large defects, exacerbating mechanical mesh sieve and consolidated into a green compact by uniaxial pressing
properties, particularly flexural strength and Weibull modulus [17]. and subsequent cold isostatic pressing (CIP) at 200 MPa for 5 min. The
Recently, another approach has been proposed for improving green compacts were moved to a BN crucible and embedded in a
thermal conductivity without excessive grain growth: controlling packing powder of 50Si3N4/50BN (by weight). Afterwards, the cruci-
thermal conductivity of the grain boundary phases. For the hot-pressed bles were placed to a graphite heating furnace for gas pressure
SiC system, Kim et al. reported that an equimolar addition of 0.5 mol% sintering (GPS) at 1900 °C for 6 h in 0.9 MPa of N2 atmosphere. The
Y2O3 – 0.5 mol% Sc2O3 led to fully crystallized glassy phases not only initial α-Si3N4phase in raw materials was completely transformed to β-
in triple pockets but also in intergranular thin films, and this Si3N4 form by solution-reprecipitation mechanism during sintering at
amorphous-free microstructure achieved the extraordinary thermal high temperature.
conductivity of 234 W/mK [18]. Based on the results of the SSN experiment, certain compositions
In the case of Si3N4, however, a similar effort to control the grain were selected for use in the sintered reaction-bonded silicon nitride
boundary state in order to improve thermal properties has yet to be (SRBSN). The raw powders were commercial silicon (99.99%,
tried. Instead, many studies have investigated the devitrification of d50=1.28 µm, Worldex, Korea) which was produced from solar-grade
glassy phases in attempts to improve the high temperature mechanical silicon ingots, and the same oxide powders used in the SSN experiment
properties of Si3N4 [19–25]. The common method for devitrifaction is were also utilized for the SRBSN specimens. The green compacts were
to insert a distinct post-annealing step at moderate temperature with a prepared according to the same experimental process, and then
long soaking time, or to use particular rare earth elements as sintering nitrided in the alumina tube furnace at 1450 °C without holding time,
additives. But most of the results have found that the intergranular thin flowing 95%N2–5%H2 mixed gas at 300 sccm. Through this process,
films remained amorphous, although partial or complete devitrification the silicon-based green compacts were fully transformed to porous
occurred in triple pockets. reaction-bonded silicon nitride (RBSN) containing α/β-Si3N4 and
Then, one may raise the question of whether the Y2O3- Sc2O3 oxynitride additives without residual silicon. Afterward, post-sintering
additives would be effective for crystallizing grain boundaries in Si3N4 was performed under the same GPS conditions used for the SSN
as well. Presently, as far as we have determined, no one has examined specimens, except that the holding time at the target temperature
this co-additive composition for sintering Si3N4. In many cases, these (1900 °C) was varied from 1 h to 12 h.
two elements were investigated as either individual sintering aids, or in The relative densities of the sintered bodies were measured by
combination with light element oxides [23–25]. Sc2O3 is well-known to Archimedes method. The microstructure of the specimens was ob-
have a strong tendency of devitrifaction, so it does not need an served using a scanning electron microscope (JSM-6700F, JEOL,
additional post-annealing stage to induce crystallization. The reason Japan) after mirror-polishing and plasma etching flowing 46sccm of
for this easy crystallization can be explained by the volume change CF4 and 4sccm of O2 gases. The detailed data of the constituting grains
from amorphous to crystalline phase, and the lower accumulation of in the SEM micrographs were quantitatively examined by image
stress resulting from the volume expansion [25]. On the other hand, analysis software (Image-Pro 7.1, Image Cybernetics, USA). The atomic
Y2O3 does require a post-annealing process [19], and a higher degree of structure and composition of the oxynitride phases in the triple pockets
devitrification was achieved by using YSiAlON [20]. and grain boundaries were characterized by high-resolution transmis-
In this study, the Y2O3-Sc2O3 co-additive was first applied as a sion electron microscope (HR-TEM) and energy-dispersive X-ray
sintering aid for Si3N4, and the examination of its effects mainly spectroscopy (EDX) (JEM-2100F, JEOL, Japan). Sample preparation
focused on microstructure and thermal conductivity. By altering the for TEM was performed by either focused ion beam (DB-FIB, Helios
total amount and ratio of Y2O3 and Sc2O3 in the SSN experiment, the NanoLab, FEI, Netherland) or mechanical grinding, dimpling and ion-
optimal composition for high thermal conductivity was selected and beam thinning method.
applied to SRBSN for further improvement of thermal conductivity. In The phase identification was conducted by X-ray diffactometer (D/
conclusion, we found that a particular combination of Y2O3-Sc2O3 Max 2200, Rigaku, Japan) using Cu α radiation at 40 kV and 100 mA.
resulted in higher thermal conductivity, even though it led to a similar The thermal conductivity was calculated as the product of thermal
microstructure and lattice oxygen content as other compositions. Based diffusivity, heat capacity, and density. The thermal diffusivity was
on the microstructure analysis and thermal conductivity measure- measured at room temperature by laser flashing method (LFA447,
ments, a hypothesis that this composition enables rapid heat transport Netsch, Germany) and heat capacity was set constant as a known value
across grain boundaries will be discussed with the aid of TEM in the literature (0.68 J/gK). The lattice oxygen content was obtained
observations, and microstructure-thermal conductivity modeling. by hot gas extraction method [6]. For this analysis, the sintered

5442
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

Table 1. S22. S44 conspicuously revealed unimodal distribution in the grain

Summary of properties of the SSN specimen sintered at 1900 °C for 6 h. sizes, containing bigger matrix grains and a higher portion of liquid
phase in the multigrain pockets, without any abnormal grains.
Composition ρrel (%TD) dN (μm) dA (μm) WLO (wt%) κ (W/mK)
These microstructural features were quantitatively evaluated by
S11 99.5 0.88 3.42 0.324 54.0 counting the frequency of each grain size and area via image analysis.
S40 97.9 1.14 2.57 0.583 67.2 The mean grain sizes were calculated by different methods, that is, the
S31 98.6 1.14 2.49 0.288 67.5
number-weighted mean grain size (dN) and area-weighted mean grain
S22 97.9 1.18 2.58 0.275 65.7
S13 98.5 1.14 2.55 0.316 78.8 size (dA) according to the equations given below,
S04 99.6 1.16 3.06 0.240 70.8 n
1
S80 95.7 1.40 2.84 0.693 70.6 dN = ∑ di
S62 99.3 1.43 2.73 0.244 73.6 n i =0 (1)
S44 98.9 1.51 2.68 0.292 73.7
S26 99.4 1.34 3.06 0.247 74.1 n
1
S08 98.7 1.22 2.58 0.265 72.7 dA = n ∑ Ai di
SYM 98.3 0.77 2.32 0.275 62.9 ∑i =0 Ai i =0 (2)

where n is total number of grains, di and Ai are the size and area of i-th
specimens were crushed by agate mortar, and then oxynitride glasses grain. The number-weighted mean grain sizes were calculated to be
were rinsed out by HF (48–51% solution, A.C.S. Reagent, J.T.Baker) at 0.88 µm, 1.18 µm, and 1.51 µm for S11, S22 and S44, respectively,
60 °C for 3 h and H2SO4 (95–98% solution, reagent grade, Sigma with the same ranking as the total additive amount. On the other hand,
Aldrich) at 120 °C for 2 h. The acid-rinsed powder was washed by the area-weighted average grain size did not follow such a sequence,
centrifuge with distilled water and ethanol for 2 times each, and then resulting in 3.42 µm, 2.58 µm and 2.68 µm for S11, S22 and S44,
dried at 105 °C in an oven overnight. The lattice oxygen content was respectively. The unexpectedly large dA value of S11 is ascribed to the
measured by ONH determinator (ONH-2000, ELTRA, Germany), with presence of abnormal grains, which compensate the effects of small
peak deconvolution according to Kitayama et al. [6]. grains by multiplying grain sizes by the corresponding area fraction.
When 4 mol% additive was added, the different ratios of Y2O3 and
Sc2O3 had scarse influence on microstructure, although the total
3. Results and discussion additive volume fraction varied from 4.09% to 3.28% with increasing
scandia proportion. Also, the dN of all the 4 mol% specimens were
3.1. Si3N4-Y2O3-Sc2O3 system prepared by SSN process nearly constant, ranging from 1.14 μm to 1.18 µm, while dA values
showed slight fluctuation. On the other hand, for the 8 mol% additive
The overall data of the SSN experiment are summarized in Table 1. system, compositional variation caused a change in frequency counts of
The SEM micrographs of samples containing an equimolar ratio of size and area of each grain. In the number-weighted evaluation, the
Y2O3 and Sc2O3 are presented in Fig. 1, which clearly shows that most coarse microstructure revealed was S44, whereas the scandia-
increasing the total volume fraction of the additives led to significant added composition, S08, exhibited the smallest dN value among the
microstructural changes. Specifically, the microstructure of S11 con- 8 mol% additive compositions. Nevertheless, the compositional varia-
sists of small, irregular matrix grains with a few number of exaggerated tion had less influence on microstructure as compared with the
elongated grains, which grew to several microns in size. It was more remarkable change that resulted from different additive content.
difficult to distinguish abnormal grains from the small matrix grains in Prior to the evaluation of thermal conductivity for the Si3N4- Y2O3-

(a) (b)

10μm 10μm

(c)

10μm

Fig. 1. SEM micrographs of SSN specimens with equimolar Y2O3-Sc2O3 additives: (a) S11; (b) S22; (c) S44.

5443
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

Sc2O3 system, the SYM composition, a conventional additive composi- both crystallized and amorphous phases had different chemical ele-
tion for high thermal conductivity Si3N4, was examined for comparison ments in S22 and S13, as summarized in Table 2. The devitrified phase
with the previous values in the literature. The thermal conductivity of contained a higher proportion of scandium and zirconium whereas the
SYM was measured to be 62.9 W/mK, which is lower than the reported amorphous phases were always silicon and yttrium rich phases.
conductivity (91 W/mK) [26]. The low thermal conductivity in this Considering the tendency of scandium for devitrification, these com-
study might stem from different processing conditions such as the positional results are in good accordance with previous reports [23].
purity of starting materials and packing powder, milling media, gas- Fig. 3 demonstrates the grain boundaries of S13 and SYM as
inlet temperature and cooling rate, and degree of reducing atmosphere observed by high-resolution TEM. Fig. 3(c) indicates a typical amor-
in the sintering furnace. In any case, it is reasonable to consider the phous grain boundary observed in the SYM specimen, with a thickness
thermal conductivity values measured in this study to be relative values of 1.7 nm, showing no clear spots of atoms. On the other hand,
rather than absolute and optimized ones. Fig. 3(a) shows the grain boundary of the S13 specimen, which
The thermal conductivity of each specimen is summurized in contained clear and bright atomic spots. Although there was no lattice
Table 1. For equimolar compositions, thermal conductivity exhibited fringe observed, the clear arrangement is more or less similar to the
a tendency to grow with the increasing total amount of sintering rare-earth ion arrangement along the basal and prismatic planes [27],
additives, with thermal conductivity values of 54.0 W/mK, 65.7 W/mK which implies more ordered and organized atomic distribution in the
and 73.7 W/mK for S11, S22 and S44, respectively. Although increas- grain boundary phase. Also, we observed a clear grain boundary with
ing the additive content resulted in a higher volume fraction of glassy negligible thickness of films as shown in Fig. 3(b). Thus, it is inferred
phase and a smaller area-weighted average grain size, S44 yielded that a higher degree of devitrification at the triple pockets and the
higher thermal conductivity, attributed to the increase in dN value, i.e., probable existence of more ordered and clean grain boundaries would
the decrease in the total grain boundary area. The effect of area-based be advantageous for rapid heat transport across the grain boundaries of
grain size is likely to be undermined due to the overall fine micro- S13 compared to the amorphous glassy phases in the entire range of
structure and large volume fraction of intergranular phases. other specimens. In the next section, further evidence of the different
On the other hand, it is noteworthy that despite having a micro- grain boundary properties is examined in the SRBSN experiments.
structure similar to the 4 mol% additive compositions, S13 exhibited
the highest thermal conductivity in this experiment, at 78.8 W/mK. In 3.2. Microstructure and thermal conductivity of SRBSNs
particular, because no difference in microstructure was observed for
the 4 mol% additive specimens, the improvement in thermal conduc- For the SRBSN fabrication, three compositions were selected based
tivity of S13 likely resulted from either a lattice variation or devitrifica- on the SSN experiments: 2 mol% Y2O3 – 5 mol% MgO (RYM), 0.5 mol
tion of the glassy phases in triple pockets and grain boundaries. % Y2O3 – 1.5 mol% Sc2O3 (RYS1) and 1 mol% Y2O3 – 3 mol% Sc2O3
To confirm the factors affecting thermal conductivity, the lattice (RYS2). A standard RYM composition was selected, consistent with the
oxygen contents of the sintered specimens were analyzed by the hot gas literature. RYS2 had the same composition as S13, which exhibited the
extraction method and summurized in Table 1. The use of single Y2O3 extraordinary κ value, while the RYS1 composition contained half of
or Sc2O3 additives proved ineffective for eliminating lattice oxygen the additives of RYS2. All of the compositions showed good nitridation
regardless of the total amount of sintering aids. In the 4 mol% system, behavior, yielding a nitridation rate of over 90% based on the weight
the mixture of additives resulted in similar values of lattice oxygen gain from silicon to silicon nitride after nitridation. According to the
content, which were less than the specimens containing solely Y2O3 or XRD analysis of the powders pulverized from the RBSN compacts, no
Sc2O3 as an additive. For the 8 mol% total additive, the lattice oxygen residual silicon was detected, meaning that all of the silicon was
content somewhat fluctuated in the Sc2O3-containing system, while the transformed to silicon nitride. Also, the β/(α+β) ratio did not display
Y2O3 single addition resulted in a higher lattice oxygen. any remarkable difference with respect to composition, recording 8.4%,
On the other hand, it seems that the coarse microstructure did not 8.8% and 5.4% for RYS1, RYS2 and RYM, respectively. After the post-
necessarily reduce the lattice oxygen as expected based on the dissolu- sintering at 1900 °C, only β-Si3N4 phase was detected in XRD patterns
tion-reprecipitation process during grain growth. According to other without any remaining α-Si3N4 phase, regardless of the length of
reports, the microstructure evolution had a clear influence on the holding time and additive compositions.
lattice oxygen content when lattice oxygen content was measured for Fig. 4 presents the SEM micrograph of those samples sintered for
specimens with fixed composition, and fabricated by different sintering 1 h and 12 h, respectively, and the number-weighted and area-
schedule [11,26]. In this study, however, a diversity of additive weighted mean grain sizes with respect to sintering time are summar-
compositions were applied, which generated dissimilar efficiencies in ized in Table 3. When the specimens were sintered for 1 h, all the
lattice purification, because the oxygen activity in the glassy phases specimen exhibited a limited degree of grain growth, and among them
changes with respect to the oxygen concentration. In this regard, a RYM showed the largest mean grain size in both evaluation methods,
larger grain size might experience a lower degree of purification which can be attributed to the formation of a liquid phase at low
efficiency due to the relatively excessive amount of oxide additives. temperature due to MgO. However, grain growth rapidly progressed in
Meanwhile, lattice oxygen content cannot explain the unusual the case of RYS2, followed by RYS1 and RYM in terms of the number-
thermal conductivity of the S13 composition, which exhibited lattice based approach. On the other hand, sintering for 12 h produced a
oxygen content similar to the 4 mol% compositions, as well as having reversal of dA, so that the largest area-weighted mean grain size is for
almost the same microstructure. In this regard, detailed observations of RYS1, which shows abrupt grain growth from 3.3 µm after 6 h
the state of the glassy phase in S13 are presented in Fig. 2 and Fig. 3, in sintering, to 10.4 µm after 12 h sintering. Also, RYM exhibited an
comparison with other compositions. Fig. 2 shows the glassy phases almost four times increase in grain size, from 2.1 µm after 6 h sintering
observed at triple or multigrain pockets in the S22 and S13 composi- to 8.0 µm after 12 h sintering. Compared to these sharp jumps in area-
tions. The standard SYM specimen was excluded from this figure since weighted grain size, RYS2 showed linear and steady grain growth
it did not contain any crystallized phase in the TEM observation. For during the extended sintering time.
both S22 and S13 in Fig. 2, the black area centered in both figures The measured thermal conductivity values (κ0), and the porosity-
indicates the crystallized pocket phases. A noticeable feature is that S22 calibrated thermal conductivity (κcal) are presented in Table 3, with κcal
shows a somewhat truncated secondary phase whose corner is not given by the equation [28],
directly connected with the Si3N4 grains due to the residual glassy
2 + Vp
phases. On the other hand, the crystallized triple pocket in S13 has κcal = κ0
more sharp corners compared to S22. The EDS analysis revealed that 2(1 − Vp ) (3)

5444
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

(a) (b)
S
S S
A
C

C S
S
S
A S

Fig. 2. TEM micrographs of multigrain junctions in (a) S22 and (b) S13. The capital C stands for the crystallized secondary phase, A for the amorphous phase, S for the silicon nitride
grain. Note the corners of the crystallized pockets. At the right-bottom of each micrograph is the selected area electron diffraction (SAED) pattern of the crystallized secondary phase.

where Vp is volume fraction of the pores. It is noteworthy that RYS2 Table 2.

sintered for 1 h exhibited a much higher thermal conductivity value TEM/EDS analysis of crystallized secondary phases and residual amorphous
phases at triple pockets for the S22 and S13 specimens.
compared to other specimens sintered for the same time. Considering
that sintering for 1 h apparently resulted in similar microstructures Element (at%) Si Y Sc Zr O N
without any noticeable grain growth, it is not easy to explain this
extraordinary thermal conductivity by conventional approaches such as S22 Crystalline 7.6 12.7 22.6 11.9 41.0 4.0
Amorphous 24.1 25.0 15.0 1.9 34.0 –
grain growth or lattice defects.
Also, another noticeable feature is that the thermal conductivity gap S13 Crystalline 8.9 9.6 23.3 7.2 51.0 –
between RYS2 and the other compositions became smaller as sintering Amorphous 27.7 22.1 5.7 1.7 42.8 –
time increased even though the difference in the number-weighted
mean grain size got larger. In this case, the area-weighted grain size
should be taken into account, since there was a reversal in the area- thermal conductivity.
weighted grain size after 12 h sintering. The rapid grain growth in
abnormal grains for RYS1 and RYM could improve the thermal 3.3. A microstructure-thermal conductivity model
conductivity of a major portion of the large grains, which would
outweigh the limited fraction of small matrix grains having lower As was closely investigated in the SSN and SRBSN experiments, it

(a) (b)

1.3 nm

β-Si3N4 β-Si3N4 β-Si3N4 β-Si3N4

(c)

β-Si3N4

1.7 nm

β-Si3N4

Fig. 3. h-TEM images of grain boundary interfaces for (a-b) S13 and for (c) SYM. The figures in the upper-left corner are enlarged images of the grain boundary in the center figure.

5445
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

(a) (b) (c)

5μm

(d) (e) (f)

Fig. 4. SEM microstructure of SRBSN specimens sintered for 1 h and 12 h: (a) RYS1 for 1 h; (b) RYS2 for 1 h; (c) RYM for 1 h; (d) RYS1 for 12 h; (e) RYS2 for 12 h; (f) RYM for 12 h.

Table 3. grain of diameter D and height H was treated as having isotropic

Summary of properties of the SRBSN specimen sintered at 1900 °C for thermal conductivity. The volume ratio of the grain boundary film and
various soaking times.
grains can be calculated as below,
Composition Sintering ρrel (% dN (μm) dA (μm) κ0 (W/ κcal (W/ V1gb
time (hr) TD) mK) mK) Vgb = Vs
V1s (4)
RYS1 1 96.0 0.40 0.88 43.5 46.2
3 99.2 0.64 1.60 60.3 61.1 4π
V1s = ac
6 100.1 0.86 3.28 78.4 78.2 3 (5)
12 99.0 1.15 10.37 114.1 115.9
⎛ ⎛ 2 ⎞⎞
c2 c − a 2 ⎟⎟
RYS2 1 97.4 0.39 1.20 73.3 76.3 S1s = 2πa 2⎜⎜1 + sin−1⎜⎜ ⎟⎟
3 99.4 0.80 3.16 84.4 85.1 ⎝ a c2 − a2 ⎝ c ⎠⎠ (6)
6 99.0 1.02 4.12 88.8 90.1
12 98.6 1.84 7.47 114.4 116.9 S1s⋅δ
V1gb ≈
2 (7)
RYM 1 93.3 0.51 1.23 50.4 55.9
3 98.2 0.61 2.35 64.8 66.6 where Vgb and Vs are volume fractions of the grain boundary film and
6 98.9 0.63 2.12 69.2 70.3 Si3N4 with respect to the total volume of the system, respectively, V1 s
12 98.2 0.95 8.01 105.1 108.1
and S1 s are the volume and surface area of a Si3N4 spheroid,
respectively, and a, c, δ are the semi-diameter of the short and long
was found that the addition of a specific content and ratio of Y2O3 and axes (D/2, H/2), and the thickness of the grain boundary. First, the
Sc2O3, i.e., the values used in S13 and RYS2, led to an unusual matrix phase consisting of Si3N4 and the grain boundary film is
improvement in thermal conductivity. This result is at odds with calculated, and then the isolated triple pocket phase is embedded in
conventional understanding about the factors that enable high thermal the matrix phase. The thermal conductivity of the matrix phase can be
conductivity, such as coarse microstructure and lower lattice oxygen expressed as below,
content. Another plausible explanation for this phenomenon is that the ⎛ ⎞
V′(
s Kgb − K s)[(Ks − K gb)(2S33 + S11) + 3K gb]
change in thermal resistance across grain boundaries affected the K m = K gb⎜⎜1 − 2
⎟
2⎟
thermal conductivity of the whole system. It was suggested that these ⎝ s S11S33 + K gb(Ks − K gb)R + 3K gb ⎠
3(Ks − K gb) (1 − V′)
compositions might possess different characteristics at the interfaces (8)
due to the higher degree of devitrification, and different features at
R = 3(S11 + S33) − V′(2S
s 11 + S33) (9)
grain boundaries, as observed by TEM.
In order to elucidate the effects of microstructure on thermal ⎡ ⎤
a2c ⎛ 2 ⎞
conductivity, several models have already been reported for Si3N4 S11 = S22 = ⋅⎢ c ⎜ c − 1⎟ − cosh−1 c ⎥
3 ⎢
a ⎝ a2 ⎠ a ⎥⎦
[16,28,29], but these commonly did not take note of the grain 2( c2 − a2 ) ⎣ (10)
boundary characteristics. We propose a microstructure-thermal con-
ductivity model based on the Hatta and Taya model for a coated fiber where Vs’ is the volume fraction of Si3N4 with respect to the total
composite [30], combined with Bruggeman's equation [31]. In the matrix volume (=Vs/(Vs+Vgb)), Ks is the mean isotropic thermal
model, prolate spheroid shaped Si3N4 grains are assumed to be conductivity of a Si3N4 grain, Kgb is the thermal conductivity of the
randomly distributed in a continuous grain boundary film. The model grain boundary film. Then, the total thermal conductivity of the system
can be calculated by Bruggeman's equation, assuming that the triple

5446
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

pockets are isotropic spheres dispersed homogeneously in the matrix Fig. 5(a-c) indicates the total thermal conductivity with respect to
phases. the glass volume fraction and the thermal conductivity of the glassy
phase in the multigrain junctions and grain boundaries. Ks was
⎡ K ⎤1/3 K tp − K tot presumed to be 120 W/mK. As shown in Fig. 5, increasing the volume
1 − Vtp = ⎢ m ⎥ ⋅ fraction of the glassy phases resulted in a linear decrease in the thermal
⎣ K tot ⎦ K tp − K m (11)
conductivity of the total system due to the low thermal conductivity of
where Vtp is the volume fraction of the triple pockets with respect to the the glassy phase (1 W/mK). This is also consistent with the reported
total volume fraction, Ktp and Ktot are the thermal conductivities of the approach for volatile sintering additives such as Li2O, MgO, and
triple pocket and the total system. fluoride, which enhanced thermal conductivity by decreasing the liquid
First of all, several constant values should be properly set up for the phase volume at elevated temperature, after densification and grain
calculation. According to the image analysis, all of the compositions growth [11,12,28]. Nevertheless, considering the glassy phase volume
exhibited a similar aspect ratio of 2, reflecting the major number of to be from 4.6 to 10.7 vol% in this experiment, the variation in thermal
small matrix grains with lower aspect ratio. The total volume of the conductivity due to the volume difference is relatively trivial, falling
glassy phase is expected to be 6 vol% considering the SiO2 layers on the within 9.3% of margin.
Si3N4 raw powders as well as the sintering additive. Obviously, the total Fig. 5(a) illustrates the effect of the thermal conductivity of the
glass volume depends on not only the additive content we selected but triple pocket phases. Regardless of the Ktp variation, the thermal
also the starting materials and fabrication processes for the SSN and conductivity of the entire state remained at a similar level. As noted
SRBSN. However, according to the calculation results which will be by Kitayama et al., the pocket phases have less influence on the total
dealt with, the effect of the total additive content was somewhat thermal conductivity, because of their isolated distribution in the
marginal compared to other variables; thus, Vgb was set to 0.06 if a matrix phase [16]. In contrast, Fig. 5(b) shows that the grain boundary
special notation was not provided. phases, which are continuously linked to the whole system, had a
The grain boundary thickness, δ, was chosen to be 1.5 nm. The decisive effect, even though the volume fraction was much smaller than
thermal conductivity of the glassy phase at the triple pocket and grain the triple pockets. When Kgb increased from 0.5 W/mK to 1.0 W/mK,
boundary is generally regarded to be 1 W/mK, however, that is a the total thermal conductivity improved as much as 20%, and the
somewhat overestimated value for the grain boundary film, taking the margin of increase gradually decreased as higher Kgb was applied.
thickness effect of an amorphous material on the thermal conductivity Thus, it can be said that controlling the grain boundary properties is
into account. According to Lee and Cahill, the thermal conductivity of more effective for improving the thermal conductivity of the total
amorphous SiO2 and SiNx significantly decreased when the film system than changing the Ktp values.
thickness was less than 200 nm due to an increase in thermal resistivity Fig. 5(c) shows the effect of grain boundary thickness on the overall
at the crystalline-amorphous interfaces [32]. The thermal conductivity thermal conductivity. It is clearly seen that the narrower grain
of the thin film bottomed at nearly 0.5 W/mK for both SiO2 and SiNx at boundary thickness enables higher thermal conductivity. In particular,
room temperature. Based on this report, the thermal conductivity of decreasing the thickness from 1.5 to 0.8 nm produced an effect that
the glass phase film at the grain boundary was assumed to be 0.5 W/ was equivalent to increasing the thermal conductivity of the grain
mK while Ktp was assumed to be 1 W/mK. boundary from 0.5 to 1.0 W/mK. This indicates that a higher Kgb value

Fig. 5. Calculation results of a model with respect to the volume faction of the glassy phase (Vgl) and grain diameter: (a) the effect of the thermal conductivity of the triple pockets where
D=0.5 µm and Kgb=0.5 W/mK; (b) the effect of the thermal conductivity of the grain boundaries where D=0.5 µm and Ktp=1 W/mK; (c) the effect of the grain boundary thickness where
D=0.5 µm, Ktp=1 W/mK and Kgb=0.5 W/mK; (d) the effect of the grain boundary thermal conductivity with varying grain diameters, where Vgl=0.06 and Ktp=1 W/mK. In all plots, the
aspect ratio was assumed to be 2, and the thermal conductivity of the Si3N4 grain to be 120 W/mK, while the thickness of the grain boundaries was set to 1.5 nm, except plot (c).

5447
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

120 120
(a) x+y=2 Ks=200

Thermal Conductivity (W/mK)

(b)

Thermal Conductivity (W/mK)

x+y=4 Ks=140
100 x+y=8 Ks=120
100
SYM
S13 Ks=95 Ks=120
80 80
Ks=95
RYS1
60 60 RYS2
RYM
Kgb=0.5
Kgb=0.5
40 Kgb=1.0 40
Kgb=1.0
0.2 0.4 0.6 0.8 1.0 1.2 1.4 0.2 0.4 0.6 0.8 1.0 1.2 1.4
Grain Diameter (µm) Grain Diameter (µm)
Fig. 6. Plot of model calculation data and experimentally measured thermal conductivity: (a) SSN process; (b) SRBSN process. The x and y values in the legend of SSN plot indicates “x
mol% Y2O3 – y mol% Sc2O3”. The Ks values for each calculated data were chosen to fit the experimental data.

and a smaller δ play the same role in reducing thermal resistance and the lower degree of area-weighted mean the grain size can cause
across the grain boundaries. the lower Ks value of RYS2. It is expected that further sintering may
Fig. 5(d) demonstrates how grain size affects thermal conductivity result in the reversal of the thermal conductivity of these two groups.
with respect to Kgb. In all cases, the larger grain diameter obviously However, grain growth concomitant with longer sintering time must
yields higher thermal conductivity, and this is not attributed to the also cause a deterioration in strength.
cleansing effect of the Si3N4 grain because of the constant Ks value, but Thus, the proper control of Kgb is preferred in order to achieve both
to the decrease in the volume fraction of the grain boundary phases. high thermal conductivity and better mechanical properties. Also,
Since Vgl remains constant, the volume fraction of the triple pockets is thermal conductivity can be further enhanced by adopting more
expected to increase, but the influence of Vtp is less crucial than that of reducing processing routes, as mentioned by Zhou et al. [14], which
Vgb as shown in Fig. 5(a). Interestingly, the higher Kgb value can will expel lattice oxygen efficiently and provide higher Ks values.
enhance the thermal conductivity of the total system where grain Comparing Ks=95 W/mK and Ks=200 W/mK, one notices that adopt-
growth is not sufficient. In other words, the thermal conductivity plot of ing a higher Kgb results in a more remarkable improvement than can be
the lower Kgb (0.5 W/mK) rapidly decreases as the grain size becomes achieved when the Ks value becomes higher.
smaller, while the decreasing slope of the plot for higher Kgb (1.0 W/
mK) is less steep. This indicates that it is possible to maintain a high κ
value over an extended range and that a finer microstructure with
4. Conclusion
excellent mechanical properties can be compatible with higher thermal
conductivity by accelerating heat transport across the grain boundaries.
The effect of sintering additives Y2O3-Sc2O3 on the thermal con-
Based on the calculation, in Fig. 6 we plotted all of the experimental
ductivity of Si3N4 was examined, for different amounts and ratios of
thermal conductivity values of the SSN and SRBSN specimens with
each additive. Two different processing routes, using SSN and SRBSN,
calculated data fitted to the measured data. Because the plot needs an
were applied to determine the optimal composition for achieving
average grain diameter, which is the short boundary of the elongated
higher thermal conductivity. In the SSN experiment, the total additive
grains, the average grain diameter of each specimen was evaluated by
content had a more decisive effect on microstructure and thermal
image analysis. For the SSN plots in Fig. 6(a), most of the thermal
conductivity than the Y2O3-Sc2O3 ratio, except for the S13 composition.
conductivity values are laid on the Ks=95 W/mK, Kgb=0.5 W/mK lines,
The overall tendency in thermal conductivity was explained by micro-
except for S11 and S13. It is not clear why S11 is far from the line,
structure and lattice oxygen content, but the thermal conductivity of
though it may be partly attributed to the lower liquid volume. But for
the S13 specimens may be affected by intergranular phases with
S13, the higher Kgb value (1 W/mK) and the same Ks value applied, and
different degree of crystallinity and thickness. In particular, in
fitted well to the high κ value of S13.
SRBSN experiment, sintering for 1 h showed scarce grain growth, but
Furthermore, concerning the SRBSN results in Fig. 6(b), the RYS1
the RYS2 composition exhibited extraordinary thermal conductivity. A
and RYM points are on the Kgb=0.5 W/mK line with Ks=120 W/mK
thermal conductivity model considering the microstructure and ther-
and 200 W/mK, respectively, whereas the RYS2 values are far from the
mal conductivity of the grain boundary was consistent with the
fitting lines. Instead, RYS2 can be fitted by regarding Kgb to be 1 W/
experimental tendency, showing that the intergranular thin films have
mK, in particular where the grain diameter is small. If RYS2 is fitted to
great influence on the total thermal conductivity of the system. Also, it
the Kgb=0.5 W/mK lines, then a much higher Ks value must be
was determined that the impact of the grain boundary becomes more
assumed, which is not consistent with the experimental conditions of
critical for finer microstructure. The experimental data obtained for
the same sintering temperature and similar soaking times, or with the
grain diameter and thermal conductivity were well fitted by the plot of
similar or lower efficiency of lattice purification.
the model calculation data.
Inversely, if RYS1 and RYM are fitted by Kgb=1 W/mK, the Ks value
should be quite lower than that of RYS2, which is not intuitively
acceptable either. Thus, it seems reasonable to say that the large
discrepancy between RYS1, RYM and RYS2 can be attributed to the Acknowledgments
different heat transport rate at the grain boundaries, and that a
reasonable set of Ks and Kgb are fitted to the experimental results. This work was supported by Materials & Components Technology
On the other hand, after being sintered for 12 h of soaking time, Development (MCTD) Program of the Ministry of Trade, Industry &
RYS1 and RYM can be fitted at Ks=200 W/mK, and RYS2 at nearly Energy (MOTIE) of Korea (Grant no. 10048968) and Fundamental
Ks=140 W/mK. Although the grain growth of RYS2 progressed more Research Program of Korean Institute of Materials Science (KIMS)
than other compositions, the higher oxygen activity in the glassy phases (Grant no. PNK4671).

5448
J.-M. Kim et al. Ceramics International 43 (2017) 5441–5449

References [17] N. Hirosaki, Y. Akimune, M. Mitomo, Effect of grain growth of β-silicon nitride on
strength, Weibull modulus, and fracture toughness, J. Am. Ceram. Soc. 76 (1993)
1892–1894.
[1] C. Buttay, D. Planson, B. Allard, D. Bergogne, P. Bevilacqua, C. Joubert, M. Lazar, [18] Y.-W. Kim, K.-Y. Lim, W.-S. Seo, Microstructure and thermal conductivity of silicon
C. Martin, H. Morel, D. Tournier, State of the art of high temperature power carbide with yttria and scandia, J. Am. Ceram. Soc. 97 (2014) 923–928.
electronics, Mater. Sci. Eng. B-Adv. 176 (2011) 283–288. [19] M.K. Cinibulk, G. Thomas, S.M. Johnson, Fabrication and secondary-phase
[2] H. Lu, C. Bailey, C. Yin, Design for reliability of power electronics modules, crystallization of rare-earth disilicate–silicon nitride ceramics, J. Am. Ceram. Soc.
Microelectron. Reliab. 49 (2009) 1250–1255. 75 (1992) 2037–2043.
[3] J.S. Haggerty, A. Lightfoot, Opportunities for enhancing the thermal conductivities [20] M.K. Cinibulk, G. Thomas, S.M. Johnson, Grain-boundary-phase crystallization
of SiC and Si3N4 ceramics through improved processing, Ceram. Eng. Sci. Proc. 16 and strength of silicon nitride sintered with a YSiAlON glass, J. Am. Ceram. Soc. 73
(1995) 475–487. (1990) 1606–1612.
[4] N. Hirosaki, S. Ogata, C. Kocer, H. Kitagawa, Y. Nakamura, Molecular dynamics [21] J.S. Vetrano, H.-J. Kleebe, E. Hampp, M.J. Hoffmann, M. Rühle, R.M. Cannon,
calculation of the ideal thermal conductivity of single-crystal α-and β−Si3N4, Phys. Yb2O3-fluxed sintered silicon nitride, J. Mater. Sci. 28 (1993) 3529–3538.
Rev. B 65 (2002) 134110. [22] S. Guo, N. Hirosaki, Y. Yamamoto, T. Nishimura, Y. Kitami, M. Mitomo,
[5] B. Li, L. Pottier, J. Roger, D. Fournier, K. Watari, K. Hirao, Measuring the Microstructural characterization and high-temperature strength of hot-pressed
anisotropic thermal diffusivity of silicon nitride grains by thermoreflectance silicon nitride ceramics with Lu2O3 additives, Philos. Mag. Lett. 83 (2003)
microscopy, J. Eur. Ceram. Soc. 19 (1999) 1631–1639. 357–365.
[6] M. Kitayama, K. Hirao, A. Tsuge, M. Toriyama, S. Kanzaki, Oxygen content in β- [23] M.L. Mecartney, Enhanced crystallization of a glassy phase in silicon nitride by the
Si3N4 crystal lattice, J. Am. Ceram. Soc. 82 (1999) 3263–3265. addition of scandia, J. Am. Ceram. Soc. 70 (1987).
[7] M. Kitayama, K. Hirao, A. Tsuge, K. Watari, M. Toriyama, S. Kanzaki, Thermal [24] D.-S. Cheong, W.A. Sanders, High-temperature deformation and microstructural
conductivity of β-Si3N4: ii, Effect of lattice oxygen, J. Am. Ceram. Soc. 83 (2000) analysis for silicon nitride–scandium (III) oxide, J. Am. Ceram. Soc. 75 (1992)
1985–1992. 3331–3336.
[8] H. Yokota, H. Abe, M. Ibukiyama, Effect of lattice defects on the thermal [25] H.-J. Kleebe, G. Pezzotti, G. Ziegler, Microstructure and fracture toughness of Si3N4
conductivity of β-Si3N4, J. Eur. Ceram. Soc. 23 (2003) 1751–1759. ceramics: combined roles of grain morphology and secondary phase chemistry, J.
[9] H. Yokota, M. Ibukiyama, Effect of lattice impurities on the thermal conductivity of Am. Ceram. Soc. 82 (1999) 1857–1867.
β-Si3N4, J. Eur. Ceram. Soc. 23 (2003) 55–60. [26] Y. Zhou, H. Hyuga, D. Kusano, Y. Yoshizawa, T. Ohji, K. Hirao, Development of
[10] M. Kitayama, K. Hirao, M. Toriyama, S. Kanzaki, Thermal conductivity of β-Si3N4: high-thermal-conductivity silicon nitride ceramics, J. Asian Ceram. Soc. 3 (2015)
i, Effects of various microstructural factors, J. Am. Ceram. Soc. 82 (1999) 221–229.
3105–3112. [27] N. Shibata, G.S. Painter, R. Satet, M. Hoffmann, S.J. Pennycook, P.F. Becher, Rare-
[11] H. Hayashi, K. Hirao, M. Toriyama, S. Kanzaki, K. Itatani, MgSiN2 addition as a earth adsorption at intergranular interfaces in silicon nitride ceramics: subnan-
means of increasing the thermal conductivity of β-silicon nitride, J. Am. Ceram. ometer observations and theory, Phys. Rev. B 72 (2005) 140101.
Soc. 84 (2001) 3060–3062. [28] X. Zhu, Y. Sakka, Y. Zhou, K. Hirao, Processing and properties of sintered reaction-
[12] B. Bing, T. Fu, X.S. Ning, Thermal conductivity and mechanical property of Si3N4 bonded silicon nitride with Y2O3–MgSiN2: effects of Si powder and Li2O addition,
ceramics sintered with CeF3/LaF3 additives, Adv. Mat. Res. 105 (2010) 171–174. Acta Mater. 55 (2007) 5581–5591.
[13] X. Zhu, Y. Zhou, K. Hirao, Development of sintered reaction-bonded Si3N4 [29] H. Yokota, S. Yamada, M. Ibukiyama, Effect of large β-Si3N4 particles on the
materials with high thermal conductivity, J. Ceram. Soc. Jpn. 112 (2004) thermal conductivity of β-Si3N4 ceramics, J. Eur. Ceram. Soc. 23 (2003)
S410–S414. 1175–1182.
[14] Y. Zhou, H. Hyuga, D. Kusano, Y. Yoshizawa, K. Hirao, A tough silicon nitride [30] H. Hatta, M. Taya, Effective thermal conductivity of a misoriented short fiber
ceramic with high thermal conductivity, Adv. Mater. 23 (2011) 4563–4567. composite, J. Appl. Phys. 58 (1985) 2478–2486.
[15] Y.-J. Park, M.-J. Park, J.-M. Kim, J.-W. Lee, J.-W. Ko, H.-D. Kim, Sintered [31] V.D. Bruggeman, Berechnung verschiedener physikalischer Konstanten von het-
reaction-bonded silicon nitrides with high thermal conductivity: the effect of the erogenen Substanzen. I. Dielektrizitätskonstanten und Leitfähigkeiten der
starting Si powder and Si3N4 diluents, J. Eur. Ceram. Soc. 34 (2014) 1105–1113. Mischkörper aus isotropen Substanzen, Ann. Phys.-Berl. 416 (1935) 636–664.
[16] M. Kitayama, K. Hirao, M. Toriyama, S. Kanzaki, Thermal conductivity of β-Si3N4: [32] S.M. Lee, D.G. Cahill, Heat transport in thin dielectric films, J. Appl. Phys. 81
i, effects of various microstructural factors, J. Am. Ceram. Soc. 82 (1999) (1997) 2590–2595.
3105–3112.

5449