Effects of β-Si₃N₄ Seeds on Microstructure and Performance of Si₃N₄ Ceramics in Semiconductor Package

Abstract

Among the various ceramic substrate materials, Si₃N₄ ceramics have demonstrated high thermal conductivity, good thermal shock resistance, and excellent corrosion resistance. As a result, they are well-suited for semiconductor substrates in high-power and harsh conditions encountered in automobiles, high-speed rail, aerospace, and wind power. In this work, Si₃N₄ ceramics with various ratios of α-Si₃N₄ and β-Si₃N₄ in raw powder form were prepared by spark plasma sintering (SPS) at 1650 °C for 30 min under 30 MPa. When the content of β-Si₃N₄ was lower than 20%, with the increase in β-Si₃N₄ content, the ceramic grain size changed gradually from 1.5 μm to 1 μm and finally resulted in 2 μm mixed grains. However, As the content of β-Si₃N₄ seed crystal increased from 20% to 50%, with the increase in β-Si₃N₄ content, the ceramic grain size changed gradually from 1 μm and 2 μm to 1.5 μm. Therefore, when the content of β-Si₃N₄ in the raw powder is 20%, the sintered ceramics exhibited a double-peak structure distribution and the best overall performance with a density of 97.5%, fracture toughness of 12.1 MPa·m^1/2, and a Vickers hardness of 14.5 GPa. The results of this study are expected to provide a new way of studying the fracture toughness of silicon nitride ceramic substrates.

1. Introduction

The development of third-generation semiconductors has induced a huge demand for ceramic substrates for power-integrated circuits [1,2,3,4]. Conventionally, Al₂O₃ has been utilized as the substrate in power-integrated circuits due to its cost-effectiveness. However, Al₂O₃ has a low toughness (3–5 MPa·m^1/2) and thermal conductivity (18–24 W∙m⁻¹∙K⁻¹) and cannot effectively satisfy the requirements of high-power demand in energy and other emerging fields [5]. AlN is another ceramic substrate material that shows high thermal conductivity (150–270 W∙m⁻¹∙K⁻¹), but it has the drawback of low fracture toughness (typically 3–3.5 MPa·m^1/2). When used in harsh working conditions of thermal shock and impact, such as those encountered in automobiles and wind power, AlN substrates are typically used in combination with plastic shock absorbers, which affects device miniaturization [6]. Due to their high theoretical thermal conductivity (320 W∙m⁻¹∙K⁻¹) and good fracture toughness (~10 MPa·m^1/2) [7], Si₃N₄ ceramics have been regarded as the most promising alternatives for Al₂O₃ and AlN. However, the performances of Si₃N₄ ceramics currently available are far from theoretical optimization, which has limited their further applications.

The performance of Si₃N₄ ceramics is affected by various factors, among which their microstructure is a prominent factor [8,9]. The microstructure has been shown to directly influence the hardness and fracture toughness of Si₃N₄ [10,11,12,13,14]. At present, many researchers increase the fracture toughness of silicon nitride ceramics by adding reinforcements [15,16,17]. Guo et al. [18] prepared silicon nitride ceramics with high fracture toughness (9.7 MPa·m^1/2) by taking Si₃N₄ as raw material and adding Lu₂O₃ as the reinforcing phase. However, adding the reinforcing phase cannot fundamentally solve the problem of low fracture toughness and high cost of silicon nitride ceramics.

Since the microstructure of Si₃N₄ ceramics is generally complex and uncontrolled, the effects of the phase composition of raw materials on the performances of Si₃N₄ have been thoroughly investigated [19,20]. Becher et al. [21] prepared Si₃N₄ ceramics with high fracture toughness (~10 MPa·m^1/2) by sintering at 1850 °C under 1 MPa N₂ atmosphere for 6 h, with α-Si₃N₄ and 2 vol% β-Si₃N₄ as raw materials. In order to ascertain the effects of β-Si₃N₄ seed crystal on the performances of Si₃N₄ ceramics, Peillon et al. [22] prepared Si₃N₄ ceramics with 2 vol% and 5 vol% of β-Si₃N₄ seed crystal and found that Si₃N₄ ceramics with 5 vol% of β-Si₃N₄ exhibited improved fracture toughness (8.4 MPa·m^1/2). In their study, the fracture toughness increased by 30% as the sintering time increased by 3 h. Lee et al. [23] reported the synthesis of nanoscale α-Si₃N₄ with grain size distributions with two peaks when sintering was performed with the inclusion of β-Si₃N₄ (0%, 50%, 100%). The addition of β-Si₃N₄ led to improved performance of Si₃N₄ ceramics, where the sample with 50% of β-Si₃N₄ seed crystal exhibited the highest fracture toughness (7.9 MPa·m^1/2). However, there is a need to further investigate the optimization of the β-Si₃N₄ seed crystal content since the high costs of nanoscale β-Si₃N₄ are not suitable for industrial applications.

Spark plasma sintering (SPS) or plasma-activated sintering, is a novel rapid hot-pressing sintering technique where particles are sintered using a pulse current, where plasma is generated by particle discharge under pulse current. Owing to its unique heating pattern, spark plasma can achieve rapid heating, which reduces the sintering time. Due to this feature, samples can be prepared with ultra-fine grain sizes [24,25]. Liu et al. [26] reported synthesis of Ybα-SiAlON with high density (3.42 g/cm³) and high fracture toughness (6.2 MPa·m^1/2) by spark plasma sintering (1600 °C, 2 MPa, 5 min) with 5 wt% of Ybα-SiAlON as the seed crystal. Zamula et al. [27] prepared high-performance Si₃N₄ by spark plasma sintering (1800 °C, 20 min), wherein a complete transition from α-Si₃N₄ to β-Si₃N₄ was observed. The compactness of the prepared Si₃N₄ exceeded 98%, and its fracture toughness reached ~5.7 MPa·m^1/2.

In this study, Si₃N₄ was prepared by SPS of α-Si₃N₄ (1650 °C, 30 MPa, 30 min) using a uniform-sized seed crystal of β-Si₃N₄ seed crystal. The microstructure of Si₃N₄ samples with different contents of β-Si₃N₄ seed crystal was investigated along with its correlation with hardness and fracture toughness. Insights obtained from this study can facilitate the design and preparation of Si₃N₄ with improved stability and effectiveness for applications in ceramic substrates.

2. Experimental

2.1. Materials

Mixtures of Si₃N₄ and sintering aid were used in this study. Two types of silicon nitride powders, 1# (α-Si₃N₄-rich powder) and 2# (β-Si₃N₄-rich powder) were used in the experiment. The powders had a nominal purity of over 90% and were purchased from Hebei Badu Metal Materials Co., Ltd. (Shijiazhuang, China). Nanoscale Y₂O₃ (purity > 99.99%), which served as the sintering aid, was purchased from Shanghai Yaoyi Alloy Materials Co., Ltd. (Shanghai, China). Al₂O₃ (0.2–0.4 μm, purity > 99.99%) was purchased from Hebei Yigui Welding Materials Co., Ltd. (Xingtai, China). PEG 400 (polyethylene glycol, purity > 99.99%), PAA (polyacrylic acid, purity > 99.99%), and ammonium citrate (purity > 99.99%) were purchased from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China).

2.2. Method

A mixture comprising 20 g of 1# and 2# Si₃N₄ was introduced into a beaker as indicated in

Table 1. Compositions of different Si₃N₄ samples.

Sample	Fraction of 1#	Fraction of 2#
SN0	100%	0%
SN1	90%	10%
SN2	80%	20%
SN3	70%	30%
SN4	50%	50%

. Subsequently, 360 μL of a solution containing PEG 400, ammonium citrate (dispersant), and 600 μL of PAA (adhesive) were added. To facilitate the sintering process, a total of 1.4 g of Y₂O₃ and 0.6 g of Al₂O₃ were introduced. The solution’s pH of 9.8 was modified and subsequently introduced into a polyurethane ball milling tank, containing zirconia grinding balls in a ball-to-powder weight ratio of 3:1. The wet-mixing process was carried out using pure water as the medium, and ball milling for 24 h. The compositions of various samples are provided in Table 1. Following ball milling, the slurry underwent vacuum drying at 120 °C for 24 h. After that, an 80-mesh sieve was utilized to obtain the Si₃N₄ composite precursor. The precursor was then subjected to pre-sintering at 800 °C for 2 h in a tube furnace, under an atmosphere of N₂ (OTF- 1200X, Hefei Kejing Materials Co., Ltd., China). After pre-sintering, the product was again sieved through an 80-mesh sieve to acquire Si₃N₄ composite particles. Finally, the as-prepared Si₃N₄ composite was inserted into a graphite mold and subjected to Spark Plasma Sintering (SPS) at 1650 °C for 30 min while being subjected to a pressure of 30 MPa (SPS-5T-5-III, Shanghai Huachen Technology Co., Ltd., Shanghai, China). As shown in Figure 1.

2.3. Measurement

The sample density was measured by using the Archimedes method. The samples were analyzed using a Bruker D8 X-ray diffractometer equipped with an energy dispersive spectrometer (XRD, Karlsruhe, Germany) and a field-emission scanning electron microscope (SEM, FEI-Nova Nano 450, Hillsboro, OR, USA) to ascertain their phase compositions and fracture morphologies, respectively. The hardness of polished Si₃N₄ samples was tested using a Vickers hardness meter (HVS-50, Shanghai Wanheng Precision Instrument Co., Ltd., Shanghai, China) under a pressure of 1000 N for 15 s. Each sample was exposed to five replicates and the average value was taken as the final result. The fracture toughness of Si₃N₄ (K_IC) was determined by using the Vickers hardness indentation method. The crack lengths of the samples were also determined. The fracture toughness can be calculated by:

$$K_{IC}=0.075\times P\times c^{\frac{3}{2}}$$

(1)

where K_IC is the fracture toughness (kgf·m^−3/2), P is the load (kgf), and c is the diagonal length of the indentation crack (mm). 1 kgf·m^−3/2 = 0.31 MPa·m^1/2.

3. Results and Discussion

3.1. Phase Composition of Seed Crystal and Microstructure of Samples

XRD analyses were utilized for investigating the phase composition of the raw powders, as shown in Figure 2. The results indicate that both 1# and 2# samples were composed of hexagonal α-Si₃N₄ (JCPDS No.74-0554, a = b = 7.765 nm, c = 5.622 nm, α = β = 120°, γ = 90°) and hexagonal β-Si₃N₄ (JCPDS No.72-1308, a = b = 7.608 nm, c = 2.911 nm, α = β = 120°, γ = 90°). No other phases were observed. Among them, the main phase of 1# Si₃N₄ powder was α-Si₃N₄, while the main phase of 2# Si₃N₄ powder was β-Si₃N₄.

Figure 3 shows SEM images of raw powder samples of α-rich Si₃N₄ and β-rich Si₃N₄. Raw powder samples of α-rich Si₃N₄ comprised irregularly shaped particles that were observed to aggregate (size = 0.5–1 μm), as shown in Figure 3a. Combined with the analysis of XRD patterns (Figure 2), it can be concluded that the observed rod-like powder was β-Si₃N₄. Raw powder samples of β-rich Si₃N₄ also comprised irregularly shaped particles with aggregation (size = 0.5–1 μm), as shown in Figure 3b. Overall, the two powders had consistent shapes and sizes, and the particles were found to be well dispersed with negligible agglomeration.

3.2. Effects of Seed Crystal Composition on Phase Composition and Microstructure of Ceramic Samples

XRD patterns of samples with different contents of β-Si₃N₄ seed crystal (SN0~SN4) prepared by SPS at 1100 °C exhibited characteristic peaks corresponding to hexagonal α-Si₃N₄ (JCPDS No.74-0554, a = b = 7.765 nm, c = 5.622 nm, α = β = 120°, γ = 90°) and hexagonal β-Si₃N₄ (JCPDS No.72-1308, a = b = 7.608 nm, c = 2.911 nm, α = β = 120°, γ = 90°), as shown in Figure 4.

Figure 5 illustrates the grain size changes in SN0~SN4 ceramic samples prepared by SPS at 1100 °C. In the cases with few amounts of β-Si₃N₄ seeds, e.g., sample SN0 (Figure 5a), the grain size is relatively small; in the cases with a few amounts of β-Si₃N₄, e.g., sample SN1~SN3 (Figure 5b–d), some grains become smaller while the other grains grew, the grain size variation in SN4 samples is small.

XRD patterns of samples with different contents of β-Si₃N₄ seed crystal (SN0~SN4) prepared by SPS at 1650 °C exhibited characteristic peaks corresponding to hexagonal β-Si₃N₄ (JCPDS No.72-1308) and hexagonal Y₄SiAlO₈N (JCPDS No.48-1630), as shown in Figure 6. No characteristic peaks of Al₂O₃ and Y₂O₃ sintering aids were detected. Complete phase transition of α-Si₃N₄ into β-Si₃N₄ and transition of Al₂O₃ and Y₂O₃ into Y₄SiAlO₈N solid solution were observed in the composite powder. Since the bond lengths of Si-N and Al-O were 0.174 nm and 0.175 nm, respectively, it is speculated that the Al-O bond may replace the Si-N bond during the sintering process. This resulted in the generation of Y₄SiAlO₈N solid solution via reaction with Y₂O₃, suggesting that Y, Al, and O were all present in the lattice structure of β-Si₃N₄ giving rise to an interfacial phase. This speculation is consistent with the observations from earlier reports [28,29].

The grain size distribution of Si₃N₄ significantly impacts its fracture toughness. Figure 7 illustrates the fracture morphologies of SN0-SN4 samples prepared by SPS at 1650 °C. As can be observed in Figure 7a, the SN0 sample comprised matrix grains (with ~1.4 μm granular grains), large grains (rod-like grains with a length of about 2 μm, width of about 1 μm), and abundant pores (~2 μm), a uniform distribution of large and small grains was observed. The SN1 sample comprised matrix grains (~1.3 μm granular grains), large grains (rod-like grains with a length of about 2 μm, width of about 1 μm), and a small number of pores (~1 μm), as shown in Figure 7b, the size distribution exhibited two peaks, where again the large and small grains were uniformly distributed. The SN2 sample comprised matrix grains (with ~1.3 μm granular grains and a small amount of rod-like grains with a length of about 2.2 μm, width of about 1 μm), large grains (rod-like grains with a length of about 6 μm, width of about 2.3 μm), and a small number of pores (~1 μm), as shown in Figure 7c, the size distribution exhibited two peaks, herein, the large and small grains were uniformly distributed in an interlocked structure. The SN3 sample comprised matrix grains (with ~1.3 μm granular grains and a small amount of rod-like grains with a length of about 1 μm, width of about 0.5 μm), large grains (rod-like grains (length of about 8 μm, width of about 2.2 μm)), and a small number of pores (~1 μm), as shown in Figure 7d, similar to the case of SN2, large and small grains were uniformly distributed in an interlocked structure. The SN4 sample comprised matrix grains (with granular grains (~1 μm) and rod-like grains with a length of about 1 μm, width of about 0.5 μm), large grains (with granular grains (~1.5 μm) and rod-like grains with a length of about 1.5 μm, width of about 0.5 μm), and a small number of pores (~1 μm), as shown in Figure 5e, similar to the other samples, the large and small grains were uniformly distributed. Additionally, sample failure was dominated by intergranular fracture.

With the increase in β-Si₃N₄ seed content, the grain distribution gradually changed from a single peak distribution (Figure 7a) to a bimodal distribution (Figure 7c). When excess β-Si₃N₄ seeds were introduced, the grain distribution changed from a bimodal distribution (Figure 7c) back to a unimodal distribution (Figure 7e).

The elemental composition of the SN0–SN4 samples was verified through the utilization of an EDX analysis, as shown in

Table 2. SEM-EDX relative content of various elements in silicon nitride ceramics prepared by SPS at 1650 °C.

Sample	Si (%)	N (%)	O (%)	Y (%)	Al (%)
SN0	46.3	43.7	6.6	1.9	1.5
SN1	47.2	43.8	6.2	1.7	1.1
SN2	46.1	43.8	6.7	1.8	1.6
SN3	46.7	43.5	6.4	2	1.4
SN4	46.9	43.2	6.6	1.7	1.6

. The Table demonstrates a uniform distribution of Si, N, Al, Y, and O elements within the silicon nitride matrix. The proportion of each element in SN0~SN4 ceramic samples is similar.

3.3. Effect of β-Si₃N₄ Seed on the Sintering and Grain Growth of Silicon Nitride Ceramics

During the liquid phase sintering process, α-Si₃N₄ in the original powder gradually dissolves into the liquid phase aid and diffuses. Its reprecipitation mainly occurs in two ways. The first is through uniform nucleation and growth, where small-sized β-Si₃N₄ particles are formed. The second involves non-uniform nucleation and growth in the high-energy crystal plane (rod-shaped axis) of β-Si₃N₄ seeds, which causes the β-Si₃N₄ seeds to grow into rod-shaped crystals.

The influence of crystal seed phase composition on the microstructure of ceramics is shown in Figure 7. As shown in Figure 8a, when 0% β-Si₃N₄ crystal seeds are introduced, α-Si₃N₄ mainly transforms into β-Si₃N₄ following a uniform nucleation growth mode, which results in the formation of uniform grains with fine grain size (Figure 7a).

As shown in Figure 8b, when an appropriate amount (20%) of β-Si₃N₄ seeds is introduced, during the dissolution precipitation process, α-Si₃N₄ exhibits both uniform nucleation and growth as well as non-uniform nucleation. According to the Ostwald ripening theory [30,31], β-Si₃N₄ particles in close contact with α-Si₃N₄ precipitate on the surface of the original β-Si₃N₄ seeds, leading to the formation of large grains. At the same time, a portion of α-Si₃N₄ precipitates and grows uniformly, forming an interlocking structure with the generated large extension grains (Figure 7c).

As shown in Figure 8c, when an excess amount (50%) of β-Si₃N₄ crystal seeds is introduced, non-uniform nucleation of β-Si₃N₄ occurs during the dissolution precipitation process. Due to the introduction of excessive β-Si₃N₄ crystal seeds into the original powder, the β-Si₃N₄ crystal seeds come into contact with each other, resulting in a decrease in α-Si₃N₄ on the surface of the β-Si₃N₄ crystal seeds. The large particles formed by precipitation on the surface of the β-Si₃N₄ crystal seeds are reduced, which is similar in size to the β-Si₃N₄ particles formed by uniform nucleation and growth, resulting in a more uniform size of grain growth (Figure 7e).

3.4. Effects of Seed Crystal Composition on Sample Performances

Figure 9 shows the trends of relative density of SN0~SN4 prepared by SPS at 1650 °C. As the content of β-Si₃N₄ seed crystal increased from 0% to 20%, the relative density of Si₃N₄ samples increased from 96.7% to 97.9%. This was suggestive of the fact that grain size distribution with two peaks favors high compactness of ceramics (see fracture morphology, Figure 7). As the content of β-Si₃N₄ seed crystal increased from 20% to 50%, the relative density of Si₃N₄ samples decreased from 97.9% to 96.5%, which may be attributed to the degrading grain size distribution with two peaks (see fracture morphology, Figure 7).

Figure 10 shows the Vickers hardness of SN0–SN4 samples prepared by SPS at 1650 °C. As the content of β-Si₃N₄ seed crystal was increased from 0% to 20%, the Vickers hardness of Si₃N₄ samples increased from 13.8 GPa to 14.5 GPa, emphasizing the significance of two peaks in the grain size distribution (see fracture morphology, Figure 7). When the content of β-Si₃N₄ seed crystal was increased from 20% to 50%, the Vickers hardness of Si₃N₄ samples decreased from 14.5 GPa to 13.4 GPa, which was attributed to degradation in the grain size distribution with two peaks and grain interlocking structure (see fracture morphology, Figure 7). Hence, it can be concluded that the two peaks observed in the grain size distribution have a major influence on the mechanical properties of the samples (see fracture morphology, Figure 4). Among all the samples tested, the SN2 sample exhibited the highest Vickers hardness (14.5 GPa).

Figure 11 shows the fracture toughness of SN0–SN4 samples prepared by SPS at 1650 °C. As the content of β-Si₃N₄ seed crystal was increased from 0% to 20%, the fracture toughness of Si₃N₄ samples increased from 9.2 MPa·m^1/2 to 12.1 MPa·m^1/2. This was ascribed to the emergence of intergranular fracture (see fracture morphology, Figure 7). As the content of β-Si₃N₄ seed crystal was increased from 20% to 50%, the fracture toughness of Si₃N₄ samples decreased from 12.1 MPa·m^1/2 to 9.5 MPa·m^1/2 (see fracture morphology, Figure 7). The decrease in Vickers hardness may be related to the degradation in the grain size distribution with two peaks and grain interlocking structure. Among all samples examined, the SN2 sample exhibited the highest fracture toughness (12.1 MPa·m^1/2). This was attributed to its unique grain size distribution with two peaks and grain interlocking structure.

In the SN0 sample, due to the presence of small and equiaxed β-Si₃N₄ grains, the crack deflection during fracturing was low, which resulted in low fracture toughness. In the SN1 sample, the β-Si₃N₄ grains consisted of small and equiaxed rod-like grains, which also led to low fracture toughness. However, for the SN2 sample, the grain size distribution exhibited two peaks, where the large and small grains formed an interlocking structure. This feature resulted in low porosity and imparted the sample with high relative density (97.9%) and the highest Vickers hardness (14.5 GPa) among all the samples, as shown in Figure 7. Herein, the heterogeneous precipitation of α-Si₃N₄ on the surface of β-Si₃N₄ seed crystals led to the formation of large, rod-like grains, which activated grain bridging and extrusion, and ultimately resulted in the self-toughening of Si₃N₄ samples. The fracture toughness of the SN2 sample reached 12.1 MPa·m^1/2, as shown in Figure 10. With excess β-Si₃N₄ seed crystals (30–50%), the original β-Si₃N₄ seed crystals had consistent sizes with β-Si₃N₄ particles generated by homogeneous phase transition of α-Si₃N₄, resulting in low crack deflection during fracturing, which drastically reduced the fracture toughness.

4. Conclusions

Si₃N₄ ceramics with different contents of β-Si₃N₄ seed crystal were prepared using the spark plasma sintering method, and the correlation of the β-Si₃N₄ content with hardness and fracture toughness was investigated. The grain growth in Si₃N₄ was found to be controlled by the dissolution-precipitation process. All the as-prepared samples comprised small particles (diameter about 1 μm) and large particles (about 2 μm). Additionally, the large and small grains were observed to be uniformly distributed. Results showed that a suitable content of (≤20%) β-Si₃N₄ seed crystal could induce the phase transition from α-Si₃N₄ to rod-like β-Si₃N₄, which resulted in the grain size distribution exhibiting two peaks. At this optimized β-Si₃N₄ seed crystal content of 20%, the best overall mechanical performance was achieved. At higher content of β-Si₃N₄ seed crystal (>50%), the β-Si₃N₄ particles showed a tendency to be mutually constrained during sintering, with uniform size distribution and lower performance. Among samples prepared by the spark plasma sintering method (1650 °C, 30 MPa, 30 min), the sample with a β-Si₃N₄ seed crystal content of 20% exhibited the best overall performance with a density of 97.9%, hardness of 14.5 GPa, and fracture toughness of 12.1 MPa·m^1/2.

The optimized Si₃N₄ in this work (using 20% β-Si₃N₄ as seed crystal) presents the highest toughness but the lowest hardness among previous works [32,33,34,35] as shown in

Table 3. Comparison of fracture toughness with other silicon nitride ceramics.

Raw Material	Preparation Method	Relative Density (%)	Hardness (GPa)	Toughness (MPa·m1/2)	Reference
80% α-Si3N4 and 20% β-Si3N4	SPS, 1650 °C, 30 MPa, 30 min	97.9	14.5	12.1	this paper
α-Si3N4	HP, 1620 °C, 30 MPa, 3 h	92.1	12.5	6.02	[32]
90% α-Si3N4 and 10% SiC ceramic composites	HP, 1680 °C, 34 MPa, 4 h	97.9	16.4	8.2	[33]
70% α-Si3N4 and 30% ZrO2 ceramic composites	SPS, 1600 °C, 30 MPa, 10 min	99	13.2	7.1	[34]
90% α-Si3N4 and 10% Ti(C, N) ceramic composites	SPS, 1700 °C, 50 MPa, 6 min	99.7	15.6	8.3	[35]

Note: SPS (spark plasma sintering); HP (Hot pressing sintering).

. This work may provide a potential application of ceramics substrate in high-speed rail and new energy vehicles. Further verification of the actual use of silicon nitride ceramics by researchers in the field of automotive development is needed.